]> 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
Change way of printing name of a process
[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"
91 " section for reshape - sorry.\n");
92 if (!backup_file)
93 pr_err("Possibly you need"
94 " to specify a --backup-file\n");
95 return 1;
96 }
97
98 dprintf("restore_backup() returns status OK.\n");
99 return 0;
100 }
101
102 int Grow_Add_device(char *devname, int fd, char *newdev)
103 {
104 /* Add a device to an active array.
105 * Currently, just extend a linear array.
106 * This requires writing a new superblock on the
107 * new device, calling the kernel to add the device,
108 * and if that succeeds, update the superblock on
109 * all other devices.
110 * This means that we need to *find* all other devices.
111 */
112 struct mdinfo info;
113
114 struct stat stb;
115 int nfd, fd2;
116 int d, nd;
117 struct supertype *st = NULL;
118 char *subarray = NULL;
119
120 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
121 pr_err("cannot get array info for %s\n", devname);
122 return 1;
123 }
124
125 if (info.array.level != -1) {
126 pr_err("can only add devices to linear arrays\n");
127 return 1;
128 }
129
130 st = super_by_fd(fd, &subarray);
131 if (!st) {
132 pr_err("cannot handle arrays with superblock version %d\n",
133 info.array.major_version);
134 return 1;
135 }
136
137 if (subarray) {
138 pr_err("Cannot grow linear sub-arrays yet\n");
139 free(subarray);
140 free(st);
141 return 1;
142 }
143
144 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
145 if (nfd < 0) {
146 pr_err("cannot open %s\n", newdev);
147 free(st);
148 return 1;
149 }
150 fstat(nfd, &stb);
151 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
152 pr_err("%s is not a block device!\n", newdev);
153 close(nfd);
154 free(st);
155 return 1;
156 }
157 /* now check out all the devices and make sure we can read the
158 * superblock */
159 for (d=0 ; d < info.array.raid_disks ; d++) {
160 mdu_disk_info_t disk;
161 char *dv;
162
163 st->ss->free_super(st);
164
165 disk.number = d;
166 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
167 pr_err("cannot get device detail for device %d\n",
168 d);
169 close(nfd);
170 free(st);
171 return 1;
172 }
173 dv = map_dev(disk.major, disk.minor, 1);
174 if (!dv) {
175 pr_err("cannot find device file for device %d\n",
176 d);
177 close(nfd);
178 free(st);
179 return 1;
180 }
181 fd2 = dev_open(dv, O_RDWR);
182 if (fd2 < 0) {
183 pr_err("cannot open device file %s\n", dv);
184 close(nfd);
185 free(st);
186 return 1;
187 }
188
189 if (st->ss->load_super(st, fd2, NULL)) {
190 pr_err("cannot find super block on %s\n", dv);
191 close(nfd);
192 close(fd2);
193 free(st);
194 return 1;
195 }
196 close(fd2);
197 }
198 /* Ok, looks good. Lets update the superblock and write it out to
199 * newdev.
200 */
201
202 info.disk.number = d;
203 info.disk.major = major(stb.st_rdev);
204 info.disk.minor = minor(stb.st_rdev);
205 info.disk.raid_disk = d;
206 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
207 st->ss->update_super(st, &info, "linear-grow-new", newdev,
208 0, 0, NULL);
209
210 if (st->ss->store_super(st, nfd)) {
211 pr_err("Cannot store new superblock on %s\n",
212 newdev);
213 close(nfd);
214 return 1;
215 }
216 close(nfd);
217
218 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
219 pr_err("Cannot add new disk to this array\n");
220 return 1;
221 }
222 /* Well, that seems to have worked.
