]> git.ipfire.org Git - thirdparty/mdadm.git/blob - Grow.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
raid6check: various cleanup/fixes
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2013 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24 #include "mdadm.h"
25 #include "dlink.h"
26 #include <sys/mman.h>
27 #include <stddef.h>
28 #include <stdint.h>
29 #include <signal.h>
30 #include <sys/wait.h>
31
32 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
33 #error no endian defined
34 #endif
35 #include "md_u.h"
36 #include "md_p.h"
37
38 int restore_backup(struct supertype *st,
39 struct mdinfo *content,
40 int working_disks,
41 int next_spare,
42 char **backup_filep,
43 int verbose)
44 {
45 int i;
46 int *fdlist;
47 struct mdinfo *dev;
48 int err;
49 int disk_count = next_spare + working_disks;
50 char *backup_file = *backup_filep;
51
52 dprintf("Called restore_backup()\n");
53 fdlist = xmalloc(sizeof(int) * disk_count);
54
55 enable_fds(next_spare);
56 for (i = 0; i < next_spare; i++)
57 fdlist[i] = -1;
58 for (dev = content->devs; dev; dev = dev->next) {
59 char buf[22];
60 int fd;
61 sprintf(buf, "%d:%d",
62 dev->disk.major,
63 dev->disk.minor);
64 fd = dev_open(buf, O_RDWR);
65
66 if (dev->disk.raid_disk >= 0)
67 fdlist[dev->disk.raid_disk] = fd;
68 else
69 fdlist[next_spare++] = fd;
70 }
71
72 if (!backup_file) {
73 backup_file = locate_backup(content->sys_name);
74 *backup_filep = backup_file;
75 }
76
77 if (st->ss->external && st->ss->recover_backup)
78 err = st->ss->recover_backup(st, content);
79 else
80 err = Grow_restart(st, content, fdlist, next_spare,
81 backup_file, verbose > 0);
82
83 while (next_spare > 0) {
84 next_spare--;
85 if (fdlist[next_spare] >= 0)
86 close(fdlist[next_spare]);
87 }
88 free(fdlist);
89 if (err) {
90 pr_err("Failed to restore critical section for reshape - sorry.\n");
91 if (!backup_file)
92 pr_err("Possibly you need to specify a --backup-file\n");
93 return 1;
94 }
95
96 dprintf("restore_backup() returns status OK.\n");
97 return 0;
98 }
99
100 int Grow_Add_device(char *devname, int fd, char *newdev)
101 {
102 /* Add a device to an active array.
103 * Currently, just extend a linear array.
104 * This requires writing a new superblock on the
105 * new device, calling the kernel to add the device,
106 * and if that succeeds, update the superblock on
107 * all other devices.
108 * This means that we need to *find* all other devices.
109 */
110 struct mdinfo info;
111
112 struct stat stb;
113 int nfd, fd2;
114 int d, nd;
115 struct supertype *st = NULL;
116 char *subarray = NULL;
117
118 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
119 pr_err("cannot get array info for %s\n", devname);
120 return 1;
121 }
122
123 if (info.array.level != -1) {
124 pr_err("can only add devices to linear arrays\n");
125 return 1;
126 }
127
128 st = super_by_fd(fd, &subarray);
129 if (!st) {
130 pr_err("cannot handle arrays with superblock version %d\n",
131 info.array.major_version);
132 return 1;
133 }
134
135 if (subarray) {
136 pr_err("Cannot grow linear sub-arrays yet\n");
137 free(subarray);
138 free(st);
139 return 1;
140 }
141
142 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
143 if (nfd < 0) {
144 pr_err("cannot open %s\n", newdev);
145 free(st);
146 return 1;
147 }
148 fstat(nfd, &stb);
149 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
150 pr_err("%s is not a block device!\n", newdev);
151 close(nfd);
152 free(st);
153 return 1;
154 }
155 /* now check out all the devices and make sure we can read the
156 * superblock */
157 for (d=0 ; d < info.array.raid_disks ; d++) {
158 mdu_disk_info_t disk;
159 char *dv;
160
161 st->ss->free_super(st);
162
163 disk.number = d;
164 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
165 pr_err("cannot get device detail for device %d\n",
166 d);
167 close(nfd);
168 free(st);
169 return 1;
170 }
171 dv = map_dev(disk.major, disk.minor, 1);
172 if (!dv) {
173 pr_err("cannot find device file for device %d\n",
174 d);
175 close(nfd);
176 free(st);
177 return 1;
178 }
179 fd2 = dev_open(dv, O_RDWR);
180 if (fd2 < 0) {
181 pr_err("cannot open device file %s\n", dv);
182 close(nfd);
183 free(st);
184 return 1;
185 }
186
187 if (st->ss->load_super(st, fd2, NULL)) {
188 pr_err("cannot find super block on %s\n", dv);
189 close(nfd);
190 close(fd2);
191 free(st);
192 return 1;
193 }
194 close(fd2);
195 }
196 /* Ok, looks good. Lets update the superblock and write it out to
197 * newdev.
198 */
199
200 info.disk.number = d;
201 info.disk.major = major(stb.st_rdev);
202 info.disk.minor = minor(stb.st_rdev);
203 info.disk.raid_disk = d;
204 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
205 st->ss->update_super(st, &info, "linear-grow-new", newdev,
206 0, 0, NULL);
207
208 if (st->ss->store_super(st, nfd)) {
209 pr_err("Cannot store new superblock on %s\n",
210 newdev);
211 close(nfd);
212 return 1;
213 }
214 close(nfd);
215
216 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
217 pr_err("Cannot add new disk to this array\n");
218 return 1;
219 }
220 /* Well, that seems to have worked.
221 * Now go through and update all superblocks
222 */
223
224 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
225 pr_err("cannot get array info for %s\n", devname);
226 return 1;
227 }
228
229 nd = d;
230 for (d=0 ; d < info.array.raid_disks ; d++) {
231 mdu_disk_info_t disk;
232 char *dv;
233
234 disk.number = d;
235 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
236 pr_err("cannot get device detail for device %d\n",
237 d);
238 return 1;
239 }
240 dv = map_dev(disk.major, disk.minor, 1);
241 if (!dv) {
242 pr_err("cannot find device file for device %d\n",
243 d);
244 return 1;
245 }
246 fd2 = dev_open(dv, O_RDWR);
247 if (fd2 < 0) {
248 pr_err("cannot open device file %s\n", dv);
249 return 1;
250 }
251 if (st->ss->load_super(st, fd2, NULL)) {
252 pr_err("cannot find super block on %s\n", dv);
253 close(fd);
254 return 1;
255 }
256 info.array.raid_disks = nd+1;
257 info.array.nr_disks = nd+1;
258 info.array.active_disks = nd+1;
259 info.array.working_disks = nd+1;
260
261 st->ss->update_super(st, &info, "linear-grow-update", dv,
262 0, 0, NULL);
263
264 if (st->ss->store_super(st, fd2)) {
265 pr_err("Cannot store new superblock on %s\n", dv);
266 close(fd2);
267 return 1;
268 }
269 close(fd2);
270 }
271
272 return 0;
273 }
274
275 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
276 {
277 /*
278 * First check that array doesn't have a bitmap
279 * Then create the bitmap
280 * Then add it
281 *
282 * For internal bitmaps, we need to check the version,
283 * find all the active devices, and write the bitmap block
284 * to all devices
285 */
286 mdu_bitmap_file_t bmf;
287 mdu_array_info_t array;
288 struct supertype *st;
289 char *subarray = NULL;
290 int major = BITMAP_MAJOR_HI;
291 int vers = md_get_version(fd);
292 unsigned long long bitmapsize, array_size;
293
294 if (vers < 9003) {
295 major = BITMAP_MAJOR_HOSTENDIAN;
296 pr_err("Warning - bitmaps created on this kernel are not portable\n"
297 " between different architectures. Consider upgrading the Linux kernel.\n");
298 }
299
300 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
301 if (errno == ENOMEM)
302 pr_err("Memory allocation failure.\n");
303 else
304 pr_err("bitmaps not supported by this kernel.\n");
305 return 1;
306 }
307 if (bmf.pathname[0]) {
308 if (strcmp(s->bitmap_file,"none")==0) {
309 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
310 pr_err("failed to remove bitmap %s\n",
311 bmf.pathname);
312 return 1;
313 }
314 return 0;
315 }
316 pr_err("%s already has a bitmap (%s)\n",
317 devname, bmf.pathname);
318 return 1;
319 }
320 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
321 pr_err("cannot get array status for %s\n", devname);
322 return 1;
323 }
324 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
325 if (strcmp(s->bitmap_file, "none")==0) {
326 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
327 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
328 pr_err("failed to remove internal bitmap.\n");
329 return 1;
330 }
331 return 0;
332 }
333 pr_err("Internal bitmap already present on %s\n",
334 devname);
335 return 1;
336 }
337
338 if (strcmp(s->bitmap_file, "none") == 0) {
339 pr_err("no bitmap found on %s\n", devname);
340 return 1;
341 }
342 if (array.level <= 0) {
343 pr_err("Bitmaps not meaningful with level %s\n",
344 map_num(pers, array.level)?:"of this array");
345 return 1;
346 }
347 bitmapsize = array.size;
348 bitmapsize <<= 1;
349 if (get_dev_size(fd, NULL, &array_size) &&
350 array_size > (0x7fffffffULL<<9)) {
351 /* Array is big enough that we cannot trust array.size
352 * try other approaches
353 */
354 bitmapsize = get_component_size(fd);
355 }
356 if (bitmapsize == 0) {
357 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
358 return 1;
359 }
360
361 if (array.level == 10) {
362 int ncopies = (array.layout&255)*((array.layout>>8)&255);
363 bitmapsize = bitmapsize * array.raid_disks / ncopies;
364 }
365
366 st = super_by_fd(fd, &subarray);
367 if (!st) {
368 pr_err("Cannot understand version %d.%d\n",
369 array.major_version, array.minor_version);
370 return 1;
371 }
372 if (subarray) {
373 pr_err("Cannot add bitmaps to sub-arrays yet\n");
374 free(subarray);
375 free(st);
376 return 1;
377 }
378 if (strcmp(s->bitmap_file, "internal") == 0) {
379 int rv;
380 int d;
381 int offset_setable = 0;
382 struct mdinfo *mdi;
383 if (st->ss->add_internal_bitmap == NULL) {
384 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
385 return 1;
386 }
387 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
388 if (mdi)
389 offset_setable = 1;
390 for (d=0; d< st->max_devs; d++) {
391 mdu_disk_info_t disk;
392 char *dv;
393 disk.number = d;
394 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
395 continue;
396 if (disk.major == 0 &&
397 disk.minor == 0)
398 continue;
399 if ((disk.state & (1<<MD_DISK_SYNC))==0)
400 continue;
401 dv = map_dev(disk.major, disk.minor, 1);
402 if (dv) {
403 int fd2 = dev_open(dv, O_RDWR);
404 if (fd2 < 0)
405 continue;
406 if (st->ss->load_super(st, fd2, NULL)==0) {
407 if (st->ss->add_internal_bitmap(
408 st,
409 &s->bitmap_chunk, c->delay, s->write_behind,
410 bitmapsize, offset_setable,
411 major)
412 )
413 st->ss->write_bitmap(st, fd2);
414 else {
415 pr_err("failed to create internal bitmap - chunksize problem.\n");
416 close(fd2);
417 return 1;
418 }
419 }
420 close(fd2);
421 }
422 }
423 if (offset_setable) {
424 st->ss->getinfo_super(st, mdi, NULL);
425 sysfs_init(mdi, fd, NULL);
426 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
427 mdi->bitmap_offset);
428 } else {
429 array.state |= (1<<MD_SB_BITMAP_PRESENT);
430 rv = ioctl(fd, SET_ARRAY_INFO, &array);
431 }
432 if (rv < 0) {
433 if (errno == EBUSY)
434 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
435 pr_err("failed to set internal bitmap.\n");
436 return 1;
437 }
438 } else {
439 int uuid[4];
440 int bitmap_fd;
441 int d;
442 int max_devs = st->max_devs;
443
444 /* try to load a superblock */
445 for (d = 0; d < max_devs; d++) {
446 mdu_disk_info_t disk;
447 char *dv;
448 int fd2;
449 disk.number = d;
450 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
451 continue;
452 if ((disk.major==0 && disk.minor==0) ||
453 (disk.state & (1<<MD_DISK_REMOVED)))
454 continue;
455 dv = map_dev(disk.major, disk.minor, 1);
456 if (!dv)
457 continue;
458 fd2 = dev_open(dv, O_RDONLY);
459 if (fd2 >= 0) {
460 if (st->ss->load_super(st, fd2, NULL) == 0) {
461 close(fd2);
462 st->ss->uuid_from_super(st, uuid);
463 break;
464 }
465 close(fd2);
466 }
467 }
468 if (d == max_devs) {
469 pr_err("cannot find UUID for array!\n");
470 return 1;
471 }
472 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
473 c->delay, s->write_behind, bitmapsize, major)) {
474 return 1;
475 }
476 bitmap_fd = open(s->bitmap_file, O_RDWR);
477 if (bitmap_fd < 0) {
478 pr_err("weird: %s cannot be opened\n",
479 s->bitmap_file);
480 return 1;
481 }
482 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
483 int err = errno;
484 if (errno == EBUSY)
485 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
486 pr_err("Cannot set bitmap file for %s: %s\n",
487 devname, strerror(err));
488 return 1;
489 }
490 }
491
492 return 0;
493 }
494
495 /*
496 * When reshaping an array we might need to backup some data.
497 * This is written to all spares with a 'super_block' describing it.
498 * The superblock goes 4K from the end of the used space on the
499 * device.
500 * It if written after the backup is complete.
501 * It has the following structure.
502 */
503
504 static struct mdp_backup_super {
505 char magic[16]; /* md_backup_data-1 or -2 */
506 __u8 set_uuid[16];
507 __u64 mtime;
508 /* start/sizes in 512byte sectors */
509 __u64 devstart; /* address on backup device/file of data */
510 __u64 arraystart;
511 __u64 length;
512 __u32 sb_csum; /* csum of preceeding bytes. */
513 __u32 pad1;
514 __u64 devstart2; /* offset in to data of second section */
515 __u64 arraystart2;
516 __u64 length2;
517 __u32 sb_csum2; /* csum of preceeding bytes. */
518 __u8 pad[512-68-32];
519 } __attribute__((aligned(512))) bsb, bsb2;
520
521 static __u32 bsb_csum(char *buf, int len)
522 {
523 int i;
524 int csum = 0;
525 for (i = 0; i < len; i++)
526 csum = (csum<<3) + buf[0];
527 return __cpu_to_le32(csum);
528 }
529
530 static int check_idle(struct supertype *st)
531 {
532 /* Check that all member arrays for this container, or the
533 * container of this array, are idle
534 */
535 char *container = (st->container_devnm[0]
536 ? st->container_devnm : st->devnm);
537 struct mdstat_ent *ent, *e;
538 int is_idle = 1;
539
540 ent = mdstat_read(0, 0);
541 for (e = ent ; e; e = e->next) {
542 if (!is_container_member(e, container))
543 continue;
544 if (e->percent >= 0) {
545 is_idle = 0;
546 break;
547 }
548 }
549 free_mdstat(ent);
550 return is_idle;
551 }
552
553 static int freeze_container(struct supertype *st)
554 {
555 char *container = (st->container_devnm[0]
556 ? st->container_devnm : st->devnm);
557
558 if (!check_idle(st))
559 return -1;
560
561 if (block_monitor(container, 1)) {
562 pr_err("failed to freeze container\n");
563 return -2;
564 }
565
566 return 1;
567 }
568
569 static void unfreeze_container(struct supertype *st)
570 {
571 char *container = (st->container_devnm[0]
572 ? st->container_devnm : st->devnm);
573
574 unblock_monitor(container, 1);
575 }
576
577 static int freeze(struct supertype *st)
578 {
579 /* Try to freeze resync/rebuild on this array/container.
580 * Return -1 if the array is busy,
581 * return -2 container cannot be frozen,
582 * return 0 if this kernel doesn't support 'frozen'
583 * return 1 if it worked.
584 */
585 if (st->ss->external)
586 return freeze_container(st);
587 else {
588 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
589 int err;
590 char buf[20];
591
592 if (!sra)
593 return -1;
594 /* Need to clear any 'read-auto' status */
595 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
596 strncmp(buf, "read-auto", 9) == 0)
597 sysfs_set_str(sra, NULL, "array_state", "clean");
598
599 err = sysfs_freeze_array(sra);
600 sysfs_free(sra);
601 return err;
602 }
603 }
604
605 static void unfreeze(struct supertype *st)
606 {
607 if (st->ss->external)
608 return unfreeze_container(st);
609 else {
610 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
611 char buf[20];
612
613 if (sra &&
614 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
615 && strcmp(buf, "frozen\n") == 0)
616 sysfs_set_str(sra, NULL, "sync_action", "idle");
617 sysfs_free(sra);
618 }
619 }
620
621 static void wait_reshape(struct mdinfo *sra)
622 {
623 int fd = sysfs_get_fd(sra, NULL, "sync_action");
624 char action[20];
625
626 if (fd < 0)
627 return;
628
629 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
630 strncmp(action, "reshape", 7) == 0)
631 sysfs_wait(fd, NULL);
632 close(fd);
633 }
634
635 static int reshape_super(struct supertype *st, unsigned long long size,
636 int level, int layout, int chunksize, int raid_disks,
637 int delta_disks, char *backup_file, char *dev,
638 int direction, int verbose)
639 {
640 /* nothing extra to check in the native case */
641 if (!st->ss->external)
642 return 0;
643 if (!st->ss->reshape_super ||
644 !st->ss->manage_reshape) {
645 pr_err("%s metadata does not support reshape\n",
646 st->ss->name);
647 return 1;
648 }
649
650 return st->ss->reshape_super(st, size, level, layout, chunksize,
651 raid_disks, delta_disks, backup_file, dev,
652 direction, verbose);
653 }
654
655 static void sync_metadata(struct supertype *st)
656 {
657 if (st->ss->external) {
658 if (st->update_tail) {
659 flush_metadata_updates(st);
660 st->update_tail = &st->updates;
661 } else
662 st->ss->sync_metadata(st);
663 }
664 }
665
666 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
667 {
668 /* when dealing with external metadata subarrays we need to be
669 * prepared to handle EAGAIN. The kernel may need to wait for
670 * mdmon to mark the array active so the kernel can handle
671 * allocations/writeback when preparing the reshape action
672 * (md_allow_write()). We temporarily disable safe_mode_delay
673 * to close a race with the array_state going clean before the
674 * next write to raid_disks / stripe_cache_size
675 */
676 char safe[50];
677 int rc;
678
679 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
680 if (!container ||
681 (strcmp(name, "raid_disks") != 0 &&
682 strcmp(name, "stripe_cache_size") != 0))
683 return sysfs_set_num(sra, NULL, name, n);
684
685 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
686 if (rc <= 0)
687 return -1;
688 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
689 rc = sysfs_set_num(sra, NULL, name, n);
690 if (rc < 0 && errno == EAGAIN) {
691 ping_monitor(container);
692 /* if we get EAGAIN here then the monitor is not active
693 * so stop trying
694 */
695 rc = sysfs_set_num(sra, NULL, name, n);
696 }
697 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
698 return rc;
699 }
700
701 int start_reshape(struct mdinfo *sra, int already_running,
702 int before_data_disks, int data_disks)
703 {
704 int err;
705 unsigned long long sync_max_to_set;
706
707 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
708 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
709 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
710 sra->reshape_progress);
711 if (before_data_disks <= data_disks)
712 sync_max_to_set = sra->reshape_progress / data_disks;
713 else
714 sync_max_to_set = (sra->component_size * data_disks
715 - sra->reshape_progress) / data_disks;
716 if (!already_running)
717 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
718 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
719 if (!already_running && err == 0) {
720 int cnt = 5;
721 do {
722 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
723 if (err)
724 sleep(1);
725 } while (err && errno == EBUSY && cnt-- > 0);
726 }
727 return err;
728 }
729
730 void abort_reshape(struct mdinfo *sra)
731 {
732 sysfs_set_str(sra, NULL, "sync_action", "idle");
733 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
734 sysfs_set_num(sra, NULL, "suspend_hi", 0);
735 sysfs_set_num(sra, NULL, "suspend_lo", 0);
736 sysfs_set_num(sra, NULL, "sync_min", 0);
737 // It isn't safe to reset sync_max as we aren't monitoring.
