]> 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
super1: PPL support
[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 struct mdinfo *mdi;
294
295 if (vers < 9003) {
296 major = BITMAP_MAJOR_HOSTENDIAN;
297 pr_err("Warning - bitmaps created on this kernel are not portable\n"
298 " between different architectures. Consider upgrading the Linux kernel.\n");
299 }
300
301 /*
302 * We only ever get called if s->bitmap_file is != NULL, so this check
303 * is just here to quiet down static code checkers.
304 */
305 if (!s->bitmap_file)
306 return 1;
307
308 if (strcmp(s->bitmap_file, "clustered") == 0)
309 major = BITMAP_MAJOR_CLUSTERED;
310
311 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
312 if (errno == ENOMEM)
313 pr_err("Memory allocation failure.\n");
314 else
315 pr_err("bitmaps not supported by this kernel.\n");
316 return 1;
317 }
318 if (bmf.pathname[0]) {
319 if (strcmp(s->bitmap_file,"none") == 0) {
320 if (ioctl(fd, SET_BITMAP_FILE, -1) != 0) {
321 pr_err("failed to remove bitmap %s\n",
322 bmf.pathname);
323 return 1;
324 }
325 return 0;
326 }
327 pr_err("%s already has a bitmap (%s)\n",
328 devname, bmf.pathname);
329 return 1;
330 }
331 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
332 pr_err("cannot get array status for %s\n", devname);
333 return 1;
334 }
335 if (array.state & (1 << MD_SB_BITMAP_PRESENT)) {
336 if (strcmp(s->bitmap_file, "none")==0) {
337 array.state &= ~(1 << MD_SB_BITMAP_PRESENT);
338 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
339 if (array.state & (1 << MD_SB_CLUSTERED))
340 pr_err("failed to remove clustered bitmap.\n");
341 else
342 pr_err("failed to remove internal bitmap.\n");
343 return 1;
344 }
345 return 0;
346 }
347 pr_err("bitmap already present on %s\n", devname);
348 return 1;
349 }
350
351 if (strcmp(s->bitmap_file, "none") == 0) {
352 pr_err("no bitmap found on %s\n", devname);
353 return 1;
354 }
355 if (array.level <= 0) {
356 pr_err("Bitmaps not meaningful with level %s\n",
357 map_num(pers, array.level)?:"of this array");
358 return 1;
359 }
360 bitmapsize = array.size;
361 bitmapsize <<= 1;
362 if (get_dev_size(fd, NULL, &array_size) &&
363 array_size > (0x7fffffffULL << 9)) {
364 /* Array is big enough that we cannot trust array.size
365 * try other approaches
366 */
367 bitmapsize = get_component_size(fd);
368 }
369 if (bitmapsize == 0) {
370 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
371 return 1;
372 }
373
374 if (array.level == 10) {
375 int ncopies;
376
377 ncopies = (array.layout & 255) * ((array.layout >> 8) & 255);
378 bitmapsize = bitmapsize * array.raid_disks / ncopies;
379 }
380
381 st = super_by_fd(fd, &subarray);
382 if (!st) {
383 pr_err("Cannot understand version %d.%d\n",
384 array.major_version, array.minor_version);
385 return 1;
386 }
387 if (subarray) {
388 pr_err("Cannot add bitmaps to sub-arrays yet\n");
389 free(subarray);
390 free(st);
391 return 1;
392 }
393
394 mdi = sysfs_read(fd, NULL, GET_CONSISTENCY_POLICY);
395 if (mdi) {
396 if (mdi->consistency_policy == CONSISTENCY_POLICY_PPL) {
397 pr_err("Cannot add bitmap to array with PPL\n");
398 free(mdi);
399 free(st);
400 return 1;
401 }
402 free(mdi);
403 }
404
405 if (strcmp(s->bitmap_file, "internal") == 0 ||
406 strcmp(s->bitmap_file, "clustered") == 0) {
407 int rv;
408 int d;
409 int offset_setable = 0;
410 if (st->ss->add_internal_bitmap == NULL) {
411 pr_err("Internal bitmaps not supported with %s metadata\n", st->ss->name);
412 return 1;
413 }
414 st->nodes = c->nodes;
415 st->cluster_name = c->homecluster;
416 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
417 if (mdi)
418 offset_setable = 1;
419 for (d = 0; d < st->max_devs; d++) {
420 mdu_disk_info_t disk;
421 char *dv;
422 int fd2;
423
424 disk.number = d;
425 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
426 continue;
427 if (disk.major == 0 && disk.minor == 0)
428 continue;
429 if ((disk.state & (1 << MD_DISK_SYNC)) == 0)
430 continue;
431 dv = map_dev(disk.major, disk.minor, 1);
432 if (!dv)
433 continue;
434 fd2 = dev_open(dv, O_RDWR);
435 if (fd2 < 0)
436 continue;
437 rv = st->ss->load_super(st, fd2, NULL);
438 if (!rv) {
439 rv = st->ss->add_internal_bitmap(
440 st, &s->bitmap_chunk, c->delay,
441 s->write_behind, bitmapsize,
442 offset_setable, major);
443 if (!rv) {
444 st->ss->write_bitmap(st, fd2,
445 NodeNumUpdate);
446 } else {
447 pr_err("failed to create internal bitmap - chunksize problem.\n");
448 }
449 } else {
450 pr_err("failed to load super-block.\n");
451 }
452 close(fd2);
453 if (rv)
454 return 1;
455 }
456 if (offset_setable) {
457 st->ss->getinfo_super(st, mdi, NULL);
458 sysfs_init(mdi, fd, NULL);
459 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
460 mdi->bitmap_offset);
461 free(mdi);
462 } else {
463 if (strcmp(s->bitmap_file, "clustered") == 0)
464 array.state |= (1 << MD_SB_CLUSTERED);
465 array.state |= (1 << MD_SB_BITMAP_PRESENT);
466 rv = ioctl(fd, SET_ARRAY_INFO, &array);
467 }
468 if (rv < 0) {
469 if (errno == EBUSY)
470 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
471 pr_err("failed to set internal bitmap.\n");
472 return 1;
473 }
474 } else {
475 int uuid[4];
476 int bitmap_fd;
477 int d;
478 int max_devs = st->max_devs;
479
480 /* try to load a superblock */
481 for (d = 0; d < max_devs; d++) {
482 mdu_disk_info_t disk;
483 char *dv;
484 int fd2;
485 disk.number = d;
486 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
487 continue;
488 if ((disk.major==0 && disk.minor == 0) ||
489 (disk.state & (1 << MD_DISK_REMOVED)))
490 continue;
491 dv = map_dev(disk.major, disk.minor, 1);
492 if (!dv)
493 continue;
494 fd2 = dev_open(dv, O_RDONLY);
495 if (fd2 >= 0) {
496 if (st->ss->load_super(st, fd2, NULL) == 0) {
497 close(fd2);
498 st->ss->uuid_from_super(st, uuid);
499 break;
500 }
501 close(fd2);
502 }
503 }
504 if (d == max_devs) {
505 pr_err("cannot find UUID for array!\n");
506 return 1;
507 }
508 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid,
509 s->bitmap_chunk, c->delay, s->write_behind,
510 bitmapsize, major)) {
511 return 1;
512 }
513 bitmap_fd = open(s->bitmap_file, O_RDWR);
514 if (bitmap_fd < 0) {
515 pr_err("weird: %s cannot be opened\n", s->bitmap_file);
516 return 1;
517 }
518 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
519 int err = errno;
520 if (errno == EBUSY)
521 pr_err("Cannot add bitmap while array is resyncing or reshaping etc.\n");
522 pr_err("Cannot set bitmap file for %s: %s\n",
523 devname, strerror(err));
524 return 1;
525 }
526 }
527
528 return 0;
529 }
530
531 /*
532 * When reshaping an array we might need to backup some data.
533 * This is written to all spares with a 'super_block' describing it.
534 * The superblock goes 4K from the end of the used space on the
535 * device.
536 * It if written after the backup is complete.
537 * It has the following structure.
538 */
539
540 static struct mdp_backup_super {
541 char magic[16]; /* md_backup_data-1 or -2 */
542 __u8 set_uuid[16];
543 __u64 mtime;
544 /* start/sizes in 512byte sectors */
545 __u64 devstart; /* address on backup device/file of data */
546 __u64 arraystart;
547 __u64 length;
548 __u32 sb_csum; /* csum of preceeding bytes. */
549 __u32 pad1;
550 __u64 devstart2; /* offset in to data of second section */
551 __u64 arraystart2;
552 __u64 length2;
553 __u32 sb_csum2; /* csum of preceeding bytes. */
554 __u8 pad[512-68-32];
555 } __attribute__((aligned(512))) bsb, bsb2;
556
557 static __u32 bsb_csum(char *buf, int len)
558 {
559 int i;
560 int csum = 0;
561 for (i = 0; i < len; i++)
562 csum = (csum<<3) + buf[0];
563 return __cpu_to_le32(csum);
564 }
565
566 static int check_idle(struct supertype *st)
567 {
568 /* Check that all member arrays for this container, or the
569 * container of this array, are idle
570 */
571 char *container = (st->container_devnm[0]
572 ? st->container_devnm : st->devnm);
573 struct mdstat_ent *ent, *e;
574 int is_idle = 1;
575
576 ent = mdstat_read(0, 0);
577 for (e = ent ; e; e = e->next) {
578 if (!is_container_member(e, container))
579 continue;
580 if (e->percent >= 0) {
581 is_idle = 0;
582 break;
583 }
584 }
585 free_mdstat(ent);
586 return is_idle;
587 }
588
589 static int freeze_container(struct supertype *st)
590 {
591 char *container = (st->container_devnm[0]
592 ? st->container_devnm : st->devnm);
593
594 if (!check_idle(st))
595 return -1;
596
597 if (block_monitor(container, 1)) {
598 pr_err("failed to freeze container\n");
599 return -2;
600 }
601
602 return 1;
603 }
604
605 static void unfreeze_container(struct supertype *st)
606 {
607 char *container = (st->container_devnm[0]
608 ? st->container_devnm : st->devnm);
609
610 unblock_monitor(container, 1);
611 }
612
613 static int freeze(struct supertype *st)
614 {
615 /* Try to freeze resync/rebuild on this array/container.
616 * Return -1 if the array is busy,
617 * return -2 container cannot be frozen,
618 * return 0 if this kernel doesn't support 'frozen'
619 * return 1 if it worked.
620 */
621 if (st->ss->external)
622 return freeze_container(st);
623 else {
624 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
625 int err;
626 char buf[20];
627
628 if (!sra)
629 return -1;
630 /* Need to clear any 'read-auto' status */
631 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
632 strncmp(buf, "read-auto", 9) == 0)
633 sysfs_set_str(sra, NULL, "array_state", "clean");
634
635 err = sysfs_freeze_array(sra);
636 sysfs_free(sra);
637 return err;
638 }
639 }
640
641 static void unfreeze(struct supertype *st)
642 {
643 if (st->ss->external)
644 return unfreeze_container(st);
645 else {
646 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
647 char buf[20];
648
649 if (sra &&
650 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
651 && strcmp(buf, "frozen\n") == 0)
652 sysfs_set_str(sra, NULL, "sync_action", "idle");
653 sysfs_free(sra);
654 }
655 }
656
657 static void wait_reshape(struct mdinfo *sra)
658 {
659 int fd = sysfs_get_fd(sra, NULL, "sync_action");
660 char action[20];
661
662 if (fd < 0)
663 return;
664
665 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
666 strncmp(action, "reshape", 7) == 0)
667 sysfs_wait(fd, NULL);
668 close(fd);
669 }
670
671 static int reshape_super(struct supertype *st, unsigned long long size,
672 int level, int layout, int chunksize, int raid_disks,
673 int delta_disks, char *backup_file, char *dev,
674 int direction, int verbose)
675 {
676 /* nothing extra to check in the native case */
677 if (!st->ss->external)
678 return 0;
679 if (!st->ss->reshape_super ||
680 !st->ss->manage_reshape) {
681 pr_err("%s metadata does not support reshape\n",
682 st->ss->name);
683 return 1;
684 }
685
686 return st->ss->reshape_super(st, size, level, layout, chunksize,
687 raid_disks, delta_disks, backup_file, dev,
688 direction, verbose);
689 }
690
691 static void sync_metadata(struct supertype *st)
692 {
693 if (st->ss->external) {
694 if (st->update_tail) {
695 flush_metadata_updates(st);
696 st->update_tail = &st->updates;
697 } else
698 st->ss->sync_metadata(st);
699 }
700 }
701
702 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
703 {
704 /* when dealing with external metadata subarrays we need to be
705 * prepared to handle EAGAIN. The kernel may need to wait for
706 * mdmon to mark the array active so the kernel can handle
707 * allocations/writeback when preparing the reshape action
708 * (md_allow_write()). We temporarily disable safe_mode_delay
709 * to close a race with the array_state going clean before the
710 * next write to raid_disks / stripe_cache_size
711 */
712 char safe[50];
713 int rc;
714
715 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
716 if (!container ||
717 (strcmp(name, "raid_disks") != 0 &&
718 strcmp(name, "stripe_cache_size") != 0))
719 return sysfs_set_num(sra, NULL, name, n);
720
721 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
722 if (rc <= 0)
723 return -1;
724 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
725 rc = sysfs_set_num(sra, NULL, name, n);
726 if (rc < 0 && errno == EAGAIN) {
727 ping_monitor(container);
728 /* if we get EAGAIN here then the monitor is not active
729 * so stop trying
730 */
731 rc = sysfs_set_num(sra, NULL, name, n);
732 }
733 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
734 return rc;
735 }
736
737 int start_reshape(struct mdinfo *sra, int already_running,
738 int before_data_disks, int data_disks)
739 {
740 int err;
741 unsigned long long sync_max_to_set;
742
743 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
744 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
745 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
746 sra->reshape_progress);
747 if (before_data_disks <= data_disks)
748 sync_max_to_set = sra->reshape_progress / data_disks;
749 else
750 sync_max_to_set = (sra->component_size * data_disks
751 - sra->reshape_progress) / data_disks;
752 if (!already_running)
753 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
754 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
755 if (!already_running && err == 0) {
756 int cnt = 5;
757 do {
758 err = sysfs_set_str(sra, NULL, "sync_action", "reshape");
759 if (err)
760 sleep(1);
761 } while (err && errno == EBUSY && cnt-- > 0);
762 }
763 return err;
764 }
765
766 void abort_reshape(struct mdinfo *sra)
767 {
768 sysfs_set_str(sra, NULL, "sync_action", "idle");
769 /*
770 * Prior to kernel commit: 23ddff3792f6 ("md: allow suspend_lo and
771 * suspend_hi to decrease as well as increase.")
772 * you could only increase suspend_{lo,hi} unless the region they
773 * covered was empty. So to reset to 0, you need to push suspend_lo
774 * up past suspend_hi first. So to maximize the chance of mdadm
775 * working on all kernels, we want to keep doing that.
776 */
777 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
778 sysfs_set_num(sra, NULL, "suspend_hi", 0);
779 sysfs_set_num(sra, NULL, "suspend_lo", 0);
780 sysfs_set_num(sra, NULL, "sync_min", 0);
781 // It isn't safe to reset sync_max as we aren't monitoring.
782 // Array really should be stopped at this point.
783 }
784
785 int remove_disks_for_takeover(struct supertype *st,
786 struct mdinfo *sra,
787 int layout)
788 {
789 int nr_of_copies;
790 struct mdinfo *remaining;
791 int slot;
792
793 if (st->ss->external) {
794 int rv = 0;
795 struct mdinfo *arrays = st->ss->container_content(st, NULL);
796 /*
797 * containter_content returns list of arrays in container
798 * If arrays->next is not NULL it means that there are
799 * 2 arrays in container and operation should be blocked
800 */
801 if (arrays) {
802 if (arrays->next)
803 rv = 1;
804 sysfs_free(arrays);
805 if (rv) {
806 pr_err("Error. Cannot perform operation on /dev/%s\n", st->devnm);
807 pr_err("For this operation it MUST be single array in container\n");
808 return rv;
809 }
810 }
811 }
812
813 if (sra->array.level == 10)
814 nr_of_copies = layout & 0xff;
815 else if (sra->array.level == 1)
816 nr_of_copies = sra->array.raid_disks;
817 else
818 return 1;
819
820 remaining = sra->devs;
821 sra->devs = NULL;
822 /* for each 'copy', select one device and remove from the list. */
823 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
824 struct mdinfo **diskp;
825 int found = 0;
826
827 /* Find a working device to keep */
828 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
829 struct mdinfo *disk = *diskp;
830
831 if (disk->disk.raid_disk < slot)
832 continue;
833 if (disk->disk.raid_disk >= slot + nr_of_copies)
834 continue;
835 if (disk->disk.state & (1<<MD_DISK_REMOVED))
836 continue;
837 if (disk->disk.state & (1<<MD_DISK_FAULTY))
838 continue;
839 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
840 continue;
841
842 /* We have found a good disk to use! */
843 *diskp = disk->next;
844 disk->next = sra->devs;
845 sra->devs = disk;
846 found = 1;
847 break;
848 }
849 if (!found)
850 break;
851 }
852
853 if (slot < sra->array.raid_disks) {
854 /* didn't find all slots */
855 struct mdinfo **e;
856 e = &remaining;
857 while (*e)
858 e = &(*e)->next;
859 *e = sra->devs;
860 sra->devs = remaining;
861 return 1;
862 }
863
864 /* Remove all 'remaining' devices from the array */
865 while (remaining) {
866 struct mdinfo *sd = remaining;
867 remaining = sd->next;
868
869 sysfs_set_str(sra, sd, "state", "faulty");
870 sysfs_set_str(sra, sd, "slot", "none");
871 /* for external metadata disks should be removed in mdmon */
872 if (!st->ss->external)
873 sysfs_set_str(sra, sd, "state", "remove");
874 sd->disk.state |= (1<<MD_DISK_REMOVED);
875 sd->disk.state &= ~(1<<MD_DISK_SYNC);
876 sd->next = sra->devs;
877 sra->devs = sd;
878 }
879 return 0;
880 }
881
882 void reshape_free_fdlist(int *fdlist,
883 unsigned long long *offsets,
884 int size)
885 {
886 int i;
887
888 for (i = 0; i < size; i++)
889 if (fdlist[i] >= 0)
890 close(fdlist[i]);
891
892 free(fdlist);
893 free(offsets);
894 }
895
896 int reshape_prepare_fdlist(char *devname,
897 struct mdinfo *sra,
898 int raid_disks,
899 int nrdisks,
900 unsigned long blocks,
901 char *backup_file,
902 int *fdlist,
903 unsigned long long *offsets)
904 {
905 int d = 0;
906 struct mdinfo *sd;
907
908 enable_fds(nrdisks);
909 for (d = 0; d <= nrdisks; d++)
910 fdlist[d] = -1;
911 d = raid_disks;
912 for (sd = sra->devs; sd; sd = sd->next) {
913 if (sd->disk.state & (1<<MD_DISK_FAULTY))
914 continue;
915 if (sd->disk.state & (1<<MD_DISK_SYNC) &&
916 sd->disk.raid_disk < raid_disks) {
917 char *dn = map_dev(sd->disk.major,
918 sd->disk.minor, 1);
919 fdlist[sd->disk.raid_disk]
920 = dev_open(dn, O_RDONLY);
921 offsets[sd->disk.raid_disk] = sd->data_offset*512;
922 if (fdlist[sd->disk.raid_disk] < 0) {
923 pr_err("%s: cannot open component %s\n",
924 devname, dn ? dn : "-unknown-");
925 d = -1;
926 goto release;
927 }
928 } else if (backup_file == NULL) {
929 /* spare */
930 char *dn = map_dev(sd->disk.major,
931 sd->disk.minor, 1);
932 fdlist[d] = dev_open(dn, O_RDWR);
933 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
934 if (fdlist[d] < 0) {
935 pr_err("%s: cannot open component %s\n",
936 devname, dn ? dn : "-unknown-");
937 d = -1;
938 goto release;
939 }
940 d++;
941 }
942 }
943 release:
944 return d;
945 }
946
947 int reshape_open_backup_file(char *backup_file,
948 int fd,
949 char *devname,
950 long blocks,
951 int *fdlist,
952 unsigned long long *offsets,
953 char *sys_name,
954 int restart)
955 {
956 /* Return 1 on success, 0 on any form of failure */
957 /* need to check backup file is large enough */
958 char buf[512];
959 struct stat stb;
960 unsigned int dev;
961 int i;
962
963 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
964 S_IRUSR | S_IWUSR);
965 *offsets = 8 * 512;
966 if (*fdlist < 0) {
967 pr_err("%s: cannot create backup file %s: %s\n",
968 devname, backup_file, strerror(errno));
969 return 0;
970 }
971 /* Guard against backup file being on array device.