223 * Now go through and update all superblocks
224 */
225
226 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
227 pr_err("cannot get array info for %s\n", devname);
228 return 1;
229 }
230
231 nd = d;
232 for (d=0 ; d < info.array.raid_disks ; d++) {
233 mdu_disk_info_t disk;
234 char *dv;
235
236 disk.number = d;
237 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
238 pr_err("cannot get device detail for device %d\n",
239 d);
240 return 1;
241 }
242 dv = map_dev(disk.major, disk.minor, 1);
243 if (!dv) {
244 pr_err("cannot find device file for device %d\n",
245 d);
246 return 1;
247 }
248 fd2 = dev_open(dv, O_RDWR);
249 if (fd2 < 0) {
250 pr_err("cannot open device file %s\n", dv);
251 return 1;
252 }
253 if (st->ss->load_super(st, fd2, NULL)) {
254 pr_err("cannot find super block on %s\n", dv);
255 close(fd);
256 return 1;
257 }
258 info.array.raid_disks = nd+1;
259 info.array.nr_disks = nd+1;
260 info.array.active_disks = nd+1;
261 info.array.working_disks = nd+1;
262
263 st->ss->update_super(st, &info, "linear-grow-update", dv,
264 0, 0, NULL);
265
266 if (st->ss->store_super(st, fd2)) {
267 pr_err("Cannot store new superblock on %s\n", dv);
268 close(fd2);
269 return 1;
270 }
271 close(fd2);
272 }
273
274 return 0;
275 }
276
277 int Grow_addbitmap(char *devname, int fd, struct context *c, struct shape *s)
278 {
279 /*
280 * First check that array doesn't have a bitmap
281 * Then create the bitmap
282 * Then add it
283 *
284 * For internal bitmaps, we need to check the version,
285 * find all the active devices, and write the bitmap block
286 * to all devices
287 */
288 mdu_bitmap_file_t bmf;
289 mdu_array_info_t array;
290 struct supertype *st;
291 char *subarray = NULL;
292 int major = BITMAP_MAJOR_HI;
293 int vers = md_get_version(fd);
294 unsigned long long bitmapsize, array_size;
295
296 if (vers < 9003) {
297 major = BITMAP_MAJOR_HOSTENDIAN;
298 pr_err("Warning - bitmaps created on this kernel"
299 " are not portable\n"
300 " between different architectures. Consider upgrading"
301 " the Linux kernel.\n");
302 }
303
304 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
305 if (errno == ENOMEM)
306 pr_err("Memory allocation failure.\n");
307 else
308 pr_err("bitmaps not supported by this kernel.\n");
309 return 1;
310 }
311 if (bmf.pathname[0]) {
312 if (strcmp(s->bitmap_file,"none")==0) {
313 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
314 pr_err("failed to remove bitmap %s\n",
315 bmf.pathname);
316 return 1;
317 }
318 return 0;
319 }
320 pr_err("%s already has a bitmap (%s)\n",
321 devname, bmf.pathname);
322 return 1;
323 }
324 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
325 pr_err("cannot get array status for %s\n", devname);
326 return 1;
327 }
328 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
329 if (strcmp(s->bitmap_file, "none")==0) {
330 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
331 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
332 pr_err("failed to remove internal bitmap.\n");
333 return 1;
334 }
335 return 0;
336 }
337 pr_err("Internal bitmap already present on %s\n",
338 devname);
339 return 1;
340 }
341
342 if (strcmp(s->bitmap_file, "none") == 0) {
343 pr_err("no bitmap found on %s\n", devname);
344 return 1;
345 }
346 if (array.level <= 0) {
347 pr_err("Bitmaps not meaningful with level %s\n",
348 map_num(pers, array.level)?:"of this array");
349 return 1;
350 }
351 bitmapsize = array.size;
352 bitmapsize <<= 1;
353 if (get_dev_size(fd, NULL, &array_size) &&
354 array_size > (0x7fffffffULL<<9)) {
355 /* Array is big enough that we cannot trust array.size
356 * try other approaches
357 */
358 bitmapsize = get_component_size(fd);
359 }
360 if (bitmapsize == 0) {
361 pr_err("Cannot reliably determine size of array to create bitmap - sorry.\n");
362 return 1;
363 }
364
365 if (array.level == 10) {
366 int ncopies = (array.layout&255)*((array.layout>>8)&255);
367 bitmapsize = bitmapsize * array.raid_disks / ncopies;
368 }
369
370 st = super_by_fd(fd, &subarray);
371 if (!st) {
372 pr_err("Cannot understand version %d.%d\n",
373 array.major_version, array.minor_version);
374 return 1;
375 }
376 if (subarray) {
377 pr_err("Cannot add bitmaps to sub-arrays yet\n");
378 free(subarray);
379 free(st);
380 return 1;
381 }
382 if (strcmp(s->bitmap_file, "internal") == 0) {
383 int rv;
384 int d;
385 int offset_setable = 0;
386 struct mdinfo *mdi;
387 if (st->ss->add_internal_bitmap == NULL) {
388 pr_err("Internal bitmaps not supported "
389 "with %s metadata\n", st->ss->name);
390 return 1;
391 }
392 mdi = sysfs_read(fd, NULL, GET_BITMAP_LOCATION);
393 if (mdi)
394 offset_setable = 1;
395 for (d=0; d< st->max_devs; d++) {
396 mdu_disk_info_t disk;
397 char *dv;
398 disk.number = d;
399 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
400 continue;
401 if (disk.major == 0 &&
402 disk.minor == 0)
403 continue;