738 // Array really should be stopped at this point.
739 }
740
741 int remove_disks_for_takeover(struct supertype *st,
742 struct mdinfo *sra,
743 int layout)
744 {
745 int nr_of_copies;
746 struct mdinfo *remaining;
747 int slot;
748
749 if (sra->array.level == 10)
750 nr_of_copies = layout & 0xff;
751 else if (sra->array.level == 1)
752 nr_of_copies = sra->array.raid_disks;
753 else
754 return 1;
755
756 remaining = sra->devs;
757 sra->devs = NULL;
758 /* for each 'copy', select one device and remove from the list. */
759 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
760 struct mdinfo **diskp;
761 int found = 0;
762
763 /* Find a working device to keep */
764 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
765 struct mdinfo *disk = *diskp;
766
767 if (disk->disk.raid_disk < slot)
768 continue;
769 if (disk->disk.raid_disk >= slot + nr_of_copies)
770 continue;
771 if (disk->disk.state & (1<<MD_DISK_REMOVED))
772 continue;
773 if (disk->disk.state & (1<<MD_DISK_FAULTY))
774 continue;
775 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
776 continue;
777
778 /* We have found a good disk to use! */
779 *diskp = disk->next;
780 disk->next = sra->devs;
781 sra->devs = disk;
782 found = 1;
783 break;
784 }
785 if (!found)
786 break;
787 }
788
789 if (slot < sra->array.raid_disks) {
790 /* didn't find all slots */
791 struct mdinfo **e;
792 e = &remaining;
793 while (*e)
794 e = &(*e)->next;
795 *e = sra->devs;
796 sra->devs = remaining;
797 return 1;
798 }
799
800 /* Remove all 'remaining' devices from the array */
801 while (remaining) {
802 struct mdinfo *sd = remaining;
803 remaining = sd->next;
804
805 sysfs_set_str(sra, sd, "state", "faulty");
806 sysfs_set_str(sra, sd, "slot", "none");
807 /* for external metadata disks should be removed in mdmon */
808 if (!st->ss->external)
809 sysfs_set_str(sra, sd, "state", "remove");
810 sd->disk.state |= (1<<MD_DISK_REMOVED);
811 sd->disk.state &= ~(1<<MD_DISK_SYNC);
812 sd->next = sra->devs;
813 sra->devs = sd;
814 }
815 return 0;
816 }
817
818 void reshape_free_fdlist(int *fdlist,
819 unsigned long long *offsets,
820 int size)
821 {
822 int i;
823
824 for (i = 0; i < size; i++)
825 if (fdlist[i] >= 0)
826 close(fdlist[i]);
827
828 free(fdlist);
829 free(offsets);
830 }
831
832 int reshape_prepare_fdlist(char *devname,
833 struct mdinfo *sra,
834 int raid_disks,
835 int nrdisks,
836 unsigned long blocks,
837 char *backup_file,
838 int *fdlist,
839 unsigned long long *offsets)
840 {
841 int d = 0;
842 struct mdinfo *sd;
843
844 enable_fds(nrdisks);
845 for (d = 0; d <= nrdisks; d++)
846 fdlist[d] = -1;
847 d = raid_disks;
848 for (sd = sra->devs; sd; sd = sd->next) {
849 if (sd->disk.state & (1<<MD_DISK_FAULTY))
850 continue;
851 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
852 sd->disk.raid_disk < raid_disks) {
853 char *dn = map_dev(sd->disk.major,
854 sd->disk.minor, 1);
855 fdlist[sd->disk.raid_disk]
856 = dev_open(dn, O_RDONLY);
857 offsets[sd->disk.raid_disk] = sd->data_offset*512;
858 if (fdlist[sd->disk.raid_disk] < 0) {
859 pr_err("%s: cannot open component %s\n",
860 devname, dn ? dn : "-unknown-");
861 d = -1;
862 goto release;
863 }
864 } else if (backup_file == NULL) {
865 /* spare */
866 char *dn = map_dev(sd->disk.major,
867 sd->disk.minor, 1);
868 fdlist[d] = dev_open(dn, O_RDWR);
869 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
870 if (fdlist[d] < 0) {
871 pr_err("%s: cannot open component %s\n",
872 devname, dn ? dn : "-unknown-");
873 d = -1;
874 goto release;
875 }
876 d++;
877 }
878 }
879 release:
880 return d;
881 }
882
883 int reshape_open_backup_file(char *backup_file,
884 int fd,
885 char *devname,
886 long blocks,
887 int *fdlist,
888 unsigned long long *offsets,
889 char *sys_name,
890 int restart)
891 {
892 /* Return 1 on success, 0 on any form of failure */
893 /* need to check backup file is large enough */
894 char buf[512];
895 struct stat stb;
896 unsigned int dev;
897 int i;
898
899 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
900 S_IRUSR | S_IWUSR);
901 *offsets = 8 * 512;
902 if (*fdlist < 0) {
903 pr_err("%s: cannot create backup file %s: %s\n",
904 devname, backup_file, strerror(errno));
905 return 0;
906 }
907 /* Guard against backup file being on array device.
908 * If array is partitioned or if LVM etc is in the
909 * way this will not notice, but it is better than
910 * nothing.
911 */
912 fstat(*fdlist, &stb);
913 dev = stb.st_dev;
914 fstat(fd, &stb);
915 if (stb.st_rdev == dev) {
916 pr_err("backup file must NOT be on the array being reshaped.\n");
917 close(*fdlist);
918 return 0;
919 }
920
921 memset(buf, 0, 512);
922 for (i=0; i < blocks + 8 ; i++) {
923 if (write(*fdlist, buf, 512) != 512) {
924 pr_err("%s: cannot create backup file %s: %s\n",
925 devname, backup_file, strerror(errno));
926 return 0;
927 }
928 }
929 if (fsync(*fdlist) != 0) {
930 pr_err("%s: cannot create backup file %s: %s\n",
931 devname, backup_file, strerror(errno));
932 return 0;
933 }
934
935 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
936 char *bu = make_backup(sys_name);
937 if (symlink(backup_file, bu))
938 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
939 strerror(errno));
940 free(bu);
941 }
942
943 return 1;
944 }
945
946 unsigned long compute_backup_blocks(int nchunk, int ochunk,
947 unsigned int ndata, unsigned int odata)
948 {
949 unsigned long a, b, blocks;
950 /* So how much do we need to backup.
951 * We need an amount of data which is both a whole number of
952 * old stripes and a whole number of new stripes.
953 * So LCM for (chunksize*datadisks).
954 */
955 a = (ochunk/512) * odata;
956 b = (nchunk/512) * ndata;
957 /* Find GCD */
958 a = GCD(a, b);
959 /* LCM == product / GCD */
960 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
961
962 return blocks;
963 }
964
965 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
966 {
967 /* Based on the current array state in info->array and
968 * the changes in info->new_* etc, determine:
969 * - whether the change is possible
970 * - Intermediate level/raid_disks/layout
971 * - whether a restriping reshape is needed
972 * - number of sectors in minimum change unit. This
973 * will cover a whole number of stripes in 'before' and
974 * 'after'.
975 *
976 * Return message if the change should be rejected
977 * NULL if the change can be achieved
978 *
979 * This can be called as part of starting a reshape, or
980 * when assembling an array that is undergoing reshape.
981 */
982 int near, far, offset, copies;
983 int new_disks;
984 int old_chunk, new_chunk;
985 /* delta_parity records change in number of devices
986 * caused by level change
987 */
988 int delta_parity = 0;
989
990 memset(re, 0, sizeof(*re));
991
992 /* If a new level not explicitly given, we assume no-change */
993 if (info->new_level == UnSet)
994 info->new_level = info->array.level;
995
996 if (info->new_chunk)
997 switch (info->new_level) {
998 case 0:
999 case 4:
1000 case 5:
1001 case 6:
1002 case 10:
1003 /* chunk size is meaningful, must divide component_size
1004 * evenly
1005 */
1006 if (info->component_size % (info->new_chunk/512)) {
1007 unsigned long long shrink = info->component_size;
1008 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1009 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1010 info->new_chunk/1024, info->component_size/2);
1011 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1012 devname, shrink/2);
1013 pr_err("will shrink the array so the given chunk size would work.\n");
1014 return "";
1015 }
1016 break;
1017 default:
1018 return "chunk size not meaningful for this level";
1019 }
1020 else
1021 info->new_chunk = info->array.chunk_size;
1022
1023 switch (info->array.level) {
1024 default:
1025 return "No reshape is possibly for this RAID level";
1026 case LEVEL_LINEAR:
1027 if (info->delta_disks != UnSet)
1028 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1029 else
1030 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1031 case 1:
1032 /* RAID1 can convert to RAID1 with different disks, or
1033 * raid5 with 2 disks, or
1034 * raid0 with 1 disk
1035 */
1036 if (info->new_level > 1 &&
1037 (info->component_size & 7))
1038 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1039 if (info->new_level == 0) {
1040 if (info->delta_disks != UnSet &&
1041 info->delta_disks != 0)
1042 return "Cannot change number of disks with RAID1->RAID0 conversion";
1043 re->level = 0;
1044 re->before.data_disks = 1;
1045 re->after.data_disks = 1;
1046 return NULL;
1047 }
1048 if (info->new_level == 1) {
1049 if (info->delta_disks == UnSet)
1050 /* Don't know what to do */
1051 return "no change requested for Growing RAID1";
1052 re->level = 1;
1053 return NULL;
1054 }
1055 if (info->array.raid_disks == 2 &&
1056 info->new_level == 5) {
1057
1058 re->level = 5;
1059 re->before.data_disks = 1;
1060 if (info->delta_disks != UnSet &&
1061 info->delta_disks != 0)
1062 re->after.data_disks = 1 + info->delta_disks;
1063 else
1064 re->after.data_disks = 1;
1065 if (re->after.data_disks < 1)
1066 return "Number of disks too small for RAID5";
1067
1068 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1069 info->array.chunk_size = 65536;
1070 break;
1071 }
1072 /* Could do some multi-stage conversions, but leave that to
1073 * later.
1074 */
1075 return "Impossibly level change request for RAID1";
1076
1077 case 10:
1078 /* RAID10 can be converted from near mode to
1079 * RAID0 by removing some devices.
1080 * It can also be reshaped if the kernel supports
1081 * new_data_offset.
1082 */
1083 switch (info->new_level) {
1084 case 0:
1085 if ((info->array.layout & ~0xff) != 0x100)
1086 return "Cannot Grow RAID10 with far/offset layout";
1087 /* number of devices must be multiple of number of copies */
1088 if (info->array.raid_disks % (info->array.layout & 0xff))
1089 return "RAID10 layout too complex for Grow operation";
1090
1091 new_disks = (info->array.raid_disks
1092 / (info->array.layout & 0xff));
1093 if (info->delta_disks == UnSet)
1094 info->delta_disks = (new_disks
1095 - info->array.raid_disks);
1096
1097 if (info->delta_disks != new_disks - info->array.raid_disks)
1098 return "New number of raid-devices impossible for RAID10";
1099 if (info->new_chunk &&
1100 info->new_chunk != info->array.chunk_size)
1101 return "Cannot change chunk-size with RAID10 Grow";
1102
1103 /* looks good */
1104 re->level = 0;
1105 re->before.data_disks = new_disks;
1106 re->after.data_disks = re->before.data_disks;
1107 return NULL;
1108
1109 case 10:
1110 near = info->array.layout & 0xff;
1111 far = (info->array.layout >> 8) & 0xff;
1112 offset = info->array.layout & 0x10000;
1113 if (far > 1 && !offset)
1114 return "Cannot reshape RAID10 in far-mode";
1115 copies = near * far;
1116
1117 old_chunk = info->array.chunk_size * far;
1118
1119 if (info->new_layout == UnSet)
1120 info->new_layout = info->array.layout;
1121 else {
1122 near = info->new_layout & 0xff;
1123 far = (info->new_layout >> 8) & 0xff;
1124 offset = info->new_layout & 0x10000;
1125 if (far > 1 && !offset)
1126 return "Cannot reshape RAID10 to far-mode";
1127 if (near * far != copies)
1128 return "Cannot change number of copies when reshaping RAID10";
1129 }
1130 if (info->delta_disks == UnSet)
1131 info->delta_disks = 0;
1132 new_disks = (info->array.raid_disks +
1133 info->delta_disks);
1134
1135 new_chunk = info->new_chunk * far;
1136
1137 re->level = 10;
1138 re->before.layout = info->array.layout;
1139 re->before.data_disks = info->array.raid_disks;
1140 re->after.layout = info->new_layout;
1141 re->after.data_disks = new_disks;
1142 /* For RAID10 we don't do backup but do allow reshape,
1143 * so set backup_blocks to INVALID_SECTORS rather than
1144 * zero.
1145 * And there is no need to synchronise stripes on both
1146 * 'old' and 'new'. So the important
1147 * number is the minimum data_offset difference
1148 * which is the larger of (offset copies * chunk).
1149 */
1150 re->backup_blocks = INVALID_SECTORS;
1151 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1152 if (new_disks < re->before.data_disks &&
1153 info->space_after < re->min_offset_change)
1154 /* Reduce component size by one chunk */
1155 re->new_size = (info->component_size -
1156 re->min_offset_change);
1157 else
1158 re->new_size = info->component_size;
1159 re->new_size = re->new_size * new_disks / copies;
1160 return NULL;
1161
1162 default:
1163 return "RAID10 can only be changed to RAID0";
1164 }
1165 case 0:
1166 /* RAID0 can be converted to RAID10, or to RAID456 */
1167 if (info->new_level == 10) {
1168 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1169 /* Assume near=2 layout */
1170 info->new_layout = 0x102;
1171 info->delta_disks = info->array.raid_disks;
1172 }
1173 if (info->new_layout == UnSet) {
1174 int copies = 1 + (info->delta_disks
1175 / info->array.raid_disks);
1176 if (info->array.raid_disks * (copies-1)
1177 != info->delta_disks)
1178 return "Impossible number of devices for RAID0->RAID10";
1179 info->new_layout = 0x100 + copies;
1180 }
1181 if (info->delta_disks == UnSet) {
1182 int copies = info->new_layout & 0xff;
1183 if (info->new_layout != 0x100 + copies)
1184 return "New layout impossible for RAID0->RAID10";;
1185 info->delta_disks = (copies - 1) *
1186 info->array.raid_disks;
1187 }
1188 if (info->new_chunk &&
1189 info->new_chunk != info->array.chunk_size)
1190 return "Cannot change chunk-size with RAID0->RAID10";
1191 /* looks good */
1192 re->level = 10;
1193 re->before.data_disks = (info->array.raid_disks +
1194 info->delta_disks);
1195 re->after.data_disks = re->before.data_disks;
1196 re->before.layout = info->new_layout;
1197 return NULL;
1198 }
1199
1200 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1201 * a raid4 style layout of the final level.
1202 */
1203 switch (info->new_level) {
1204 case 4:
1205 delta_parity = 1;
1206 case 0:
1207 re->level = 4;
1208 re->before.layout = 0;
1209 break;
1210 case 5:
1211 delta_parity = 1;
1212 re->level = 5;
1213 re->before.layout = ALGORITHM_PARITY_N;
1214 if (info->new_layout == UnSet)
1215 info->new_layout = map_name(r5layout, "default");
1216 break;
1217 case 6:
1218 delta_parity = 2;
1219 re->level = 6;
1220 re->before.layout = ALGORITHM_PARITY_N;
1221 if (info->new_layout == UnSet)
1222 info->new_layout = map_name(r6layout, "default");
1223 break;
1224 default:
1225 return "Impossible level change requested";
1226 }
1227 re->before.data_disks = info->array.raid_disks;
1228 /* determining 'after' layout happens outside this 'switch' */
1229 break;
1230
1231 case 4:
1232 info->array.layout = ALGORITHM_PARITY_N;
1233 case 5:
1234 switch (info->new_level) {
1235 case 0:
1236 delta_parity = -1;
1237 case 4:
1238 re->level = info->array.level;
1239 re->before.data_disks = info->array.raid_disks - 1;
1240 re->before.layout = info->array.layout;
1241 break;
1242 case 5:
1243 re->level = 5;
1244 re->before.data_disks = info->array.raid_disks - 1;
1245 re->before.layout = info->array.layout;
1246 break;
1247 case 6:
1248 delta_parity = 1;
1249 re->level = 6;
1250 re->before.data_disks = info->array.raid_disks - 1;
1251 switch (info->array.layout) {
1252 case ALGORITHM_LEFT_ASYMMETRIC:
1253 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1254 break;
1255 case ALGORITHM_RIGHT_ASYMMETRIC:
1256 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1257 break;
1258 case ALGORITHM_LEFT_SYMMETRIC:
1259 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1260 break;
1261 case ALGORITHM_RIGHT_SYMMETRIC:
1262 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1263 break;
1264 case ALGORITHM_PARITY_0:
1265 re->before.layout = ALGORITHM_PARITY_0_6;
1266 break;
1267 case ALGORITHM_PARITY_N:
1268 re->before.layout = ALGORITHM_PARITY_N_6;
1269 break;
1270 default:
1271 return "Cannot convert an array with this layout";
1272 }
1273 break;
1274 case 1:
1275 if (info->array.raid_disks != 2)
1276 return "Can only convert a 2-device array to RAID1";
1277 if (info->delta_disks != UnSet &&
1278 info->delta_disks != 0)
1279 return "Cannot set raid_disk when converting RAID5->RAID1";
1280 re->level = 1;
1281 info->new_chunk = 0;
1282 return NULL;
1283 default:
1284 return "Impossible level change requested";
1285 }
1286 break;
1287 case 6:
1288 switch (info->new_level) {
1289 case 4:
1290 case 5:
1291 delta_parity = -1;
1292 case 6:
1293 re->level = 6;
1294 re->before.data_disks = info->array.raid_disks - 2;
1295 re->before.layout = info->array.layout;
1296 break;
1297 default:
1298 return "Impossible level change requested";
1299 }
1300 break;
1301 }
1302
1303 /* If we reached here then it looks like a re-stripe is
1304 * happening. We have determined the intermediate level
1305 * and initial raid_disks/layout and stored these in 're'.
1306 *
1307 * We need to deduce the final layout that can be atomically
1308 * converted to the end state.