972 * If array is partitioned or if LVM etc is in the
973 * way this will not notice, but it is better than
974 * nothing.
975 */
976 fstat(*fdlist, &stb);
977 dev = stb.st_dev;
978 fstat(fd, &stb);
979 if (stb.st_rdev == dev) {
980 pr_err("backup file must NOT be on the array being reshaped.\n");
981 close(*fdlist);
982 return 0;
983 }
984
985 memset(buf, 0, 512);
986 for (i=0; i < blocks + 8 ; i++) {
987 if (write(*fdlist, buf, 512) != 512) {
988 pr_err("%s: cannot create backup file %s: %s\n",
989 devname, backup_file, strerror(errno));
990 return 0;
991 }
992 }
993 if (fsync(*fdlist) != 0) {
994 pr_err("%s: cannot create backup file %s: %s\n",
995 devname, backup_file, strerror(errno));
996 return 0;
997 }
998
999 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
1000 char *bu = make_backup(sys_name);
1001 if (symlink(backup_file, bu))
1002 pr_err("Recording backup file in " MAP_DIR " failed: %s\n",
1003 strerror(errno));
1004 free(bu);
1005 }
1006
1007 return 1;
1008 }
1009
1010 unsigned long compute_backup_blocks(int nchunk, int ochunk,
1011 unsigned int ndata, unsigned int odata)
1012 {
1013 unsigned long a, b, blocks;
1014 /* So how much do we need to backup.
1015 * We need an amount of data which is both a whole number of
1016 * old stripes and a whole number of new stripes.
1017 * So LCM for (chunksize*datadisks).
1018 */
1019 a = (ochunk/512) * odata;
1020 b = (nchunk/512) * ndata;
1021 /* Find GCD */
1022 a = GCD(a, b);
1023 /* LCM == product / GCD */
1024 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
1025
1026 return blocks;
1027 }
1028
1029 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
1030 {
1031 /* Based on the current array state in info->array and
1032 * the changes in info->new_* etc, determine:
1033 * - whether the change is possible
1034 * - Intermediate level/raid_disks/layout
1035 * - whether a restriping reshape is needed
1036 * - number of sectors in minimum change unit. This
1037 * will cover a whole number of stripes in 'before' and
1038 * 'after'.
1039 *
1040 * Return message if the change should be rejected
1041 * NULL if the change can be achieved
1042 *
1043 * This can be called as part of starting a reshape, or
1044 * when assembling an array that is undergoing reshape.
1045 */
1046 int near, far, offset, copies;
1047 int new_disks;
1048 int old_chunk, new_chunk;
1049 /* delta_parity records change in number of devices
1050 * caused by level change
1051 */
1052 int delta_parity = 0;
1053
1054 memset(re, 0, sizeof(*re));
1055
1056 /* If a new level not explicitly given, we assume no-change */
1057 if (info->new_level == UnSet)
1058 info->new_level = info->array.level;
1059
1060 if (info->new_chunk)
1061 switch (info->new_level) {
1062 case 0:
1063 case 4:
1064 case 5:
1065 case 6:
1066 case 10:
1067 /* chunk size is meaningful, must divide component_size
1068 * evenly
1069 */
1070 if (info->component_size % (info->new_chunk/512)) {
1071 unsigned long long shrink = info->component_size;
1072 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1073 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1074 info->new_chunk/1024, info->component_size/2);
1075 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1076 devname, shrink/2);
1077 pr_err("will shrink the array so the given chunk size would work.\n");
1078 return "";
1079 }
1080 break;
1081 default:
1082 return "chunk size not meaningful for this level";
1083 }
1084 else
1085 info->new_chunk = info->array.chunk_size;
1086
1087 switch (info->array.level) {
1088 default:
1089 return "No reshape is possibly for this RAID level";
1090 case LEVEL_LINEAR:
1091 if (info->delta_disks != UnSet)
1092 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1093 else
1094 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1095 case 1:
1096 /* RAID1 can convert to RAID1 with different disks, or
1097 * raid5 with 2 disks, or
1098 * raid0 with 1 disk
1099 */
1100 if (info->new_level > 1 &&
1101 (info->component_size & 7))
1102 return "Cannot convert RAID1 of this size - reduce size to multiple of 4K first.";
1103 if (info->new_level == 0) {
1104 if (info->delta_disks != UnSet &&
1105 info->delta_disks != 0)
1106 return "Cannot change number of disks with RAID1->RAID0 conversion";
1107 re->level = 0;
1108 re->before.data_disks = 1;
1109 re->after.data_disks = 1;
1110 return NULL;
1111 }
1112 if (info->new_level == 1) {
1113 if (info->delta_disks == UnSet)
1114 /* Don't know what to do */
1115 return "no change requested for Growing RAID1";
1116 re->level = 1;
1117 return NULL;
1118 }
1119 if (info->array.raid_disks != 2 &&
1120 info->new_level == 5)
1121 return "Can only convert a 2-device array to RAID5";
1122 if (info->array.raid_disks == 2 &&
1123 info->new_level == 5) {
1124
1125 re->level = 5;
1126 re->before.data_disks = 1;
1127 if (info->delta_disks != UnSet &&
1128 info->delta_disks != 0)
1129 re->after.data_disks = 1 + info->delta_disks;
1130 else
1131 re->after.data_disks = 1;
1132 if (re->after.data_disks < 1)
1133 return "Number of disks too small for RAID5";
1134
1135 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1136 info->array.chunk_size = 65536;
1137 break;
1138 }
1139 /* Could do some multi-stage conversions, but leave that to
1140 * later.
1141 */
1142 return "Impossibly level change request for RAID1";
1143
1144 case 10:
1145 /* RAID10 can be converted from near mode to
1146 * RAID0 by removing some devices.
1147 * It can also be reshaped if the kernel supports
1148 * new_data_offset.
1149 */
1150 switch (info->new_level) {
1151 case 0:
1152 if ((info->array.layout & ~0xff) != 0x100)
1153 return "Cannot Grow RAID10 with far/offset layout";
1154 /* number of devices must be multiple of number of copies */
1155 if (info->array.raid_disks % (info->array.layout & 0xff))
1156 return "RAID10 layout too complex for Grow operation";
1157
1158 new_disks = (info->array.raid_disks
1159 / (info->array.layout & 0xff));
1160 if (info->delta_disks == UnSet)
1161 info->delta_disks = (new_disks
1162 - info->array.raid_disks);
1163
1164 if (info->delta_disks != new_disks - info->array.raid_disks)
1165 return "New number of raid-devices impossible for RAID10";
1166 if (info->new_chunk &&
1167 info->new_chunk != info->array.chunk_size)
1168 return "Cannot change chunk-size with RAID10 Grow";
1169
1170 /* looks good */
1171 re->level = 0;
1172 re->before.data_disks = new_disks;
1173 re->after.data_disks = re->before.data_disks;
1174 return NULL;
1175
1176 case 10:
1177 near = info->array.layout & 0xff;
1178 far = (info->array.layout >> 8) & 0xff;
1179 offset = info->array.layout & 0x10000;
1180 if (far > 1 && !offset)
1181 return "Cannot reshape RAID10 in far-mode";
1182 copies = near * far;
1183
1184 old_chunk = info->array.chunk_size * far;
1185
1186 if (info->new_layout == UnSet)
1187 info->new_layout = info->array.layout;
1188 else {
1189 near = info->new_layout & 0xff;
1190 far = (info->new_layout >> 8) & 0xff;
1191 offset = info->new_layout & 0x10000;
1192 if (far > 1 && !offset)
1193 return "Cannot reshape RAID10 to far-mode";
1194 if (near * far != copies)
1195 return "Cannot change number of copies when reshaping RAID10";
1196 }
1197 if (info->delta_disks == UnSet)
1198 info->delta_disks = 0;
1199 new_disks = (info->array.raid_disks +
1200 info->delta_disks);
1201
1202 new_chunk = info->new_chunk * far;
1203
1204 re->level = 10;
1205 re->before.layout = info->array.layout;
1206 re->before.data_disks = info->array.raid_disks;
1207 re->after.layout = info->new_layout;
1208 re->after.data_disks = new_disks;
1209 /* For RAID10 we don't do backup but do allow reshape,
1210 * so set backup_blocks to INVALID_SECTORS rather than
1211 * zero.
1212 * And there is no need to synchronise stripes on both
1213 * 'old' and 'new'. So the important
1214 * number is the minimum data_offset difference
1215 * which is the larger of (offset copies * chunk).
1216 */
1217 re->backup_blocks = INVALID_SECTORS;
1218 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1219 if (new_disks < re->before.data_disks &&
1220 info->space_after < re->min_offset_change)
1221 /* Reduce component size by one chunk */
1222 re->new_size = (info->component_size -
1223 re->min_offset_change);
1224 else
1225 re->new_size = info->component_size;
1226 re->new_size = re->new_size * new_disks / copies;
1227 return NULL;
1228
1229 default:
1230 return "RAID10 can only be changed to RAID0";
1231 }
1232 case 0:
1233 /* RAID0 can be converted to RAID10, or to RAID456 */
1234 if (info->new_level == 10) {
1235 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1236 /* Assume near=2 layout */
1237 info->new_layout = 0x102;
1238 info->delta_disks = info->array.raid_disks;
1239 }
1240 if (info->new_layout == UnSet) {
1241 int copies = 1 + (info->delta_disks
1242 / info->array.raid_disks);
1243 if (info->array.raid_disks * (copies-1)
1244 != info->delta_disks)
1245 return "Impossible number of devices for RAID0->RAID10";
1246 info->new_layout = 0x100 + copies;
1247 }
1248 if (info->delta_disks == UnSet) {
1249 int copies = info->new_layout & 0xff;
1250 if (info->new_layout != 0x100 + copies)
1251 return "New layout impossible for RAID0->RAID10";;
1252 info->delta_disks = (copies - 1) *
1253 info->array.raid_disks;
1254 }
1255 if (info->new_chunk &&
1256 info->new_chunk != info->array.chunk_size)
1257 return "Cannot change chunk-size with RAID0->RAID10";
1258 /* looks good */
1259 re->level = 10;
1260 re->before.data_disks = (info->array.raid_disks +
1261 info->delta_disks);
1262 re->after.data_disks = re->before.data_disks;
1263 re->before.layout = info->new_layout;
1264 return NULL;
1265 }
1266
1267 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1268 * a raid4 style layout of the final level.
1269 */
1270 switch (info->new_level) {
1271 case 4:
1272 delta_parity = 1;
1273 case 0:
1274 re->level = 4;
1275 re->before.layout = 0;
1276 break;
1277 case 5:
1278 delta_parity = 1;
1279 re->level = 5;
1280 re->before.layout = ALGORITHM_PARITY_N;
1281 if (info->new_layout == UnSet)
1282 info->new_layout = map_name(r5layout, "default");
1283 break;
1284 case 6:
1285 delta_parity = 2;
1286 re->level = 6;
1287 re->before.layout = ALGORITHM_PARITY_N;
1288 if (info->new_layout == UnSet)
1289 info->new_layout = map_name(r6layout, "default");
1290 break;
1291 default:
1292 return "Impossible level change requested";
1293 }
1294 re->before.data_disks = info->array.raid_disks;
1295 /* determining 'after' layout happens outside this 'switch' */
1296 break;
1297
1298 case 4:
1299 info->array.layout = ALGORITHM_PARITY_N;
1300 case 5:
1301 switch (info->new_level) {
1302 case 0:
1303 delta_parity = -1;
1304 case 4:
1305 re->level = info->array.level;
1306 re->before.data_disks = info->array.raid_disks - 1;
1307 re->before.layout = info->array.layout;
1308 break;
1309 case 5:
1310 re->level = 5;
1311 re->before.data_disks = info->array.raid_disks - 1;
1312 re->before.layout = info->array.layout;
1313 break;
1314 case 6:
1315 delta_parity = 1;
1316 re->level = 6;
1317 re->before.data_disks = info->array.raid_disks - 1;
1318 switch (info->array.layout) {
1319 case ALGORITHM_LEFT_ASYMMETRIC:
1320 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1321 break;
1322 case ALGORITHM_RIGHT_ASYMMETRIC:
1323 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1324 break;
1325 case ALGORITHM_LEFT_SYMMETRIC:
1326 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1327 break;
1328 case ALGORITHM_RIGHT_SYMMETRIC:
1329 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1330 break;
1331 case ALGORITHM_PARITY_0:
1332 re->before.layout = ALGORITHM_PARITY_0_6;
1333 break;
1334 case ALGORITHM_PARITY_N:
1335 re->before.layout = ALGORITHM_PARITY_N_6;
1336 break;
1337 default:
1338 return "Cannot convert an array with this layout";
1339 }
1340 break;
1341 case 1:
1342 if (info->array.raid_disks != 2)
1343 return "Can only convert a 2-device array to RAID1";
1344 if (info->delta_disks != UnSet &&
1345 info->delta_disks != 0)
1346 return "Cannot set raid_disk when converting RAID5->RAID1";
1347 re->level = 1;
1348 info->new_chunk = 0;
1349 return NULL;
1350 default:
1351 return "Impossible level change requested";
1352 }
1353 break;
1354 case 6:
1355 switch (info->new_level) {
1356 case 4:
1357 case 5:
1358 delta_parity = -1;
1359 case 6:
1360 re->level = 6;
1361 re->before.data_disks = info->array.raid_disks - 2;
1362 re->before.layout = info->array.layout;
1363 break;
1364 default:
1365 return "Impossible level change requested";
1366 }
1367 break;
1368 }
1369
1370 /* If we reached here then it looks like a re-stripe is
1371 * happening. We have determined the intermediate level
1372 * and initial raid_disks/layout and stored these in 're'.
1373 *
1374 * We need to deduce the final layout that can be atomically
1375 * converted to the end state.
1376 */
1377 switch (info->new_level) {
1378 case 0:
1379 /* We can only get to RAID0 from RAID4 or RAID5
1380 * with appropriate layout and one extra device
1381 */
1382 if (re->level != 4 && re->level != 5)
1383 return "Cannot covert to RAID0 from this level";
1384
1385 switch (re->level) {
1386 case 4:
1387 re->before.layout = 0;
1388 re->after.layout = 0;
1389 break;
1390 case 5:
1391 re->after.layout = ALGORITHM_PARITY_N;
1392 break;
1393 }
1394 break;
1395
1396 case 4:
1397 /* We can only get to RAID4 from RAID5 */
1398 if (re->level != 4 && re->level != 5)
1399 return "Cannot convert to RAID4 from this level";
1400
1401 switch (re->level) {
1402 case 4:
1403 re->after.layout = 0;
1404 break;
1405 case 5:
1406 re->after.layout = ALGORITHM_PARITY_N;
1407 break;
1408 }
1409 break;
1410
1411 case 5:
1412 /* We get to RAID5 from RAID5 or RAID6 */
1413 if (re->level != 5 && re->level != 6)
1414 return "Cannot convert to RAID5 from this level";
1415
1416 switch (re->level) {
1417 case 5:
1418 if (info->new_layout == UnSet)
1419 re->after.layout = re->before.layout;
1420 else
1421 re->after.layout = info->new_layout;
1422 break;
1423 case 6:
1424 if (info->new_layout == UnSet)
1425 info->new_layout = re->before.layout;
1426
1427 /* after.layout needs to be raid6 version of new_layout */
1428 if (info->new_layout == ALGORITHM_PARITY_N)
1429 re->after.layout = ALGORITHM_PARITY_N;
1430 else {
1431 char layout[40];
1432 char *ls = map_num(r5layout, info->new_layout);
1433 int l;
1434 if (ls) {
1435 /* Current RAID6 layout has a RAID5
1436 * equivalent - good
1437 */
1438 strcat(strcpy(layout, ls), "-6");
1439 l = map_name(r6layout, layout);
1440 if (l == UnSet)
1441 return "Cannot find RAID6 layout to convert to";
1442 } else {
1443 /* Current RAID6 has no equivalent.
1444 * If it is already a '-6' layout we
1445 * can leave it unchanged, else we must
1446 * fail
1447 */
1448 ls = map_num(r6layout, info->new_layout);
1449 if (!ls ||
1450 strcmp(ls+strlen(ls)-2, "-6") != 0)
1451 return "Please specify new layout";
1452 l = info->new_layout;
1453 }
1454 re->after.layout = l;
1455 }
1456 }
1457 break;
1458
1459 case 6:
1460 /* We must already be at level 6 */
1461 if (re->level != 6)
1462 return "Impossible level change";
1463 if (info->new_layout == UnSet)
1464 re->after.layout = info->array.layout;
1465 else
1466 re->after.layout = info->new_layout;
1467 break;
1468 default:
1469 return "Impossible level change requested";
1470 }
1471 if (info->delta_disks == UnSet)
1472 info->delta_disks = delta_parity;
1473
1474 re->after.data_disks = (re->before.data_disks
1475 + info->delta_disks
1476 - delta_parity);
1477 switch (re->level) {
1478 case 6: re->parity = 2;
1479 break;
1480 case 4:
1481 case 5: re->parity = 1;
1482 break;
1483 default: re->parity = 0;
1484 break;
1485 }
1486 /* So we have a restripe operation, we need to calculate the number
1487 * of blocks per reshape operation.