404 if ((disk.state & (1<<MD_DISK_SYNC))==0)
405 continue;
406 dv = map_dev(disk.major, disk.minor, 1);
407 if (dv) {
408 int fd2 = dev_open(dv, O_RDWR);
409 if (fd2 < 0)
410 continue;
411 if (st->ss->load_super(st, fd2, NULL)==0) {
412 if (st->ss->add_internal_bitmap(
413 st,
414 &s->bitmap_chunk, c->delay, s->write_behind,
415 bitmapsize, offset_setable,
416 major)
417 )
418 st->ss->write_bitmap(st, fd2);
419 else {
420 pr_err("failed to create internal bitmap"
421 " - chunksize problem.\n");
422 close(fd2);
423 return 1;
424 }
425 }
426 close(fd2);
427 }
428 }
429 if (offset_setable) {
430 st->ss->getinfo_super(st, mdi, NULL);
431 sysfs_init(mdi, fd, NULL);
432 rv = sysfs_set_num_signed(mdi, NULL, "bitmap/location",
433 mdi->bitmap_offset);
434 } else {
435 array.state |= (1<<MD_SB_BITMAP_PRESENT);
436 rv = ioctl(fd, SET_ARRAY_INFO, &array);
437 }
438 if (rv < 0) {
439 if (errno == EBUSY)
440 pr_err("Cannot add bitmap while array is"
441 " resyncing or reshaping etc.\n");
442 pr_err("failed to set internal bitmap.\n");
443 return 1;
444 }
445 } else {
446 int uuid[4];
447 int bitmap_fd;
448 int d;
449 int max_devs = st->max_devs;
450
451 /* try to load a superblock */
452 for (d = 0; d < max_devs; d++) {
453 mdu_disk_info_t disk;
454 char *dv;
455 int fd2;
456 disk.number = d;
457 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
458 continue;
459 if ((disk.major==0 && disk.minor==0) ||
460 (disk.state & (1<<MD_DISK_REMOVED)))
461 continue;
462 dv = map_dev(disk.major, disk.minor, 1);
463 if (!dv)
464 continue;
465 fd2 = dev_open(dv, O_RDONLY);
466 if (fd2 >= 0) {
467 if (st->ss->load_super(st, fd2, NULL) == 0) {
468 close(fd2);
469 st->ss->uuid_from_super(st, uuid);
470 break;
471 }
472 close(fd2);
473 }
474 }
475 if (d == max_devs) {
476 pr_err("cannot find UUID for array!\n");
477 return 1;
478 }
479 if (CreateBitmap(s->bitmap_file, c->force, (char*)uuid, s->bitmap_chunk,
480 c->delay, s->write_behind, bitmapsize, major)) {
481 return 1;
482 }
483 bitmap_fd = open(s->bitmap_file, O_RDWR);
484 if (bitmap_fd < 0) {
485 pr_err("weird: %s cannot be opened\n",
486 s->bitmap_file);
487 return 1;
488 }
489 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
490 int err = errno;
491 if (errno == EBUSY)
492 pr_err("Cannot add bitmap while array is"
493 " resyncing or reshaping etc.\n");
494 pr_err("Cannot set bitmap file for %s: %s\n",
495 devname, strerror(err));
496 return 1;
497 }
498 }
499
500 return 0;
501 }
502
503 /*
504 * When reshaping an array we might need to backup some data.
505 * This is written to all spares with a 'super_block' describing it.
506 * The superblock goes 4K from the end of the used space on the
507 * device.
508 * It if written after the backup is complete.
509 * It has the following structure.
510 */
511
512 static struct mdp_backup_super {
513 char magic[16]; /* md_backup_data-1 or -2 */
514 __u8 set_uuid[16];
515 __u64 mtime;
516 /* start/sizes in 512byte sectors */
517 __u64 devstart; /* address on backup device/file of data */
518 __u64 arraystart;
519 __u64 length;
520 __u32 sb_csum; /* csum of preceeding bytes. */
521 __u32 pad1;
522 __u64 devstart2; /* offset in to data of second section */
523 __u64 arraystart2;
524 __u64 length2;
525 __u32 sb_csum2; /* csum of preceeding bytes. */
526 __u8 pad[512-68-32];
527 } __attribute__((aligned(512))) bsb, bsb2;
528
529 static __u32 bsb_csum(char *buf, int len)
530 {
531 int i;
532 int csum = 0;
533 for (i = 0; i < len; i++)
534 csum = (csum<<3) + buf[0];
535 return __cpu_to_le32(csum);
536 }
537
538 static int check_idle(struct supertype *st)
539 {
540 /* Check that all member arrays for this container, or the
541 * container of this array, are idle
542 */
543 char *container = (st->container_devnm[0]
544 ? st->container_devnm : st->devnm);
545 struct mdstat_ent *ent, *e;
546 int is_idle = 1;
547
548 ent = mdstat_read(0, 0);
549 for (e = ent ; e; e = e->next) {
550 if (!is_container_member(e, container))
551 continue;
552 if (e->percent >= 0) {
553 is_idle = 0;
554 break;
555 }
556 }
557 free_mdstat(ent);
558 return is_idle;
559 }
560
561 static int freeze_container(struct supertype *st)
562 {
563 char *container = (st->container_devnm[0]
564 ? st->container_devnm : st->devnm);
565
566 if (!check_idle(st))
567 return -1;
568
569 if (block_monitor(container, 1)) {
570 pr_err("failed to freeze container\n");
571 return -2;
572 }
573
574 return 1;
575 }
576
577 static void unfreeze_container(struct supertype *st)
578 {
579 char *container = (st->container_devnm[0]
580 ? st->container_devnm : st->devnm);
581
582 unblock_monitor(container, 1);
583 }
584
585 static int freeze(struct supertype *st)
586 {
587 /* Try to freeze resync/rebuild on this array/container.
588 * Return -1 if the array is busy,
589 * return -2 container cannot be frozen,
590 * return 0 if this kernel doesn't support 'frozen'
591 * return 1 if it worked.