1309 */
1310 switch (info->new_level) {
1311 case 0:
1312 /* We can only get to RAID0 from RAID4 or RAID5
1313 * with appropriate layout and one extra device
1314 */
1315 if (re->level != 4 && re->level != 5)
1316 return "Cannot covert to RAID0 from this level";
1317
1318 switch (re->level) {
1319 case 4:
1320 re->before.layout = 0;
1321 re->after.layout = 0;
1322 break;
1323 case 5:
1324 re->after.layout = ALGORITHM_PARITY_N;
1325 break;
1326 }
1327 break;
1328
1329 case 4:
1330 /* We can only get to RAID4 from RAID5 */
1331 if (re->level != 4 && re->level != 5)
1332 return "Cannot convert to RAID4 from this level";
1333
1334 switch (re->level) {
1335 case 4:
1336 re->after.layout = 0;
1337 break;
1338 case 5:
1339 re->after.layout = ALGORITHM_PARITY_N;
1340 break;
1341 }
1342 break;
1343
1344 case 5:
1345 /* We get to RAID5 from RAID5 or RAID6 */
1346 if (re->level != 5 && re->level != 6)
1347 return "Cannot convert to RAID5 from this level";
1348
1349 switch (re->level) {
1350 case 5:
1351 if (info->new_layout == UnSet)
1352 re->after.layout = re->before.layout;
1353 else
1354 re->after.layout = info->new_layout;
1355 break;
1356 case 6:
1357 if (info->new_layout == UnSet)
1358 info->new_layout = re->before.layout;
1359
1360 /* after.layout needs to be raid6 version of new_layout */
1361 if (info->new_layout == ALGORITHM_PARITY_N)
1362 re->after.layout = ALGORITHM_PARITY_N;
1363 else {
1364 char layout[40];
1365 char *ls = map_num(r5layout, info->new_layout);
1366 int l;
1367 if (ls) {
1368 /* Current RAID6 layout has a RAID5
1369 * equivalent - good
1370 */
1371 strcat(strcpy(layout, ls), "-6");
1372 l = map_name(r6layout, layout);
1373 if (l == UnSet)
1374 return "Cannot find RAID6 layout to convert to";
1375 } else {
1376 /* Current RAID6 has no equivalent.
1377 * If it is already a '-6' layout we
1378 * can leave it unchanged, else we must
1379 * fail
1380 */
1381 ls = map_num(r6layout, info->new_layout);
1382 if (!ls ||
1383 strcmp(ls+strlen(ls)-2, "-6") != 0)
1384 return "Please specify new layout";
1385 l = info->new_layout;
1386 }
1387 re->after.layout = l;
1388 }
1389 }
1390 break;
1391
1392 case 6:
1393 /* We must already be at level 6 */
1394 if (re->level != 6)
1395 return "Impossible level change";
1396 if (info->new_layout == UnSet)
1397 re->after.layout = info->array.layout;
1398 else
1399 re->after.layout = info->new_layout;
1400 break;
1401 default:
1402 return "Impossible level change requested";
1403 }
1404 if (info->delta_disks == UnSet)
1405 info->delta_disks = delta_parity;
1406
1407 re->after.data_disks = (re->before.data_disks
1408 + info->delta_disks
1409 - delta_parity);
1410 switch (re->level) {
1411 case 6: re->parity = 2;
1412 break;
1413 case 4:
1414 case 5: re->parity = 1;
1415 break;
1416 default: re->parity = 0;
1417 break;
1418 }
1419 /* So we have a restripe operation, we need to calculate the number
1420 * of blocks per reshape operation.
1421 */
1422 re->new_size = info->component_size * re->before.data_disks;
1423 if (info->new_chunk == 0)
1424 info->new_chunk = info->array.chunk_size;
1425 if (re->after.data_disks == re->before.data_disks &&
1426 re->after.layout == re->before.layout &&
1427 info->new_chunk == info->array.chunk_size) {
1428 /* Nothing to change, can change level immediately. */
1429 re->level = info->new_level;
1430 re->backup_blocks = 0;
1431 return NULL;
1432 }
1433 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1434 /* chunk and layout changes make no difference */
1435 re->level = info->new_level;
1436 re->backup_blocks = 0;
1437 return NULL;
1438 }
1439
1440 if (re->after.data_disks == re->before.data_disks &&
1441 get_linux_version() < 2006032)
1442 return "in-place reshape is not safe before 2.6.32 - sorry.";
1443
1444 if (re->after.data_disks < re->before.data_disks &&
1445 get_linux_version() < 2006030)
1446 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1447
1448 re->backup_blocks = compute_backup_blocks(
1449 info->new_chunk, info->array.chunk_size,
1450 re->after.data_disks,
1451 re->before.data_disks);
1452 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1453
1454 re->new_size = info->component_size * re->after.data_disks;
1455 return NULL;
1456 }
1457
1458 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1459 char *text_version)
1460 {
1461 struct mdinfo *info;
1462 char *subarray;
1463 int ret_val = -1;
1464
1465 if ((st == NULL) || (sra == NULL))
1466 return ret_val;
1467
1468 if (text_version == NULL)
1469 text_version = sra->text_version;
1470 subarray = strchr(text_version+1, '/')+1;
1471 info = st->ss->container_content(st, subarray);
1472 if (info) {
1473 unsigned long long current_size = 0;
1474 unsigned long long new_size =
1475 info->custom_array_size/2;
1476
1477 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1478 new_size > current_size) {
1479 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1480 < 0)
1481 dprintf("Error: Cannot set array size");
1482 else {
1483 ret_val = 0;
1484 dprintf("Array size changed");
1485 }
1486 dprintf_cont(" from %llu to %llu.\n",
1487 current_size, new_size);
1488 }
1489 sysfs_free(info);
1490 } else
1491 dprintf("Error: set_array_size(): info pointer in NULL\n");
1492
1493 return ret_val;
1494 }
1495
1496 static int reshape_array(char *container, int fd, char *devname,
1497 struct supertype *st, struct mdinfo *info,
1498 int force, struct mddev_dev *devlist,
1499 unsigned long long data_offset,
1500 char *backup_file, int verbose, int forked,
1501 int restart, int freeze_reshape);
1502 static int reshape_container(char *container, char *devname,
1503 int mdfd,
1504 struct supertype *st,
1505 struct mdinfo *info,
1506 int force,
1507 char *backup_file, int verbose,
1508 int forked, int restart, int freeze_reshape);
1509
1510 int Grow_reshape(char *devname, int fd,
1511 struct mddev_dev *devlist,
1512 unsigned long long data_offset,
1513 struct context *c, struct shape *s)
1514 {
1515 /* Make some changes in the shape of an array.
1516 * The kernel must support the change.
1517 *
1518 * There are three different changes. Each can trigger
1519 * a resync or recovery so we freeze that until we have
1520 * requested everything (if kernel supports freezing - 2.6.30).
1521 * The steps are:
1522 * - change size (i.e. component_size)
1523 * - change level
1524 * - change layout/chunksize/ndisks
1525 *
1526 * The last can require a reshape. It is different on different
1527 * levels so we need to check the level before actioning it.
1528 * Some times the level change needs to be requested after the
1529 * reshape (e.g. raid6->raid5, raid5->raid0)
1530 *
1531 */
1532 struct mdu_array_info_s array;
1533 int rv = 0;
1534 struct supertype *st;
1535 char *subarray = NULL;
1536
1537 int frozen;
1538 int changed = 0;
1539 char *container = NULL;
1540 int cfd = -1;
1541
1542 struct mddev_dev *dv;
1543 int added_disks;
1544
1545 struct mdinfo info;
1546 struct mdinfo *sra;
1547
1548 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1549 pr_err("%s is not an active md array - aborting\n",
1550 devname);
1551 return 1;
1552 }
1553 if (data_offset != INVALID_SECTORS && array.level != 10
1554 && (array.level < 4 || array.level > 6)) {
1555 pr_err("--grow --data-offset not yet supported\n");
1556 return 1;
1557 }
1558
1559 if (s->size > 0 &&
1560 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1561 pr_err("cannot change component size at the same time as other changes.\n"
1562 " Change size first, then check data is intact before making other changes.\n");
1563 return 1;
1564 }
1565
1566 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1567 get_linux_version() < 2006032 &&
1568 !check_env("MDADM_FORCE_FEWER")) {
1569 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1570 " Please use a newer kernel\n");
1571 return 1;
1572 }
1573
1574 st = super_by_fd(fd, &subarray);
1575 if (!st) {
1576 pr_err("Unable to determine metadata format for %s\n", devname);
1577 return 1;
1578 }
1579 if (s->raiddisks > st->max_devs) {
1580 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1581 return 1;
1582 }
1583
1584 /* in the external case we need to check that the requested reshape is
1585 * supported, and perform an initial check that the container holds the
1586 * pre-requisite spare devices (mdmon owns final validation)
1587 */
1588 if (st->ss->external) {
1589 int rv;
1590
1591 if (subarray) {
1592 container = st->container_devnm;
1593 cfd = open_dev_excl(st->container_devnm);
1594 } else {
1595 container = st->devnm;
1596 close(fd);
1597 cfd = open_dev_excl(st->devnm);
1598 fd = cfd;
1599 }
1600 if (cfd < 0) {
1601 pr_err("Unable to open container for %s\n",
1602 devname);
1603 free(subarray);
1604 return 1;
1605 }
1606
1607 rv = st->ss->load_container(st, cfd, NULL);
1608
1609 if (rv) {
1610 pr_err("Cannot read superblock for %s\n",
1611 devname);
1612 free(subarray);
1613 return 1;
1614 }
1615
1616 /* check if operation is supported for metadata handler */
1617 if (st->ss->container_content) {
1618 struct mdinfo *cc = NULL;
1619 struct mdinfo *content = NULL;
1620
1621 cc = st->ss->container_content(st, subarray);
1622 for (content = cc; content ; content = content->next) {
1623 int allow_reshape = 1;
1624
1625 /* check if reshape is allowed based on metadata
1626 * indications stored in content.array.status
1627 */
1628 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1629 allow_reshape = 0;
1630 if (content->array.state
1631 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1632 allow_reshape = 0;
1633 if (!allow_reshape) {
1634 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1635 devname, container);
1636 sysfs_free(cc);
1637 free(subarray);
1638 return 1;
1639 }
1640 }
1641 sysfs_free(cc);
1642 }
1643 if (mdmon_running(container))
1644 st->update_tail = &st->updates;
1645 }
1646
1647 added_disks = 0;
1648 for (dv = devlist; dv; dv = dv->next)
1649 added_disks++;
1650 if (s->raiddisks > array.raid_disks &&
1651 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1652 !c->force) {
1653 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1654 " Use --force to over-ride this check.\n",
1655 s->raiddisks - array.raid_disks,
1656 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1657 array.spare_disks + added_disks);
1658 return 1;
1659 }
1660
1661 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1662 | GET_STATE | GET_VERSION);
1663 if (sra) {
1664 if (st->ss->external && subarray == NULL) {
1665 array.level = LEVEL_CONTAINER;
1666 sra->array.level = LEVEL_CONTAINER;
1667 }
1668 } else {
1669 pr_err("failed to read sysfs parameters for %s\n",
1670 devname);
1671 return 1;
1672 }
1673 frozen = freeze(st);
1674 if (frozen < -1) {
1675 /* freeze() already spewed the reason */
1676 sysfs_free(sra);
1677 return 1;
1678 } else if (frozen < 0) {
1679 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1680 sysfs_free(sra);
1681 return 1;
1682 }
1683
1684 /* ========= set size =============== */
1685 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1686 unsigned long long orig_size = get_component_size(fd)/2;
1687 unsigned long long min_csize;
1688 struct mdinfo *mdi;
1689 int raid0_takeover = 0;
1690
1691 if (orig_size == 0)
1692 orig_size = (unsigned) array.size;
1693
1694 if (orig_size == 0) {
1695 pr_err("Cannot set device size in this type of array.\n");
1696 rv = 1;
1697 goto release;
1698 }
1699
1700 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1701 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1702 rv = 1;
1703 goto release;
1704 }
1705 sync_metadata(st);
1706 if (st->ss->external) {
1707 /* metadata can have size limitation
1708 * update size value according to metadata information
1709 */
1710 struct mdinfo *sizeinfo =
1711 st->ss->container_content(st, subarray);
1712 if (sizeinfo) {
1713 unsigned long long new_size =
1714 sizeinfo->custom_array_size/2;
1715 int data_disks = get_data_disks(
1716 sizeinfo->array.level,
1717 sizeinfo->array.layout,
1718 sizeinfo->array.raid_disks);
1719 new_size /= data_disks;
1720 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size, new_size,
1721 new_size * data_disks);
1722 s->size = new_size;
1723 sysfs_free(sizeinfo);
1724 }
1725 }
1726
1727 /* Update the size of each member device in case
1728 * they have been resized. This will never reduce
1729 * below the current used-size. The "size" attribute
1730 * understands '0' to mean 'max'.
1731 */
1732 min_csize = 0;
1733 rv = 0;
1734 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1735 if (sysfs_set_num(sra, mdi, "size",
1736 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1737 /* Probably kernel refusing to let us
1738 * reduce the size - not an error.
1739 */
1740 break;
1741 }
1742 if (array.not_persistent == 0 &&
1743 array.major_version == 0 &&
1744 get_linux_version() < 3001000) {
1745 /* Dangerous to allow size to exceed 2TB */
1746 unsigned long long csize;
1747 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1748 if (csize >= 2ULL*1024*1024*1024)
1749 csize = 2ULL*1024*1024*1024;
1750 if ((min_csize == 0 || (min_csize
1751 > csize)))
1752 min_csize = csize;
1753 }
1754 }
1755 }
1756 if (rv) {
1757 pr_err("Cannot set size on 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 use more than 2TB per device on this kernel.\n");
1762 rv = 1;
1763 goto size_change_error;
1764 }
1765 if (min_csize && s->size == MAX_SIZE) {
1766 /* Don't let the kernel choose a size - it will get
1767 * it wrong
1768 */
1769 pr_err("Limited v0.90 array to 2TB per device\n");
1770 s->size = min_csize;
1771 }
1772 if (st->ss->external) {
1773 if (sra->array.level == 0) {
1774 rv = sysfs_set_str(sra, NULL, "level",
1775 "raid5");
1776 if (!rv) {
1777 raid0_takeover = 1;
1778 /* get array parameters after takeover
1779 * to change one parameter at time only
1780 */
1781 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1782 }
1783 }
1784 /* make sure mdmon is
1785 * aware of the new level */
1786 if (!mdmon_running(st->container_devnm))
1787 start_mdmon(st->container_devnm);
1788 ping_monitor(container);
1789 if (mdmon_running(st->container_devnm) &&
1790 st->update_tail == NULL)
1791 st->update_tail = &st->updates;
1792 }
1793
1794 if (s->size == MAX_SIZE)
1795 s->size = 0;
1796 array.size = s->size;
1797 if (s->size & ~INT32_MAX) {
1798 /* got truncated to 32bit, write to
1799 * component_size instead
1800 */
1801 if (sra)
1802 rv = sysfs_set_num(sra, NULL,
1803 "component_size", s->size);
1804 else
1805 rv = -1;
1806 } else {
1807 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1808
1809 /* manage array size when it is managed externally
1810 */
1811 if ((rv == 0) && st->ss->external)
1812 rv = set_array_size(st, sra, sra->text_version);
1813 }
1814
1815 if (raid0_takeover) {
1816 /* do not recync non-existing parity,
1817 * we will drop it anyway
1818 */
1819 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1820 /* go back to raid0, drop parity disk
1821 */
1822 sysfs_set_str(sra, NULL, "level", "raid0");
1823 ioctl(fd, GET_ARRAY_INFO, &array);
1824 }
1825
1826 size_change_error:
1827 if (rv != 0) {
1828 int err = errno;
1829
1830 /* restore metadata */
1831 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1832 UnSet, NULL, devname,
1833 ROLLBACK_METADATA_CHANGES,
1834 c->verbose) == 0)
1835 sync_metadata(st);
1836 pr_err("Cannot set device size for %s: %s\n",
1837 devname, strerror(err));
1838 if (err == EBUSY &&
1839 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1840 cont_err("Bitmap must be removed before size can be changed\n");
1841 rv = 1;
1842 goto release;
1843 }
1844 if (s->assume_clean) {
1845 /* This will fail on kernels older than 3.0 unless
1846 * a backport has been arranged.
1847 */
1848 if (sra == NULL ||
1849 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1850 pr_err("--assume-clean not supported with --grow on this kernel\n");
1851 }
1852 ioctl(fd, GET_ARRAY_INFO, &array);
1853 s->size = get_component_size(fd)/2;
1854 if (s->size == 0)
1855 s->size = array.size;
1856 if (c->verbose >= 0) {
1857 if (s->size == orig_size)
1858 pr_err("component size of %s unchanged at %lluK\n",
1859 devname, s->size);
1860 else
1861 pr_err("component size of %s has been set to %lluK\n",
1862 devname, s->size);
1863 }
1864 changed = 1;
1865 } else if (array.level != LEVEL_CONTAINER) {
1866 s->size = get_component_size(fd)/2;
1867 if (s->size == 0)
1868 s->size = array.size;
1869 }
1870
1871 /* See if there is anything else to do */
1872 if ((s->level == UnSet || s->level == array.level) &&
1873 (s->layout_str == NULL) &&
1874 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1875 data_offset == INVALID_SECTORS &&
1876 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1877 /* Nothing more to do */
1878 if (!changed && c->verbose >= 0)
1879 pr_err("%s: no change requested\n",
1880 devname);
1881 goto release;
1882 }
1883
1884 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1885 * current implementation assumes that following conditions must be met:
1886 * - RAID10:
1887 * - far_copies == 1
1888 * - near_copies == 2
1889 */
1890 if ((s->level == 0 && array.level == 10 && sra &&
1891 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1892 (s->level == 0 && array.level == 1 && sra)) {
1893 int err;
1894 err = remove_disks_for_takeover(st, sra, array.layout);
1895 if (err) {
1896 dprintf("Array cannot be reshaped\n");
1897 if (cfd > -1)
1898 close(cfd);
1899 rv = 1;
1900 goto release;
1901 }
1902 /* Make sure mdmon has seen the device removal
1903 * and updated metadata before we continue with
1904 * level change
1905 */
1906 if (container)
1907 ping_monitor(container);
1908 }
1909
1910 memset(&info, 0, sizeof(info));
1911 info.array = array;
1912 sysfs_init(&info, fd, NULL);
1913 strcpy(info.text_version, sra->text_version);
1914 info.component_size = s->size*2;
1915 info.new_level = s->level;
1916 info.new_chunk = s->chunk * 1024;
1917 if (info.array.level == LEVEL_CONTAINER) {
1918 info.delta_disks = UnSet;
1919 info.array.raid_disks = s->raiddisks;
1920 } else if (s->raiddisks)
1921 info.delta_disks = s->raiddisks - info.array.raid_disks;
1922 else
1923 info.delta_disks = UnSet;
1924 if (s->layout_str == NULL) {
1925 info.new_layout = UnSet;
1926 if (info.array.level == 6 &&
1927 (info.new_level == 6 || info.new_level == UnSet) &&
1928 info.array.layout >= 16) {
1929 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
1930 cont_err("during the reshape, please specify --layout=preserve\n");
1931 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
1932 rv = 1;
1933 goto release;
1934 }
1935 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1936 strcmp(s->layout_str, "normalize") == 0) {
1937 /* If we have a -6 RAID6 layout, remove the '-6'. */
1938 info.new_layout = UnSet;
1939 if (info.array.level == 6 && info.new_level == UnSet) {
1940 char l[40], *h;
1941 strcpy(l, map_num(r6layout, info.array.layout));
1942 h = strrchr(l, '-');
1943 if (h && strcmp(h, "-6") == 0) {
1944 *h = 0;
1945 info.new_layout = map_name(r6layout, l);
1946 }
1947 } else {
1948 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
1949 rv = 1;
1950 goto release;
1951 }
1952 } else if (strcmp(s->layout_str, "preserve") == 0) {
1953 /* This means that a non-standard RAID6 layout
1954 * is OK.