1488 */
1489 re->new_size = info->component_size * re->before.data_disks;
1490 if (info->new_chunk == 0)
1491 info->new_chunk = info->array.chunk_size;
1492 if (re->after.data_disks == re->before.data_disks &&
1493 re->after.layout == re->before.layout &&
1494 info->new_chunk == info->array.chunk_size) {
1495 /* Nothing to change, can change level immediately. */
1496 re->level = info->new_level;
1497 re->backup_blocks = 0;
1498 return NULL;
1499 }
1500 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1501 /* chunk and layout changes make no difference */
1502 re->level = info->new_level;
1503 re->backup_blocks = 0;
1504 return NULL;
1505 }
1506
1507 if (re->after.data_disks == re->before.data_disks &&
1508 get_linux_version() < 2006032)
1509 return "in-place reshape is not safe before 2.6.32 - sorry.";
1510
1511 if (re->after.data_disks < re->before.data_disks &&
1512 get_linux_version() < 2006030)
1513 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1514
1515 re->backup_blocks = compute_backup_blocks(
1516 info->new_chunk, info->array.chunk_size,
1517 re->after.data_disks,
1518 re->before.data_disks);
1519 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1520
1521 re->new_size = info->component_size * re->after.data_disks;
1522 return NULL;
1523 }
1524
1525 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1526 char *text_version)
1527 {
1528 struct mdinfo *info;
1529 char *subarray;
1530 int ret_val = -1;
1531
1532 if ((st == NULL) || (sra == NULL))
1533 return ret_val;
1534
1535 if (text_version == NULL)
1536 text_version = sra->text_version;
1537 subarray = strchr(text_version+1, '/')+1;
1538 info = st->ss->container_content(st, subarray);
1539 if (info) {
1540 unsigned long long current_size = 0;
1541 unsigned long long new_size =
1542 info->custom_array_size/2;
1543
1544 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1545 new_size > current_size) {
1546 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1547 < 0)
1548 dprintf("Error: Cannot set array size");
1549 else {
1550 ret_val = 0;
1551 dprintf("Array size changed");
1552 }
1553 dprintf_cont(" from %llu to %llu.\n",
1554 current_size, new_size);
1555 }
1556 sysfs_free(info);
1557 } else
1558 dprintf("Error: set_array_size(): info pointer in NULL\n");
1559
1560 return ret_val;
1561 }
1562
1563 static int reshape_array(char *container, int fd, char *devname,
1564 struct supertype *st, struct mdinfo *info,
1565 int force, struct mddev_dev *devlist,
1566 unsigned long long data_offset,
1567 char *backup_file, int verbose, int forked,
1568 int restart, int freeze_reshape);
1569 static int reshape_container(char *container, char *devname,
1570 int mdfd,
1571 struct supertype *st,
1572 struct mdinfo *info,
1573 int force,
1574 char *backup_file, int verbose,
1575 int forked, int restart, int freeze_reshape);
1576
1577 int Grow_reshape(char *devname, int fd,
1578 struct mddev_dev *devlist,
1579 unsigned long long data_offset,
1580 struct context *c, struct shape *s)
1581 {
1582 /* Make some changes in the shape of an array.
1583 * The kernel must support the change.
1584 *
1585 * There are three different changes. Each can trigger
1586 * a resync or recovery so we freeze that until we have
1587 * requested everything (if kernel supports freezing - 2.6.30).
1588 * The steps are:
1589 * - change size (i.e. component_size)
1590 * - change level
1591 * - change layout/chunksize/ndisks
1592 *
1593 * The last can require a reshape. It is different on different
1594 * levels so we need to check the level before actioning it.
1595 * Some times the level change needs to be requested after the
1596 * reshape (e.g. raid6->raid5, raid5->raid0)
1597 *
1598 */
1599 struct mdu_array_info_s array;
1600 int rv = 0;
1601 struct supertype *st;
1602 char *subarray = NULL;
1603
1604 int frozen;
1605 int changed = 0;
1606 char *container = NULL;
1607 int cfd = -1;
1608
1609 struct mddev_dev *dv;
1610 int added_disks;
1611
1612 struct mdinfo info;
1613 struct mdinfo *sra;
1614
1615 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1616 pr_err("%s is not an active md array - aborting\n",
1617 devname);
1618 return 1;
1619 }
1620 if (data_offset != INVALID_SECTORS && array.level != 10
1621 && (array.level < 4 || array.level > 6)) {
1622 pr_err("--grow --data-offset not yet supported\n");
1623 return 1;
1624 }
1625
1626 if (s->size > 0 &&
1627 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1628 pr_err("cannot change component size at the same time as other changes.\n"
1629 " Change size first, then check data is intact before making other changes.\n");
1630 return 1;
1631 }
1632
1633 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1634 get_linux_version() < 2006032 &&
1635 !check_env("MDADM_FORCE_FEWER")) {
1636 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1637 " Please use a newer kernel\n");
1638 return 1;
1639 }
1640
1641 st = super_by_fd(fd, &subarray);
1642 if (!st) {
1643 pr_err("Unable to determine metadata format for %s\n", devname);
1644 return 1;
1645 }
1646 if (s->raiddisks > st->max_devs) {
1647 pr_err("Cannot increase raid-disks on this array beyond %d\n", st->max_devs);
1648 return 1;
1649 }
1650 if (s->level == 0 &&
1651 (array.state & (1<<MD_SB_BITMAP_PRESENT)) &&
1652 !(array.state & (1<<MD_SB_CLUSTERED))) {
1653 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
1654 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
1655 pr_err("failed to remove internal bitmap.\n");
1656 return 1;
1657 }
1658 }
1659
1660 /* in the external case we need to check that the requested reshape is
1661 * supported, and perform an initial check that the container holds the
1662 * pre-requisite spare devices (mdmon owns final validation)
1663 */
1664 if (st->ss->external) {
1665 int rv;
1666
1667 if (subarray) {
1668 container = st->container_devnm;
1669 cfd = open_dev_excl(st->container_devnm);
1670 } else {
1671 container = st->devnm;
1672 close(fd);
1673 cfd = open_dev_excl(st->devnm);
1674 fd = cfd;
1675 }
1676 if (cfd < 0) {
1677 pr_err("Unable to open container for %s\n",
1678 devname);
1679 free(subarray);
1680 return 1;
1681 }
1682
1683 rv = st->ss->load_container(st, cfd, NULL);
1684
1685 if (rv) {
1686 pr_err("Cannot read superblock for %s\n",
1687 devname);
1688 free(subarray);
1689 return 1;
1690 }
1691
1692 /* check if operation is supported for metadata handler */
1693 if (st->ss->container_content) {
1694 struct mdinfo *cc = NULL;
1695 struct mdinfo *content = NULL;
1696
1697 cc = st->ss->container_content(st, subarray);
1698 for (content = cc; content ; content = content->next) {
1699 int allow_reshape = 1;
1700
1701 /* check if reshape is allowed based on metadata
1702 * indications stored in content.array.status
1703 */
1704 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1705 allow_reshape = 0;
1706 if (content->array.state
1707 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1708 allow_reshape = 0;
1709 if (!allow_reshape) {
1710 pr_err("cannot reshape arrays in container with unsupported metadata: %s(%s)\n",
1711 devname, container);
1712 sysfs_free(cc);
1713 free(subarray);
1714 return 1;
1715 }
1716 }
1717 sysfs_free(cc);
1718 }
1719 if (mdmon_running(container))
1720 st->update_tail = &st->updates;
1721 }
1722
1723 added_disks = 0;
1724 for (dv = devlist; dv; dv = dv->next)
1725 added_disks++;
1726 if (s->raiddisks > array.raid_disks &&
1727 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1728 !c->force) {
1729 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
1730 " Use --force to over-ride this check.\n",
1731 s->raiddisks - array.raid_disks,
1732 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1733 array.spare_disks + added_disks);
1734 return 1;
1735 }
1736
1737 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1738 | GET_STATE | GET_VERSION);
1739 if (sra) {
1740 if (st->ss->external && subarray == NULL) {
1741 array.level = LEVEL_CONTAINER;
1742 sra->array.level = LEVEL_CONTAINER;
1743 }
1744 } else {
1745 pr_err("failed to read sysfs parameters for %s\n",
1746 devname);
1747 return 1;
1748 }
1749 frozen = freeze(st);
1750 if (frozen < -1) {
1751 /* freeze() already spewed the reason */
1752 sysfs_free(sra);
1753 return 1;
1754 } else if (frozen < 0) {
1755 pr_err("%s is performing resync/recovery and cannot be reshaped\n", devname);
1756 sysfs_free(sra);
1757 return 1;
1758 }
1759
1760 /* ========= set size =============== */
1761 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1762 unsigned long long orig_size = get_component_size(fd)/2;
1763 unsigned long long min_csize;
1764 struct mdinfo *mdi;
1765 int raid0_takeover = 0;
1766
1767 if (orig_size == 0)
1768 orig_size = (unsigned) array.size;
1769
1770 if (orig_size == 0) {
1771 pr_err("Cannot set device size in this type of array.\n");
1772 rv = 1;
1773 goto release;
1774 }
1775
1776 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1777 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1778 rv = 1;
1779 goto release;
1780 }
1781 sync_metadata(st);
1782 if (st->ss->external) {
1783 /* metadata can have size limitation
1784 * update size value according to metadata information
1785 */
1786 struct mdinfo *sizeinfo =
1787 st->ss->container_content(st, subarray);
1788 if (sizeinfo) {
1789 unsigned long long new_size =
1790 sizeinfo->custom_array_size/2;
1791 int data_disks = get_data_disks(
1792 sizeinfo->array.level,
1793 sizeinfo->array.layout,
1794 sizeinfo->array.raid_disks);
1795 new_size /= data_disks;
1796 dprintf("Metadata size correction from %llu to %llu (%llu)\n", orig_size, new_size,
1797 new_size * data_disks);
1798 s->size = new_size;
1799 sysfs_free(sizeinfo);
1800 }
1801 }
1802
1803 /* Update the size of each member device in case
1804 * they have been resized. This will never reduce
1805 * below the current used-size. The "size" attribute
1806 * understands '0' to mean 'max'.
1807 */
1808 min_csize = 0;
1809 rv = 0;
1810 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1811 if (sysfs_set_num(sra, mdi, "size",
1812 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1813 /* Probably kernel refusing to let us
1814 * reduce the size - not an error.
1815 */
1816 break;
1817 }
1818 if (array.not_persistent == 0 &&
1819 array.major_version == 0 &&
1820 get_linux_version() < 3001000) {
1821 /* Dangerous to allow size to exceed 2TB */
1822 unsigned long long csize;
1823 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1824 if (csize >= 2ULL*1024*1024*1024)
1825 csize = 2ULL*1024*1024*1024;
1826 if ((min_csize == 0 || (min_csize
1827 > csize)))
1828 min_csize = csize;
1829 }
1830 }
1831 }
1832 if (rv) {
1833 pr_err("Cannot set size on array members.\n");
1834 goto size_change_error;
1835 }
1836 if (min_csize && s->size > min_csize) {
1837 pr_err("Cannot safely make this array use more than 2TB per device on this kernel.\n");
1838 rv = 1;
1839 goto size_change_error;
1840 }
1841 if (min_csize && s->size == MAX_SIZE) {
1842 /* Don't let the kernel choose a size - it will get
1843 * it wrong
1844 */
1845 pr_err("Limited v0.90 array to 2TB per device\n");
1846 s->size = min_csize;
1847 }
1848 if (st->ss->external) {
1849 if (sra->array.level == 0) {
1850 rv = sysfs_set_str(sra, NULL, "level",
1851 "raid5");
1852 if (!rv) {
1853 raid0_takeover = 1;
1854 /* get array parameters after takeover
1855 * to change one parameter at time only
1856 */
1857 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1858 }
1859 }
1860 /* make sure mdmon is
1861 * aware of the new level */
1862 if (!mdmon_running(st->container_devnm))
1863 start_mdmon(st->container_devnm);
1864 ping_monitor(container);
1865 if (mdmon_running(st->container_devnm) &&
1866 st->update_tail == NULL)
1867 st->update_tail = &st->updates;
1868 }
1869
1870 if (s->size == MAX_SIZE)
1871 s->size = 0;
1872 array.size = s->size;
1873 if (s->size & ~INT32_MAX) {
1874 /* got truncated to 32bit, write to
1875 * component_size instead
1876 */
1877 if (sra)
1878 rv = sysfs_set_num(sra, NULL,
1879 "component_size", s->size);
1880 else
1881 rv = -1;
1882 } else {
1883 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1884
1885 /* manage array size when it is managed externally
1886 */
1887 if ((rv == 0) && st->ss->external)
1888 rv = set_array_size(st, sra, sra->text_version);
1889 }
1890
1891 if (raid0_takeover) {
1892 /* do not recync non-existing parity,
1893 * we will drop it anyway
1894 */
1895 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1896 /* go back to raid0, drop parity disk
1897 */
1898 sysfs_set_str(sra, NULL, "level", "raid0");
1899 ioctl(fd, GET_ARRAY_INFO, &array);
1900 }
1901
1902 size_change_error:
1903 if (rv != 0) {
1904 int err = errno;
1905
1906 /* restore metadata */
1907 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1908 UnSet, NULL, devname,
1909 ROLLBACK_METADATA_CHANGES,
1910 c->verbose) == 0)
1911 sync_metadata(st);
1912 pr_err("Cannot set device size for %s: %s\n",
1913 devname, strerror(err));
1914 if (err == EBUSY &&
1915 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1916 cont_err("Bitmap must be removed before size can be changed\n");
1917 rv = 1;
1918 goto release;
1919 }
1920 if (s->assume_clean) {
1921 /* This will fail on kernels older than 3.0 unless
1922 * a backport has been arranged.
1923 */
1924 if (sra == NULL ||
1925 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1926 pr_err("--assume-clean not supported with --grow on this kernel\n");
1927 }
1928 ioctl(fd, GET_ARRAY_INFO, &array);
1929 s->size = get_component_size(fd)/2;
1930 if (s->size == 0)
1931 s->size = array.size;
1932 if (c->verbose >= 0) {
1933 if (s->size == orig_size)
1934 pr_err("component size of %s unchanged at %lluK\n",
1935 devname, s->size);
1936 else
1937 pr_err("component size of %s has been set to %lluK\n",
1938 devname, s->size);
1939 }
1940 changed = 1;
1941 } else if (array.level != LEVEL_CONTAINER) {
1942 s->size = get_component_size(fd)/2;
1943 if (s->size == 0)
1944 s->size = array.size;
1945 }
1946
1947 /* See if there is anything else to do */
1948 if ((s->level == UnSet || s->level == array.level) &&
1949 (s->layout_str == NULL) &&
1950 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1951 data_offset == INVALID_SECTORS &&
1952 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1953 /* Nothing more to do */
1954 if (!changed && c->verbose >= 0)
1955 pr_err("%s: no change requested\n",
1956 devname);
1957 goto release;
1958 }
1959
1960 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1961 * current implementation assumes that following conditions must be met:
1962 * - RAID10:
1963 * - far_copies == 1
1964 * - near_copies == 2
1965 */
1966 if ((s->level == 0 && array.level == 10 && sra &&
1967 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1968 (s->level == 0 && array.level == 1 && sra)) {
1969 int err;
1970 err = remove_disks_for_takeover(st, sra, array.layout);
1971 if (err) {
1972 dprintf("Array cannot be reshaped\n");
1973 if (cfd > -1)
1974 close(cfd);
1975 rv = 1;
1976 goto release;
1977 }
1978 /* Make sure mdmon has seen the device removal
1979 * and updated metadata before we continue with
1980 * level change
1981 */
1982 if (container)
1983 ping_monitor(container);
1984 }
1985
1986 memset(&info, 0, sizeof(info));
1987 info.array = array;
1988 sysfs_init(&info, fd, NULL);
1989 strcpy(info.text_version, sra->text_version);
1990 info.component_size = s->size*2;
1991 info.new_level = s->level;
1992 info.new_chunk = s->chunk * 1024;
1993 if (info.array.level == LEVEL_CONTAINER) {
1994 info.delta_disks = UnSet;
1995 info.array.raid_disks = s->raiddisks;
1996 } else if (s->raiddisks)
1997 info.delta_disks = s->raiddisks - info.array.raid_disks;
1998 else
1999 info.delta_disks = UnSet;
2000 if (s->layout_str == NULL) {
2001 info.new_layout = UnSet;
2002 if (info.array.level == 6 &&
2003 (info.new_level == 6 || info.new_level == UnSet) &&
2004 info.array.layout >= 16) {
2005 pr_err("%s has a non-standard layout. If you wish to preserve this\n", devname);
2006 cont_err("during the reshape, please specify --layout=preserve\n");
2007 cont_err("If you want to change it, specify a layout or use --layout=normalise\n");
2008 rv = 1;
2009 goto release;
2010 }
2011 } else if (strcmp(s->layout_str, "normalise") == 0 ||
2012 strcmp(s->layout_str, "normalize") == 0) {
2013 /* If we have a -6 RAID6 layout, remove the '-6'. */
2014 info.new_layout = UnSet;
2015 if (info.array.level == 6 && info.new_level == UnSet) {
2016 char l[40], *h;
2017 strcpy(l, map_num(r6layout, info.array.layout));
2018 h = strrchr(l, '-');
2019 if (h && strcmp(h, "-6") == 0) {
2020 *h = 0;
2021 info.new_layout = map_name(r6layout, l);
2022 }
2023 } else {
2024 pr_err("%s is only meaningful when reshaping a RAID6 array.\n", s->layout_str);
2025 rv = 1;
2026 goto release;
2027 }
2028 } else if (strcmp(s->layout_str, "preserve") == 0) {
2029 /* This means that a non-standard RAID6 layout
2030 * is OK.
2031 * In particular:
2032 * - When reshape a RAID6 (e.g. adding a device)
2033 * which is in a non-standard layout, it is OK
2034 * to preserve that layout.
2035 * - When converting a RAID5 to RAID6, leave it in
2036 * the XXX-6 layout, don't re-layout.
2037 */
2038 if (info.array.level == 6 && info.new_level == UnSet)
2039 info.new_layout = info.array.layout;
2040 else if (info.array.level == 5 && info.new_level == 6) {
2041 char l[40];
2042 strcpy(l, map_num(r5layout, info.array.layout));
2043 strcat(l, "-6");
2044 info.new_layout = map_name(r6layout, l);
2045 } else {
2046 pr_err("%s in only meaningful when reshaping to RAID6\n", s->layout_str);
2047 rv = 1;
2048 goto release;
2049 }
2050 } else {
2051 int l = info.new_level;
2052 if (l == UnSet)
2053 l = info.array.level;
2054 switch (l) {
2055 case 5:
2056 info.new_layout = map_name(r5layout, s->layout_str);
2057 break;
2058 case 6:
2059 info.new_layout = map_name(r6layout, s->layout_str);
2060 break;
2061 case 10:
2062 info.new_layout = parse_layout_10(s->layout_str);
2063 break;
2064 case LEVEL_FAULTY:
2065 info.new_layout = parse_layout_faulty(s->layout_str);
2066 break;
2067 default:
2068 pr_err("layout not meaningful with this level\n");
2069 rv = 1;
2070 goto release;
2071 }
2072 if (info.new_layout == UnSet) {
2073 pr_err("layout %s not understood for this level\n",
2074 s->layout_str);
2075 rv = 1;
2076 goto release;
2077 }
2078 }
2079
2080 if (array.level == LEVEL_FAULTY) {
2081 if (s->level != UnSet && s->level != array.level) {
2082 pr_err("cannot change level of Faulty device\n");
2083 rv =1 ;
2084 }
2085 if (s->chunk) {
2086 pr_err("cannot set chunksize of Faulty device\n");
2087 rv =1 ;
2088 }
2089 if (s->raiddisks && s->raiddisks != 1) {
2090 pr_err("cannot set raid_disks of Faulty device\n");
2091 rv =1 ;
2092 }
2093 if (s->layout_str) {
2094 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2095 dprintf("Cannot get array information.\n");
2096 goto release;
2097 }
2098 array.layout = info.new_layout;
2099 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2100 pr_err("failed to set new layout\n");
2101 rv = 1;
2102 } else if (c->verbose >= 0)
2103 printf("layout for %s set to %d\n",
2104 devname, array.layout);
2105 }
2106 } else if (array.level == LEVEL_CONTAINER) {
2107 /* This change is to be applied to every array in the
2108 * container. This is only needed when the metadata imposes
2109 * restraints of the various arrays in the container.