592 */
593 if (st->ss->external)
594 return freeze_container(st);
595 else {
596 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
597 int err;
598 char buf[20];
599
600 if (!sra)
601 return -1;
602 /* Need to clear any 'read-auto' status */
603 if (sysfs_get_str(sra, NULL, "array_state", buf, 20) > 0 &&
604 strncmp(buf, "read-auto", 9) == 0)
605 sysfs_set_str(sra, NULL, "array_state", "clean");
606
607 err = sysfs_freeze_array(sra);
608 sysfs_free(sra);
609 return err;
610 }
611 }
612
613 static void unfreeze(struct supertype *st)
614 {
615 if (st->ss->external)
616 return unfreeze_container(st);
617 else {
618 struct mdinfo *sra = sysfs_read(-1, st->devnm, GET_VERSION);
619 char buf[20];
620
621 if (sra &&
622 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0
623 && strcmp(buf, "frozen\n") == 0) {
624 printf("unfreeze\n");
625 sysfs_set_str(sra, NULL, "sync_action", "idle");
626 }
627 sysfs_free(sra);
628 }
629 }
630
631 static void wait_reshape(struct mdinfo *sra)
632 {
633 int fd = sysfs_get_fd(sra, NULL, "sync_action");
634 char action[20];
635
636 if (fd < 0)
637 return;
638
639 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
640 strncmp(action, "reshape", 7) == 0)
641 sysfs_wait(fd, NULL);
642 close(fd);
643 }
644
645 static int reshape_super(struct supertype *st, unsigned long long size,
646 int level, int layout, int chunksize, int raid_disks,
647 int delta_disks, char *backup_file, char *dev,
648 int direction, int verbose)
649 {
650 /* nothing extra to check in the native case */
651 if (!st->ss->external)
652 return 0;
653 if (!st->ss->reshape_super ||
654 !st->ss->manage_reshape) {
655 pr_err("%s metadata does not support reshape\n",
656 st->ss->name);
657 return 1;
658 }
659
660 return st->ss->reshape_super(st, size, level, layout, chunksize,
661 raid_disks, delta_disks, backup_file, dev,
662 direction, verbose);
663 }
664
665 static void sync_metadata(struct supertype *st)
666 {
667 if (st->ss->external) {
668 if (st->update_tail) {
669 flush_metadata_updates(st);
670 st->update_tail = &st->updates;
671 } else
672 st->ss->sync_metadata(st);
673 }
674 }
675
676 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
677 {
678 /* when dealing with external metadata subarrays we need to be
679 * prepared to handle EAGAIN. The kernel may need to wait for
680 * mdmon to mark the array active so the kernel can handle
681 * allocations/writeback when preparing the reshape action
682 * (md_allow_write()). We temporarily disable safe_mode_delay
683 * to close a race with the array_state going clean before the
684 * next write to raid_disks / stripe_cache_size
685 */
686 char safe[50];
687 int rc;
688
689 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
690 if (!container ||
691 (strcmp(name, "raid_disks") != 0 &&
692 strcmp(name, "stripe_cache_size") != 0))
693 return sysfs_set_num(sra, NULL, name, n);
694
695 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
696 if (rc <= 0)
697 return -1;
698 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
699 rc = sysfs_set_num(sra, NULL, name, n);
700 if (rc < 0 && errno == EAGAIN) {
701 ping_monitor(container);
702 /* if we get EAGAIN here then the monitor is not active
703 * so stop trying
704 */
705 rc = sysfs_set_num(sra, NULL, name, n);
706 }
707 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
708 return rc;
709 }
710
711 int start_reshape(struct mdinfo *sra, int already_running,
712 int before_data_disks, int data_disks)
713 {
714 int err;
715 unsigned long long sync_max_to_set;
716
717 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
718 err = sysfs_set_num(sra, NULL, "suspend_hi", sra->reshape_progress);
719 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo",
720 sra->reshape_progress);
721 if (before_data_disks <= data_disks)
722 sync_max_to_set = sra->reshape_progress / data_disks;
723 else
724 sync_max_to_set = (sra->component_size * data_disks
725 - sra->reshape_progress) / data_disks;
726 if (!already_running)
727 sysfs_set_num(sra, NULL, "sync_min", sync_max_to_set);
728 err = err ?: sysfs_set_num(sra, NULL, "sync_max", sync_max_to_set);
729 if (!already_running)
730 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
731
732 return err;
733 }
734
735 void abort_reshape(struct mdinfo *sra)
736 {
737 sysfs_set_str(sra, NULL, "sync_action", "idle");
738 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
739 sysfs_set_num(sra, NULL, "suspend_hi", 0);
740 sysfs_set_num(sra, NULL, "suspend_lo", 0);
741 sysfs_set_num(sra, NULL, "sync_min", 0);
742 // It isn't safe to reset sync_max as we aren't monitoring.
743 // Array really should be stopped at this point.