1955 * In particular:
1956 * - When reshape a RAID6 (e.g. adding a device)
1957 * which is in a non-standard layout, it is OK
1958 * to preserve that layout.
1959 * - When converting a RAID5 to RAID6, leave it in
1960 * the XXX-6 layout, don't re-layout.
1961 */
1962 if (info.array.level == 6 && info.new_level == UnSet)
1963 info.new_layout = info.array.layout;
1964 else if (info.array.level == 5 && info.new_level == 6) {
1965 char l[40];
1966 strcpy(l, map_num(r5layout, info.array.layout));
1967 strcat(l, "-6");
1968 info.new_layout = map_name(r6layout, l);
1969 } else {
1970 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
1971 rv = 1;
1972 goto release;
1973 }
1974 } else {
1975 int l = info.new_level;
1976 if (l == UnSet)
1977 l = info.array.level;
1978 switch (l) {
1979 case 5:
1980 info.new_layout = map_name(r5layout, s->layout_str);
1981 break;
1982 case 6:
1983 info.new_layout = map_name(r6layout, s->layout_str);
1984 break;
1985 case 10:
1986 info.new_layout = parse_layout_10(s->layout_str);
1987 break;
1988 case LEVEL_FAULTY:
1989 info.new_layout = parse_layout_faulty(s->layout_str);
1990 break;
1991 default:
1992 pr_err("layout not meaningful with this level\n");
1993 rv = 1;
1994 goto release;
1995 }
1996 if (info.new_layout == UnSet) {
1997 pr_err("layout %s not understood for this level\n",
1998 s->layout_str);
1999 rv = 1;
2000 goto release;
2001 }
2002 }
2003
2004 if (array.level == LEVEL_FAULTY) {
2005 if (s->level != UnSet && s->level != array.level) {
2006 pr_err("cannot change level of Faulty device\n");
2007 rv =1 ;
2008 }
2009 if (s->chunk) {
2010 pr_err("cannot set chunksize of Faulty device\n");
2011 rv =1 ;
2012 }
2013 if (s->raiddisks && s->raiddisks != 1) {
2014 pr_err("cannot set raid_disks of Faulty device\n");
2015 rv =1 ;
2016 }
2017 if (s->layout_str) {
2018 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2019 dprintf("Cannot get array information.\n");
2020 goto release;
2021 }
2022 array.layout = info.new_layout;
2023 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2024 pr_err("failed to set new layout\n");
2025 rv = 1;
2026 } else if (c->verbose >= 0)
2027 printf("layout for %s set to %d\n",
2028 devname, array.layout);
2029 }
2030 } else if (array.level == LEVEL_CONTAINER) {
2031 /* This change is to be applied to every array in the
2032 * container. This is only needed when the metadata imposes
2033 * restraints of the various arrays in the container.
2034 * Currently we only know that IMSM requires all arrays
2035 * to have the same number of devices so changing the
2036 * number of devices (On-Line Capacity Expansion) must be
2037 * performed at the level of the container
2038 */
2039 rv = reshape_container(container, devname, -1, st, &info,
2040 c->force, c->backup_file, c->verbose, 0, 0, 0);
2041 frozen = 0;
2042 } else {
2043 /* get spare devices from external metadata
2044 */
2045 if (st->ss->external) {
2046 struct mdinfo *info2;
2047
2048 info2 = st->ss->container_content(st, subarray);
2049 if (info2) {
2050 info.array.spare_disks =
2051 info2->array.spare_disks;
2052 sysfs_free(info2);
2053 }
2054 }
2055
2056 /* Impose these changes on a single array. First
2057 * check that the metadata is OK with the change. */
2058
2059 if (reshape_super(st, 0, info.new_level,
2060 info.new_layout, info.new_chunk,
2061 info.array.raid_disks, info.delta_disks,
2062 c->backup_file, devname, APPLY_METADATA_CHANGES,
2063 c->verbose)) {
2064 rv = 1;
2065 goto release;
2066 }
2067 sync_metadata(st);
2068 rv = reshape_array(container, fd, devname, st, &info, c->force,
2069 devlist, data_offset, c->backup_file, c->verbose,
2070 0, 0, 0);
2071 frozen = 0;
2072 }
2073 release:
2074 sysfs_free(sra);
2075 if (frozen > 0)
2076 unfreeze(st);
2077 return rv;
2078 }
2079
2080 /* verify_reshape_position()
2081 * Function checks if reshape position in metadata is not farther
2082 * than position in md.
2083 * Return value:
2084 * 0 : not valid sysfs entry
2085 * it can be caused by not started reshape, it should be started
2086 * by reshape array or raid0 array is before takeover
2087 * -1 : error, reshape position is obviously wrong
2088 * 1 : success, reshape progress correct or updated
2089 */
2090 static int verify_reshape_position(struct mdinfo *info, int level)
2091 {
2092 int ret_val = 0;
2093 char buf[40];
2094 int rv;
2095
2096 /* read sync_max, failure can mean raid0 array */
2097 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2098
2099 if (rv > 0) {
2100 char *ep;
2101 unsigned long long position = strtoull(buf, &ep, 0);
2102
2103 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2104 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2105 position *= get_data_disks(level,
2106 info->new_layout,
2107 info->array.raid_disks);
2108 if (info->reshape_progress < position) {
2109 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2110 info->reshape_progress, position);
2111 info->reshape_progress = position;
2112 ret_val = 1;
2113 } else if (info->reshape_progress > position) {
2114 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2115 position, info->reshape_progress);
2116 ret_val = -1;
2117 } else {
2118 dprintf("Reshape position in md and metadata are the same;");
2119 ret_val = 1;
2120 }
2121 }
2122 } else if (rv == 0) {
2123 /* for valid sysfs entry, 0-length content
2124 * should be indicated as error
2125 */
2126 ret_val = -1;
2127 }
2128
2129 return ret_val;
2130 }
2131
2132 static unsigned long long choose_offset(unsigned long long lo,
2133 unsigned long long hi,
2134 unsigned long long min,
2135 unsigned long long max)
2136 {
2137 /* Choose a new offset between hi and lo.
2138 * It must be between min and max, but
2139 * we would prefer something near the middle of hi/lo, and also
2140 * prefer to be aligned to a big power of 2.
2141 *
2142 * So we start with the middle, then for each bit,
2143 * starting at '1' and increasing, if it is set, we either
2144 * add it or subtract it if possible, preferring the option
2145 * which is furthest from the boundary.
2146 *
2147 * We stop once we get a 1MB alignment. As units are in sectors,
2148 * 1MB = 2*1024 sectors.
2149 */
2150 unsigned long long choice = (lo + hi) / 2;
2151 unsigned long long bit = 1;
2152
2153 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2154 unsigned long long bigger, smaller;
2155 if (! (bit & choice))
2156 continue;
2157 bigger = choice + bit;
2158 smaller = choice - bit;
2159 if (bigger > max && smaller < min)
2160 break;
2161 if (bigger > max)
2162 choice = smaller;
2163 else if (smaller < min)
2164 choice = bigger;
2165 else if (hi - bigger > smaller - lo)
2166 choice = bigger;
2167 else
2168 choice = smaller;
2169 }
2170 return choice;
2171 }
2172
2173 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2174 char *devname, int delta_disks,
2175 unsigned long long data_offset,
2176 unsigned long long min,
2177 int can_fallback)
2178 {
2179 struct mdinfo *sd;
2180 int dir = 0;
2181 int err = 0;
2182 unsigned long long before, after;
2183
2184 /* Need to find min space before and after so same is used
2185 * on all devices
2186 */
2187 before = UINT64_MAX;
2188 after = UINT64_MAX;
2189 for (sd = sra->devs; sd; sd = sd->next) {
2190 char *dn;
2191 int dfd;
2192 int rv;
2193 struct supertype *st2;
2194 struct mdinfo info2;
2195
2196 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2197 continue;
2198 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2199 dfd = dev_open(dn, O_RDONLY);
2200 if (dfd < 0) {
2201 pr_err("%s: cannot open component %s\n",
2202 devname, dn ? dn : "-unknown-");
2203 goto release;
2204 }
2205 st2 = dup_super(st);
2206 rv = st2->ss->load_super(st2,dfd, NULL);
2207 close(dfd);
2208 if (rv) {
2209 free(st2);
2210 pr_err("%s: cannot get superblock from %s\n",
2211 devname, dn);
2212 goto release;
2213 }
2214 st2->ss->getinfo_super(st2, &info2, NULL);
2215 st2->ss->free_super(st2);
2216 free(st2);
2217 if (info2.space_before == 0 &&
2218 info2.space_after == 0) {
2219 /* Metadata doesn't support data_offset changes */
2220 if (!can_fallback)
2221 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2222 devname);
2223 goto fallback;
2224 }
2225 if (before > info2.space_before)
2226 before = info2.space_before;
2227 if (after > info2.space_after)
2228 after = info2.space_after;
2229
2230 if (data_offset != INVALID_SECTORS) {
2231 if (dir == 0) {
2232 if (info2.data_offset == data_offset) {
2233 pr_err("%s: already has that data_offset\n",
2234 dn);
2235 goto release;
2236 }
2237 if (data_offset < info2.data_offset)
2238 dir = -1;
2239 else
2240 dir = 1;
2241 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2242 (data_offset >= info2.data_offset && dir == -1)) {
2243 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2244 dn);
2245 goto release;
2246 }
2247 }
2248 }
2249 if (before == UINT64_MAX)
2250 /* impossible really, there must be no devices */
2251 return 1;
2252
2253 for (sd = sra->devs; sd; sd = sd->next) {
2254 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2255 unsigned long long new_data_offset;
2256
2257 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2258 continue;
2259 if (delta_disks < 0) {
2260 /* Don't need any space as array is shrinking
2261 * just move data_offset up by min
2262 */
2263 if (data_offset == INVALID_SECTORS)
2264 new_data_offset = sd->data_offset + min;
2265 else {
2266 if (data_offset < sd->data_offset + min) {
2267 pr_err("--data-offset too small for %s\n",
2268 dn);
2269 goto release;
2270 }
2271 new_data_offset = data_offset;
2272 }
2273 } else if (delta_disks > 0) {
2274 /* need space before */
2275 if (before < min) {
2276 if (can_fallback)
2277 goto fallback;
2278 pr_err("Insufficient head-space for reshape on %s\n",
2279 dn);
2280 goto release;
2281 }
2282 if (data_offset == INVALID_SECTORS)
2283 new_data_offset = sd->data_offset - min;
2284 else {
2285 if (data_offset > sd->data_offset - min) {
2286 pr_err("--data-offset too large for %s\n",
2287 dn);
2288 goto release;
2289 }
2290 new_data_offset = data_offset;
2291 }
2292 } else {
2293 if (dir == 0) {
2294 /* can move up or down. If 'data_offset'
2295 * was set we would have already decided,
2296 * so just choose direction with most space.
2297 */
2298 if (before > after)
2299 dir = -1;
2300 else
2301 dir = 1;
2302 }
2303 sysfs_set_str(sra, NULL, "reshape_direction",
2304 dir == 1 ? "backwards" : "forwards");
2305 if (dir > 0) {
2306 /* Increase data offset */
2307 if (after < min) {
2308 if (can_fallback)
2309 goto fallback;
2310 pr_err("Insufficient tail-space for reshape on %s\n",
2311 dn);
2312 goto release;
2313 }
2314 if (data_offset != INVALID_SECTORS &&
2315 data_offset < sd->data_offset + min) {
2316 pr_err("--data-offset too small on %s\n",
2317 dn);
2318 goto release;
2319 }
2320 if (data_offset != INVALID_SECTORS)
2321 new_data_offset = data_offset;
2322 else
2323 new_data_offset = choose_offset(sd->data_offset,
2324 sd->data_offset + after,
2325 sd->data_offset + min,
2326 sd->data_offset + after);
2327 } else {
2328 /* Decrease data offset */
2329 if (before < min) {
2330 if (can_fallback)
2331 goto fallback;
2332 pr_err("insufficient head-room on %s\n",
2333 dn);
2334 goto release;
2335 }
2336 if (data_offset != INVALID_SECTORS &&
2337 data_offset < sd->data_offset - min) {
2338 pr_err("--data-offset too small on %s\n",
2339 dn);
2340 goto release;
2341 }
2342 if (data_offset != INVALID_SECTORS)
2343 new_data_offset = data_offset;
2344 else
2345 new_data_offset = choose_offset(sd->data_offset - before,
2346 sd->data_offset,
2347 sd->data_offset - before,
2348 sd->data_offset - min);
2349 }
2350 }
2351 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2352 if (err < 0 && errno == E2BIG) {
2353 /* try again after increasing data size to max */
2354 err = sysfs_set_num(sra, sd, "size", 0);
2355 if (err < 0 && errno == EINVAL &&
2356 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2357 /* some kernels have a bug where you cannot
2358 * use '0' on spare devices. */
2359 sysfs_set_num(sra, sd, "size",
2360 (sra->component_size + after)/2);
2361 }
2362 err = sysfs_set_num(sra, sd, "new_offset",
2363 new_data_offset);
2364 }
2365 if (err < 0) {
2366 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2367 pr_err("data-offset is too big for %s\n",
2368 dn);
2369 goto release;
2370 }
2371 if (sd == sra->devs &&
2372 (errno == ENOENT || errno == E2BIG))
2373 /* Early kernel, no 'new_offset' file,
2374 * or kernel doesn't like us.
2375 * For RAID5/6 this is not fatal
2376 */
2377 return 1;
2378 pr_err("Cannot set new_offset for %s\n",
2379 dn);
2380 break;
2381 }
2382 }
2383 return err;
2384 release:
2385 return -1;
2386 fallback:
2387 /* Just use a backup file */
2388 return 1;
2389 }
2390
2391 static int raid10_reshape(char *container, int fd, char *devname,
2392 struct supertype *st, struct mdinfo *info,
2393 struct reshape *reshape,
2394 unsigned long long data_offset,
2395 int force, int verbose)
2396 {
2397 /* Changing raid_disks, layout, chunksize or possibly
2398 * just data_offset for a RAID10.
2399 * We must always change data_offset. We change by at least
2400 * ->min_offset_change which is the largest of the old and new
2401 * chunk sizes.
2402 * If raid_disks is increasing, then data_offset must decrease
2403 * by at least this copy size.
2404 * If raid_disks is unchanged, data_offset must increase or
2405 * decrease by at least min_offset_change but preferably by much more.
2406 * We choose half of the available space.
2407 * If raid_disks is decreasing, data_offset must increase by
2408 * at least min_offset_change. To allow of this, component_size
2409 * must be decreased by the same amount.
2410 *
2411 * So we calculate the required minimum and direction, possibly
2412 * reduce the component_size, then iterate through the devices
2413 * and set the new_data_offset.
2414 * If that all works, we set chunk_size, layout, raid_disks, and start
2415 * 'reshape'
2416 */
2417 struct mdinfo *sra;
2418 unsigned long long min;
2419 int err = 0;
2420
2421 sra = sysfs_read(fd, NULL,
2422 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2423 );
2424 if (!sra) {
2425 pr_err("%s: Cannot get array details from sysfs\n",
2426 devname);
2427 goto release;
2428 }
2429 min = reshape->min_offset_change;
2430
2431 if (info->delta_disks)
2432 sysfs_set_str(sra, NULL, "reshape_direction",
2433 info->delta_disks < 0 ? "backwards" : "forwards");
2434 if (info->delta_disks < 0 &&
2435 info->space_after < min) {
2436 int rv = sysfs_set_num(sra, NULL, "component_size",
2437 (sra->component_size -
2438 min)/2);
2439 if (rv) {
2440 pr_err("cannot reduce component size\n");
2441 goto release;
2442 }
2443 }
2444 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2445 min, 0);
2446 if (err == 1) {
2447 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2448 cont_err("supported on this kernel\n");
2449 err = -1;
2450 }
2451 if (err < 0)
2452 goto release;
2453
2454 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2455 err = errno;
2456 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2457 err = errno;
2458 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2459 info->array.raid_disks + info->delta_disks) < 0)
2460 err = errno;
2461 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2462 err = errno;
2463 if (err) {
2464 pr_err("Cannot set array shape for %s\n",
2465 devname);
2466 if (err == EBUSY &&
2467 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2468 cont_err(" Bitmap must be removed before shape can be changed\n");
2469 goto release;
2470 }
2471 sysfs_free(sra);
2472 return 0;
2473 release:
2474 sysfs_free(sra);
2475 return 1;
2476 }
2477
2478 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2479 {
2480 struct mdinfo *sra, *sd;
2481 /* Initialisation to silence compiler warning */
2482 unsigned long long min_space_before = 0, min_space_after = 0;
2483 int first = 1;
2484
2485 sra = sysfs_read(fd, NULL, GET_DEVS);
2486 if (!sra)
2487 return;
2488 for (sd = sra->devs; sd; sd = sd->next) {
2489 char *dn;
2490 int dfd;
2491 struct supertype *st2;
2492 struct mdinfo info2;
2493
2494 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2495 continue;
2496 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2497 dfd = dev_open(dn, O_RDONLY);
2498 if (dfd < 0)
2499 break;
2500 st2 = dup_super(st);
2501 if (st2->ss->load_super(st2,dfd, NULL)) {
2502 close(dfd);
2503 free(st2);
2504 break;
2505 }
2506 close(dfd);
2507 st2->ss->getinfo_super(st2, &info2, NULL);
2508 st2->ss->free_super(st2);
2509 free(st2);
2510 if (first ||
2511 min_space_before > info2.space_before)
2512 min_space_before = info2.space_before;
2513 if (first ||
2514 min_space_after > info2.space_after)
2515 min_space_after = info2.space_after;
2516 first = 0;
2517 }
2518 if (sd == NULL && !first) {
2519 info->space_after = min_space_after;
2520 info->space_before = min_space_before;
2521 }
2522 sysfs_free(sra);
2523 }
2524
2525 static void update_cache_size(char *container, struct mdinfo *sra,
2526 struct mdinfo *info,
2527 int disks, unsigned long long blocks)
2528 {
2529 /* Check that the internal stripe cache is
2530 * large enough, or it won't work.
2531 * It must hold at least 4 stripes of the larger
2532 * chunk size
2533 */
2534 unsigned long cache;
2535 cache = max(info->array.chunk_size, info->new_chunk);
2536 cache *= 4; /* 4 stripes minimum */
2537 cache /= 512; /* convert to sectors */
2538 /* make sure there is room for 'blocks' with a bit to spare */
2539 if (cache < 16 + blocks / disks)
2540 cache = 16 + blocks / disks;
2541 cache /= (4096/512); /* Convert from sectors to pages */
2542
2543 if (sra->cache_size < cache)
2544 subarray_set_num(container, sra, "stripe_cache_size",
2545 cache+1);
2546 }
2547
2548 static int impose_reshape(struct mdinfo *sra,
2549 struct mdinfo *info,
2550 struct supertype *st,
2551 int fd,
2552 int restart,
2553 char *devname, char *container,
2554 struct reshape *reshape)
2555 {
2556 struct mdu_array_info_s array;
2557
2558 sra->new_chunk = info->new_chunk;
2559
2560 if (restart) {
2561 /* for external metadata checkpoint saved by mdmon can be lost
2562 * or missed /due to e.g. crash/. Check if md is not during
2563 * restart farther than metadata points to.
2564 * If so, this means metadata information is obsolete.