2110 * Currently we only know that IMSM requires all arrays
2111 * to have the same number of devices so changing the
2112 * number of devices (On-Line Capacity Expansion) must be
2113 * performed at the level of the container
2114 */
2115 if (fd > 0) {
2116 close(fd);
2117 fd = -1;
2118 }
2119 rv = reshape_container(container, devname, -1, st, &info,
2120 c->force, c->backup_file, c->verbose, 0, 0, 0);
2121 frozen = 0;
2122 } else {
2123 /* get spare devices from external metadata
2124 */
2125 if (st->ss->external) {
2126 struct mdinfo *info2;
2127
2128 info2 = st->ss->container_content(st, subarray);
2129 if (info2) {
2130 info.array.spare_disks =
2131 info2->array.spare_disks;
2132 sysfs_free(info2);
2133 }
2134 }
2135
2136 /* Impose these changes on a single array. First
2137 * check that the metadata is OK with the change. */
2138
2139 if (reshape_super(st, 0, info.new_level,
2140 info.new_layout, info.new_chunk,
2141 info.array.raid_disks, info.delta_disks,
2142 c->backup_file, devname, APPLY_METADATA_CHANGES,
2143 c->verbose)) {
2144 rv = 1;
2145 goto release;
2146 }
2147 sync_metadata(st);
2148 rv = reshape_array(container, fd, devname, st, &info, c->force,
2149 devlist, data_offset, c->backup_file, c->verbose,
2150 0, 0, 0);
2151 frozen = 0;
2152 }
2153 release:
2154 sysfs_free(sra);
2155 if (frozen > 0)
2156 unfreeze(st);
2157 return rv;
2158 }
2159
2160 /* verify_reshape_position()
2161 * Function checks if reshape position in metadata is not farther
2162 * than position in md.
2163 * Return value:
2164 * 0 : not valid sysfs entry
2165 * it can be caused by not started reshape, it should be started
2166 * by reshape array or raid0 array is before takeover
2167 * -1 : error, reshape position is obviously wrong
2168 * 1 : success, reshape progress correct or updated
2169 */
2170 static int verify_reshape_position(struct mdinfo *info, int level)
2171 {
2172 int ret_val = 0;
2173 char buf[40];
2174 int rv;
2175
2176 /* read sync_max, failure can mean raid0 array */
2177 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2178
2179 if (rv > 0) {
2180 char *ep;
2181 unsigned long long position = strtoull(buf, &ep, 0);
2182
2183 dprintf("Read sync_max sysfs entry is: %s\n", buf);
2184 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2185 position *= get_data_disks(level,
2186 info->new_layout,
2187 info->array.raid_disks);
2188 if (info->reshape_progress < position) {
2189 dprintf("Corrected reshape progress (%llu) to md position (%llu)\n",
2190 info->reshape_progress, position);
2191 info->reshape_progress = position;
2192 ret_val = 1;
2193 } else if (info->reshape_progress > position) {
2194 pr_err("Fatal error: array reshape was not properly frozen (expected reshape position is %llu, but reshape progress is %llu.\n",
2195 position, info->reshape_progress);
2196 ret_val = -1;
2197 } else {
2198 dprintf("Reshape position in md and metadata are the same;");
2199 ret_val = 1;
2200 }
2201 }
2202 } else if (rv == 0) {
2203 /* for valid sysfs entry, 0-length content
2204 * should be indicated as error
2205 */
2206 ret_val = -1;
2207 }
2208
2209 return ret_val;
2210 }
2211
2212 static unsigned long long choose_offset(unsigned long long lo,
2213 unsigned long long hi,
2214 unsigned long long min,
2215 unsigned long long max)
2216 {
2217 /* Choose a new offset between hi and lo.
2218 * It must be between min and max, but
2219 * we would prefer something near the middle of hi/lo, and also
2220 * prefer to be aligned to a big power of 2.
2221 *
2222 * So we start with the middle, then for each bit,
2223 * starting at '1' and increasing, if it is set, we either
2224 * add it or subtract it if possible, preferring the option
2225 * which is furthest from the boundary.
2226 *
2227 * We stop once we get a 1MB alignment. As units are in sectors,
2228 * 1MB = 2*1024 sectors.
2229 */
2230 unsigned long long choice = (lo + hi) / 2;
2231 unsigned long long bit = 1;
2232
2233 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2234 unsigned long long bigger, smaller;
2235 if (! (bit & choice))
2236 continue;
2237 bigger = choice + bit;
2238 smaller = choice - bit;
2239 if (bigger > max && smaller < min)
2240 break;
2241 if (bigger > max)
2242 choice = smaller;
2243 else if (smaller < min)
2244 choice = bigger;
2245 else if (hi - bigger > smaller - lo)
2246 choice = bigger;
2247 else
2248 choice = smaller;
2249 }
2250 return choice;
2251 }
2252
2253 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2254 char *devname, int delta_disks,
2255 unsigned long long data_offset,
2256 unsigned long long min,
2257 int can_fallback)
2258 {
2259 struct mdinfo *sd;
2260 int dir = 0;
2261 int err = 0;
2262 unsigned long long before, after;
2263
2264 /* Need to find min space before and after so same is used
2265 * on all devices
2266 */
2267 before = UINT64_MAX;
2268 after = UINT64_MAX;
2269 for (sd = sra->devs; sd; sd = sd->next) {
2270 char *dn;
2271 int dfd;
2272 int rv;
2273 struct supertype *st2;
2274 struct mdinfo info2;
2275
2276 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2277 continue;
2278 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2279 dfd = dev_open(dn, O_RDONLY);
2280 if (dfd < 0) {
2281 pr_err("%s: cannot open component %s\n",
2282 devname, dn ? dn : "-unknown-");
2283 goto release;
2284 }
2285 st2 = dup_super(st);
2286 rv = st2->ss->load_super(st2,dfd, NULL);
2287 close(dfd);
2288 if (rv) {
2289 free(st2);
2290 pr_err("%s: cannot get superblock from %s\n",
2291 devname, dn);
2292 goto release;
2293 }
2294 st2->ss->getinfo_super(st2, &info2, NULL);
2295 st2->ss->free_super(st2);
2296 free(st2);
2297 if (info2.space_before == 0 &&
2298 info2.space_after == 0) {
2299 /* Metadata doesn't support data_offset changes */
2300 if (!can_fallback)
2301 pr_err("%s: Metadata version doesn't support data_offset changes\n",
2302 devname);
2303 goto fallback;
2304 }
2305 if (before > info2.space_before)
2306 before = info2.space_before;
2307 if (after > info2.space_after)
2308 after = info2.space_after;
2309
2310 if (data_offset != INVALID_SECTORS) {
2311 if (dir == 0) {
2312 if (info2.data_offset == data_offset) {
2313 pr_err("%s: already has that data_offset\n",
2314 dn);
2315 goto release;
2316 }
2317 if (data_offset < info2.data_offset)
2318 dir = -1;
2319 else
2320 dir = 1;
2321 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2322 (data_offset >= info2.data_offset && dir == -1)) {
2323 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2324 dn);
2325 goto release;
2326 }
2327 }
2328 }
2329 if (before == UINT64_MAX)
2330 /* impossible really, there must be no devices */
2331 return 1;
2332
2333 for (sd = sra->devs; sd; sd = sd->next) {
2334 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2335 unsigned long long new_data_offset;
2336
2337 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2338 continue;
2339 if (delta_disks < 0) {
2340 /* Don't need any space as array is shrinking
2341 * just move data_offset up by min
2342 */
2343 if (data_offset == INVALID_SECTORS)
2344 new_data_offset = sd->data_offset + min;
2345 else {
2346 if (data_offset < sd->data_offset + min) {
2347 pr_err("--data-offset too small for %s\n",
2348 dn);
2349 goto release;
2350 }
2351 new_data_offset = data_offset;
2352 }
2353 } else if (delta_disks > 0) {
2354 /* need space before */
2355 if (before < min) {
2356 if (can_fallback)
2357 goto fallback;
2358 pr_err("Insufficient head-space for reshape on %s\n",
2359 dn);
2360 goto release;
2361 }
2362 if (data_offset == INVALID_SECTORS)
2363 new_data_offset = sd->data_offset - min;
2364 else {
2365 if (data_offset > sd->data_offset - min) {
2366 pr_err("--data-offset too large for %s\n",
2367 dn);
2368 goto release;
2369 }
2370 new_data_offset = data_offset;
2371 }
2372 } else {
2373 if (dir == 0) {
2374 /* can move up or down. If 'data_offset'
2375 * was set we would have already decided,
2376 * so just choose direction with most space.
2377 */
2378 if (before > after)
2379 dir = -1;
2380 else
2381 dir = 1;
2382 }
2383 sysfs_set_str(sra, NULL, "reshape_direction",
2384 dir == 1 ? "backwards" : "forwards");
2385 if (dir > 0) {
2386 /* Increase data offset */
2387 if (after < min) {
2388 if (can_fallback)
2389 goto fallback;
2390 pr_err("Insufficient tail-space for reshape on %s\n",
2391 dn);
2392 goto release;
2393 }
2394 if (data_offset != INVALID_SECTORS &&
2395 data_offset < sd->data_offset + min) {
2396 pr_err("--data-offset too small on %s\n",
2397 dn);
2398 goto release;
2399 }
2400 if (data_offset != INVALID_SECTORS)
2401 new_data_offset = data_offset;
2402 else
2403 new_data_offset = choose_offset(sd->data_offset,
2404 sd->data_offset + after,
2405 sd->data_offset + min,
2406 sd->data_offset + after);
2407 } else {
2408 /* Decrease data offset */
2409 if (before < min) {
2410 if (can_fallback)
2411 goto fallback;
2412 pr_err("insufficient head-room on %s\n",
2413 dn);
2414 goto release;
2415 }
2416 if (data_offset != INVALID_SECTORS &&
2417 data_offset < sd->data_offset - min) {
2418 pr_err("--data-offset too small on %s\n",
2419 dn);
2420 goto release;
2421 }
2422 if (data_offset != INVALID_SECTORS)
2423 new_data_offset = data_offset;
2424 else
2425 new_data_offset = choose_offset(sd->data_offset - before,
2426 sd->data_offset,
2427 sd->data_offset - before,
2428 sd->data_offset - min);
2429 }
2430 }
2431 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2432 if (err < 0 && errno == E2BIG) {
2433 /* try again after increasing data size to max */
2434 err = sysfs_set_num(sra, sd, "size", 0);
2435 if (err < 0 && errno == EINVAL &&
2436 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2437 /* some kernels have a bug where you cannot
2438 * use '0' on spare devices. */
2439 sysfs_set_num(sra, sd, "size",
2440 (sra->component_size + after)/2);
2441 }
2442 err = sysfs_set_num(sra, sd, "new_offset",
2443 new_data_offset);
2444 }
2445 if (err < 0) {
2446 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2447 pr_err("data-offset is too big for %s\n",
2448 dn);
2449 goto release;
2450 }
2451 if (sd == sra->devs &&
2452 (errno == ENOENT || errno == E2BIG))
2453 /* Early kernel, no 'new_offset' file,
2454 * or kernel doesn't like us.
2455 * For RAID5/6 this is not fatal
2456 */
2457 return 1;
2458 pr_err("Cannot set new_offset for %s\n",
2459 dn);
2460 break;
2461 }
2462 }
2463 return err;
2464 release:
2465 return -1;
2466 fallback:
2467 /* Just use a backup file */
2468 return 1;
2469 }
2470
2471 static int raid10_reshape(char *container, int fd, char *devname,
2472 struct supertype *st, struct mdinfo *info,
2473 struct reshape *reshape,
2474 unsigned long long data_offset,
2475 int force, int verbose)
2476 {
2477 /* Changing raid_disks, layout, chunksize or possibly
2478 * just data_offset for a RAID10.
2479 * We must always change data_offset. We change by at least
2480 * ->min_offset_change which is the largest of the old and new
2481 * chunk sizes.
2482 * If raid_disks is increasing, then data_offset must decrease
2483 * by at least this copy size.
2484 * If raid_disks is unchanged, data_offset must increase or
2485 * decrease by at least min_offset_change but preferably by much more.
2486 * We choose half of the available space.
2487 * If raid_disks is decreasing, data_offset must increase by
2488 * at least min_offset_change. To allow of this, component_size
2489 * must be decreased by the same amount.
2490 *
2491 * So we calculate the required minimum and direction, possibly
2492 * reduce the component_size, then iterate through the devices
2493 * and set the new_data_offset.
2494 * If that all works, we set chunk_size, layout, raid_disks, and start
2495 * 'reshape'
2496 */
2497 struct mdinfo *sra;
2498 unsigned long long min;
2499 int err = 0;
2500
2501 sra = sysfs_read(fd, NULL,
2502 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2503 );
2504 if (!sra) {
2505 pr_err("%s: Cannot get array details from sysfs\n",
2506 devname);
2507 goto release;
2508 }
2509 min = reshape->min_offset_change;
2510
2511 if (info->delta_disks)
2512 sysfs_set_str(sra, NULL, "reshape_direction",
2513 info->delta_disks < 0 ? "backwards" : "forwards");
2514 if (info->delta_disks < 0 &&
2515 info->space_after < min) {
2516 int rv = sysfs_set_num(sra, NULL, "component_size",
2517 (sra->component_size -
2518 min)/2);
2519 if (rv) {
2520 pr_err("cannot reduce component size\n");
2521 goto release;
2522 }
2523 }
2524 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2525 min, 0);
2526 if (err == 1) {
2527 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2528 cont_err("supported on this kernel\n");
2529 err = -1;
2530 }
2531 if (err < 0)
2532 goto release;
2533
2534 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2535 err = errno;
2536 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2537 err = errno;
2538 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2539 info->array.raid_disks + info->delta_disks) < 0)
2540 err = errno;
2541 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2542 err = errno;
2543 if (err) {
2544 pr_err("Cannot set array shape for %s\n",
2545 devname);
2546 if (err == EBUSY &&
2547 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2548 cont_err(" Bitmap must be removed before shape can be changed\n");
2549 goto release;
2550 }
2551 sysfs_free(sra);
2552 return 0;
2553 release:
2554 sysfs_free(sra);
2555 return 1;
2556 }
2557
2558 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2559 {
2560 struct mdinfo *sra, *sd;
2561 /* Initialisation to silence compiler warning */
2562 unsigned long long min_space_before = 0, min_space_after = 0;
2563 int first = 1;
2564
2565 sra = sysfs_read(fd, NULL, GET_DEVS);
2566 if (!sra)
2567 return;
2568 for (sd = sra->devs; sd; sd = sd->next) {
2569 char *dn;
2570 int dfd;
2571 struct supertype *st2;
2572 struct mdinfo info2;
2573
2574 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2575 continue;
2576 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2577 dfd = dev_open(dn, O_RDONLY);
2578 if (dfd < 0)
2579 break;
2580 st2 = dup_super(st);
2581 if (st2->ss->load_super(st2,dfd, NULL)) {
2582 close(dfd);
2583 free(st2);
2584 break;
2585 }
2586 close(dfd);
2587 st2->ss->getinfo_super(st2, &info2, NULL);
2588 st2->ss->free_super(st2);
2589 free(st2);
2590 if (first ||
2591 min_space_before > info2.space_before)
2592 min_space_before = info2.space_before;
2593 if (first ||
2594 min_space_after > info2.space_after)
2595 min_space_after = info2.space_after;
2596 first = 0;
2597 }
2598 if (sd == NULL && !first) {
2599 info->space_after = min_space_after;
2600 info->space_before = min_space_before;
2601 }
2602 sysfs_free(sra);
2603 }
2604
2605 static void update_cache_size(char *container, struct mdinfo *sra,
2606 struct mdinfo *info,
2607 int disks, unsigned long long blocks)
2608 {
2609 /* Check that the internal stripe cache is
2610 * large enough, or it won't work.
2611 * It must hold at least 4 stripes of the larger
2612 * chunk size
2613 */
2614 unsigned long cache;
2615 cache = max(info->array.chunk_size, info->new_chunk);
2616 cache *= 4; /* 4 stripes minimum */
2617 cache /= 512; /* convert to sectors */
2618 /* make sure there is room for 'blocks' with a bit to spare */
2619 if (cache < 16 + blocks / disks)
2620 cache = 16 + blocks / disks;
2621 cache /= (4096/512); /* Convert from sectors to pages */
2622
2623 if (sra->cache_size < cache)
2624 subarray_set_num(container, sra, "stripe_cache_size",
2625 cache+1);
2626 }
2627
2628 static int impose_reshape(struct mdinfo *sra,
2629 struct mdinfo *info,
2630 struct supertype *st,
2631 int fd,
2632 int restart,
2633 char *devname, char *container,
2634 struct reshape *reshape)
2635 {
2636 struct mdu_array_info_s array;
2637
2638 sra->new_chunk = info->new_chunk;
2639
2640 if (restart) {
2641 /* for external metadata checkpoint saved by mdmon can be lost
2642 * or missed /due to e.g. crash/. Check if md is not during
2643 * restart farther than metadata points to.
2644 * If so, this means metadata information is obsolete.
2645 */
2646 if (st->ss->external)
2647 verify_reshape_position(info, reshape->level);
2648 sra->reshape_progress = info->reshape_progress;
2649 } else {
2650 sra->reshape_progress = 0;
2651 if (reshape->after.data_disks < reshape->before.data_disks)
2652 /* start from the end of the new array */
2653 sra->reshape_progress = (sra->component_size
2654 * reshape->after.data_disks);
2655 }
2656
2657 ioctl(fd, GET_ARRAY_INFO, &array);
2658 if (info->array.chunk_size == info->new_chunk &&
2659 reshape->before.layout == reshape->after.layout &&
2660 st->ss->external == 0) {
2661 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2662 array.raid_disks = reshape->after.data_disks + reshape->parity;
2663 if (!restart &&
2664 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2665 int err = errno;
2666
2667 pr_err("Cannot set device shape for %s: %s\n",
2668 devname, strerror(errno));
2669
2670 if (err == EBUSY &&
2671 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2672 cont_err("Bitmap must be removed before shape can be changed\n");
2673
2674 goto release;
2675 }
2676 } else if (!restart) {
2677 /* set them all just in case some old 'new_*' value
2678 * persists from some earlier problem.