744 }
745
746 int remove_disks_for_takeover(struct supertype *st,
747 struct mdinfo *sra,
748 int layout)
749 {
750 int nr_of_copies;
751 struct mdinfo *remaining;
752 int slot;
753
754 if (sra->array.level == 10)
755 nr_of_copies = layout & 0xff;
756 else if (sra->array.level == 1)
757 nr_of_copies = sra->array.raid_disks;
758 else
759 return 1;
760
761 remaining = sra->devs;
762 sra->devs = NULL;
763 /* for each 'copy', select one device and remove from the list. */
764 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
765 struct mdinfo **diskp;
766 int found = 0;
767
768 /* Find a working device to keep */
769 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
770 struct mdinfo *disk = *diskp;
771
772 if (disk->disk.raid_disk < slot)
773 continue;
774 if (disk->disk.raid_disk >= slot + nr_of_copies)
775 continue;
776 if (disk->disk.state & (1<<MD_DISK_REMOVED))
777 continue;
778 if (disk->disk.state & (1<<MD_DISK_FAULTY))
779 continue;
780 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
781 continue;
782
783 /* We have found a good disk to use! */
784 *diskp = disk->next;
785 disk->next = sra->devs;
786 sra->devs = disk;
787 found = 1;
788 break;
789 }
790 if (!found)
791 break;
792 }
793
794 if (slot < sra->array.raid_disks) {
795 /* didn't find all slots */
796 struct mdinfo **e;
797 e = &remaining;
798 while (*e)
799 e = &(*e)->next;
800 *e = sra->devs;
801 sra->devs = remaining;
802 return 1;
803 }
804
805 /* Remove all 'remaining' devices from the array */
806 while (remaining) {
807 struct mdinfo *sd = remaining;
808 remaining = sd->next;
809
810 sysfs_set_str(sra, sd, "state", "faulty");
811 sysfs_set_str(sra, sd, "slot", "none");
812 /* for external metadata disks should be removed in mdmon */
813 if (!st->ss->external)
814 sysfs_set_str(sra, sd, "state", "remove");
815 sd->disk.state |= (1<<MD_DISK_REMOVED);
816 sd->disk.state &= ~(1<<MD_DISK_SYNC);
817 sd->next = sra->devs;
818 sra->devs = sd;
819 }
820 return 0;
821 }
822
823 void reshape_free_fdlist(int *fdlist,
824 unsigned long long *offsets,
825 int size)
826 {
827 int i;
828
829 for (i = 0; i < size; i++)
830 if (fdlist[i] >= 0)
831 close(fdlist[i]);
832
833 free(fdlist);
834 free(offsets);
835 }
836
837 int reshape_prepare_fdlist(char *devname,
838 struct mdinfo *sra,
839 int raid_disks,
840 int nrdisks,
841 unsigned long blocks,
842 char *backup_file,
843 int *fdlist,
844 unsigned long long *offsets)
845 {
846 int d = 0;
847 struct mdinfo *sd;
848
849 enable_fds(nrdisks);
850 for (d = 0; d <= nrdisks; d++)
851 fdlist[d] = -1;
852 d = raid_disks;
853 for (sd = sra->devs; sd; sd = sd->next) {
854 if (sd->disk.state & (1<<MD_DISK_FAULTY))
855 continue;
856 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
857 char *dn = map_dev(sd->disk.major,
858 sd->disk.minor, 1);
859 fdlist[sd->disk.raid_disk]
860 = dev_open(dn, O_RDONLY);
861 offsets[sd->disk.raid_disk] = sd->data_offset*512;
862 if (fdlist[sd->disk.raid_disk] < 0) {
863 pr_err("%s: cannot open component %s\n",
864 devname, dn ? dn : "-unknown-");
865 d = -1;
866 goto release;
867 }
868 } else if (backup_file == NULL) {
869 /* spare */
870 char *dn = map_dev(sd->disk.major,
871 sd->disk.minor, 1);
872 fdlist[d] = dev_open(dn, O_RDWR);
873 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
874 if (fdlist[d] < 0) {
875 pr_err("%s: cannot open component %s\n",
876 devname, dn ? dn : "-unknown-");
877 d = -1;
878 goto release;
879 }
880 d++;
881 }
882 }
883 release:
884 return d;
885 }
886
887 int reshape_open_backup_file(char *backup_file,
888 int fd,
889 char *devname,
890 long blocks,
891 int *fdlist,
892 unsigned long long *offsets,
893 char *sys_name,
894 int restart)
895 {
896 /* Return 1 on success, 0 on any form of failure */
897 /* need to check backup file is large enough */
898 char buf[512];
899 struct stat stb;
900 unsigned int dev;
901 int i;
902
903 *fdlist = open(backup_file, O_RDWR|O_CREAT|(restart ? O_TRUNC : O_EXCL),
904 S_IRUSR | S_IWUSR);
905 *offsets = 8 * 512;
906 if (*fdlist < 0) {
907 pr_err("%s: cannot create backup file %s: %s\n",
908 devname, backup_file, strerror(errno));
909 return 0;
910 }
911 /* Guard against backup file being on array device.
912 * If array is partitioned or if LVM etc is in the
913 * way this will not notice, but it is better than
914 * nothing.