2565 */
2566 if (st->ss->external)
2567 verify_reshape_position(info, reshape->level);
2568 sra->reshape_progress = info->reshape_progress;
2569 } else {
2570 sra->reshape_progress = 0;
2571 if (reshape->after.data_disks < reshape->before.data_disks)
2572 /* start from the end of the new array */
2573 sra->reshape_progress = (sra->component_size
2574 * reshape->after.data_disks);
2575 }
2576
2577 ioctl(fd, GET_ARRAY_INFO, &array);
2578 if (info->array.chunk_size == info->new_chunk &&
2579 reshape->before.layout == reshape->after.layout &&
2580 st->ss->external == 0) {
2581 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2582 array.raid_disks = reshape->after.data_disks + reshape->parity;
2583 if (!restart &&
2584 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2585 int err = errno;
2586
2587 pr_err("Cannot set device shape for %s: %s\n",
2588 devname, strerror(errno));
2589
2590 if (err == EBUSY &&
2591 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2592 cont_err("Bitmap must be removed before shape can be changed\n");
2593
2594 goto release;
2595 }
2596 } else if (!restart) {
2597 /* set them all just in case some old 'new_*' value
2598 * persists from some earlier problem.
2599 */
2600 int err = 0;
2601 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2602 err = errno;
2603 if (!err && sysfs_set_num(sra, NULL, "layout",
2604 reshape->after.layout) < 0)
2605 err = errno;
2606 if (!err && subarray_set_num(container, sra, "raid_disks",
2607 reshape->after.data_disks +
2608 reshape->parity) < 0)
2609 err = errno;
2610 if (err) {
2611 pr_err("Cannot set device shape for %s\n",
2612 devname);
2613
2614 if (err == EBUSY &&
2615 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2616 cont_err("Bitmap must be removed before shape can be changed\n");
2617 goto release;
2618 }
2619 }
2620 return 0;
2621 release:
2622 return -1;
2623 }
2624
2625 static int impose_level(int fd, int level, char *devname, int verbose)
2626 {
2627 char *c;
2628 struct mdu_array_info_s array;
2629 struct mdinfo info;
2630 sysfs_init(&info, fd, NULL);
2631
2632 ioctl(fd, GET_ARRAY_INFO, &array);
2633 if (level == 0 &&
2634 (array.level >= 4 && array.level <= 6)) {
2635 /* To convert to RAID0 we need to fail and
2636 * remove any non-data devices. */
2637 int found = 0;
2638 int d;
2639 int data_disks = array.raid_disks - 1;
2640 if (array.level == 6)
2641 data_disks -= 1;
2642 if (array.level == 5 &&
2643 array.layout != ALGORITHM_PARITY_N)
2644 return -1;
2645 if (array.level == 6 &&
2646 array.layout != ALGORITHM_PARITY_N_6)
2647 return -1;
2648 sysfs_set_str(&info, NULL,"sync_action", "idle");
2649 /* First remove any spares so no recovery starts */
2650 for (d = 0, found = 0;
2651 d < MAX_DISKS && found < array.nr_disks;
2652 d++) {
2653 mdu_disk_info_t disk;
2654 disk.number = d;
2655 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2656 continue;
2657 if (disk.major == 0 && disk.minor == 0)
2658 continue;
2659 found++;
2660 if ((disk.state & (1 << MD_DISK_ACTIVE))
2661 && disk.raid_disk < data_disks)
2662 /* keep this */
2663 continue;
2664 ioctl(fd, HOT_REMOVE_DISK,
2665 makedev(disk.major, disk.minor));
2666 }
2667 /* Now fail anything left */
2668 ioctl(fd, GET_ARRAY_INFO, &array);
2669 for (d = 0, found = 0;
2670 d < MAX_DISKS && found < array.nr_disks;
2671 d++) {
2672 int cnt;
2673 mdu_disk_info_t disk;
2674 disk.number = d;
2675 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2676 continue;
2677 if (disk.major == 0 && disk.minor == 0)
2678 continue;
2679 found++;
2680 if ((disk.state & (1 << MD_DISK_ACTIVE))
2681 && disk.raid_disk < data_disks)
2682 /* keep this */
2683 continue;
2684 ioctl(fd, SET_DISK_FAULTY,
2685 makedev(disk.major, disk.minor));
2686 cnt = 5;
2687 while (ioctl(fd, HOT_REMOVE_DISK,
2688 makedev(disk.major, disk.minor)) < 0
2689 && errno == EBUSY
2690 && cnt--) {
2691 usleep(10000);
2692 }
2693 }
2694 }
2695 c = map_num(pers, level);
2696 if (c) {
2697 int err = sysfs_set_str(&info, NULL, "level", c);
2698 if (err) {
2699 err = errno;
2700 pr_err("%s: could not set level to %s\n",
2701 devname, c);
2702 if (err == EBUSY &&
2703 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2704 cont_err("Bitmap must be removed before level can be changed\n");
2705 return err;
2706 }
2707 if (verbose >= 0)
2708 pr_err("level of %s changed to %s\n",
2709 devname, c);
2710 }
2711 return 0;
2712 }
2713
2714 int sigterm = 0;
2715 static void catch_term(int sig)
2716 {
2717 sigterm = 1;
2718 }
2719
2720 static int continue_via_systemd(char *devnm)
2721 {
2722 int skipped, i, pid, status;
2723 char pathbuf[1024];
2724 /* In a systemd/udev world, it is best to get systemd to
2725 * run "mdadm --grow --continue" rather than running in the
2726 * background.
2727 */
2728 switch(fork()) {
2729 case 0:
2730 /* FIXME yuk. CLOSE_EXEC?? */
2731 skipped = 0;
2732 for (i = 3; skipped < 20; i++)
2733 if (close(i) < 0)
2734 skipped++;
2735 else
2736 skipped = 0;
2737
2738 /* Don't want to see error messages from
2739 * systemctl. If the service doesn't exist,
2740 * we fork ourselves.
2741 */
2742 close(2);
2743 open("/dev/null", O_WRONLY);
2744 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2745 devnm);
2746 status = execl("/usr/bin/systemctl", "systemctl",
2747 "start",
2748 pathbuf, NULL);
2749 status = execl("/bin/systemctl", "systemctl", "start",
2750 pathbuf, NULL);
2751 exit(1);
2752 case -1: /* Just do it ourselves. */
2753 break;
2754 default: /* parent - good */
2755 pid = wait(&status);
2756 if (pid >= 0 && status == 0)
2757 return 1;
2758 }
2759 return 0;
2760 }
2761
2762 static int reshape_array(char *container, int fd, char *devname,
2763 struct supertype *st, struct mdinfo *info,
2764 int force, struct mddev_dev *devlist,
2765 unsigned long long data_offset,
2766 char *backup_file, int verbose, int forked,
2767 int restart, int freeze_reshape)
2768 {
2769 struct reshape reshape;
2770 int spares_needed;
2771 char *msg;
2772 int orig_level = UnSet;
2773 int odisks;
2774 int delayed;
2775
2776 struct mdu_array_info_s array;
2777 char *c;
2778
2779 struct mddev_dev *dv;
2780 int added_disks;
2781
2782 int *fdlist = NULL;
2783 unsigned long long *offsets = NULL;
2784 int d;
2785 int nrdisks;
2786 int err;
2787 unsigned long blocks;
2788 unsigned long long array_size;
2789 int done;
2790 struct mdinfo *sra = NULL;
2791 char buf[20];
2792
2793 /* when reshaping a RAID0, the component_size might be zero.
2794 * So try to fix that up.
2795 */
2796 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2797 dprintf("Cannot get array information.\n");
2798 goto release;
2799 }
2800 if (array.level == 0 && info->component_size == 0) {
2801 get_dev_size(fd, NULL, &array_size);
2802 info->component_size = array_size / array.raid_disks;
2803 }
2804
2805 if (array.level == 10)
2806 /* Need space_after info */
2807 get_space_after(fd, st, info);
2808
2809 if (info->reshape_active) {
2810 int new_level = info->new_level;
2811 info->new_level = UnSet;
2812 if (info->delta_disks > 0)
2813 info->array.raid_disks -= info->delta_disks;
2814 msg = analyse_change(devname, info, &reshape);
2815 info->new_level = new_level;
2816 if (info->delta_disks > 0)
2817 info->array.raid_disks += info->delta_disks;
2818 if (!restart)
2819 /* Make sure the array isn't read-only */
2820 ioctl(fd, RESTART_ARRAY_RW, 0);
2821 } else
2822 msg = analyse_change(devname, info, &reshape);
2823 if (msg) {
2824 /* if msg == "", error has already been printed */
2825 if (msg[0])
2826 pr_err("%s\n", msg);
2827 goto release;
2828 }
2829 if (restart &&
2830 (reshape.level != info->array.level ||
2831 reshape.before.layout != info->array.layout ||
2832 reshape.before.data_disks + reshape.parity
2833 != info->array.raid_disks - max(0, info->delta_disks))) {
2834 pr_err("reshape info is not in native format - cannot continue.\n");
2835 goto release;
2836 }
2837
2838 if (st->ss->external && restart && (info->reshape_progress == 0) &&
2839 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
2840 (strncmp(buf, "reshape", 7) == 0))) {
2841 /* When reshape is restarted from '0', very begin of array
2842 * it is possible that for external metadata reshape and array
2843 * configuration doesn't happen.
2844 * Check if md has the same opinion, and reshape is restarted
2845 * from 0. If so, this is regular reshape start after reshape
2846 * switch in metadata to next array only.
2847 */
2848 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2849 (info->reshape_progress == 0))
2850 restart = 0;
2851 }
2852 if (restart) {
2853 /* reshape already started. just skip to monitoring the reshape */
2854 if (reshape.backup_blocks == 0)
2855 return 0;
2856 if (restart & RESHAPE_NO_BACKUP)
2857 return 0;
2858
2859 /* Need 'sra' down at 'started:' */
2860 sra = sysfs_read(fd, NULL,
2861 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2862 GET_CACHE);
2863 if (!sra) {
2864 pr_err("%s: Cannot get array details from sysfs\n",
2865 devname);
2866 goto release;
2867 }
2868
2869 if (!backup_file)
2870 backup_file = locate_backup(sra->sys_name);
2871
2872 goto started;
2873 }
2874 /* The container is frozen but the array may not be.
2875 * So freeze the array so spares don't get put to the wrong use
2876 * FIXME there should probably be a cleaner separation between
2877 * freeze_array and freeze_container.
2878 */
2879 sysfs_freeze_array(info);
2880 /* Check we have enough spares to not be degraded */
2881 added_disks = 0;
2882 for (dv = devlist; dv ; dv=dv->next)
2883 added_disks++;
2884 spares_needed = max(reshape.before.data_disks,
2885 reshape.after.data_disks)
2886 + reshape.parity - array.raid_disks;
2887
2888 if (!force &&
2889 info->new_level > 1 && info->array.level > 1 &&
2890 spares_needed > info->array.spare_disks + added_disks) {
2891 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2892 " Use --force to over-ride this check.\n",
2893 spares_needed,
2894 spares_needed == 1 ? "" : "s",
2895 info->array.spare_disks + added_disks);
2896 goto release;
2897 }
2898 /* Check we have enough spares to not fail */
2899 spares_needed = max(reshape.before.data_disks,
2900 reshape.after.data_disks)
2901 - array.raid_disks;
2902 if ((info->new_level > 1 || info->new_level == 0) &&
2903 spares_needed > info->array.spare_disks +added_disks) {
2904 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2905 spares_needed,
2906 spares_needed == 1 ? "" : "s",
2907 info->array.spare_disks + added_disks);
2908 goto release;
2909 }
2910
2911 if (reshape.level != array.level) {
2912 int err = impose_level(fd, reshape.level, devname, verbose);
2913 if (err)
2914 goto release;
2915 info->new_layout = UnSet; /* after level change,
2916 * layout is meaningless */
2917 orig_level = array.level;
2918 sysfs_freeze_array(info);
2919
2920 if (reshape.level > 0 && st->ss->external) {
2921 /* make sure mdmon is aware of the new level */
2922 if (mdmon_running(container))
2923 flush_mdmon(container);
2924
2925 if (!mdmon_running(container))
2926 start_mdmon(container);
2927 ping_monitor(container);
2928 if (mdmon_running(container) &&
2929 st->update_tail == NULL)
2930 st->update_tail = &st->updates;
2931 }
2932 }
2933 /* ->reshape_super might have chosen some spares from the
2934 * container that it wants to be part of the new array.
2935 * We can collect them with ->container_content and give
2936 * them to the kernel.
2937 */
2938 if (st->ss->reshape_super && st->ss->container_content) {
2939 char *subarray = strchr(info->text_version+1, '/')+1;
2940 struct mdinfo *info2 =
2941 st->ss->container_content(st, subarray);
2942 struct mdinfo *d;
2943
2944 if (info2) {
2945 sysfs_init(info2, fd, st->devnm);
2946 /* When increasing number of devices, we need to set
2947 * new raid_disks before adding these, or they might
2948 * be rejected.
2949 */
2950 if (reshape.backup_blocks &&
2951 reshape.after.data_disks > reshape.before.data_disks)
2952 subarray_set_num(container, info2, "raid_disks",
2953 reshape.after.data_disks +
2954 reshape.parity);
2955 for (d = info2->devs; d; d = d->next) {
2956 if (d->disk.state == 0 &&
2957 d->disk.raid_disk >= 0) {
2958 /* This is a spare that wants to
2959 * be part of the array.
2960 */
2961 add_disk(fd, st, info2, d);
2962 }
2963 }
2964 sysfs_free(info2);
2965 }
2966 }
2967 /* We might have been given some devices to add to the
2968 * array. Now that the array has been changed to the right
2969 * level and frozen, we can safely add them.
2970 */
2971 if (devlist)
2972 Manage_subdevs(devname, fd, devlist, verbose,
2973 0,NULL, 0);
2974
2975 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
2976 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
2977 if (reshape.backup_blocks == 0) {
2978 /* No restriping needed, but we might need to impose
2979 * some more changes: layout, raid_disks, chunk_size
2980 */
2981 /* read current array info */
2982 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2983 dprintf("Cannot get array information.\n");
2984 goto release;
2985 }
2986 /* compare current array info with new values and if
2987 * it is different update them to new */
2988 if (info->new_layout != UnSet &&
2989 info->new_layout != array.layout) {
2990 array.layout = info->new_layout;
2991 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2992 pr_err("failed to set new layout\n");
2993 goto release;
2994 } else if (verbose >= 0)
2995 printf("layout for %s set to %d\n",
2996 devname, array.layout);
2997 }
2998 if (info->delta_disks != UnSet &&
2999 info->delta_disks != 0 &&
3000 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3001 array.raid_disks += info->delta_disks;
3002 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3003 pr_err("failed to set raid disks\n");
3004 goto release;
3005 } else if (verbose >= 0) {
3006 printf("raid_disks for %s set to %d\n",
3007 devname, array.raid_disks);
3008 }
3009 }
3010 if (info->new_chunk != 0 &&
3011 info->new_chunk != array.chunk_size) {
3012 if (sysfs_set_num(info, NULL,
3013 "chunk_size", info->new_chunk) != 0) {
3014 pr_err("failed to set chunk size\n");
3015 goto release;
3016 } else if (verbose >= 0)
3017 printf("chunk size for %s set to %d\n",
3018 devname, array.chunk_size);
3019 }
3020 unfreeze(st);
3021 return 0;
3022 }
3023
3024 /*
3025 * There are three possibilities.
3026 * 1/ The array will shrink.
3027 * We need to ensure the reshape will pause before reaching
3028 * the 'critical section'. We also need to fork and wait for
3029 * that to happen. When it does we
3030 * suspend/backup/complete/unfreeze
3031 *
3032 * 2/ The array will not change size.
3033 * This requires that we keep a backup of a sliding window
3034 * so that we can restore data after a crash. So we need
3035 * to fork and monitor progress.
3036 * In future we will allow the data_offset to change, so
3037 * a sliding backup becomes unnecessary.
3038 *
3039 * 3/ The array will grow. This is relatively easy.
3040 * However the kernel's restripe routines will cheerfully
3041 * overwrite some early data before it is safe. So we
3042 * need to make a backup of the early parts of the array
3043 * and be ready to restore it if rebuild aborts very early.
3044 * For externally managed metadata, we still need a forked
3045 * child to monitor the reshape and suspend IO over the region
3046 * that is being reshaped.
3047 *
3048 * We backup data by writing it to one spare, or to a
3049 * file which was given on command line.
3050 *
3051 * In each case, we first make sure that storage is available
3052 * for the required backup.
3053 * Then we:
3054 * - request the shape change.
3055 * - fork to handle backup etc.
3056 */
3057 /* Check that we can hold all the data */
3058 get_dev_size(fd, NULL, &array_size);
3059 if (reshape.new_size < (array_size/512)) {
3060 pr_err("this change will reduce the size of the array.\n"
3061 " use --grow --array-size first to truncate array.\n"
3062 " e.g. mdadm --grow %s --array-size %llu\n",
3063 devname, reshape.new_size/2);
3064 goto release;
3065 }
3066
3067 if (array.level == 10) {
3068 /* Reshaping RAID10 does not require any data backup by
3069 * user-space. Instead it requires that the data_offset
3070 * is changed to avoid the need for backup.
3071 * So this is handled very separately
3072 */
3073 if (restart)
3074 /* Nothing to do. */
3075 return 0;
3076 return raid10_reshape(container, fd, devname, st, info,
3077 &reshape, data_offset,
3078 force, verbose);
3079 }
3080 sra = sysfs_read(fd, NULL,
3081 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3082 GET_CACHE);
3083 if (!sra) {
3084 pr_err("%s: Cannot get array details from sysfs\n",
3085 devname);
3086 goto release;
3087 }
3088
3089 if (!backup_file)
3090 switch(set_new_data_offset(sra, st, devname,
3091 reshape.after.data_disks - reshape.before.data_disks,
3092 data_offset,
3093 reshape.min_offset_change, 1)) {
3094 case -1:
3095 goto release;
3096 case 0:
3097 /* Updated data_offset, so it's easy now */
3098 update_cache_size(container, sra, info,
3099 min(reshape.before.data_disks,
3100 reshape.after.data_disks),
3101 reshape.backup_blocks);
3102
3103 /* Right, everything seems fine. Let's kick things off.
3104 */
3105 sync_metadata(st);
3106
3107 if (impose_reshape(sra, info, st, fd, restart,
3108 devname, container, &reshape) < 0)
3109 goto release;
3110 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3111 struct mdinfo *sd;
3112 if (errno != EINVAL) {
3113 pr_err("Failed to initiate reshape!\n");
3114 goto release;
3115 }
3116 /* revert data_offset and try the old way */
3117 for (sd = sra->devs; sd; sd = sd->next) {
3118 sysfs_set_num(sra, sd, "new_offset",
3119 sd->data_offset);
3120 sysfs_set_str(sra, NULL, "reshape_direction",
3121 "forwards");
3122 }
3123 break;
3124 }
3125 if (info->new_level == reshape.level)
3126 return 0;
3127 /* need to adjust level when reshape completes */
3128 switch(fork()) {
3129 case -1: /* ignore error, but don't wait */
3130 return 0;
3131 default: /* parent */
3132 return 0;
3133 case 0:
3134 map_fork();
3135 break;
3136 }
3137 close(fd);
3138 wait_reshape(sra);
3139 fd = open_dev(sra->sys_name);
3140 if (fd >= 0)
3141 impose_level(fd, info->new_level, devname, verbose);
3142 return 0;
3143 case 1: /* Couldn't set data_offset, try the old way */
3144 if (data_offset != INVALID_SECTORS) {
3145 pr_err("Cannot update data_offset on this array\n");
3146 goto release;
3147 }
3148 break;
3149 }
3150
3151 started:
3152 /* Decide how many blocks (sectors) for a reshape
3153 * unit. The number we have so far is just a minimum
3154 */
3155 blocks = reshape.backup_blocks;
3156 if (reshape.before.data_disks ==
3157 reshape.after.data_disks) {
3158 /* Make 'blocks' bigger for better throughput, but
3159 * not so big that we reject it below.