2679 */
2680 int err = 0;
2681 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2682 err = errno;
2683 if (!err && sysfs_set_num(sra, NULL, "layout",
2684 reshape->after.layout) < 0)
2685 err = errno;
2686 if (!err && subarray_set_num(container, sra, "raid_disks",
2687 reshape->after.data_disks +
2688 reshape->parity) < 0)
2689 err = errno;
2690 if (err) {
2691 pr_err("Cannot set device shape for %s\n",
2692 devname);
2693
2694 if (err == EBUSY &&
2695 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2696 cont_err("Bitmap must be removed before shape can be changed\n");
2697 goto release;
2698 }
2699 }
2700 return 0;
2701 release:
2702 return -1;
2703 }
2704
2705 static int impose_level(int fd, int level, char *devname, int verbose)
2706 {
2707 char *c;
2708 struct mdu_array_info_s array;
2709 struct mdinfo info;
2710 sysfs_init(&info, fd, NULL);
2711
2712 ioctl(fd, GET_ARRAY_INFO, &array);
2713 if (level == 0 &&
2714 (array.level >= 4 && array.level <= 6)) {
2715 /* To convert to RAID0 we need to fail and
2716 * remove any non-data devices. */
2717 int found = 0;
2718 int d;
2719 int data_disks = array.raid_disks - 1;
2720 if (array.level == 6)
2721 data_disks -= 1;
2722 if (array.level == 5 &&
2723 array.layout != ALGORITHM_PARITY_N)
2724 return -1;
2725 if (array.level == 6 &&
2726 array.layout != ALGORITHM_PARITY_N_6)
2727 return -1;
2728 sysfs_set_str(&info, NULL,"sync_action", "idle");
2729 /* First remove any spares so no recovery starts */
2730 for (d = 0, found = 0;
2731 d < MAX_DISKS && found < array.nr_disks;
2732 d++) {
2733 mdu_disk_info_t disk;
2734 disk.number = d;
2735 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2736 continue;
2737 if (disk.major == 0 && disk.minor == 0)
2738 continue;
2739 found++;
2740 if ((disk.state & (1 << MD_DISK_ACTIVE))
2741 && disk.raid_disk < data_disks)
2742 /* keep this */
2743 continue;
2744 ioctl(fd, HOT_REMOVE_DISK,
2745 makedev(disk.major, disk.minor));
2746 }
2747 /* Now fail anything left */
2748 ioctl(fd, GET_ARRAY_INFO, &array);
2749 for (d = 0, found = 0;
2750 d < MAX_DISKS && found < array.nr_disks;
2751 d++) {
2752 mdu_disk_info_t disk;
2753 disk.number = d;
2754 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2755 continue;
2756 if (disk.major == 0 && disk.minor == 0)
2757 continue;
2758 found++;
2759 if ((disk.state & (1 << MD_DISK_ACTIVE))
2760 && disk.raid_disk < data_disks)
2761 /* keep this */
2762 continue;
2763 ioctl(fd, SET_DISK_FAULTY,
2764 makedev(disk.major, disk.minor));
2765 hot_remove_disk(fd, makedev(disk.major, disk.minor), 1);
2766 }
2767 }
2768 c = map_num(pers, level);
2769 if (c) {
2770 int err = sysfs_set_str(&info, NULL, "level", c);
2771 if (err) {
2772 err = errno;
2773 pr_err("%s: could not set level to %s\n",
2774 devname, c);
2775 if (err == EBUSY &&
2776 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2777 cont_err("Bitmap must be removed before level can be changed\n");
2778 return err;
2779 }
2780 if (verbose >= 0)
2781 pr_err("level of %s changed to %s\n",
2782 devname, c);
2783 }
2784 return 0;
2785 }
2786
2787 int sigterm = 0;
2788 static void catch_term(int sig)
2789 {
2790 sigterm = 1;
2791 }
2792
2793 static int continue_via_systemd(char *devnm)
2794 {
2795 int skipped, i, pid, status;
2796 char pathbuf[1024];
2797 /* In a systemd/udev world, it is best to get systemd to
2798 * run "mdadm --grow --continue" rather than running in the
2799 * background.
2800 */
2801 switch(fork()) {
2802 case 0:
2803 /* FIXME yuk. CLOSE_EXEC?? */
2804 skipped = 0;
2805 for (i = 3; skipped < 20; i++)
2806 if (close(i) < 0)
2807 skipped++;
2808 else
2809 skipped = 0;
2810
2811 /* Don't want to see error messages from
2812 * systemctl. If the service doesn't exist,
2813 * we fork ourselves.
2814 */
2815 close(2);
2816 open("/dev/null", O_WRONLY);
2817 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2818 devnm);
2819 status = execl("/usr/bin/systemctl", "systemctl",
2820 "start",
2821 pathbuf, NULL);
2822 status = execl("/bin/systemctl", "systemctl", "start",
2823 pathbuf, NULL);
2824 exit(1);
2825 case -1: /* Just do it ourselves. */
2826 break;
2827 default: /* parent - good */
2828 pid = wait(&status);
2829 if (pid >= 0 && status == 0)
2830 return 1;
2831 }
2832 return 0;
2833 }
2834
2835 static int reshape_array(char *container, int fd, char *devname,
2836 struct supertype *st, struct mdinfo *info,
2837 int force, struct mddev_dev *devlist,
2838 unsigned long long data_offset,
2839 char *backup_file, int verbose, int forked,
2840 int restart, int freeze_reshape)
2841 {
2842 struct reshape reshape;
2843 int spares_needed;
2844 char *msg;
2845 int orig_level = UnSet;
2846 int odisks;
2847 int delayed;
2848
2849 struct mdu_array_info_s array;
2850 char *c;
2851
2852 struct mddev_dev *dv;
2853 int added_disks;
2854
2855 int *fdlist = NULL;
2856 unsigned long long *offsets = NULL;
2857 int d;
2858 int nrdisks;
2859 int err;
2860 unsigned long blocks;
2861 unsigned long long array_size;
2862 int done;
2863 struct mdinfo *sra = NULL;
2864 char buf[20];
2865
2866 /* when reshaping a RAID0, the component_size might be zero.
2867 * So try to fix that up.
2868 */
2869 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2870 dprintf("Cannot get array information.\n");
2871 goto release;
2872 }
2873 if (array.level == 0 && info->component_size == 0) {
2874 get_dev_size(fd, NULL, &array_size);
2875 info->component_size = array_size / array.raid_disks;
2876 }
2877
2878 if (array.level == 10)
2879 /* Need space_after info */
2880 get_space_after(fd, st, info);
2881
2882 if (info->reshape_active) {
2883 int new_level = info->new_level;
2884 info->new_level = UnSet;
2885 if (info->delta_disks > 0)
2886 info->array.raid_disks -= info->delta_disks;
2887 msg = analyse_change(devname, info, &reshape);
2888 info->new_level = new_level;
2889 if (info->delta_disks > 0)
2890 info->array.raid_disks += info->delta_disks;
2891 if (!restart)
2892 /* Make sure the array isn't read-only */
2893 ioctl(fd, RESTART_ARRAY_RW, 0);
2894 } else
2895 msg = analyse_change(devname, info, &reshape);
2896 if (msg) {
2897 /* if msg == "", error has already been printed */
2898 if (msg[0])
2899 pr_err("%s\n", msg);
2900 goto release;
2901 }
2902 if (restart &&
2903 (reshape.level != info->array.level ||
2904 reshape.before.layout != info->array.layout ||
2905 reshape.before.data_disks + reshape.parity
2906 != info->array.raid_disks - max(0, info->delta_disks))) {
2907 pr_err("reshape info is not in native format - cannot continue.\n");
2908 goto release;
2909 }
2910
2911 if (st->ss->external && restart && (info->reshape_progress == 0) &&
2912 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
2913 (strncmp(buf, "reshape", 7) == 0))) {
2914 /* When reshape is restarted from '0', very begin of array
2915 * it is possible that for external metadata reshape and array
2916 * configuration doesn't happen.
2917 * Check if md has the same opinion, and reshape is restarted
2918 * from 0. If so, this is regular reshape start after reshape
2919 * switch in metadata to next array only.
2920 */
2921 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2922 (info->reshape_progress == 0))
2923 restart = 0;
2924 }
2925 if (restart) {
2926 /* reshape already started. just skip to monitoring the reshape */
2927 if (reshape.backup_blocks == 0)
2928 return 0;
2929 if (restart & RESHAPE_NO_BACKUP)
2930 return 0;
2931
2932 /* Need 'sra' down at 'started:' */
2933 sra = sysfs_read(fd, NULL,
2934 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2935 GET_CACHE);
2936 if (!sra) {
2937 pr_err("%s: Cannot get array details from sysfs\n",
2938 devname);
2939 goto release;
2940 }
2941
2942 if (!backup_file)
2943 backup_file = locate_backup(sra->sys_name);
2944
2945 goto started;
2946 }
2947 /* The container is frozen but the array may not be.
2948 * So freeze the array so spares don't get put to the wrong use
2949 * FIXME there should probably be a cleaner separation between
2950 * freeze_array and freeze_container.
2951 */
2952 sysfs_freeze_array(info);
2953 /* Check we have enough spares to not be degraded */
2954 added_disks = 0;
2955 for (dv = devlist; dv ; dv=dv->next)
2956 added_disks++;
2957 spares_needed = max(reshape.before.data_disks,
2958 reshape.after.data_disks)
2959 + reshape.parity - array.raid_disks;
2960
2961 if (!force &&
2962 info->new_level > 1 && info->array.level > 1 &&
2963 spares_needed > info->array.spare_disks + added_disks) {
2964 pr_err("Need %d spare%s to avoid degraded array, and only have %d.\n"
2965 " Use --force to over-ride this check.\n",
2966 spares_needed,
2967 spares_needed == 1 ? "" : "s",
2968 info->array.spare_disks + added_disks);
2969 goto release;
2970 }
2971 /* Check we have enough spares to not fail */
2972 spares_needed = max(reshape.before.data_disks,
2973 reshape.after.data_disks)
2974 - array.raid_disks;
2975 if ((info->new_level > 1 || info->new_level == 0) &&
2976 spares_needed > info->array.spare_disks +added_disks) {
2977 pr_err("Need %d spare%s to create working array, and only have %d.\n",
2978 spares_needed,
2979 spares_needed == 1 ? "" : "s",
2980 info->array.spare_disks + added_disks);
2981 goto release;
2982 }
2983
2984 if (reshape.level != array.level) {
2985 int err = impose_level(fd, reshape.level, devname, verbose);
2986 if (err)
2987 goto release;
2988 info->new_layout = UnSet; /* after level change,
2989 * layout is meaningless */
2990 orig_level = array.level;
2991 sysfs_freeze_array(info);
2992
2993 if (reshape.level > 0 && st->ss->external) {
2994 /* make sure mdmon is aware of the new level */
2995 if (mdmon_running(container))
2996 flush_mdmon(container);
2997
2998 if (!mdmon_running(container))
2999 start_mdmon(container);
3000 ping_monitor(container);
3001 if (mdmon_running(container) &&
3002 st->update_tail == NULL)
3003 st->update_tail = &st->updates;
3004 }
3005 }
3006 /* ->reshape_super might have chosen some spares from the
3007 * container that it wants to be part of the new array.
3008 * We can collect them with ->container_content and give
3009 * them to the kernel.
3010 */
3011 if (st->ss->reshape_super && st->ss->container_content) {
3012 char *subarray = strchr(info->text_version+1, '/')+1;
3013 struct mdinfo *info2 =
3014 st->ss->container_content(st, subarray);
3015 struct mdinfo *d;
3016
3017 if (info2) {
3018 sysfs_init(info2, fd, st->devnm);
3019 /* When increasing number of devices, we need to set
3020 * new raid_disks before adding these, or they might
3021 * be rejected.
3022 */
3023 if (reshape.backup_blocks &&
3024 reshape.after.data_disks > reshape.before.data_disks)
3025 subarray_set_num(container, info2, "raid_disks",
3026 reshape.after.data_disks +
3027 reshape.parity);
3028 for (d = info2->devs; d; d = d->next) {
3029 if (d->disk.state == 0 &&
3030 d->disk.raid_disk >= 0) {
3031 /* This is a spare that wants to
3032 * be part of the array.
3033 */
3034 add_disk(fd, st, info2, d);
3035 }
3036 }
3037 sysfs_free(info2);
3038 }
3039 }
3040 /* We might have been given some devices to add to the
3041 * array. Now that the array has been changed to the right
3042 * level and frozen, we can safely add them.
3043 */
3044 if (devlist) {
3045 if (Manage_subdevs(devname, fd, devlist, verbose,
3046 0, NULL, 0))
3047 goto release;
3048 }
3049
3050 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3051 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3052 if (reshape.backup_blocks == 0) {
3053 /* No restriping needed, but we might need to impose
3054 * some more changes: layout, raid_disks, chunk_size
3055 */
3056 /* read current array info */
3057 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
3058 dprintf("Cannot get array information.\n");
3059 goto release;
3060 }
3061 /* compare current array info with new values and if
3062 * it is different update them to new */
3063 if (info->new_layout != UnSet &&
3064 info->new_layout != array.layout) {
3065 array.layout = info->new_layout;
3066 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3067 pr_err("failed to set new layout\n");
3068 goto release;
3069 } else if (verbose >= 0)
3070 printf("layout for %s set to %d\n",
3071 devname, array.layout);
3072 }
3073 if (info->delta_disks != UnSet &&
3074 info->delta_disks != 0 &&
3075 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3076 array.raid_disks += info->delta_disks;
3077 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3078 pr_err("failed to set raid disks\n");
3079 goto release;
3080 } else if (verbose >= 0) {
3081 printf("raid_disks for %s set to %d\n",
3082 devname, array.raid_disks);
3083 }
3084 }
3085 if (info->new_chunk != 0 &&
3086 info->new_chunk != array.chunk_size) {
3087 if (sysfs_set_num(info, NULL,
3088 "chunk_size", info->new_chunk) != 0) {
3089 pr_err("failed to set chunk size\n");
3090 goto release;
3091 } else if (verbose >= 0)
3092 printf("chunk size for %s set to %d\n",
3093 devname, array.chunk_size);
3094 }
3095 unfreeze(st);
3096 return 0;
3097 }
3098
3099 /*
3100 * There are three possibilities.
3101 * 1/ The array will shrink.
3102 * We need to ensure the reshape will pause before reaching
3103 * the 'critical section'. We also need to fork and wait for
3104 * that to happen. When it does we
3105 * suspend/backup/complete/unfreeze
3106 *
3107 * 2/ The array will not change size.
3108 * This requires that we keep a backup of a sliding window
3109 * so that we can restore data after a crash. So we need
3110 * to fork and monitor progress.
3111 * In future we will allow the data_offset to change, so
3112 * a sliding backup becomes unnecessary.
3113 *
3114 * 3/ The array will grow. This is relatively easy.
3115 * However the kernel's restripe routines will cheerfully
3116 * overwrite some early data before it is safe. So we
3117 * need to make a backup of the early parts of the array
3118 * and be ready to restore it if rebuild aborts very early.
3119 * For externally managed metadata, we still need a forked
3120 * child to monitor the reshape and suspend IO over the region
3121 * that is being reshaped.
3122 *
3123 * We backup data by writing it to one spare, or to a
3124 * file which was given on command line.
3125 *
3126 * In each case, we first make sure that storage is available
3127 * for the required backup.
3128 * Then we:
3129 * - request the shape change.
3130 * - fork to handle backup etc.
3131 */
3132 /* Check that we can hold all the data */
3133 get_dev_size(fd, NULL, &array_size);
3134 if (reshape.new_size < (array_size/512)) {
3135 pr_err("this change will reduce the size of the array.\n"
3136 " use --grow --array-size first to truncate array.\n"
3137 " e.g. mdadm --grow %s --array-size %llu\n",
3138 devname, reshape.new_size/2);
3139 goto release;
3140 }
3141
3142 if (array.level == 10) {
3143 /* Reshaping RAID10 does not require any data backup by
3144 * user-space. Instead it requires that the data_offset
3145 * is changed to avoid the need for backup.
3146 * So this is handled very separately
3147 */
3148 if (restart)
3149 /* Nothing to do. */
3150 return 0;
3151 return raid10_reshape(container, fd, devname, st, info,
3152 &reshape, data_offset,
3153 force, verbose);
3154 }
3155 sra = sysfs_read(fd, NULL,
3156 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3157 GET_CACHE);
3158 if (!sra) {
3159 pr_err("%s: Cannot get array details from sysfs\n",
3160 devname);
3161 goto release;
3162 }
3163
3164 if (!backup_file)
3165 switch(set_new_data_offset(sra, st, devname,
3166 reshape.after.data_disks - reshape.before.data_disks,
3167 data_offset,
3168 reshape.min_offset_change, 1)) {
3169 case -1:
3170 goto release;
3171 case 0:
3172 /* Updated data_offset, so it's easy now */
3173 update_cache_size(container, sra, info,
3174 min(reshape.before.data_disks,
3175 reshape.after.data_disks),
3176 reshape.backup_blocks);
3177
3178 /* Right, everything seems fine. Let's kick things off.
3179 */
3180 sync_metadata(st);
3181
3182 if (impose_reshape(sra, info, st, fd, restart,
3183 devname, container, &reshape) < 0)
3184 goto release;
3185 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3186 struct mdinfo *sd;
3187 if (errno != EINVAL) {
3188 pr_err("Failed to initiate reshape!\n");
3189 goto release;
3190 }
3191 /* revert data_offset and try the old way */
3192 for (sd = sra->devs; sd; sd = sd->next) {
3193 sysfs_set_num(sra, sd, "new_offset",
3194 sd->data_offset);
3195 sysfs_set_str(sra, NULL, "reshape_direction",
3196 "forwards");
3197 }
3198 break;
3199 }
3200 if (info->new_level == reshape.level)
3201 return 0;
3202 /* need to adjust level when reshape completes */
3203 switch(fork()) {
3204 case -1: /* ignore error, but don't wait */
3205 return 0;
3206 default: /* parent */
3207 return 0;
3208 case 0:
3209 map_fork();
3210 break;
3211 }
3212 close(fd);
3213 wait_reshape(sra);
3214 fd = open_dev(sra->sys_name);
3215 if (fd >= 0)
3216 impose_level(fd, info->new_level, devname, verbose);
3217 return 0;
3218 case 1: /* Couldn't set data_offset, try the old way */
3219 if (data_offset != INVALID_SECTORS) {
3220 pr_err("Cannot update data_offset on this array\n");
3221 goto release;
3222 }
3223 break;
3224 }
3225
3226 started:
3227 /* Decide how many blocks (sectors) for a reshape
3228 * unit. The number we have so far is just a minimum
3229 */
3230 blocks = reshape.backup_blocks;
3231 if (reshape.before.data_disks ==
3232 reshape.after.data_disks) {
3233 /* Make 'blocks' bigger for better throughput, but
3234 * not so big that we reject it below.
3235 * Try for 16 megabytes
3236 */
3237 while (blocks * 32 < sra->component_size &&
3238 blocks < 16*1024*2)
3239 blocks *= 2;
3240 } else
3241 pr_err("Need to backup %luK of critical section..\n", blocks/2);
3242
3243 if (blocks >= sra->component_size/2) {
3244 pr_err("%s: Something wrong - reshape aborted\n",
3245 devname);
3246 goto release;
3247 }
3248
3249 /* Now we need to open all these devices so we can read/write.
3250 */
3251 nrdisks = max(reshape.before.data_disks,
3252 reshape.after.data_disks) + reshape.parity
3253 + sra->array.spare_disks;
3254 fdlist = xcalloc((1+nrdisks), sizeof(int));
3255 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3256
3257 odisks = reshape.before.data_disks + reshape.parity;
3258 d = reshape_prepare_fdlist(devname, sra, odisks,
3259 nrdisks, blocks, backup_file,
3260 fdlist, offsets);
3261 if (d < odisks) {
3262 goto release;
3263 }
3264 if ((st->ss->manage_reshape == NULL) ||
3265 (st->ss->recover_backup == NULL)) {
3266 if (backup_file == NULL) {
3267 if (reshape.after.data_disks <=
3268 reshape.before.data_disks) {
3269 pr_err("%s: Cannot grow - need backup-file\n",
3270 devname);
3271 pr_err(" Please provide one with \"--backup=...\"\n");
3272 goto release;
3273 } else if (d == odisks) {
3274 pr_err("%s: Cannot grow - need a spare or backup-file to backup critical section\n", devname);
3275 goto release;
3276 }
3277 } else {
3278 if (!reshape_open_backup_file(backup_file, fd, devname,
3279 (signed)blocks,
3280 fdlist+d, offsets+d,
3281 sra->sys_name,
3282 restart)) {
3283 goto release;
3284 }
3285 d++;
3286 }
3287 }
3288
3289 update_cache_size(container, sra, info,
3290 min(reshape.before.data_disks, reshape.after.data_disks),
3291 blocks);
3292
3293 /* Right, everything seems fine. Let's kick things off.