915 */
916 fstat(*fdlist, &stb);
917 dev = stb.st_dev;
918 fstat(fd, &stb);
919 if (stb.st_rdev == dev) {
920 pr_err("backup file must NOT be"
921 " on the array being reshaped.\n");
922 close(*fdlist);
923 return 0;
924 }
925
926 memset(buf, 0, 512);
927 for (i=0; i < blocks + 8 ; i++) {
928 if (write(*fdlist, buf, 512) != 512) {
929 pr_err("%s: cannot create"
930 " backup file %s: %s\n",
931 devname, backup_file, strerror(errno));
932 return 0;
933 }
934 }
935 if (fsync(*fdlist) != 0) {
936 pr_err("%s: cannot create backup file %s: %s\n",
937 devname, backup_file, strerror(errno));
938 return 0;
939 }
940
941 if (!restart && strncmp(backup_file, MAP_DIR, strlen(MAP_DIR)) != 0) {
942 char *bu = make_backup(sys_name);
943 if (symlink(backup_file, bu))
944 pr_err("Recording backup file in " MAP_DIR "failed: %s\n",
945 strerror(errno));
946 free(bu);
947 }
948
949 return 1;
950 }
951
952 unsigned long compute_backup_blocks(int nchunk, int ochunk,
953 unsigned int ndata, unsigned int odata)
954 {
955 unsigned long a, b, blocks;
956 /* So how much do we need to backup.
957 * We need an amount of data which is both a whole number of
958 * old stripes and a whole number of new stripes.
959 * So LCM for (chunksize*datadisks).
960 */
961 a = (ochunk/512) * odata;
962 b = (nchunk/512) * ndata;
963 /* Find GCD */
964 a = GCD(a, b);
965 /* LCM == product / GCD */
966 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
967
968 return blocks;
969 }
970
971 char *analyse_change(char *devname, struct mdinfo *info, struct reshape *re)
972 {
973 /* Based on the current array state in info->array and
974 * the changes in info->new_* etc, determine:
975 * - whether the change is possible
976 * - Intermediate level/raid_disks/layout
977 * - whether a restriping reshape is needed
978 * - number of sectors in minimum change unit. This
979 * will cover a whole number of stripes in 'before' and
980 * 'after'.
981 *
982 * Return message if the change should be rejected
983 * NULL if the change can be achieved
984 *
985 * This can be called as part of starting a reshape, or
986 * when assembling an array that is undergoing reshape.
987 */
988 int near, far, offset, copies;
989 int new_disks;
990 int old_chunk, new_chunk;
991 /* delta_parity records change in number of devices
992 * caused by level change
993 */
994 int delta_parity = 0;
995
996 memset(re, 0, sizeof(*re));
997
998 /* If a new level not explicitly given, we assume no-change */
999 if (info->new_level == UnSet)
1000 info->new_level = info->array.level;
1001
1002 if (info->new_chunk)
1003 switch (info->new_level) {
1004 case 0:
1005 case 4:
1006 case 5:
1007 case 6:
1008 case 10:
1009 /* chunk size is meaningful, must divide component_size
1010 * evenly
1011 */
1012 if (info->component_size % (info->new_chunk/512)) {
1013 unsigned long long shrink = info->component_size;
1014 shrink &= ~(unsigned long long)(info->new_chunk/512-1);
1015 pr_err("New chunk size (%dK) does not evenly divide device size (%lluk)\n",
1016 info->new_chunk/1024, info->component_size/2);
1017 pr_err("After shrinking any filesystem, \"mdadm --grow %s --size %llu\"\n",
1018 devname, shrink/2);
1019 pr_err("will shrink the array so the given chunk size would work.\n");
1020 return "";
1021 }
1022 break;
1023 default:
1024 return "chunk size not meaningful for this level";
1025 }
1026 else
1027 info->new_chunk = info->array.chunk_size;
1028
1029 switch (info->array.level) {
1030 default:
1031 return "No reshape is possibly for this RAID level";
1032 case LEVEL_LINEAR:
1033 if (info->delta_disks != UnSet)
1034 return "Only --add is supported for LINEAR, setting --raid-disks is not needed";
1035 else
1036 return "Only --add is supported for LINEAR, other --grow options are not meaningful";
1037 case 1:
1038 /* RAID1 can convert to RAID1 with different disks, or
1039 * raid5 with 2 disks, or
1040 * raid0 with 1 disk
1041 */
1042 if (info->new_level > 1 &&
1043 (info->component_size & 7))
1044 return "Cannot convert RAID1 of this size - "
1045 "reduce size to multiple of 4K first.";
1046 if (info->new_level == 0) {
1047 if (info->delta_disks != UnSet &&
1048 info->delta_disks != 0)
1049 return "Cannot change number of disks "
1050 "with RAID1->RAID0 conversion";
1051 re->level = 0;
1052 re->before.data_disks = 1;
1053 re->after.data_disks = 1;
1054 return NULL;
1055 }
1056 if (info->new_level == 1) {
1057 if (info->delta_disks == UnSet)
1058 /* Don't know what to do */
1059 return "no change requested for Growing RAID1";
1060 re->level = 1;
1061 return NULL;
1062 }
1063 if (info->array.raid_disks == 2 &&
1064 info->new_level == 5) {
1065
1066 re->level = 5;
1067 re->before.data_disks = 1;
1068 if (info->delta_disks != UnSet &&
1069 info->delta_disks != 0)
1070 re->after.data_disks = 1 + info->delta_disks;
1071 else
1072 re->after.data_disks = 1;
1073 if (re->after.data_disks < 1)
1074 return "Number of disks too small for RAID5";
1075
1076 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
1077 info->array.chunk_size = 65536;
1078 break;
1079 }
1080 /* Could do some multi-stage conversions, but leave that to
1081 * later.