3160 * Try for 16 megabytes
3161 */
3162 while (blocks * 32 < sra->component_size &&
3163 blocks < 16*1024*2)
3164 blocks *= 2;
3165 } else
3166 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3167
3168 if (blocks >= sra->component_size/2) {
3169 pr_err("%s: Something wrong - reshape aborted\n",
3170 devname);
3171 goto release;
3172 }
3173
3174 /* Now we need to open all these devices so we can read/write.
3175 */
3176 nrdisks = max(reshape.before.data_disks,
3177 reshape.after.data_disks) + reshape.parity
3178 + sra->array.spare_disks;
3179 fdlist = xcalloc((1+nrdisks), sizeof(int));
3180 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3181
3182 odisks = reshape.before.data_disks + reshape.parity;
3183 d = reshape_prepare_fdlist(devname, sra, odisks,
3184 nrdisks, blocks, backup_file,
3185 fdlist, offsets);
3186 if (d < odisks) {
3187 goto release;
3188 }
3189 if ((st->ss->manage_reshape == NULL) ||
3190 (st->ss->recover_backup == NULL)) {
3191 if (backup_file == NULL) {
3192 if (reshape.after.data_disks <=
3193 reshape.before.data_disks) {
3194 pr_err("%s: Cannot grow - need backup-file\n",
3195 devname);
3196 pr_err(" Please provide one with \"--backup=...\"\n");
3197 goto release;
3198 } else if (d == odisks) {
3199 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3200 goto release;
3201 }
3202 } else {
3203 if (!reshape_open_backup_file(backup_file, fd, devname,
3204 (signed)blocks,
3205 fdlist+d, offsets+d,
3206 sra->sys_name,
3207 restart)) {
3208 goto release;
3209 }
3210 d++;
3211 }
3212 }
3213
3214 update_cache_size(container, sra, info,
3215 min(reshape.before.data_disks, reshape.after.data_disks),
3216 blocks);
3217
3218 /* Right, everything seems fine. Let's kick things off.
3219 * If only changing raid_disks, use ioctl, else use
3220 * sysfs.
3221 */
3222 sync_metadata(st);
3223
3224 if (impose_reshape(sra, info, st, fd, restart,
3225 devname, container, &reshape) < 0)
3226 goto release;
3227
3228 err = start_reshape(sra, restart, reshape.before.data_disks,
3229 reshape.after.data_disks);
3230 if (err) {
3231 pr_err("Cannot %s reshape for %s\n",
3232 restart ? "continue" : "start",
3233 devname);
3234 goto release;
3235 }
3236 if (restart)
3237 sysfs_set_str(sra, NULL, "array_state", "active");
3238 if (freeze_reshape) {
3239 free(fdlist);
3240 free(offsets);
3241 sysfs_free(sra);
3242 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3243 sra->reshape_progress);
3244 return 1;
3245 }
3246
3247 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3248 if (continue_via_systemd(container ?: sra->sys_name)) {
3249 free(fdlist);
3250 free(offsets);
3251 sysfs_free(sra);
3252 return 0;
3253 }
3254
3255 /* Now we just need to kick off the reshape and watch, while
3256 * handling backups of the data...
3257 * This is all done by a forked background process.
3258 */
3259 switch(forked ? 0 : fork()) {
3260 case -1:
3261 pr_err("Cannot run child to monitor reshape: %s\n",
3262 strerror(errno));
3263 abort_reshape(sra);
3264 goto release;
3265 default:
3266 free(fdlist);
3267 free(offsets);
3268 sysfs_free(sra);
3269 return 0;
3270 case 0:
3271 map_fork();
3272 break;
3273 }
3274
3275 /* If another array on the same devices is busy, the
3276 * reshape will wait for them. This would mean that
3277 * the first section that we suspend will stay suspended
3278 * for a long time. So check on that possibility
3279 * by looking for "DELAYED" in /proc/mdstat, and if found,
3280 * wait a while
3281 */
3282 do {
3283 struct mdstat_ent *mds, *m;
3284 delayed = 0;
3285 mds = mdstat_read(1, 0);
3286 for (m = mds; m; m = m->next)
3287 if (strcmp(m->devnm, sra->sys_name) == 0) {
3288 if (m->resync &&
3289 m->percent == RESYNC_DELAYED)
3290 delayed = 1;
3291 if (m->resync == 0)
3292 /* Haven't started the reshape thread
3293 * yet, wait a bit
3294 */
3295 delayed = 2;
3296 break;
3297 }
3298 free_mdstat(mds);
3299 if (delayed == 1 && get_linux_version() < 3007000) {
3300 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3301 " You might experience problems until other reshapes complete.\n");
3302 delayed = 0;
3303 }
3304 if (delayed)
3305 mdstat_wait(30 - (delayed-1) * 25);
3306 } while (delayed);
3307 mdstat_close();
3308 close(fd);
3309 if (check_env("MDADM_GROW_VERIFY"))
3310 fd = open(devname, O_RDONLY | O_DIRECT);
3311 else
3312 fd = -1;
3313 mlockall(MCL_FUTURE);
3314
3315 signal(SIGTERM, catch_term);
3316
3317 if (st->ss->external) {
3318 /* metadata handler takes it from here */
3319 done = st->ss->manage_reshape(
3320 fd, sra, &reshape, st, blocks,
3321 fdlist, offsets,
3322 d - odisks, fdlist+odisks,
3323 offsets+odisks);
3324 } else
3325 done = child_monitor(
3326 fd, sra, &reshape, st, blocks,
3327 fdlist, offsets,
3328 d - odisks, fdlist+odisks,
3329 offsets+odisks);
3330
3331 free(fdlist);
3332 free(offsets);
3333
3334 if (backup_file && done) {
3335 char *bul;
3336 bul = make_backup(sra->sys_name);
3337 if (bul) {
3338 char buf[1024];
3339 int l = readlink(bul, buf, sizeof(buf) - 1);
3340 if (l > 0) {
3341 buf[l]=0;
3342 unlink(buf);
3343 }
3344 unlink(bul);
3345 free(bul);
3346 }
3347 unlink(backup_file);
3348 }
3349 if (!done) {
3350 abort_reshape(sra);
3351 goto out;
3352 }
3353
3354 if (!st->ss->external &&
3355 !(reshape.before.data_disks != reshape.after.data_disks
3356 && info->custom_array_size) &&
3357 info->new_level == reshape.level &&
3358 !forked) {
3359 /* no need to wait for the reshape to finish as
3360 * there is nothing more to do.
3361 */
3362 sysfs_free(sra);
3363 exit(0);
3364 }
3365 wait_reshape(sra);
3366
3367 if (st->ss->external) {
3368 /* Re-load the metadata as much could have changed */
3369 int cfd = open_dev(st->container_devnm);
3370 if (cfd >= 0) {
3371 flush_mdmon(container);
3372 st->ss->free_super(st);
3373 st->ss->load_container(st, cfd, container);
3374 close(cfd);
3375 }
3376 }
3377
3378 /* set new array size if required customer_array_size is used
3379 * by this metadata.
3380 */
3381 if (reshape.before.data_disks !=
3382 reshape.after.data_disks &&
3383 info->custom_array_size)
3384 set_array_size(st, info, info->text_version);
3385
3386 if (info->new_level != reshape.level) {
3387 if (fd < 0)
3388 fd = open(devname, O_RDONLY);
3389 impose_level(fd, info->new_level, devname, verbose);
3390 close(fd);
3391 if (info->new_level == 0)
3392 st->update_tail = NULL;
3393 }
3394 out:
3395 sysfs_free(sra);
3396 if (forked)
3397 return 0;
3398 unfreeze(st);
3399 exit(0);
3400
3401 release:
3402 free(fdlist);
3403 free(offsets);
3404 if (orig_level != UnSet && sra) {
3405 c = map_num(pers, orig_level);
3406 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3407 pr_err("aborting level change\n");
3408 }
3409 sysfs_free(sra);
3410 if (!forked)
3411 unfreeze(st);
3412 return 1;
3413 }
3414
3415 /* mdfd handle is passed to be closed in child process (after fork).
3416 */
3417 int reshape_container(char *container, char *devname,
3418 int mdfd,
3419 struct supertype *st,
3420 struct mdinfo *info,
3421 int force,
3422 char *backup_file, int verbose,
3423 int forked, int restart, int freeze_reshape)
3424 {
3425 struct mdinfo *cc = NULL;
3426 int rv = restart;
3427 char last_devnm[32] = "";
3428
3429 /* component_size is not meaningful for a container,
3430 * so pass '0' meaning 'no change'
3431 */
3432 if (!restart &&
3433 reshape_super(st, 0, info->new_level,
3434 info->new_layout, info->new_chunk,
3435 info->array.raid_disks, info->delta_disks,
3436 backup_file, devname, APPLY_METADATA_CHANGES,
3437 verbose)) {
3438 unfreeze(st);
3439 return 1;
3440 }
3441
3442 sync_metadata(st);
3443
3444 /* ping monitor to be sure that update is on disk
3445 */
3446 ping_monitor(container);
3447
3448 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3449 if (continue_via_systemd(container))
3450 return 0;
3451
3452 switch (forked ? 0 : fork()) {
3453 case -1: /* error */
3454 perror("Cannot fork to complete reshape\n");
3455 unfreeze(st);
3456 return 1;
3457 default: /* parent */
3458 if (!freeze_reshape)
3459 printf("%s: multi-array reshape continues in background\n", Name);
3460 return 0;
3461 case 0: /* child */
3462 map_fork();
3463 break;
3464 }
3465
3466 /* close unused handle in child process
3467 */
3468 if (mdfd > -1)
3469 close(mdfd);
3470
3471 while(1) {
3472 /* For each member array with reshape_active,
3473 * we need to perform the reshape.
3474 * We pick the first array that needs reshaping and
3475 * reshape it. reshape_array() will re-read the metadata
3476 * so the next time through a different array should be
3477 * ready for reshape.
3478 * It is possible that the 'different' array will not
3479 * be assembled yet. In that case we simple exit.
3480 * When it is assembled, the mdadm which assembles it
3481 * will take over the reshape.
3482 */
3483 struct mdinfo *content;
3484 int fd;
3485 struct mdstat_ent *mdstat;
3486 char *adev;
3487 int devid;
3488
3489 sysfs_free(cc);
3490
3491 cc = st->ss->container_content(st, NULL);
3492
3493 for (content = cc; content ; content = content->next) {
3494 char *subarray;
3495 if (!content->reshape_active)
3496 continue;
3497
3498 subarray = strchr(content->text_version+1, '/')+1;
3499 mdstat = mdstat_by_subdev(subarray, container);
3500 if (!mdstat)
3501 continue;
3502 if (mdstat->active == 0) {
3503 pr_err("Skipping inactive array %s.\n",
3504 mdstat->devnm);
3505 free_mdstat(mdstat);
3506 mdstat = NULL;
3507 continue;
3508 }
3509 break;
3510 }
3511 if (!content)
3512 break;
3513
3514 devid = devnm2devid(mdstat->devnm);
3515 adev = map_dev(major(devid), minor(devid), 0);
3516 if (!adev)
3517 adev = content->text_version;
3518
3519 fd = open_dev(mdstat->devnm);
3520 if (fd < 0) {
3521 pr_err("Device %s cannot be opened for reshape.\n", adev);
3522 break;
3523 }
3524
3525 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3526 /* Do not allow for multiple reshape_array() calls for
3527 * the same array.
3528 * It can happen when reshape_array() returns without
3529 * error, when reshape is not finished (wrong reshape
3530 * starting/continuation conditions). Mdmon doesn't
3531 * switch to next array in container and reentry
3532 * conditions for the same array occur.
3533 * This is possibly interim until the behaviour of
3534 * reshape_array is resolved().
3535 */
3536 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3537 close(fd);
3538 break;
3539 }
3540 strcpy(last_devnm, mdstat->devnm);
3541
3542 sysfs_init(content, fd, mdstat->devnm);
3543
3544 if (mdmon_running(container))
3545 flush_mdmon(container);
3546
3547 rv = reshape_array(container, fd, adev, st,
3548 content, force, NULL, INVALID_SECTORS,
3549 backup_file, verbose, 1, restart,
3550 freeze_reshape);
3551 close(fd);
3552
3553 if (freeze_reshape) {
3554 sysfs_free(cc);
3555 exit(0);
3556 }
3557
3558 restart = 0;
3559 if (rv)
3560 break;
3561
3562 if (mdmon_running(container))
3563 flush_mdmon(container);
3564 }
3565 if (!rv)
3566 unfreeze(st);
3567 sysfs_free(cc);
3568 exit(0);
3569 }
3570
3571 /*
3572 * We run a child process in the background which performs the following
3573 * steps:
3574 * - wait for resync to reach a certain point
3575 * - suspend io to the following section
3576 * - backup that section
3577 * - allow resync to proceed further
3578 * - resume io
3579 * - discard the backup.
3580 *
3581 * When are combined in slightly different ways in the three cases.
3582 * Grow:
3583 * - suspend/backup/allow/wait/resume/discard
3584 * Shrink:
3585 * - allow/wait/suspend/backup/allow/wait/resume/discard
3586 * same-size:
3587 * - wait/resume/discard/suspend/backup/allow
3588 *
3589 * suspend/backup/allow always come together
3590 * wait/resume/discard do too.
3591 * For the same-size case we have two backups to improve flow.
3592 *
3593 */
3594
3595 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3596 unsigned long long backup_point,
3597 unsigned long long wait_point,
3598 unsigned long long *suspend_point,
3599 unsigned long long *reshape_completed, int *frozen)
3600 {
3601 /* This function is called repeatedly by the reshape manager.
3602 * It determines how much progress can safely be made and allows
3603 * that progress.
3604 * - 'info' identifies the array and particularly records in
3605 * ->reshape_progress the metadata's knowledge of progress
3606 * This is a sector offset from the start of the array
3607 * of the next array block to be relocated. This number
3608 * may increase from 0 or decrease from array_size, depending
3609 * on the type of reshape that is happening.
3610 * Note that in contrast, 'sync_completed' is a block count of the
3611 * reshape so far. It gives the distance between the start point
3612 * (head or tail of device) and the next place that data will be
3613 * written. It always increases.
3614 * - 'reshape' is the structure created by analyse_change
3615 * - 'backup_point' shows how much the metadata manager has backed-up
3616 * data. For reshapes with increasing progress, it is the next address
3617 * to be backed up, previous addresses have been backed-up. For
3618 * decreasing progress, it is the earliest address that has been
3619 * backed up - later address are also backed up.
3620 * So addresses between reshape_progress and backup_point are
3621 * backed up providing those are in the 'correct' order.
3622 * - 'wait_point' is an array address. When reshape_completed
3623 * passes this point, progress_reshape should return. It might
3624 * return earlier if it determines that ->reshape_progress needs
3625 * to be updated or further backup is needed.
3626 * - suspend_point is maintained by progress_reshape and the caller
3627 * should not touch it except to initialise to zero.
3628 * It is an array address and it only increases in 2.6.37 and earlier.
3629 * This makes it difficult to handle reducing reshapes with
3630 * external metadata.
3631 * However: it is similar to backup_point in that it records the
3632 * other end of a suspended region from reshape_progress.
3633 * it is moved to extend the region that is safe to backup and/or
3634 * reshape
3635 * - reshape_completed is read from sysfs and returned. The caller
3636 * should copy this into ->reshape_progress when it has reason to
3637 * believe that the metadata knows this, and any backup outside this
3638 * has been erased.
3639 *
3640 * Return value is:
3641 * 1 if more data from backup_point - but only as far as suspend_point,
3642 * should be backed up
3643 * 0 if things are progressing smoothly
3644 * -1 if the reshape is finished because it is all done,
3645 * -2 if the reshape is finished due to an error.
3646 */
3647
3648 int advancing = (reshape->after.data_disks
3649 >= reshape->before.data_disks);
3650 unsigned long long need_backup; /* All data between start of array and
3651 * here will at some point need to
3652 * be backed up.
3653 */
3654 unsigned long long read_offset, write_offset;
3655 unsigned long long write_range;
3656 unsigned long long max_progress, target, completed;
3657 unsigned long long array_size = (info->component_size
3658 * reshape->before.data_disks);
3659 int fd;
3660 char buf[20];
3661
3662 /* First, we unsuspend any region that is now known to be safe.
3663 * If suspend_point is on the 'wrong' side of reshape_progress, then
3664 * we don't have or need suspension at the moment. This is true for
3665 * native metadata when we don't need to back-up.
3666 */
3667 if (advancing) {
3668 if (info->reshape_progress <= *suspend_point)
3669 sysfs_set_num(info, NULL, "suspend_lo",
3670 info->reshape_progress);
3671 } else {
3672 /* Note: this won't work in 2.6.37 and before.
3673 * Something somewhere should make sure we don't need it!
3674 */
3675 if (info->reshape_progress >= *suspend_point)
3676 sysfs_set_num(info, NULL, "suspend_hi",
3677 info->reshape_progress);
3678 }
3679
3680 /* Now work out how far it is safe to progress.
3681 * If the read_offset for ->reshape_progress is less than
3682 * 'blocks' beyond the write_offset, we can only progress as far
3683 * as a backup.
3684 * Otherwise we can progress until the write_offset for the new location
3685 * reaches (within 'blocks' of) the read_offset at the current location.
3686 * However that region must be suspended unless we are using native
3687 * metadata.
3688 * If we need to suspend more, we limit it to 128M per device, which is
3689 * rather arbitrary and should be some time-based calculation.
3690 */
3691 read_offset = info->reshape_progress / reshape->before.data_disks;
3692 write_offset = info->reshape_progress / reshape->after.data_disks;
3693 write_range = info->new_chunk/512;
3694 if (reshape->before.data_disks == reshape->after.data_disks)
3695 need_backup = array_size;
3696 else
3697 need_backup = reshape->backup_blocks;
3698 if (advancing) {
3699 if (read_offset < write_offset + write_range)
3700 max_progress = backup_point;
3701 else
3702 max_progress =
3703 read_offset *
3704 reshape->after.data_disks;
3705 } else {
3706 if (read_offset > write_offset - write_range)
3707 /* Can only progress as far as has been backed up,
3708 * which must be suspended */
3709 max_progress = backup_point;
3710 else if (info->reshape_progress <= need_backup)
3711 max_progress = backup_point;
3712 else {
3713 if (info->array.major_version >= 0)
3714 /* Can progress until backup is needed */
3715 max_progress = need_backup;
3716 else {
3717 /* Can progress until metadata update is required */
3718 max_progress =
3719 read_offset *
3720 reshape->after.data_disks;
3721 /* but data must be suspended */
3722 if (max_progress < *suspend_point)
3723 max_progress = *suspend_point;
3724 }
3725 }
3726 }
3727
3728 /* We know it is safe to progress to 'max_progress' providing
3729 * it is suspended or we are using native metadata.
3730 * Consider extending suspend_point 128M per device if it
3731 * is less than 64M per device beyond reshape_progress.
3732 * But always do a multiple of 'blocks'
3733 * FIXME this is too big - it takes to long to complete
3734 * this much.
3735 */
3736 target = 64*1024*2 * min(reshape->before.data_disks,
3737 reshape->after.data_disks);
3738 target /= reshape->backup_blocks;
3739 if (target < 2)
3740 target = 2;
3741 target *= reshape->backup_blocks;
3742
3743 /* For externally managed metadata we always need to suspend IO to
3744 * the area being reshaped so we regularly push suspend_point forward.