3294 * If only changing raid_disks, use ioctl, else use
3295 * sysfs.
3296 */
3297 sync_metadata(st);
3298
3299 if (impose_reshape(sra, info, st, fd, restart,
3300 devname, container, &reshape) < 0)
3301 goto release;
3302
3303 err = start_reshape(sra, restart, reshape.before.data_disks,
3304 reshape.after.data_disks);
3305 if (err) {
3306 pr_err("Cannot %s reshape for %s\n",
3307 restart ? "continue" : "start",
3308 devname);
3309 goto release;
3310 }
3311 if (restart)
3312 sysfs_set_str(sra, NULL, "array_state", "active");
3313 if (freeze_reshape) {
3314 free(fdlist);
3315 free(offsets);
3316 sysfs_free(sra);
3317 pr_err("Reshape has to be continued from location %llu when root filesystem has been mounted.\n",
3318 sra->reshape_progress);
3319 return 1;
3320 }
3321
3322 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3323 if (continue_via_systemd(container ?: sra->sys_name)) {
3324 free(fdlist);
3325 free(offsets);
3326 sysfs_free(sra);
3327 return 0;
3328 }
3329
3330 /* Now we just need to kick off the reshape and watch, while
3331 * handling backups of the data...
3332 * This is all done by a forked background process.
3333 */
3334 switch(forked ? 0 : fork()) {
3335 case -1:
3336 pr_err("Cannot run child to monitor reshape: %s\n",
3337 strerror(errno));
3338 abort_reshape(sra);
3339 goto release;
3340 default:
3341 free(fdlist);
3342 free(offsets);
3343 sysfs_free(sra);
3344 return 0;
3345 case 0:
3346 map_fork();
3347 break;
3348 }
3349
3350 /* If another array on the same devices is busy, the
3351 * reshape will wait for them. This would mean that
3352 * the first section that we suspend will stay suspended
3353 * for a long time. So check on that possibility
3354 * by looking for "DELAYED" in /proc/mdstat, and if found,
3355 * wait a while
3356 */
3357 do {
3358 struct mdstat_ent *mds, *m;
3359 delayed = 0;
3360 mds = mdstat_read(1, 0);
3361 for (m = mds; m; m = m->next)
3362 if (strcmp(m->devnm, sra->sys_name) == 0) {
3363 if (m->resync &&
3364 m->percent == RESYNC_DELAYED)
3365 delayed = 1;
3366 if (m->resync == 0)
3367 /* Haven't started the reshape thread
3368 * yet, wait a bit
3369 */
3370 delayed = 2;
3371 break;
3372 }
3373 free_mdstat(mds);
3374 if (delayed == 1 && get_linux_version() < 3007000) {
3375 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3376 " You might experience problems until other reshapes complete.\n");
3377 delayed = 0;
3378 }
3379 if (delayed)
3380 mdstat_wait(30 - (delayed-1) * 25);
3381 } while (delayed);
3382 mdstat_close();
3383 close(fd);
3384 if (check_env("MDADM_GROW_VERIFY"))
3385 fd = open(devname, O_RDONLY | O_DIRECT);
3386 else
3387 fd = -1;
3388 mlockall(MCL_FUTURE);
3389
3390 signal(SIGTERM, catch_term);
3391
3392 if (st->ss->external) {
3393 /* metadata handler takes it from here */
3394 done = st->ss->manage_reshape(
3395 fd, sra, &reshape, st, blocks,
3396 fdlist, offsets,
3397 d - odisks, fdlist+odisks,
3398 offsets+odisks);
3399 } else
3400 done = child_monitor(
3401 fd, sra, &reshape, st, blocks,
3402 fdlist, offsets,
3403 d - odisks, fdlist+odisks,
3404 offsets+odisks);
3405
3406 free(fdlist);
3407 free(offsets);
3408
3409 if (backup_file && done) {
3410 char *bul;
3411 bul = make_backup(sra->sys_name);
3412 if (bul) {
3413 char buf[1024];
3414 int l = readlink(bul, buf, sizeof(buf) - 1);
3415 if (l > 0) {
3416 buf[l]=0;
3417 unlink(buf);
3418 }
3419 unlink(bul);
3420 free(bul);
3421 }
3422 unlink(backup_file);
3423 }
3424 if (!done) {
3425 abort_reshape(sra);
3426 goto out;
3427 }
3428
3429 if (!st->ss->external &&
3430 !(reshape.before.data_disks != reshape.after.data_disks
3431 && info->custom_array_size) &&
3432 info->new_level == reshape.level &&
3433 !forked) {
3434 /* no need to wait for the reshape to finish as
3435 * there is nothing more to do.
3436 */
3437 sysfs_free(sra);
3438 exit(0);
3439 }
3440 wait_reshape(sra);
3441
3442 if (st->ss->external) {
3443 /* Re-load the metadata as much could have changed */
3444 int cfd = open_dev(st->container_devnm);
3445 if (cfd >= 0) {
3446 flush_mdmon(container);
3447 st->ss->free_super(st);
3448 st->ss->load_container(st, cfd, container);
3449 close(cfd);
3450 }
3451 }
3452
3453 /* set new array size if required customer_array_size is used
3454 * by this metadata.
3455 */
3456 if (reshape.before.data_disks !=
3457 reshape.after.data_disks &&
3458 info->custom_array_size)
3459 set_array_size(st, info, info->text_version);
3460
3461 if (info->new_level != reshape.level) {
3462 if (fd < 0)
3463 fd = open(devname, O_RDONLY);
3464 impose_level(fd, info->new_level, devname, verbose);
3465 close(fd);
3466 if (info->new_level == 0)
3467 st->update_tail = NULL;
3468 }
3469 out:
3470 sysfs_free(sra);
3471 if (forked)
3472 return 0;
3473 unfreeze(st);
3474 exit(0);
3475
3476 release:
3477 free(fdlist);
3478 free(offsets);
3479 if (orig_level != UnSet && sra) {
3480 c = map_num(pers, orig_level);
3481 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3482 pr_err("aborting level change\n");
3483 }
3484 sysfs_free(sra);
3485 if (!forked)
3486 unfreeze(st);
3487 return 1;
3488 }
3489
3490 /* mdfd handle is passed to be closed in child process (after fork).
3491 */
3492 int reshape_container(char *container, char *devname,
3493 int mdfd,
3494 struct supertype *st,
3495 struct mdinfo *info,
3496 int force,
3497 char *backup_file, int verbose,
3498 int forked, int restart, int freeze_reshape)
3499 {
3500 struct mdinfo *cc = NULL;
3501 int rv = restart;
3502 char last_devnm[32] = "";
3503
3504 /* component_size is not meaningful for a container,
3505 * so pass '0' meaning 'no change'
3506 */
3507 if (!restart &&
3508 reshape_super(st, 0, info->new_level,
3509 info->new_layout, info->new_chunk,
3510 info->array.raid_disks, info->delta_disks,
3511 backup_file, devname, APPLY_METADATA_CHANGES,
3512 verbose)) {
3513 unfreeze(st);
3514 return 1;
3515 }
3516
3517 sync_metadata(st);
3518
3519 /* ping monitor to be sure that update is on disk
3520 */
3521 ping_monitor(container);
3522
3523 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3524 if (continue_via_systemd(container))
3525 return 0;
3526
3527 switch (forked ? 0 : fork()) {
3528 case -1: /* error */
3529 perror("Cannot fork to complete reshape\n");
3530 unfreeze(st);
3531 return 1;
3532 default: /* parent */
3533 if (!freeze_reshape)
3534 printf("%s: multi-array reshape continues in background\n", Name);
3535 return 0;
3536 case 0: /* child */
3537 map_fork();
3538 break;
3539 }
3540
3541 /* close unused handle in child process
3542 */
3543 if (mdfd > -1)
3544 close(mdfd);
3545
3546 while(1) {
3547 /* For each member array with reshape_active,
3548 * we need to perform the reshape.
3549 * We pick the first array that needs reshaping and
3550 * reshape it. reshape_array() will re-read the metadata
3551 * so the next time through a different array should be
3552 * ready for reshape.
3553 * It is possible that the 'different' array will not
3554 * be assembled yet. In that case we simple exit.
3555 * When it is assembled, the mdadm which assembles it
3556 * will take over the reshape.
3557 */
3558 struct mdinfo *content;
3559 int fd;
3560 struct mdstat_ent *mdstat;
3561 char *adev;
3562 dev_t devid;
3563
3564 sysfs_free(cc);
3565
3566 cc = st->ss->container_content(st, NULL);
3567
3568 for (content = cc; content ; content = content->next) {
3569 char *subarray;
3570 if (!content->reshape_active)
3571 continue;
3572
3573 subarray = strchr(content->text_version+1, '/')+1;
3574 mdstat = mdstat_by_subdev(subarray, container);
3575 if (!mdstat)
3576 continue;
3577 if (mdstat->active == 0) {
3578 pr_err("Skipping inactive array %s.\n",
3579 mdstat->devnm);
3580 free_mdstat(mdstat);
3581 mdstat = NULL;
3582 continue;
3583 }
3584 break;
3585 }
3586 if (!content)
3587 break;
3588
3589 devid = devnm2devid(mdstat->devnm);
3590 adev = map_dev(major(devid), minor(devid), 0);
3591 if (!adev)
3592 adev = content->text_version;
3593
3594 fd = open_dev(mdstat->devnm);
3595 if (fd < 0) {
3596 pr_err("Device %s cannot be opened for reshape.\n", adev);
3597 break;
3598 }
3599
3600 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3601 /* Do not allow for multiple reshape_array() calls for
3602 * the same array.
3603 * It can happen when reshape_array() returns without
3604 * error, when reshape is not finished (wrong reshape
3605 * starting/continuation conditions). Mdmon doesn't
3606 * switch to next array in container and reentry
3607 * conditions for the same array occur.
3608 * This is possibly interim until the behaviour of
3609 * reshape_array is resolved().
3610 */
3611 printf("%s: Multiple reshape execution detected for device %s.\n", Name, adev);
3612 close(fd);
3613 break;
3614 }
3615 strcpy(last_devnm, mdstat->devnm);
3616
3617 sysfs_init(content, fd, mdstat->devnm);
3618
3619 if (mdmon_running(container))
3620 flush_mdmon(container);
3621
3622 rv = reshape_array(container, fd, adev, st,
3623 content, force, NULL, INVALID_SECTORS,
3624 backup_file, verbose, 1, restart,
3625 freeze_reshape);
3626 close(fd);
3627
3628 if (freeze_reshape) {
3629 sysfs_free(cc);
3630 exit(0);
3631 }
3632
3633 restart = 0;
3634 if (rv)
3635 break;
3636
3637 if (mdmon_running(container))
3638 flush_mdmon(container);
3639 }
3640 if (!rv)
3641 unfreeze(st);
3642 sysfs_free(cc);
3643 exit(0);
3644 }
3645
3646 /*
3647 * We run a child process in the background which performs the following
3648 * steps:
3649 * - wait for resync to reach a certain point
3650 * - suspend io to the following section
3651 * - backup that section
3652 * - allow resync to proceed further
3653 * - resume io
3654 * - discard the backup.
3655 *
3656 * When are combined in slightly different ways in the three cases.
3657 * Grow:
3658 * - suspend/backup/allow/wait/resume/discard
3659 * Shrink:
3660 * - allow/wait/suspend/backup/allow/wait/resume/discard
3661 * same-size:
3662 * - wait/resume/discard/suspend/backup/allow
3663 *
3664 * suspend/backup/allow always come together
3665 * wait/resume/discard do too.
3666 * For the same-size case we have two backups to improve flow.
3667 *
3668 */
3669
3670 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3671 unsigned long long backup_point,
3672 unsigned long long wait_point,
3673 unsigned long long *suspend_point,
3674 unsigned long long *reshape_completed, int *frozen)
3675 {
3676 /* This function is called repeatedly by the reshape manager.
3677 * It determines how much progress can safely be made and allows
3678 * that progress.
3679 * - 'info' identifies the array and particularly records in
3680 * ->reshape_progress the metadata's knowledge of progress
3681 * This is a sector offset from the start of the array
3682 * of the next array block to be relocated. This number
3683 * may increase from 0 or decrease from array_size, depending
3684 * on the type of reshape that is happening.
3685 * Note that in contrast, 'sync_completed' is a block count of the
3686 * reshape so far. It gives the distance between the start point
3687 * (head or tail of device) and the next place that data will be
3688 * written. It always increases.
3689 * - 'reshape' is the structure created by analyse_change
3690 * - 'backup_point' shows how much the metadata manager has backed-up
3691 * data. For reshapes with increasing progress, it is the next address
3692 * to be backed up, previous addresses have been backed-up. For
3693 * decreasing progress, it is the earliest address that has been
3694 * backed up - later address are also backed up.
3695 * So addresses between reshape_progress and backup_point are
3696 * backed up providing those are in the 'correct' order.
3697 * - 'wait_point' is an array address. When reshape_completed
3698 * passes this point, progress_reshape should return. It might
3699 * return earlier if it determines that ->reshape_progress needs
3700 * to be updated or further backup is needed.
3701 * - suspend_point is maintained by progress_reshape and the caller
3702 * should not touch it except to initialise to zero.
3703 * It is an array address and it only increases in 2.6.37 and earlier.
3704 * This makes it difficult to handle reducing reshapes with
3705 * external metadata.
3706 * However: it is similar to backup_point in that it records the
3707 * other end of a suspended region from reshape_progress.
3708 * it is moved to extend the region that is safe to backup and/or
3709 * reshape
3710 * - reshape_completed is read from sysfs and returned. The caller
3711 * should copy this into ->reshape_progress when it has reason to
3712 * believe that the metadata knows this, and any backup outside this
3713 * has been erased.
3714 *
3715 * Return value is:
3716 * 1 if more data from backup_point - but only as far as suspend_point,
3717 * should be backed up
3718 * 0 if things are progressing smoothly
3719 * -1 if the reshape is finished because it is all done,
3720 * -2 if the reshape is finished due to an error.
3721 */
3722
3723 int advancing = (reshape->after.data_disks
3724 >= reshape->before.data_disks);
3725 unsigned long long need_backup; /* All data between start of array and
3726 * here will at some point need to
3727 * be backed up.
3728 */
3729 unsigned long long read_offset, write_offset;
3730 unsigned long long write_range;
3731 unsigned long long max_progress, target, completed;
3732 unsigned long long array_size = (info->component_size
3733 * reshape->before.data_disks);
3734 int fd;
3735 char buf[20];
3736
3737 /* First, we unsuspend any region that is now known to be safe.
3738 * If suspend_point is on the 'wrong' side of reshape_progress, then
3739 * we don't have or need suspension at the moment. This is true for
3740 * native metadata when we don't need to back-up.
3741 */
3742 if (advancing) {
3743 if (info->reshape_progress <= *suspend_point)
3744 sysfs_set_num(info, NULL, "suspend_lo",
3745 info->reshape_progress);
3746 } else {
3747 /* Note: this won't work in 2.6.37 and before.
3748 * Something somewhere should make sure we don't need it!
3749 */
3750 if (info->reshape_progress >= *suspend_point)
3751 sysfs_set_num(info, NULL, "suspend_hi",
3752 info->reshape_progress);
3753 }
3754
3755 /* Now work out how far it is safe to progress.
3756 * If the read_offset for ->reshape_progress is less than
3757 * 'blocks' beyond the write_offset, we can only progress as far
3758 * as a backup.
3759 * Otherwise we can progress until the write_offset for the new location
3760 * reaches (within 'blocks' of) the read_offset at the current location.
3761 * However that region must be suspended unless we are using native
3762 * metadata.
3763 * If we need to suspend more, we limit it to 128M per device, which is
3764 * rather arbitrary and should be some time-based calculation.
3765 */
3766 read_offset = info->reshape_progress / reshape->before.data_disks;
3767 write_offset = info->reshape_progress / reshape->after.data_disks;
3768 write_range = info->new_chunk/512;
3769 if (reshape->before.data_disks == reshape->after.data_disks)
3770 need_backup = array_size;
3771 else
3772 need_backup = reshape->backup_blocks;
3773 if (advancing) {
3774 if (read_offset < write_offset + write_range)
3775 max_progress = backup_point;
3776 else
3777 max_progress =
3778 read_offset *
3779 reshape->after.data_disks;
3780 } else {
3781 if (read_offset > write_offset - write_range)
3782 /* Can only progress as far as has been backed up,
3783 * which must be suspended */
3784 max_progress = backup_point;
3785 else if (info->reshape_progress <= need_backup)
3786 max_progress = backup_point;
3787 else {
3788 if (info->array.major_version >= 0)
3789 /* Can progress until backup is needed */
3790 max_progress = need_backup;
3791 else {
3792 /* Can progress until metadata update is required */
3793 max_progress =
3794 read_offset *
3795 reshape->after.data_disks;
3796 /* but data must be suspended */
3797 if (max_progress < *suspend_point)
3798 max_progress = *suspend_point;
3799 }
3800 }
3801 }
3802
3803 /* We know it is safe to progress to 'max_progress' providing
3804 * it is suspended or we are using native metadata.
3805 * Consider extending suspend_point 128M per device if it
3806 * is less than 64M per device beyond reshape_progress.
3807 * But always do a multiple of 'blocks'
3808 * FIXME this is too big - it takes to long to complete
3809 * this much.
3810 */
3811 target = 64*1024*2 * min(reshape->before.data_disks,
3812 reshape->after.data_disks);
3813 target /= reshape->backup_blocks;
3814 if (target < 2)
3815 target = 2;
3816 target *= reshape->backup_blocks;
3817
3818 /* For externally managed metadata we always need to suspend IO to
3819 * the area being reshaped so we regularly push suspend_point forward.
3820 * For native metadata we only need the suspend if we are going to do
3821 * a backup.
3822 */
3823 if (advancing) {
3824 if ((need_backup > info->reshape_progress
3825 || info->array.major_version < 0) &&
3826 *suspend_point < info->reshape_progress + target) {
3827 if (need_backup < *suspend_point + 2 * target)
3828 *suspend_point = need_backup;
3829 else if (*suspend_point + 2 * target < array_size)
3830 *suspend_point += 2 * target;
3831 else
3832 *suspend_point = array_size;
3833 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3834 if (max_progress > *suspend_point)
3835 max_progress = *suspend_point;
3836 }
3837 } else {
3838 if (info->array.major_version >= 0) {
3839 /* Only need to suspend when about to backup */
3840 if (info->reshape_progress < need_backup * 2 &&
3841 *suspend_point > 0) {
3842 *suspend_point = 0;
3843 sysfs_set_num(info, NULL, "suspend_lo", 0);
3844 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3845 }
3846 } else {
3847 /* Need to suspend continually */
3848 if (info->reshape_progress < *suspend_point)
3849 *suspend_point = info->reshape_progress;
3850 if (*suspend_point + target < info->reshape_progress)
3851 /* No need to move suspend region yet */;
3852 else {
3853 if (*suspend_point >= 2 * target)
3854 *suspend_point -= 2 * target;
3855 else
3856 *suspend_point = 0;
3857 sysfs_set_num(info, NULL, "suspend_lo",
3858 *suspend_point);
3859 }
3860 if (max_progress < *suspend_point)
3861 max_progress = *suspend_point;
3862 }
3863 }
3864
3865 /* now set sync_max to allow that progress. sync_max, like
3866 * sync_completed is a count of sectors written per device, so
3867 * we find the difference between max_progress and the start point,
3868 * and divide that by after.data_disks to get a sync_max
3869 * number.