1082 */
1083 return "Impossibly level change request for RAID1";
1084
1085 case 10:
1086 /* RAID10 can be converted from near mode to
1087 * RAID0 by removing some devices.
1088 * It can also be reshaped if the kernel supports
1089 * new_data_offset.
1090 */
1091 switch (info->new_level) {
1092 case 0:
1093 if ((info->array.layout & ~0xff) != 0x100)
1094 return "Cannot Grow RAID10 with far/offset layout";
1095 /* number of devices must be multiple of number of copies */
1096 if (info->array.raid_disks % (info->array.layout & 0xff))
1097 return "RAID10 layout too complex for Grow operation";
1098
1099 new_disks = (info->array.raid_disks
1100 / (info->array.layout & 0xff));
1101 if (info->delta_disks == UnSet)
1102 info->delta_disks = (new_disks
1103 - info->array.raid_disks);
1104
1105 if (info->delta_disks != new_disks - info->array.raid_disks)
1106 return "New number of raid-devices impossible for RAID10";
1107 if (info->new_chunk &&
1108 info->new_chunk != info->array.chunk_size)
1109 return "Cannot change chunk-size with RAID10 Grow";
1110
1111 /* looks good */
1112 re->level = 0;
1113 re->before.data_disks = new_disks;
1114 re->after.data_disks = re->before.data_disks;
1115 return NULL;
1116
1117 case 10:
1118 near = info->array.layout & 0xff;
1119 far = (info->array.layout >> 8) & 0xff;
1120 offset = info->array.layout & 0x10000;
1121 if (far > 1 && !offset)
1122 return "Cannot reshape RAID10 in far-mode";
1123 copies = near * far;
1124
1125 old_chunk = info->array.chunk_size * far;
1126
1127 if (info->new_layout == UnSet)
1128 info->new_layout = info->array.layout;
1129 else {
1130 near = info->new_layout & 0xff;
1131 far = (info->new_layout >> 8) & 0xff;
1132 offset = info->new_layout & 0x10000;
1133 if (far > 1 && !offset)
1134 return "Cannot reshape RAID10 to far-mode";
1135 if (near * far != copies)
1136 return "Cannot change number of copies"
1137 " when reshaping RAID10";
1138 }
1139 if (info->delta_disks == UnSet)
1140 info->delta_disks = 0;
1141 new_disks = (info->array.raid_disks +
1142 info->delta_disks);
1143
1144 new_chunk = info->new_chunk * far;
1145
1146 re->level = 10;
1147 re->before.layout = info->array.layout;
1148 re->before.data_disks = info->array.raid_disks;
1149 re->after.layout = info->new_layout;
1150 re->after.data_disks = new_disks;
1151 /* For RAID10 we don't do backup but do allow reshape,
1152 * so set backup_blocks to INVALID_SECTORS rather than
1153 * zero.
1154 * And there is no need to synchronise stripes on both
1155 * 'old' and 'new'. So the important
1156 * number is the minimum data_offset difference
1157 * which is the larger of (offset copies * chunk).
1158 */
1159 re->backup_blocks = INVALID_SECTORS;
1160 re->min_offset_change = max(old_chunk, new_chunk) / 512;
1161 if (new_disks < re->before.data_disks &&
1162 info->space_after < re->min_offset_change)
1163 /* Reduce component size by one chunk */
1164 re->new_size = (info->component_size -
1165 re->min_offset_change);
1166 else
1167 re->new_size = info->component_size;
1168 re->new_size = re->new_size * new_disks / copies;
1169 return NULL;
1170
1171 default:
1172 return "RAID10 can only be changed to RAID0";
1173 }
1174 case 0:
1175 /* RAID0 can be converted to RAID10, or to RAID456 */
1176 if (info->new_level == 10) {
1177 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
1178 /* Assume near=2 layout */
1179 info->new_layout = 0x102;
1180 info->delta_disks = info->array.raid_disks;
1181 }
1182 if (info->new_layout == UnSet) {
1183 int copies = 1 + (info->delta_disks
1184 / info->array.raid_disks);
1185 if (info->array.raid_disks * (copies-1)
1186 != info->delta_disks)
1187 return "Impossible number of devices"
1188 " for RAID0->RAID10";
1189 info->new_layout = 0x100 + copies;
1190 }
1191 if (info->delta_disks == UnSet) {
1192 int copies = info->new_layout & 0xff;
1193 if (info->new_layout != 0x100 + copies)
1194 return "New layout impossible"
1195 " for RAID0->RAID10";;
1196 info->delta_disks = (copies - 1) *
1197 info->array.raid_disks;
1198 }
1199 if (info->new_chunk &&
1200 info->new_chunk != info->array.chunk_size)
1201 return "Cannot change chunk-size with RAID0->RAID10";
1202 /* looks good */
1203 re->level = 10;
1204 re->before.data_disks = (info->array.raid_disks +
1205 info->delta_disks);
1206 re->after.data_disks = re->before.data_disks;
1207 re->before.layout = info->new_layout;
1208 return NULL;
1209 }
1210
1211 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1212 * a raid4 style layout of the final level.