3745 * For native metadata we only need the suspend if we are going to do
3746 * a backup.
3747 */
3748 if (advancing) {
3749 if ((need_backup > info->reshape_progress
3750 || info->array.major_version < 0) &&
3751 *suspend_point < info->reshape_progress + target) {
3752 if (need_backup < *suspend_point + 2 * target)
3753 *suspend_point = need_backup;
3754 else if (*suspend_point + 2 * target < array_size)
3755 *suspend_point += 2 * target;
3756 else
3757 *suspend_point = array_size;
3758 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3759 if (max_progress > *suspend_point)
3760 max_progress = *suspend_point;
3761 }
3762 } else {
3763 if (info->array.major_version >= 0) {
3764 /* Only need to suspend when about to backup */
3765 if (info->reshape_progress < need_backup * 2 &&
3766 *suspend_point > 0) {
3767 *suspend_point = 0;
3768 sysfs_set_num(info, NULL, "suspend_lo", 0);
3769 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3770 }
3771 } else {
3772 /* Need to suspend continually */
3773 if (info->reshape_progress < *suspend_point)
3774 *suspend_point = info->reshape_progress;
3775 if (*suspend_point + target < info->reshape_progress)
3776 /* No need to move suspend region yet */;
3777 else {
3778 if (*suspend_point >= 2 * target)
3779 *suspend_point -= 2 * target;
3780 else
3781 *suspend_point = 0;
3782 sysfs_set_num(info, NULL, "suspend_lo",
3783 *suspend_point);
3784 }
3785 if (max_progress < *suspend_point)
3786 max_progress = *suspend_point;
3787 }
3788 }
3789
3790 /* now set sync_max to allow that progress. sync_max, like
3791 * sync_completed is a count of sectors written per device, so
3792 * we find the difference between max_progress and the start point,
3793 * and divide that by after.data_disks to get a sync_max
3794 * number.
3795 * At the same time we convert wait_point to a similar number
3796 * for comparing against sync_completed.
3797 */
3798 /* scale down max_progress to per_disk */
3799 max_progress /= reshape->after.data_disks;
3800 /* Round to chunk size as some kernels give an erroneously high number */
3801 max_progress /= info->new_chunk/512;
3802 max_progress *= info->new_chunk/512;
3803 /* And round to old chunk size as the kernel wants that */
3804 max_progress /= info->array.chunk_size/512;
3805 max_progress *= info->array.chunk_size/512;
3806 /* Limit progress to the whole device */
3807 if (max_progress > info->component_size)
3808 max_progress = info->component_size;
3809 wait_point /= reshape->after.data_disks;
3810 if (!advancing) {
3811 /* switch from 'device offset' to 'processed block count' */
3812 max_progress = info->component_size - max_progress;
3813 wait_point = info->component_size - wait_point;
3814 }
3815
3816 if (!*frozen)
3817 sysfs_set_num(info, NULL, "sync_max", max_progress);
3818
3819 /* Now wait. If we have already reached the point that we were
3820 * asked to wait to, don't wait at all, else wait for any change.
3821 * We need to select on 'sync_completed' as that is the place that
3822 * notifications happen, but we are really interested in
3823 * 'reshape_position'
3824 */
3825 fd = sysfs_get_fd(info, NULL, "sync_completed");
3826 if (fd < 0)
3827 goto check_progress;
3828
3829 if (sysfs_fd_get_ll(fd, &completed) < 0)
3830 goto check_progress;
3831
3832 while (completed < max_progress && completed < wait_point) {
3833 /* Check that sync_action is still 'reshape' to avoid
3834 * waiting forever on a dead array
3835 */
3836 char action[20];
3837 if (sysfs_get_str(info, NULL, "sync_action",
3838 action, 20) <= 0 ||
3839 strncmp(action, "reshape", 7) != 0)
3840 break;
3841 /* Some kernels reset 'sync_completed' to zero
3842 * before setting 'sync_action' to 'idle'.
3843 * So we need these extra tests.
3844 */
3845 if (completed == 0 && advancing
3846 && strncmp(action, "idle", 4) == 0
3847 && info->reshape_progress > 0)
3848 break;
3849 if (completed == 0 && !advancing
3850 && strncmp(action, "idle", 4) == 0
3851 && info->reshape_progress < (info->component_size
3852 * reshape->after.data_disks))
3853 break;
3854 sysfs_wait(fd, NULL);
3855 if (sysfs_fd_get_ll(fd, &completed) < 0)
3856 goto check_progress;
3857 }
3858 /* Some kernels reset 'sync_completed' to zero,
3859 * we need to have real point we are in md.
3860 * So in that case, read 'reshape_position' from sysfs.
3861 */
3862 if (completed == 0) {
3863 unsigned long long reshapep;
3864 char action[20];
3865 if (sysfs_get_str(info, NULL, "sync_action",
3866 action, 20) > 0 &&
3867 strncmp(action, "idle", 4) == 0 &&
3868 sysfs_get_ll(info, NULL,
3869 "reshape_position", &reshapep) == 0)
3870 *reshape_completed = reshapep;
3871 } else {
3872 /* some kernels can give an incorrectly high
3873 * 'completed' number, so round down */
3874 completed /= (info->new_chunk/512);
3875 completed *= (info->new_chunk/512);
3876 /* Convert 'completed' back in to a 'progress' number */
3877 completed *= reshape->after.data_disks;
3878 if (!advancing)
3879 completed = (info->component_size
3880 * reshape->after.data_disks
3881 - completed);
3882 *reshape_completed = completed;
3883 }
3884
3885 close(fd);
3886
3887 /* We return the need_backup flag. Caller will decide
3888 * how much - a multiple of ->backup_blocks up to *suspend_point
3889 */
3890 if (advancing)
3891 return need_backup > info->reshape_progress;
3892 else
3893 return need_backup >= info->reshape_progress;
3894
3895 check_progress:
3896 /* if we couldn't read a number from sync_completed, then
3897 * either the reshape did complete, or it aborted.
3898 * We can tell which by checking for 'none' in reshape_position.
3899 * If it did abort, then it might immediately restart if it
3900 * it was just a device failure that leaves us degraded but
3901 * functioning.
3902 */
3903 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3904 || strncmp(buf, "none", 4) != 0) {
3905 /* The abort might only be temporary. Wait up to 10
3906 * seconds for fd to contain a valid number again.
3907 */
3908 int wait = 10000;
3909 int rv = -2;
3910 unsigned long long new_sync_max;
3911 while (fd >= 0 && rv < 0 && wait > 0) {
3912 if (sysfs_wait(fd, &wait) != 1)
3913 break;
3914 switch (sysfs_fd_get_ll(fd, &completed)) {
3915 case 0:
3916 /* all good again */
3917 rv = 1;
3918 /* If "sync_max" is no longer max_progress
3919 * we need to freeze things
3920 */
3921 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3922 *frozen = (new_sync_max != max_progress);
3923 break;
3924 case -2: /* read error - abort */
3925 wait = 0;
3926 break;
3927 }
3928 }
3929 if (fd >= 0)
3930 close(fd);
3931 return rv; /* abort */
3932 } else {
3933 /* Maybe racing with array shutdown - check state */
3934 if (fd >= 0)
3935 close(fd);
3936 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3937 || strncmp(buf, "inactive", 8) == 0
3938 || strncmp(buf, "clear",5) == 0)
3939 return -2; /* abort */
3940 return -1; /* complete */
3941 }
3942 }
3943
3944 /* FIXME return status is never checked */
3945 static int grow_backup(struct mdinfo *sra,
3946 unsigned long long offset, /* per device */
3947 unsigned long stripes, /* per device, in old chunks */
3948 int *sources, unsigned long long *offsets,
3949 int disks, int chunk, int level, int layout,
3950 int dests, int *destfd, unsigned long long *destoffsets,
3951 int part, int *degraded,
3952 char *buf)
3953 {
3954 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3955 * to storage 'destfd' (offset 'destoffsets'), after first
3956 * suspending IO. Then allow resync to continue
3957 * over the suspended section.
3958 * Use part 'part' of the backup-super-block.
3959 */
3960 int odata = disks;
3961 int rv = 0;
3962 int i;
3963 unsigned long long ll;
3964 int new_degraded;
3965 //printf("offset %llu\n", offset);
3966 if (level >= 4)
3967 odata--;
3968 if (level == 6)
3969 odata--;
3970
3971 /* Check that array hasn't become degraded, else we might backup the wrong data */
3972 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
3973 return -1; /* FIXME this error is ignored */
3974 new_degraded = (int)ll;
3975 if (new_degraded != *degraded) {
3976 /* check each device to ensure it is still working */
3977 struct mdinfo *sd;
3978 for (sd = sra->devs ; sd ; sd = sd->next) {
3979 if (sd->disk.state & (1<<MD_DISK_FAULTY))
3980 continue;
3981 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
3982 char sbuf[20];
3983 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
3984 strstr(sbuf, "faulty") ||
3985 strstr(sbuf, "in_sync") == NULL) {
3986 /* this device is dead */
3987 sd->disk.state = (1<<MD_DISK_FAULTY);
3988 if (sd->disk.raid_disk >= 0 &&
3989 sources[sd->disk.raid_disk] >= 0) {
3990 close(sources[sd->disk.raid_disk]);
3991 sources[sd->disk.raid_disk] = -1;
3992 }
3993 }
3994 }
3995 }
3996 *degraded = new_degraded;
3997 }
3998 if (part) {
3999 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4000 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4001 } else {
4002 bsb.arraystart = __cpu_to_le64(offset * odata);
4003 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4004 }
4005 if (part)
4006 bsb.magic[15] = '2';
4007 for (i = 0; i < dests; i++)
4008 if (part)
4009 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4010 else
4011 lseek64(destfd[i], destoffsets[i], 0);
4012
4013 rv = save_stripes(sources, offsets,
4014 disks, chunk, level, layout,
4015 dests, destfd,
4016 offset*512*odata, stripes * chunk * odata,
4017 buf);
4018
4019 if (rv)
4020 return rv;
4021 bsb.mtime = __cpu_to_le64(time(0));
4022 for (i = 0; i < dests; i++) {
4023 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4024
4025 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4026 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4027 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4028 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4029
4030 rv = -1;
4031 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4032 != destoffsets[i] - 4096)
4033 break;
4034 if (write(destfd[i], &bsb, 512) != 512)
4035 break;
4036 if (destoffsets[i] > 4096) {
4037 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4038 destoffsets[i]+stripes*chunk*odata)
4039 break;
4040 if (write(destfd[i], &bsb, 512) != 512)
4041 break;
4042 }
4043 fsync(destfd[i]);
4044 rv = 0;
4045 }
4046
4047 return rv;
4048 }
4049
4050 /* in 2.6.30, the value reported by sync_completed can be
4051 * less that it should be by one stripe.
4052 * This only happens when reshape hits sync_max and pauses.
4053 * So allow wait_backup to either extent sync_max further
4054 * than strictly necessary, or return before the
4055 * sync has got quite as far as we would really like.
4056 * This is what 'blocks2' is for.
4057 * The various caller give appropriate values so that
4058 * every works.
4059 */
4060 /* FIXME return value is often ignored */
4061 static int forget_backup(int dests, int *destfd,
4062 unsigned long long *destoffsets,
4063 int part)
4064 {
4065 /*
4066 * Erase backup 'part' (which is 0 or 1)
4067 */
4068 int i;
4069 int rv;
4070
4071 if (part) {
4072 bsb.arraystart2 = __cpu_to_le64(0);
4073 bsb.length2 = __cpu_to_le64(0);
4074 } else {
4075 bsb.arraystart = __cpu_to_le64(0);
4076 bsb.length = __cpu_to_le64(0);
4077 }
4078 bsb.mtime = __cpu_to_le64(time(0));
4079 rv = 0;
4080 for (i = 0; i < dests; i++) {
4081 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4082 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4083 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4084 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4085 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4086 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4087 destoffsets[i]-4096)
4088 rv = -1;
4089 if (rv == 0 &&
4090 write(destfd[i], &bsb, 512) != 512)
4091 rv = -1;
4092 fsync(destfd[i]);
4093 }
4094 return rv;
4095 }
4096
4097 static void fail(char *msg)
4098 {
4099 int rv;
4100 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4101 rv |= (write(2, "\n", 1) != 1);
4102 exit(rv ? 1 : 2);
4103 }
4104
4105 static char *abuf, *bbuf;
4106 static unsigned long long abuflen;
4107 static void validate(int afd, int bfd, unsigned long long offset)
4108 {
4109 /* check that the data in the backup against the array.
4110 * This is only used for regression testing and should not
4111 * be used while the array is active
4112 */
4113 if (afd < 0)
4114 return;
4115 lseek64(bfd, offset - 4096, 0);
4116 if (read(bfd, &bsb2, 512) != 512)
4117 fail("cannot read bsb");
4118 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4119 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4120 fail("first csum bad");
4121 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4122 fail("magic is bad");
4123 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4124 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4125 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4126 fail("second csum bad");
4127
4128 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4129 fail("devstart is wrong");
4130
4131 if (bsb2.length) {
4132 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4133
4134 if (abuflen < len) {
4135 free(abuf);
4136 free(bbuf);
4137 abuflen = len;
4138 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4139 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4140 abuflen = 0;
4141 /* just stop validating on mem-alloc failure */
4142 return;
4143 }
4144 }
4145
4146 lseek64(bfd, offset, 0);
4147 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4148 //printf("len %llu\n", len);
4149 fail("read first backup failed");
4150 }
4151 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4152 if ((unsigned long long)read(afd, abuf, len) != len)
4153 fail("read first from array failed");
4154 if (memcmp(bbuf, abuf, len) != 0) {
4155 #if 0
4156 int i;
4157 printf("offset=%llu len=%llu\n",
4158 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4159 for (i=0; i<len; i++)
4160 if (bbuf[i] != abuf[i]) {
4161 printf("first diff byte %d\n", i);
4162 break;
4163 }
4164 #endif
4165 fail("data1 compare failed");
4166 }
4167 }
4168 if (bsb2.length2) {
4169 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4170
4171 if (abuflen < len) {
4172 free(abuf);
4173 free(bbuf);
4174 abuflen = len;
4175 abuf = xmalloc(abuflen);
4176 bbuf = xmalloc(abuflen);
4177 }
4178
4179 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4180 if ((unsigned long long)read(bfd, bbuf, len) != len)
4181 fail("read second backup failed");
4182 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4183 if ((unsigned long long)read(afd, abuf, len) != len)
4184 fail("read second from array failed");
4185 if (memcmp(bbuf, abuf, len) != 0)
4186 fail("data2 compare failed");
4187 }
4188 }
4189
4190 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4191 struct supertype *st, unsigned long blocks,
4192 int *fds, unsigned long long *offsets,
4193 int dests, int *destfd, unsigned long long *destoffsets)
4194 {
4195 /* Monitor a reshape where backup is being performed using
4196 * 'native' mechanism - either to a backup file, or
4197 * to some space in a spare.
4198 */
4199 char *buf;
4200 int degraded = -1;
4201 unsigned long long speed;
4202 unsigned long long suspend_point, array_size;
4203 unsigned long long backup_point, wait_point;
4204 unsigned long long reshape_completed;
4205 int done = 0;
4206 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4207 int part = 0; /* The next part of the backup area to fill. It may already
4208 * be full, so we need to check */
4209 int level = reshape->level;
4210 int layout = reshape->before.layout;
4211 int data = reshape->before.data_disks;
4212 int disks = reshape->before.data_disks + reshape->parity;
4213 int chunk = sra->array.chunk_size;
4214 struct mdinfo *sd;
4215 unsigned long stripes;
4216 int uuid[4];
4217 int frozen = 0;
4218
4219 /* set up the backup-super-block. This requires the
4220 * uuid from the array.
4221 */
4222 /* Find a superblock */
4223 for (sd = sra->devs; sd; sd = sd->next) {
4224 char *dn;
4225 int devfd;
4226 int ok;
4227 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4228 continue;
4229 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4230 devfd = dev_open(dn, O_RDONLY);
4231 if (devfd < 0)
4232 continue;
4233 ok = st->ss->load_super(st, devfd, NULL);
4234 close(devfd);
4235 if (ok == 0)
4236 break;
4237 }
4238 if (!sd) {
4239 pr_err("Cannot find a superblock\n");
4240 return 0;
4241 }
4242
4243 memset(&bsb, 0, 512);
4244 memcpy(bsb.magic, "md_backup_data-1", 16);
4245 st->ss->uuid_from_super(st, uuid);
4246 memcpy(bsb.set_uuid, uuid, 16);
4247 bsb.mtime = __cpu_to_le64(time(0));
4248 bsb.devstart2 = blocks;
4249
4250 stripes = blocks / (sra->array.chunk_size/512) /
4251 reshape->before.data_disks;
4252
4253 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4254 /* Don't start the 'reshape' */
4255 return 0;
4256 if (reshape->before.data_disks == reshape->after.data_disks) {
4257 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4258 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4259 }
4260
4261 if (increasing) {
4262 array_size = sra->component_size * reshape->after.data_disks;
4263 backup_point = sra->reshape_progress;
4264 suspend_point = 0;
4265 } else {
4266 array_size = sra->component_size * reshape->before.data_disks;
4267 backup_point = reshape->backup_blocks;
4268 suspend_point = array_size;
4269 }
4270
4271 while (!done) {
4272 int rv;
4273
4274 /* Want to return as soon the oldest backup slot can
4275 * be released as that allows us to start backing up
4276 * some more, providing suspend_point has been
4277 * advanced, which it should have.
4278 */
4279 if (increasing) {
4280 wait_point = array_size;
4281 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4282 wait_point = (__le64_to_cpu(bsb.arraystart) +
4283 __le64_to_cpu(bsb.length));
4284 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4285 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4286 __le64_to_cpu(bsb.length2));
4287 } else {
4288 wait_point = 0;
4289 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4290 wait_point = __le64_to_cpu(bsb.arraystart);
4291 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4292 wait_point = __le64_to_cpu(bsb.arraystart2);
4293 }
4294
4295 reshape_completed = sra->reshape_progress;
4296 rv = progress_reshape(sra, reshape,
4297 backup_point, wait_point,
4298 &suspend_point, &reshape_completed,
4299 &frozen);
4300 /* external metadata would need to ping_monitor here */
4301 sra->reshape_progress = reshape_completed;
4302
4303 /* Clear any backup region that is before 'here' */
4304 if (increasing) {
4305 if (__le64_to_cpu(bsb.length) > 0 &&
4306 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4307 __le64_to_cpu(bsb.length)))
4308 forget_backup(dests, destfd,
4309 destoffsets, 0);
4310 if (__le64_to_cpu(bsb.length2) > 0 &&
4311 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4312 __le64_to_cpu(bsb.length2)))
4313 forget_backup(dests, destfd,
4314 destoffsets, 1);
4315 } else {
4316 if (__le64_to_cpu(bsb.length) > 0 &&
4317 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4318 forget_backup(dests, destfd,
4319 destoffsets, 0);
4320 if (__le64_to_cpu(bsb.length2) > 0 &&
4321 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4322 forget_backup(dests, destfd,
4323 destoffsets, 1);
4324 }
4325 if (sigterm)
4326 rv = -2;
4327 if (rv < 0) {
4328 if (rv == -1)
4329 done = 1;
4330 break;
4331 }
4332 if (rv == 0 && increasing && !st->ss->external) {
4333 /* No longer need to monitor this reshape */
4334 sysfs_set_str(sra, NULL, "sync_max", "max");
4335 done = 1;
4336 break;
4337 }
4338
4339 while (rv) {
4340 unsigned long long offset;
4341 unsigned long actual_stripes;
4342 /* Need to backup some data.