3870 * At the same time we convert wait_point to a similar number
3871 * for comparing against sync_completed.
3872 */
3873 /* scale down max_progress to per_disk */
3874 max_progress /= reshape->after.data_disks;
3875 /* Round to chunk size as some kernels give an erroneously high number */
3876 max_progress /= info->new_chunk/512;
3877 max_progress *= info->new_chunk/512;
3878 /* And round to old chunk size as the kernel wants that */
3879 max_progress /= info->array.chunk_size/512;
3880 max_progress *= info->array.chunk_size/512;
3881 /* Limit progress to the whole device */
3882 if (max_progress > info->component_size)
3883 max_progress = info->component_size;
3884 wait_point /= reshape->after.data_disks;
3885 if (!advancing) {
3886 /* switch from 'device offset' to 'processed block count' */
3887 max_progress = info->component_size - max_progress;
3888 wait_point = info->component_size - wait_point;
3889 }
3890
3891 if (!*frozen)
3892 sysfs_set_num(info, NULL, "sync_max", max_progress);
3893
3894 /* Now wait. If we have already reached the point that we were
3895 * asked to wait to, don't wait at all, else wait for any change.
3896 * We need to select on 'sync_completed' as that is the place that
3897 * notifications happen, but we are really interested in
3898 * 'reshape_position'
3899 */
3900 fd = sysfs_get_fd(info, NULL, "sync_completed");
3901 if (fd < 0)
3902 goto check_progress;
3903
3904 if (sysfs_fd_get_ll(fd, &completed) < 0)
3905 goto check_progress;
3906
3907 while (completed < max_progress && completed < wait_point) {
3908 /* Check that sync_action is still 'reshape' to avoid
3909 * waiting forever on a dead array
3910 */
3911 char action[20];
3912 if (sysfs_get_str(info, NULL, "sync_action",
3913 action, 20) <= 0 ||
3914 strncmp(action, "reshape", 7) != 0)
3915 break;
3916 /* Some kernels reset 'sync_completed' to zero
3917 * before setting 'sync_action' to 'idle'.
3918 * So we need these extra tests.
3919 */
3920 if (completed == 0 && advancing
3921 && strncmp(action, "idle", 4) == 0
3922 && info->reshape_progress > 0)
3923 break;
3924 if (completed == 0 && !advancing
3925 && strncmp(action, "idle", 4) == 0
3926 && info->reshape_progress < (info->component_size
3927 * reshape->after.data_disks))
3928 break;
3929 sysfs_wait(fd, NULL);
3930 if (sysfs_fd_get_ll(fd, &completed) < 0)
3931 goto check_progress;
3932 }
3933 /* Some kernels reset 'sync_completed' to zero,
3934 * we need to have real point we are in md.
3935 * So in that case, read 'reshape_position' from sysfs.
3936 */
3937 if (completed == 0) {
3938 unsigned long long reshapep;
3939 char action[20];
3940 if (sysfs_get_str(info, NULL, "sync_action",
3941 action, 20) > 0 &&
3942 strncmp(action, "idle", 4) == 0 &&
3943 sysfs_get_ll(info, NULL,
3944 "reshape_position", &reshapep) == 0)
3945 *reshape_completed = reshapep;
3946 } else {
3947 /* some kernels can give an incorrectly high
3948 * 'completed' number, so round down */
3949 completed /= (info->new_chunk/512);
3950 completed *= (info->new_chunk/512);
3951 /* Convert 'completed' back in to a 'progress' number */
3952 completed *= reshape->after.data_disks;
3953 if (!advancing)
3954 completed = (info->component_size
3955 * reshape->after.data_disks
3956 - completed);
3957 *reshape_completed = completed;
3958 }
3959
3960 close(fd);
3961
3962 /* We return the need_backup flag. Caller will decide
3963 * how much - a multiple of ->backup_blocks up to *suspend_point
3964 */
3965 if (advancing)
3966 return need_backup > info->reshape_progress;
3967 else
3968 return need_backup >= info->reshape_progress;
3969
3970 check_progress:
3971 /* if we couldn't read a number from sync_completed, then
3972 * either the reshape did complete, or it aborted.
3973 * We can tell which by checking for 'none' in reshape_position.
3974 * If it did abort, then it might immediately restart if it
3975 * it was just a device failure that leaves us degraded but
3976 * functioning.
3977 */
3978 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3979 || strncmp(buf, "none", 4) != 0) {
3980 /* The abort might only be temporary. Wait up to 10
3981 * seconds for fd to contain a valid number again.
3982 */
3983 int wait = 10000;
3984 int rv = -2;
3985 unsigned long long new_sync_max;
3986 while (fd >= 0 && rv < 0 && wait > 0) {
3987 if (sysfs_wait(fd, &wait) != 1)
3988 break;
3989 switch (sysfs_fd_get_ll(fd, &completed)) {
3990 case 0:
3991 /* all good again */
3992 rv = 1;
3993 /* If "sync_max" is no longer max_progress
3994 * we need to freeze things
3995 */
3996 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3997 *frozen = (new_sync_max != max_progress);
3998 break;
3999 case -2: /* read error - abort */
4000 wait = 0;
4001 break;
4002 }
4003 }
4004 if (fd >= 0)
4005 close(fd);
4006 return rv; /* abort */
4007 } else {
4008 /* Maybe racing with array shutdown - check state */
4009 if (fd >= 0)
4010 close(fd);
4011 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
4012 || strncmp(buf, "inactive", 8) == 0
4013 || strncmp(buf, "clear",5) == 0)
4014 return -2; /* abort */
4015 return -1; /* complete */
4016 }
4017 }
4018
4019 /* FIXME return status is never checked */
4020 static int grow_backup(struct mdinfo *sra,
4021 unsigned long long offset, /* per device */
4022 unsigned long stripes, /* per device, in old chunks */
4023 int *sources, unsigned long long *offsets,
4024 int disks, int chunk, int level, int layout,
4025 int dests, int *destfd, unsigned long long *destoffsets,
4026 int part, int *degraded,
4027 char *buf)
4028 {
4029 /* Backup 'blocks' sectors at 'offset' on each device of the array,
4030 * to storage 'destfd' (offset 'destoffsets'), after first
4031 * suspending IO. Then allow resync to continue
4032 * over the suspended section.
4033 * Use part 'part' of the backup-super-block.
4034 */
4035 int odata = disks;
4036 int rv = 0;
4037 int i;
4038 unsigned long long ll;
4039 int new_degraded;
4040 //printf("offset %llu\n", offset);
4041 if (level >= 4)
4042 odata--;
4043 if (level == 6)
4044 odata--;
4045
4046 /* Check that array hasn't become degraded, else we might backup the wrong data */
4047 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4048 return -1; /* FIXME this error is ignored */
4049 new_degraded = (int)ll;
4050 if (new_degraded != *degraded) {
4051 /* check each device to ensure it is still working */
4052 struct mdinfo *sd;
4053 for (sd = sra->devs ; sd ; sd = sd->next) {
4054 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4055 continue;
4056 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4057 char sbuf[100];
4058
4059 if (sysfs_get_str(sra, sd, "state",
4060 sbuf, sizeof(sbuf)) < 0 ||
4061 strstr(sbuf, "faulty") ||
4062 strstr(sbuf, "in_sync") == NULL) {
4063 /* this device is dead */
4064 sd->disk.state = (1<<MD_DISK_FAULTY);
4065 if (sd->disk.raid_disk >= 0 &&
4066 sources[sd->disk.raid_disk] >= 0) {
4067 close(sources[sd->disk.raid_disk]);
4068 sources[sd->disk.raid_disk] = -1;
4069 }
4070 }
4071 }
4072 }
4073 *degraded = new_degraded;
4074 }
4075 if (part) {
4076 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4077 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4078 } else {
4079 bsb.arraystart = __cpu_to_le64(offset * odata);
4080 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4081 }
4082 if (part)
4083 bsb.magic[15] = '2';
4084 for (i = 0; i < dests; i++)
4085 if (part)
4086 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4087 else
4088 lseek64(destfd[i], destoffsets[i], 0);
4089
4090 rv = save_stripes(sources, offsets,
4091 disks, chunk, level, layout,
4092 dests, destfd,
4093 offset*512*odata, stripes * chunk * odata,
4094 buf);
4095
4096 if (rv)
4097 return rv;
4098 bsb.mtime = __cpu_to_le64(time(0));
4099 for (i = 0; i < dests; i++) {
4100 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4101
4102 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4103 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4104 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4105 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4106
4107 rv = -1;
4108 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4109 != destoffsets[i] - 4096)
4110 break;
4111 if (write(destfd[i], &bsb, 512) != 512)
4112 break;
4113 if (destoffsets[i] > 4096) {
4114 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4115 destoffsets[i]+stripes*chunk*odata)
4116 break;
4117 if (write(destfd[i], &bsb, 512) != 512)
4118 break;
4119 }
4120 fsync(destfd[i]);
4121 rv = 0;
4122 }
4123
4124 return rv;
4125 }
4126
4127 /* in 2.6.30, the value reported by sync_completed can be
4128 * less that it should be by one stripe.
4129 * This only happens when reshape hits sync_max and pauses.
4130 * So allow wait_backup to either extent sync_max further
4131 * than strictly necessary, or return before the
4132 * sync has got quite as far as we would really like.
4133 * This is what 'blocks2' is for.
4134 * The various caller give appropriate values so that
4135 * every works.
4136 */
4137 /* FIXME return value is often ignored */
4138 static int forget_backup(int dests, int *destfd,
4139 unsigned long long *destoffsets,
4140 int part)
4141 {
4142 /*
4143 * Erase backup 'part' (which is 0 or 1)
4144 */
4145 int i;
4146 int rv;
4147
4148 if (part) {
4149 bsb.arraystart2 = __cpu_to_le64(0);
4150 bsb.length2 = __cpu_to_le64(0);
4151 } else {
4152 bsb.arraystart = __cpu_to_le64(0);
4153 bsb.length = __cpu_to_le64(0);
4154 }
4155 bsb.mtime = __cpu_to_le64(time(0));
4156 rv = 0;
4157 for (i = 0; i < dests; i++) {
4158 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4159 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4160 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4161 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4162 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4163 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4164 destoffsets[i]-4096)
4165 rv = -1;
4166 if (rv == 0 &&
4167 write(destfd[i], &bsb, 512) != 512)
4168 rv = -1;
4169 fsync(destfd[i]);
4170 }
4171 return rv;
4172 }
4173
4174 static void fail(char *msg)
4175 {
4176 int rv;
4177 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4178 rv |= (write(2, "\n", 1) != 1);
4179 exit(rv ? 1 : 2);
4180 }
4181
4182 static char *abuf, *bbuf;
4183 static unsigned long long abuflen;
4184 static void validate(int afd, int bfd, unsigned long long offset)
4185 {
4186 /* check that the data in the backup against the array.
4187 * This is only used for regression testing and should not
4188 * be used while the array is active
4189 */
4190 if (afd < 0)
4191 return;
4192 lseek64(bfd, offset - 4096, 0);
4193 if (read(bfd, &bsb2, 512) != 512)
4194 fail("cannot read bsb");
4195 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4196 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4197 fail("first csum bad");
4198 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4199 fail("magic is bad");
4200 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4201 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4202 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4203 fail("second csum bad");
4204
4205 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4206 fail("devstart is wrong");
4207
4208 if (bsb2.length) {
4209 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4210
4211 if (abuflen < len) {
4212 free(abuf);
4213 free(bbuf);
4214 abuflen = len;
4215 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4216 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4217 abuflen = 0;
4218 /* just stop validating on mem-alloc failure */
4219 return;
4220 }
4221 }
4222
4223 lseek64(bfd, offset, 0);
4224 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4225 //printf("len %llu\n", len);
4226 fail("read first backup failed");
4227 }
4228 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4229 if ((unsigned long long)read(afd, abuf, len) != len)
4230 fail("read first from array failed");
4231 if (memcmp(bbuf, abuf, len) != 0) {
4232 #if 0
4233 int i;
4234 printf("offset=%llu len=%llu\n",
4235 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4236 for (i=0; i<len; i++)
4237 if (bbuf[i] != abuf[i]) {
4238 printf("first diff byte %d\n", i);
4239 break;
4240 }
4241 #endif
4242 fail("data1 compare failed");
4243 }
4244 }
4245 if (bsb2.length2) {
4246 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4247
4248 if (abuflen < len) {
4249 free(abuf);
4250 free(bbuf);
4251 abuflen = len;
4252 abuf = xmalloc(abuflen);
4253 bbuf = xmalloc(abuflen);
4254 }
4255
4256 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4257 if ((unsigned long long)read(bfd, bbuf, len) != len)
4258 fail("read second backup failed");
4259 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4260 if ((unsigned long long)read(afd, abuf, len) != len)
4261 fail("read second from array failed");
4262 if (memcmp(bbuf, abuf, len) != 0)
4263 fail("data2 compare failed");
4264 }
4265 }
4266
4267 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4268 struct supertype *st, unsigned long blocks,
4269 int *fds, unsigned long long *offsets,
4270 int dests, int *destfd, unsigned long long *destoffsets)
4271 {
4272 /* Monitor a reshape where backup is being performed using
4273 * 'native' mechanism - either to a backup file, or
4274 * to some space in a spare.
4275 */
4276 char *buf;
4277 int degraded = -1;
4278 unsigned long long speed;
4279 unsigned long long suspend_point, array_size;
4280 unsigned long long backup_point, wait_point;
4281 unsigned long long reshape_completed;
4282 int done = 0;
4283 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4284 int part = 0; /* The next part of the backup area to fill. It may already
4285 * be full, so we need to check */
4286 int level = reshape->level;
4287 int layout = reshape->before.layout;
4288 int data = reshape->before.data_disks;
4289 int disks = reshape->before.data_disks + reshape->parity;
4290 int chunk = sra->array.chunk_size;
4291 struct mdinfo *sd;
4292 unsigned long stripes;
4293 int uuid[4];
4294 int frozen = 0;
4295
4296 /* set up the backup-super-block. This requires the
4297 * uuid from the array.
4298 */
4299 /* Find a superblock */
4300 for (sd = sra->devs; sd; sd = sd->next) {
4301 char *dn;
4302 int devfd;
4303 int ok;
4304 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4305 continue;
4306 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4307 devfd = dev_open(dn, O_RDONLY);
4308 if (devfd < 0)
4309 continue;
4310 ok = st->ss->load_super(st, devfd, NULL);
4311 close(devfd);
4312 if (ok == 0)
4313 break;
4314 }
4315 if (!sd) {
4316 pr_err("Cannot find a superblock\n");
4317 return 0;
4318 }
4319
4320 memset(&bsb, 0, 512);
4321 memcpy(bsb.magic, "md_backup_data-1", 16);
4322 st->ss->uuid_from_super(st, uuid);
4323 memcpy(bsb.set_uuid, uuid, 16);
4324 bsb.mtime = __cpu_to_le64(time(0));
4325 bsb.devstart2 = blocks;
4326
4327 stripes = blocks / (sra->array.chunk_size/512) /
4328 reshape->before.data_disks;
4329
4330 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4331 /* Don't start the 'reshape' */
4332 return 0;
4333 if (reshape->before.data_disks == reshape->after.data_disks) {
4334 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4335 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4336 }
4337
4338 if (increasing) {
4339 array_size = sra->component_size * reshape->after.data_disks;
4340 backup_point = sra->reshape_progress;
4341 suspend_point = 0;
4342 } else {
4343 array_size = sra->component_size * reshape->before.data_disks;
4344 backup_point = reshape->backup_blocks;
4345 suspend_point = array_size;
4346 }
4347
4348 while (!done) {
4349 int rv;
4350
4351 /* Want to return as soon the oldest backup slot can
4352 * be released as that allows us to start backing up
4353 * some more, providing suspend_point has been
4354 * advanced, which it should have.
4355 */
4356 if (increasing) {
4357 wait_point = array_size;
4358 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4359 wait_point = (__le64_to_cpu(bsb.arraystart) +
4360 __le64_to_cpu(bsb.length));
4361 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4362 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4363 __le64_to_cpu(bsb.length2));
4364 } else {
4365 wait_point = 0;
4366 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4367 wait_point = __le64_to_cpu(bsb.arraystart);
4368 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4369 wait_point = __le64_to_cpu(bsb.arraystart2);
4370 }
4371
4372 reshape_completed = sra->reshape_progress;
4373 rv = progress_reshape(sra, reshape,
4374 backup_point, wait_point,
4375 &suspend_point, &reshape_completed,
4376 &frozen);
4377 /* external metadata would need to ping_monitor here */
4378 sra->reshape_progress = reshape_completed;
4379
4380 /* Clear any backup region that is before 'here' */
4381 if (increasing) {
4382 if (__le64_to_cpu(bsb.length) > 0 &&
4383 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4384 __le64_to_cpu(bsb.length)))
4385 forget_backup(dests, destfd,
4386 destoffsets, 0);
4387 if (__le64_to_cpu(bsb.length2) > 0 &&
4388 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4389 __le64_to_cpu(bsb.length2)))
4390 forget_backup(dests, destfd,
4391 destoffsets, 1);
4392 } else {
4393 if (__le64_to_cpu(bsb.length) > 0 &&
4394 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4395 forget_backup(dests, destfd,
4396 destoffsets, 0);
4397 if (__le64_to_cpu(bsb.length2) > 0 &&
4398 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4399 forget_backup(dests, destfd,
4400 destoffsets, 1);
4401 }
4402 if (sigterm)
4403 rv = -2;
4404 if (rv < 0) {
4405 if (rv == -1)
4406 done = 1;
4407 break;
4408 }
4409 if (rv == 0 && increasing && !st->ss->external) {
4410 /* No longer need to monitor this reshape */
4411 sysfs_set_str(sra, NULL, "sync_max", "max");
4412 done = 1;
4413 break;
4414 }
4415
4416 while (rv) {
4417 unsigned long long offset;
4418 unsigned long actual_stripes;
4419 /* Need to backup some data.
4420 * If 'part' is not used and the desired
4421 * backup size is suspended, do a backup,
4422 * then consider the next part.
4423 */
4424 /* Check that 'part' is unused */
4425 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4426 break;
4427 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4428 break;
4429
4430 offset = backup_point / data;
4431 actual_stripes = stripes;
4432 if (increasing) {
4433 if (offset + actual_stripes * (chunk/512) >
4434 sra->component_size)
4435 actual_stripes = ((sra->component_size - offset)
4436 / (chunk/512));
4437 if (offset + actual_stripes * (chunk/512) >
4438 suspend_point/data)
4439 break;
4440 } else {
4441 if (offset < actual_stripes * (chunk/512))
4442 actual_stripes = offset / (chunk/512);
4443 offset -= actual_stripes * (chunk/512);
4444 if (offset < suspend_point/data)
4445 break;
4446 }
4447 if (actual_stripes == 0)
4448 break;
4449 grow_backup(sra, offset, actual_stripes,
4450 fds, offsets,
4451 disks, chunk, level, layout,
4452 dests, destfd, destoffsets,
4453 part, &degraded, buf);
4454 validate(afd, destfd[0], destoffsets[0]);
4455 /* record where 'part' is up to */
4456 part = !part;
4457 if (increasing)
4458 backup_point += actual_stripes * (chunk/512) * data;
4459 else
4460 backup_point -= actual_stripes * (chunk/512) * data;
4461 }
4462 }
4463
4464 /* FIXME maybe call progress_reshape one more time instead */
4465 /* remove any remaining suspension */
4466 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4467 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4468 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4469 sysfs_set_num(sra, NULL, "sync_min", 0);
4470
4471 if (reshape->before.data_disks == reshape->after.data_disks)
4472 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4473 free(buf);
4474 return done;
4475 }
4476
4477 /*
4478 * If any spare contains md_back_data-1 which is recent wrt mtime,
4479 * write that data into the array and update the super blocks with
4480 * the new reshape_progress
4481 */
4482 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4483 char *backup_file, int verbose)
4484 {
4485 int i, j;
4486 int old_disks;
4487 unsigned long long *offsets;
4488 unsigned long long nstripe, ostripe;
4489 int ndata, odata;
4490
4491 odata = info->array.raid_disks - info->delta_disks - 1;
4492 if (info->array.level == 6) odata--; /* number of data disks */
4493 ndata = info->array.raid_disks - 1;
4494 if (info->new_level == 6) ndata--;
4495
4496 old_disks = info->array.raid_disks - info->delta_disks;
4497
4498 if (info->delta_disks <= 0)
4499 /* Didn't grow, so the backup file must have
4500 * been used
4501 */
4502 old_disks = cnt;
4503 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4504 struct mdinfo dinfo;
4505 int fd;
4506 int bsbsize;
4507 char *devname, namebuf[20];
4508 unsigned long long lo, hi;
4509
4510 /* This was a spare and may have some saved data on it.
4511 * Load the superblock, find and load the
4512 * backup_super_block.
4513 * If either fail, go on to next device.
4514 * If the backup contains no new info, just return
4515 * else restore data and update all superblocks
4516 */
4517 if (i == old_disks-1) {
4518 fd = open(backup_file, O_RDONLY);
4519 if (fd<0) {
4520 pr_err("backup file %s inaccessible: %s\n",
4521 backup_file, strerror(errno));
4522 continue;
4523 }
4524 devname = backup_file;
4525 } else {
4526 fd = fdlist[i];
4527 if (fd < 0)
4528 continue;
4529 if (st->ss->load_super(st, fd, NULL))
4530 continue;
4531
4532 st->ss->getinfo_super(st, &dinfo, NULL);
4533 st->ss->free_super(st);
4534
4535 if (lseek64(fd,
4536 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4537 0) < 0) {
4538 pr_err("Cannot seek on device %d\n", i);
4539 continue; /* Cannot seek */
4540 }
4541 sprintf(namebuf, "device-%d", i);
4542 devname = namebuf;
4543 }
4544 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4545 if (verbose)
4546 pr_err("Cannot read from %s\n", devname);
4547 continue; /* Cannot read */
4548 }
4549 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4550 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4551 if (verbose)
4552 pr_err("No backup metadata on %s\n", devname);
4553 continue;
4554 }
4555 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4556 if (verbose)
4557 pr_err("Bad backup-metadata checksum on %s\n", devname);
4558 continue; /* bad checksum */
4559 }
4560 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4561 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4562 if (verbose)
4563 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4564 continue; /* Bad second checksum */
4565 }
4566 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4567 if (verbose)
4568 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4569 continue; /* Wrong uuid */
4570 }
4571
4572 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4573 * sometimes they aren't... So allow considerable flexability in matching, and allow
4574 * this test to be overridden by an environment variable.
4575 */
4576 if(time_after(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) + 2*60*60) ||
4577 time_before(info->array.utime, (unsigned int)__le64_to_cpu(bsb.mtime) - 10*60)) {
4578 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4579 pr_err("accepting backup with timestamp %lu for array with timestamp %lu\n",
4580 (unsigned long)__le64_to_cpu(bsb.mtime),
4581 (unsigned long)info->array.utime);
4582 } else {
4583 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4584 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4585 continue; /* time stamp is too bad */
4586 }
4587 }
4588
4589 if (bsb.magic[15] == '1') {
4590 if (bsb.length == 0)
4591 continue;
4592 if (info->delta_disks >= 0) {
4593 /* reshape_progress is increasing */
4594 if (__le64_to_cpu(bsb.arraystart)
4595 + __le64_to_cpu(bsb.length)
4596 < info->reshape_progress) {
4597 nonew:
4598 if (verbose)
4599 pr_err("backup-metadata found on %s but is not needed\n", devname);
4600 continue; /* No new data here */
4601 }
4602 } else {
4603 /* reshape_progress is decreasing */
4604 if (__le64_to_cpu(bsb.arraystart) >=
4605 info->reshape_progress)
4606 goto nonew; /* No new data here */
4607 }
4608 } else {
4609 if (bsb.length == 0 && bsb.length2 == 0)
4610 continue;
4611 if (info->delta_disks >= 0) {
4612 /* reshape_progress is increasing */
4613 if ((__le64_to_cpu(bsb.arraystart)
4614 + __le64_to_cpu(bsb.length)
4615 < info->reshape_progress)
4616 &&
4617 (__le64_to_cpu(bsb.arraystart2)
4618 + __le64_to_cpu(bsb.length2)
4619 < info->reshape_progress))
4620 goto nonew; /* No new data here */
4621 } else {
4622 /* reshape_progress is decreasing */
4623 if (__le64_to_cpu(bsb.arraystart) >=
4624 info->reshape_progress &&
4625 __le64_to_cpu(bsb.arraystart2) >=
4626 info->reshape_progress)
4627 goto nonew; /* No new data here */
4628 }
4629 }
4630 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4631 second_fail:
4632 if (verbose)
4633 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4634 devname);
4635 continue; /* Cannot seek */
4636 }
4637 /* There should be a duplicate backup superblock 4k before here */
4638 if (lseek64(fd, -4096, 1) < 0 ||
4639 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4640 goto second_fail; /* Cannot find leading superblock */
4641 if (bsb.magic[15] == '1')
4642 bsbsize = offsetof(struct mdp_backup_super, pad1);
4643 else
4644 bsbsize = offsetof(struct mdp_backup_super, pad);
4645 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4646 goto second_fail; /* Cannot find leading superblock */
4647
4648 /* Now need the data offsets for all devices. */
4649 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4650 for(j=0; j<info->array.raid_disks; j++) {
4651 if (fdlist[j] < 0)
4652 continue;
4653 if (st->ss->load_super(st, fdlist[j], NULL))
4654 /* FIXME should be this be an error */
4655 continue;
4656 st->ss->getinfo_super(st, &dinfo, NULL);
4657 st->ss->free_super(st);
4658 offsets[j] = dinfo.data_offset * 512;
4659 }
4660 printf("%s: restoring critical section\n", Name);
4661
4662 if (restore_stripes(fdlist, offsets,
4663 info->array.raid_disks,
4664 info->new_chunk,
4665 info->new_level,
4666 info->new_layout,
4667 fd, __le64_to_cpu(bsb.devstart)*512,
4668 __le64_to_cpu(bsb.arraystart)*512,
4669 __le64_to_cpu(bsb.length)*512, NULL)) {
4670 /* didn't succeed, so giveup */
4671 if (verbose)
4672 pr_err("Error restoring backup from %s\n",
4673 devname);
4674 free(offsets);
4675 return 1;
4676 }
4677
4678 if (bsb.magic[15] == '2' &&
4679 restore_stripes(fdlist, offsets,
4680 info->array.raid_disks,
4681 info->new_chunk,
4682 info->new_level,
4683 info->new_layout,
4684 fd, __le64_to_cpu(bsb.devstart)*512 +
4685 __le64_to_cpu(bsb.devstart2)*512,
4686 __le64_to_cpu(bsb.arraystart2)*512,
4687 __le64_to_cpu(bsb.length2)*512, NULL)) {
4688 /* didn't succeed, so giveup */
4689 if (verbose)
4690 pr_err("Error restoring second backup from %s\n",
4691 devname);
4692 free(offsets);
4693 return 1;
4694 }
4695
4696 free(offsets);
4697
4698 /* Ok, so the data is restored. Let's update those superblocks. */
4699
4700 lo = hi = 0;
4701 if (bsb.length) {
4702 lo = __le64_to_cpu(bsb.arraystart);
4703 hi = lo + __le64_to_cpu(bsb.length);
4704 }
4705 if (bsb.magic[15] == '2' && bsb.length2) {
4706 unsigned long long lo1, hi1;
4707 lo1 = __le64_to_cpu(bsb.arraystart2);
4708 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4709 if (lo == hi) {
4710 lo = lo1;
4711 hi = hi1;
4712 } else if (lo < lo1)
4713 hi = hi1;
4714 else
4715 lo = lo1;
4716 }
4717 if (lo < hi &&
4718 (info->reshape_progress < lo ||
4719 info->reshape_progress > hi))
4720 /* backup does not affect reshape_progress*/ ;
4721 else if (info->delta_disks >= 0) {
4722 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4723 __le64_to_cpu(bsb.length);
4724 if (bsb.magic[15] == '2') {
4725 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4726 __le64_to_cpu(bsb.length2);
4727 if (p2 > info->reshape_progress)
4728 info->reshape_progress = p2;
4729 }
4730 } else {
4731 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4732 if (bsb.magic[15] == '2') {
4733 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4734 if (p2 < info->reshape_progress)
4735 info->reshape_progress = p2;
4736 }
4737 }
4738 for (j=0; j<info->array.raid_disks; j++) {
4739 if (fdlist[j] < 0)
4740 continue;
4741 if (st->ss->load_super(st, fdlist[j], NULL))
4742 continue;
4743 st->ss->getinfo_super(st, &dinfo, NULL);
4744 dinfo.reshape_progress = info->reshape_progress;
4745 st->ss->update_super(st, &dinfo,
4746 "_reshape_progress",
4747 NULL,0, 0, NULL);
4748 st->ss->store_super(st, fdlist[j]);
4749 st->ss->free_super(st);
4750 }
4751 return 0;
4752 }
4753 /* Didn't find any backup data, try to see if any
4754 * was needed.
4755 */
4756 if (info->delta_disks < 0) {
4757 /* When shrinking, the critical section is at the end.
4758 * So see if we are before the critical section.
4759 */
4760 unsigned long long first_block;
4761 nstripe = ostripe = 0;
4762 first_block = 0;
4763 while (ostripe >= nstripe) {
4764 ostripe += info->array.chunk_size / 512;
4765 first_block = ostripe * odata;
4766 nstripe = first_block / ndata / (info->new_chunk/512) *
4767 (info->new_chunk/512);
4768 }
4769
4770 if (info->reshape_progress >= first_block)
4771 return 0;
4772 }
4773 if (info->delta_disks > 0) {
4774 /* See if we are beyond the critical section. */
4775 unsigned long long last_block;
4776 nstripe = ostripe = 0;
4777 last_block = 0;
4778 while (nstripe >= ostripe) {
4779 nstripe += info->new_chunk / 512;
4780 last_block = nstripe * ndata;
4781 ostripe = last_block / odata / (info->array.chunk_size/512) *
4782 (info->array.chunk_size/512);
4783 }
4784
4785 if (info->reshape_progress >= last_block)
4786 return 0;
4787 }
4788 /* needed to recover critical section! */
4789 if (verbose)
4790 pr_err("Failed to find backup of critical section\n");
4791 return 1;
4792 }
4793
4794 int Grow_continue_command(char *devname, int fd,
4795 char *backup_file, int verbose)
4796 {
4797 int ret_val = 0;
4798 struct supertype *st = NULL;
4799 struct mdinfo *content = NULL;
4800 struct mdinfo array;
4801 char *subarray = NULL;
4802 struct mdinfo *cc = NULL;
4803 struct mdstat_ent *mdstat = NULL;
4804 int cfd = -1;
4805 int fd2;
4806
4807 dprintf("Grow continue from command line called for %s\n",
4808 devname);
4809
4810 st = super_by_fd(fd, &subarray);
4811 if (!st || !st->ss) {
4812 pr_err("Unable to determine metadata format for %s\n",
4813 devname);
4814 return 1;
4815 }
4816 dprintf("Grow continue is run for ");
4817 if (st->ss->external == 0) {
4818 int d;
4819 int cnt = 5;
4820 dprintf_cont("native array (%s)\n", devname);
4821 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4822 pr_err("%s is not an active md array - aborting\n", devname);
4823 ret_val = 1;
4824 goto Grow_continue_command_exit;
4825 }
4826 content = &array;
4827 /* Need to load a superblock.
4828 * FIXME we should really get what we need from
4829 * sysfs
4830 */
4831 do {
4832 for (d = 0; d < MAX_DISKS; d++) {
4833 mdu_disk_info_t disk;
4834 char *dv;
4835 int err;
4836 disk.number = d;
4837 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4838 continue;
4839 if (disk.major == 0 && disk.minor == 0)
4840 continue;
4841 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4842 continue;
4843 dv = map_dev(disk.major, disk.minor, 1);
4844 if (!dv)
4845 continue;
4846 fd2 = dev_open(dv, O_RDONLY);
4847 if (fd2 < 0)
4848 continue;
4849 err = st->ss->load_super(st, fd2, NULL);
4850 close(fd2);
4851 if (err)
4852 continue;
4853 break;
4854 }
4855 if (d == MAX_DISKS) {
4856 pr_err("Unable to load metadata for %s\n",
4857 devname);
4858 ret_val = 1;
4859 goto Grow_continue_command_exit;
4860 }
4861 st->ss->getinfo_super(st, content, NULL);
4862 if (!content->reshape_active)
4863 sleep(3);
4864 else
4865 break;
4866 } while (cnt-- > 0);
4867 } else {
4868 char *container;
4869
4870 if (subarray) {
4871 dprintf_cont("subarray (%s)\n", subarray);
4872 container = st->container_devnm;
4873 cfd = open_dev_excl(st->container_devnm);
4874 } else {
4875 container = st->devnm;
4876 close(fd);
4877 cfd = open_dev_excl(st->devnm);
4878 dprintf_cont("container (%s)\n", container);
4879 fd = cfd;
4880 }
4881 if (cfd < 0) {
4882 pr_err("Unable to open container for %s\n", devname);
4883 ret_val = 1;
4884 goto Grow_continue_command_exit;
4885 }
4886
4887 /* find in container array under reshape
4888 */
4889 ret_val = st->ss->load_container(st, cfd, NULL);
4890 if (ret_val) {
4891 pr_err("Cannot read superblock for %s\n",
4892 devname);
4893 ret_val = 1;
4894 goto Grow_continue_command_exit;
4895 }
4896
4897 cc = st->ss->container_content(st, subarray);
4898 for (content = cc; content ; content = content->next) {
4899 char *array;
4900 int allow_reshape = 1;
4901
4902 if (content->reshape_active == 0)
4903 continue;
4904 /* The decision about array or container wide
4905 * reshape is taken in Grow_continue based
4906 * content->reshape_active state, therefore we
4907 * need to check_reshape based on
4908 * reshape_active and subarray name
4909 */
4910 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4911 allow_reshape = 0;
4912 if (content->reshape_active == CONTAINER_RESHAPE &&
4913 (content->array.state
4914 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4915 allow_reshape = 0;
4916
4917 if (!allow_reshape) {
4918 pr_err("cannot continue reshape of an array in container with unsupported metadata: %s(%s)\n",
4919 devname, container);
4920 ret_val = 1;
4921 goto Grow_continue_command_exit;
4922 }
4923
4924 array = strchr(content->text_version+1, '/')+1;
4925 mdstat = mdstat_by_subdev(array, container);
4926 if (!mdstat)
4927 continue;
4928 if (mdstat->active == 0) {
4929 pr_err("Skipping inactive array %s.\n",
4930 mdstat->devnm);
4931 free_mdstat(mdstat);
4932 mdstat = NULL;
4933 continue;
4934 }
4935 break;
4936 }
4937 if (!content) {
4938 pr_err("Unable to determine reshaped array for %s\n", devname);
4939 ret_val = 1;
4940 goto Grow_continue_command_exit;
4941 }
4942 fd2 = open_dev(mdstat->devnm);
4943 if (fd2 < 0) {
4944 pr_err("cannot open (%s)\n", mdstat->devnm);
4945 ret_val = 1;
4946 goto Grow_continue_command_exit;
4947 }
4948
4949 sysfs_init(content, fd2, mdstat->devnm);
4950
4951 close(fd2);
4952
4953 /* start mdmon in case it is not running
4954 */
4955 if (!mdmon_running(container))
4956 start_mdmon(container);
4957 ping_monitor(container);
4958
4959 if (mdmon_running(container))
4960 st->update_tail = &st->updates;
4961 else {
4962 pr_err("No mdmon found. Grow cannot continue.\n");
4963 ret_val = 1;
4964 goto Grow_continue_command_exit;
4965 }
4966 }
4967
4968 /* verify that array under reshape is started from
4969 * correct position
4970 */
4971 if (verify_reshape_position(content, content->array.level) < 0) {
4972 ret_val = 1;
4973 goto Grow_continue_command_exit;
4974 }
4975
4976 /* continue reshape
4977 */
4978 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4979
4980 Grow_continue_command_exit:
4981 if (cfd > -1)
4982 close(cfd);
4983 st->ss->free_super(st);
4984 free_mdstat(mdstat);
4985 sysfs_free(cc);
4986 free(subarray);
4987
4988 return ret_val;
4989 }
4990
4991 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4992 char *backup_file, int forked, int freeze_reshape)
4993 {
4994 int ret_val = 2;
4995
4996 if (!info->reshape_active)
4997 return ret_val;
4998
4999 if (st->ss->external) {
5000 int cfd = open_dev(st->container_devnm);
5001
5002 if (cfd < 0)
5003 return 1;
5004
5005 st->ss->load_container(st, cfd, st->container_devnm);
5006 close(cfd);
5007 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
5008 st, info, 0, backup_file,
5009 0, forked,
5010 1 | info->reshape_active,
5011 freeze_reshape);
5012 } else
5013 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
5014 NULL, INVALID_SECTORS,
5015 backup_file, 0, forked,
5016 1 | info->reshape_active,
5017 freeze_reshape);
5018
5019 return ret_val;
5020 }
5021
5022 char *make_backup(char *name)
5023 {
5024 char *base = "backup_file-";
5025 int len;
5026 char *fname;
5027
5028 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
5029 fname = xmalloc(len);
5030 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
5031 return fname;
5032 }
5033
5034 char *locate_backup(char *name)
5035 {
5036 char *fl = make_backup(name);
5037 struct stat stb;
5038
5039 if (stat(fl, &stb) == 0 &&
5040 S_ISREG(stb.st_mode))
5041 return fl;
5042
5043 free(fl);
5044 return NULL;
5045 }
Unnamed repository; edit this file 'description' to name the repository.