4343 * If 'part' is not used and the desired
4344 * backup size is suspended, do a backup,
4345 * then consider the next part.
4346 */
4347 /* Check that 'part' is unused */
4348 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4349 break;
4350 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4351 break;
4352
4353 offset = backup_point / data;
4354 actual_stripes = stripes;
4355 if (increasing) {
4356 if (offset + actual_stripes * (chunk/512) >
4357 sra->component_size)
4358 actual_stripes = ((sra->component_size - offset)
4359 / (chunk/512));
4360 if (offset + actual_stripes * (chunk/512) >
4361 suspend_point/data)
4362 break;
4363 } else {
4364 if (offset < actual_stripes * (chunk/512))
4365 actual_stripes = offset / (chunk/512);
4366 offset -= actual_stripes * (chunk/512);
4367 if (offset < suspend_point/data)
4368 break;
4369 }
4370 if (actual_stripes == 0)
4371 break;
4372 grow_backup(sra, offset, actual_stripes,
4373 fds, offsets,
4374 disks, chunk, level, layout,
4375 dests, destfd, destoffsets,
4376 part, &degraded, buf);
4377 validate(afd, destfd[0], destoffsets[0]);
4378 /* record where 'part' is up to */
4379 part = !part;
4380 if (increasing)
4381 backup_point += actual_stripes * (chunk/512) * data;
4382 else
4383 backup_point -= actual_stripes * (chunk/512) * data;
4384 }
4385 }
4386
4387 /* FIXME maybe call progress_reshape one more time instead */
4388 /* remove any remaining suspension */
4389 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4390 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4391 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4392 sysfs_set_num(sra, NULL, "sync_min", 0);
4393
4394 if (reshape->before.data_disks == reshape->after.data_disks)
4395 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4396 free(buf);
4397 return done;
4398 }
4399
4400 /*
4401 * If any spare contains md_back_data-1 which is recent wrt mtime,
4402 * write that data into the array and update the super blocks with
4403 * the new reshape_progress
4404 */
4405 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4406 char *backup_file, int verbose)
4407 {
4408 int i, j;
4409 int old_disks;
4410 unsigned long long *offsets;
4411 unsigned long long nstripe, ostripe;
4412 int ndata, odata;
4413
4414 odata = info->array.raid_disks - info->delta_disks - 1;
4415 if (info->array.level == 6) odata--; /* number of data disks */
4416 ndata = info->array.raid_disks - 1;
4417 if (info->new_level == 6) ndata--;
4418
4419 old_disks = info->array.raid_disks - info->delta_disks;
4420
4421 if (info->delta_disks <= 0)
4422 /* Didn't grow, so the backup file must have
4423 * been used
4424 */
4425 old_disks = cnt;
4426 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4427 struct mdinfo dinfo;
4428 int fd;
4429 int bsbsize;
4430 char *devname, namebuf[20];
4431 unsigned long long lo, hi;
4432
4433 /* This was a spare and may have some saved data on it.
4434 * Load the superblock, find and load the
4435 * backup_super_block.
4436 * If either fail, go on to next device.
4437 * If the backup contains no new info, just return
4438 * else restore data and update all superblocks
4439 */
4440 if (i == old_disks-1) {
4441 fd = open(backup_file, O_RDONLY);
4442 if (fd<0) {
4443 pr_err("backup file %s inaccessible: %s\n",
4444 backup_file, strerror(errno));
4445 continue;
4446 }
4447 devname = backup_file;
4448 } else {
4449 fd = fdlist[i];
4450 if (fd < 0)
4451 continue;
4452 if (st->ss->load_super(st, fd, NULL))
4453 continue;
4454
4455 st->ss->getinfo_super(st, &dinfo, NULL);
4456 st->ss->free_super(st);
4457
4458 if (lseek64(fd,
4459 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4460 0) < 0) {
4461 pr_err("Cannot seek on device %d\n", i);
4462 continue; /* Cannot seek */
4463 }
4464 sprintf(namebuf, "device-%d", i);
4465 devname = namebuf;
4466 }
4467 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4468 if (verbose)
4469 pr_err("Cannot read from %s\n", devname);
4470 continue; /* Cannot read */
4471 }
4472 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4473 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4474 if (verbose)
4475 pr_err("No backup metadata on %s\n", devname);
4476 continue;
4477 }
4478 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4479 if (verbose)
4480 pr_err("Bad backup-metadata checksum on %s\n", devname);
4481 continue; /* bad checksum */
4482 }
4483 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4484 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4485 if (verbose)
4486 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4487 continue; /* Bad second checksum */
4488 }
4489 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4490 if (verbose)
4491 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4492 continue; /* Wrong uuid */
4493 }
4494
4495 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4496 * sometimes they aren't... So allow considerable flexability in matching, and allow
4497 * this test to be overridden by an environment variable.
4498 */
4499 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4500 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4501 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4502 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4503 (unsigned long)__le64_to_cpu(bsb.mtime),
4504 (unsigned long)info->array.utime);
4505 } else {
4506 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4507 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4508 continue; /* time stamp is too bad */
4509 }
4510 }
4511
4512 if (bsb.magic[15] == '1') {
4513 if (bsb.length == 0)
4514 continue;
4515 if (info->delta_disks >= 0) {
4516 /* reshape_progress is increasing */
4517 if (__le64_to_cpu(bsb.arraystart)
4518 + __le64_to_cpu(bsb.length)
4519 < info->reshape_progress) {
4520 nonew:
4521 if (verbose)
4522 pr_err("backup-metadata found on %s but is not needed\n", devname);
4523 continue; /* No new data here */
4524 }
4525 } else {
4526 /* reshape_progress is decreasing */
4527 if (__le64_to_cpu(bsb.arraystart) >=
4528 info->reshape_progress)
4529 goto nonew; /* No new data here */
4530 }
4531 } else {
4532 if (bsb.length == 0 && bsb.length2 == 0)
4533 continue;
4534 if (info->delta_disks >= 0) {
4535 /* reshape_progress is increasing */
4536 if ((__le64_to_cpu(bsb.arraystart)
4537 + __le64_to_cpu(bsb.length)
4538 < info->reshape_progress)
4539 &&
4540 (__le64_to_cpu(bsb.arraystart2)
4541 + __le64_to_cpu(bsb.length2)
4542 < info->reshape_progress))
4543 goto nonew; /* No new data here */
4544 } else {
4545 /* reshape_progress is decreasing */
4546 if (__le64_to_cpu(bsb.arraystart) >=
4547 info->reshape_progress &&
4548 __le64_to_cpu(bsb.arraystart2) >=
4549 info->reshape_progress)
4550 goto nonew; /* No new data here */
4551 }
4552 }
4553 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4554 second_fail:
4555 if (verbose)
4556 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4557 devname);
4558 continue; /* Cannot seek */
4559 }
4560 /* There should be a duplicate backup superblock 4k before here */
4561 if (lseek64(fd, -4096, 1) < 0 ||
4562 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4563 goto second_fail; /* Cannot find leading superblock */
4564 if (bsb.magic[15] == '1')
4565 bsbsize = offsetof(struct mdp_backup_super, pad1);
4566 else
4567 bsbsize = offsetof(struct mdp_backup_super, pad);
4568 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4569 goto second_fail; /* Cannot find leading superblock */
4570
4571 /* Now need the data offsets for all devices. */
4572 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4573 for(j=0; j<info->array.raid_disks; j++) {
4574 if (fdlist[j] < 0)
4575 continue;
4576 if (st->ss->load_super(st, fdlist[j], NULL))
4577 /* FIXME should be this be an error */
4578 continue;
4579 st->ss->getinfo_super(st, &dinfo, NULL);
4580 st->ss->free_super(st);
4581 offsets[j] = dinfo.data_offset * 512;
4582 }
4583 printf("%s: restoring critical section\n", Name);
4584
4585 if (restore_stripes(fdlist, offsets,
4586 info->array.raid_disks,
4587 info->new_chunk,
4588 info->new_level,
4589 info->new_layout,
4590 fd, __le64_to_cpu(bsb.devstart)*512,
4591 __le64_to_cpu(bsb.arraystart)*512,
4592 __le64_to_cpu(bsb.length)*512, NULL)) {
4593 /* didn't succeed, so giveup */
4594 if (verbose)
4595 pr_err("Error restoring backup from %s\n",
4596 devname);
4597 free(offsets);
4598 return 1;
4599 }
4600
4601 if (bsb.magic[15] == '2' &&
4602 restore_stripes(fdlist, offsets,
4603 info->array.raid_disks,
4604 info->new_chunk,
4605 info->new_level,
4606 info->new_layout,
4607 fd, __le64_to_cpu(bsb.devstart)*512 +
4608 __le64_to_cpu(bsb.devstart2)*512,
4609 __le64_to_cpu(bsb.arraystart2)*512,
4610 __le64_to_cpu(bsb.length2)*512, NULL)) {
4611 /* didn't succeed, so giveup */
4612 if (verbose)
4613 pr_err("Error restoring second backup from %s\n",
4614 devname);
4615 free(offsets);
4616 return 1;
4617 }
4618
4619 free(offsets);
4620
4621 /* Ok, so the data is restored. Let's update those superblocks. */
4622
4623 lo = hi = 0;
4624 if (bsb.length) {
4625 lo = __le64_to_cpu(bsb.arraystart);
4626 hi = lo + __le64_to_cpu(bsb.length);
4627 }
4628 if (bsb.magic[15] == '2' && bsb.length2) {
4629 unsigned long long lo1, hi1;
4630 lo1 = __le64_to_cpu(bsb.arraystart2);
4631 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4632 if (lo == hi) {
4633 lo = lo1;
4634 hi = hi1;
4635 } else if (lo < lo1)
4636 hi = hi1;
4637 else
4638 lo = lo1;
4639 }
4640 if (lo < hi &&
4641 (info->reshape_progress < lo ||
4642 info->reshape_progress > hi))
4643 /* backup does not affect reshape_progress*/ ;
4644 else if (info->delta_disks >= 0) {
4645 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4646 __le64_to_cpu(bsb.length);
4647 if (bsb.magic[15] == '2') {
4648 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4649 __le64_to_cpu(bsb.length2);
4650 if (p2 > info->reshape_progress)
4651 info->reshape_progress = p2;
4652 }
4653 } else {
4654 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4655 if (bsb.magic[15] == '2') {
4656 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4657 if (p2 < info->reshape_progress)
4658 info->reshape_progress = p2;
4659 }
4660 }
4661 for (j=0; j<info->array.raid_disks; j++) {
4662 if (fdlist[j] < 0)
4663 continue;
4664 if (st->ss->load_super(st, fdlist[j], NULL))
4665 continue;
4666 st->ss->getinfo_super(st, &dinfo, NULL);
4667 dinfo.reshape_progress = info->reshape_progress;
4668 st->ss->update_super(st, &dinfo,
4669 "_reshape_progress",
4670 NULL,0, 0, NULL);
4671 st->ss->store_super(st, fdlist[j]);
4672 st->ss->free_super(st);
4673 }
4674 return 0;
4675 }
4676 /* Didn't find any backup data, try to see if any
4677 * was needed.
4678 */
4679 if (info->delta_disks < 0) {
4680 /* When shrinking, the critical section is at the end.
4681 * So see if we are before the critical section.
4682 */
4683 unsigned long long first_block;
4684 nstripe = ostripe = 0;
4685 first_block = 0;
4686 while (ostripe >= nstripe) {
4687 ostripe += info->array.chunk_size / 512;
4688 first_block = ostripe * odata;
4689 nstripe = first_block / ndata / (info->new_chunk/512) *
4690 (info->new_chunk/512);
4691 }
4692
4693 if (info->reshape_progress >= first_block)
4694 return 0;
4695 }
4696 if (info->delta_disks > 0) {
4697 /* See if we are beyond the critical section. */
4698 unsigned long long last_block;
4699 nstripe = ostripe = 0;
4700 last_block = 0;
4701 while (nstripe >= ostripe) {
4702 nstripe += info->new_chunk / 512;
4703 last_block = nstripe * ndata;
4704 ostripe = last_block / odata / (info->array.chunk_size/512) *
4705 (info->array.chunk_size/512);
4706 }
4707
4708 if (info->reshape_progress >= last_block)
4709 return 0;
4710 }
4711 /* needed to recover critical section! */
4712 if (verbose)
4713 pr_err("Failed to find backup of critical section\n");
4714 return 1;
4715 }
4716
4717 int Grow_continue_command(char *devname, int fd,
4718 char *backup_file, int verbose)
4719 {
4720 int ret_val = 0;
4721 struct supertype *st = NULL;
4722 struct mdinfo *content = NULL;
4723 struct mdinfo array;
4724 char *subarray = NULL;
4725 struct mdinfo *cc = NULL;
4726 struct mdstat_ent *mdstat = NULL;
4727 int cfd = -1;
4728 int fd2 = -1;
4729
4730 dprintf("Grow continue from command line called for %s\n",
4731 devname);
4732
4733 st = super_by_fd(fd, &subarray);
4734 if (!st || !st->ss) {
4735 pr_err("Unable to determine metadata format for %s\n",
4736 devname);
4737 return 1;
4738 }
4739 dprintf("Grow continue is run for ");
4740 if (st->ss->external == 0) {
4741 int d;
4742 dprintf_cont("native array (%s)\n", devname);
4743 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4744 pr_err("%s is not an active md array - aborting\n", devname);
4745 ret_val = 1;
4746 goto Grow_continue_command_exit;
4747 }
4748 content = &array;
4749 /* Need to load a superblock.
4750 * FIXME we should really get what we need from
4751 * sysfs
4752 */
4753 for (d = 0; d < MAX_DISKS; d++) {
4754 mdu_disk_info_t disk;
4755 char *dv;
4756 int err;
4757 disk.number = d;
4758 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4759 continue;
4760 if (disk.major == 0 && disk.minor == 0)
4761 continue;
4762 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4763 continue;
4764 dv = map_dev(disk.major, disk.minor, 1);
4765 if (!dv)
4766 continue;
4767 fd2 = dev_open(dv, O_RDONLY);
4768 if (fd2 < 0)
4769 continue;
4770 err = st->ss->load_super(st, fd2, NULL);
4771 close(fd2);
4772 /* invalidate fd2 to avoid possible double close() */
4773 fd2 = -1;
4774 if (err)
4775 continue;
4776 break;
4777 }
4778 if (d == MAX_DISKS) {
4779 pr_err("Unable to load metadata for %s\n",
4780 devname);
4781 ret_val = 1;
4782 goto Grow_continue_command_exit;
4783 }
4784 st->ss->getinfo_super(st, content, NULL);
4785 } else {
4786 char *container;
4787
4788 if (subarray) {
4789 dprintf_cont("subarray (%s)\n", subarray);
4790 container = st->container_devnm;
4791 cfd = open_dev_excl(st->container_devnm);
4792 } else {
4793 container = st->devnm;
4794 close(fd);
4795 cfd = open_dev_excl(st->devnm);
4796 dprintf_cont("container (%s)\n", container);
4797 fd = cfd;
4798 }
4799 if (cfd < 0) {
4800 pr_err("Unable to open container for %s\n", devname);
4801 ret_val = 1;
4802 goto Grow_continue_command_exit;
4803 }
4804
4805 /* find in container array under reshape
4806 */
4807 ret_val = st->ss->load_container(st, cfd, NULL);
4808 if (ret_val) {
4809 pr_err("Cannot read superblock for %s\n",
4810 devname);
4811 ret_val = 1;
4812 goto Grow_continue_command_exit;
4813 }
4814
4815 cc = st->ss->container_content(st, subarray);
4816 for (content = cc; content ; content = content->next) {
4817 char *array;
4818 int allow_reshape = 1;
4819
4820 if (content->reshape_active == 0)
4821 continue;
4822 /* The decision about array or container wide
4823 * reshape is taken in Grow_continue based
4824 * content->reshape_active state, therefore we
4825 * need to check_reshape based on
4826 * reshape_active and subarray name
4827 */
4828 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4829 allow_reshape = 0;
4830 if (content->reshape_active == CONTAINER_RESHAPE &&
4831 (content->array.state
4832 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4833 allow_reshape = 0;
4834
4835 if (!allow_reshape) {
4836 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4837 devname, container);
4838 ret_val = 1;
4839 goto Grow_continue_command_exit;
4840 }
4841
4842 array = strchr(content->text_version+1, '/')+1;
4843 mdstat = mdstat_by_subdev(array, container);
4844 if (!mdstat)
4845 continue;
4846 if (mdstat->active == 0) {
4847 pr_err("Skipping inactive array %s.\n",
4848 mdstat->devnm);
4849 free_mdstat(mdstat);
4850 mdstat = NULL;
4851 continue;
4852 }
4853 break;
4854 }
4855 if (!content) {
4856 pr_err("Unable to determine reshaped array for %s\n", devname);
4857 ret_val = 1;
4858 goto Grow_continue_command_exit;
4859 }
4860 fd2 = open_dev(mdstat->devnm);
4861 if (fd2 < 0) {
4862 pr_err("cannot open (%s)\n", mdstat->devnm);
4863 ret_val = 1;
4864 goto Grow_continue_command_exit;
4865 }
4866
4867 sysfs_init(content, fd2, mdstat->devnm);
4868
4869 /* start mdmon in case it is not running
4870 */
4871 if (!mdmon_running(container))
4872 start_mdmon(container);
4873 ping_monitor(container);
4874
4875 if (mdmon_running(container))
4876 st->update_tail = &st->updates;
4877 else {
4878 pr_err("No mdmon found. Grow cannot continue.\n");
4879 ret_val = 1;
4880 goto Grow_continue_command_exit;
4881 }
4882 }
4883
4884 /* verify that array under reshape is started from
4885 * correct position
4886 */
4887 if (verify_reshape_position(content, content->array.level) < 0) {
4888 ret_val = 1;
4889 goto Grow_continue_command_exit;
4890 }
4891
4892 /* continue reshape
4893 */
4894 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4895
4896 Grow_continue_command_exit:
4897 if (fd2 > -1)
4898 close(fd2);
4899 if (cfd > -1)
4900 close(cfd);
4901 st->ss->free_super(st);
4902 free_mdstat(mdstat);
4903 sysfs_free(cc);
4904 free(subarray);
4905
4906 return ret_val;
4907 }
4908
4909 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4910 char *backup_file, int forked, int freeze_reshape)
4911 {
4912 int ret_val = 2;
4913
4914 if (!info->reshape_active)
4915 return ret_val;
4916
4917 if (st->ss->external) {
4918 int cfd = open_dev(st->container_devnm);
4919
4920 if (cfd < 0)
4921 return 1;
4922
4923 st->ss->load_container(st, cfd, st->container_devnm);
4924 close(cfd);
4925 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4926 st, info, 0, backup_file,
4927 0, forked,
4928 1 | info->reshape_active,
4929 freeze_reshape);
4930 } else
4931 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4932 NULL, INVALID_SECTORS,
4933 backup_file, 0, forked,
4934 1 | info->reshape_active,
4935 freeze_reshape);
4936
4937 return ret_val;
4938 }
4939
4940 char *make_backup(char *name)
4941 {
4942 char *base = "backup_file-";
4943 int len;
4944 char *fname;
4945
4946 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4947 fname = xmalloc(len);
4948 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4949 return fname;
4950 }
4951
4952 char *locate_backup(char *name)
4953 {
4954 char *fl = make_backup(name);
4955 struct stat stb;
4956
4957 if (stat(fl, &stb) == 0 &&
4958 S_ISREG(stb.st_mode))
4959 return fl;
4960
4961 free(fl);
4962 return NULL;
4963 }
Unnamed repository; edit this file 'description' to name the repository.