1213 */
1214 switch (info->new_level) {
1215 case 4:
1216 delta_parity = 1;
1217 case 0:
1218 re->level = 4;
1219 re->before.layout = 0;
1220 break;
1221 case 5:
1222 delta_parity = 1;
1223 re->level = 5;
1224 re->before.layout = ALGORITHM_PARITY_N;
1225 if (info->new_layout == UnSet)
1226 info->new_layout = map_name(r5layout, "default");
1227 break;
1228 case 6:
1229 delta_parity = 2;
1230 re->level = 6;
1231 re->before.layout = ALGORITHM_PARITY_N;
1232 if (info->new_layout == UnSet)
1233 info->new_layout = map_name(r6layout, "default");
1234 break;
1235 default:
1236 return "Impossible level change requested";
1237 }
1238 re->before.data_disks = info->array.raid_disks;
1239 /* determining 'after' layout happens outside this 'switch' */
1240 break;
1241
1242 case 4:
1243 info->array.layout = ALGORITHM_PARITY_N;
1244 case 5:
1245 switch (info->new_level) {
1246 case 0:
1247 delta_parity = -1;
1248 case 4:
1249 re->level = info->array.level;
1250 re->before.data_disks = info->array.raid_disks - 1;
1251 re->before.layout = info->array.layout;
1252 break;
1253 case 5:
1254 re->level = 5;
1255 re->before.data_disks = info->array.raid_disks - 1;
1256 re->before.layout = info->array.layout;
1257 break;
1258 case 6:
1259 delta_parity = 1;
1260 re->level = 6;
1261 re->before.data_disks = info->array.raid_disks - 1;
1262 switch (info->array.layout) {
1263 case ALGORITHM_LEFT_ASYMMETRIC:
1264 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1265 break;
1266 case ALGORITHM_RIGHT_ASYMMETRIC:
1267 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1268 break;
1269 case ALGORITHM_LEFT_SYMMETRIC:
1270 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1271 break;
1272 case ALGORITHM_RIGHT_SYMMETRIC:
1273 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1274 break;
1275 case ALGORITHM_PARITY_0:
1276 re->before.layout = ALGORITHM_PARITY_0_6;
1277 break;
1278 case ALGORITHM_PARITY_N:
1279 re->before.layout = ALGORITHM_PARITY_N_6;
1280 break;
1281 default:
1282 return "Cannot convert an array with this layout";
1283 }
1284 break;
1285 case 1:
1286 if (info->array.raid_disks != 2)
1287 return "Can only convert a 2-device array to RAID1";
1288 if (info->delta_disks != UnSet &&
1289 info->delta_disks != 0)
1290 return "Cannot set raid_disk when "
1291 "converting RAID5->RAID1";
1292 re->level = 1;
1293 info->new_chunk = 0;
1294 return NULL;
1295 default:
1296 return "Impossible level change requested";
1297 }
1298 break;
1299 case 6:
1300 switch (info->new_level) {
1301 case 4:
1302 case 5:
1303 delta_parity = -1;
1304 case 6:
1305 re->level = 6;
1306 re->before.data_disks = info->array.raid_disks - 2;
1307 re->before.layout = info->array.layout;
1308 break;
1309 default:
1310 return "Impossible level change requested";
1311 }
1312 break;
1313 }
1314
1315 /* If we reached here then it looks like a re-stripe is
1316 * happening. We have determined the intermediate level
1317 * and initial raid_disks/layout and stored these in 're'.
1318 *
1319 * We need to deduce the final layout that can be atomically
1320 * converted to the end state.
1321 */
1322 switch (info->new_level) {
1323 case 0:
1324 /* We can only get to RAID0 from RAID4 or RAID5
1325 * with appropriate layout and one extra device
1326 */
1327 if (re->level != 4 && re->level != 5)
1328 return "Cannot covert to RAID0 from this level";
1329
1330 switch (re->level) {
1331 case 4:
1332 re->before.layout = 0;
1333 re->after.layout = 0;
1334 break;
1335 case 5:
1336 re->after.layout = ALGORITHM_PARITY_N;
1337 break;
1338 }
1339 break;
1340
1341 case 4:
1342 /* We can only get to RAID4 from RAID5 */
1343 if (re->level != 4 && re->level != 5)
1344 return "Cannot convert to RAID4 from this level";
1345
1346 switch (re->level) {
1347 case 4:
1348 re->after.layout = 0;
1349 break;
1350 case 5:
1351 re->after.layout = ALGORITHM_PARITY_N;
1352 break;
1353 }
1354 break;
1355
1356 case 5:
1357 /* We get to RAID5 from RAID5 or RAID6 */
1358 if (re->level != 5 && re->level != 6)
1359 return "Cannot convert to RAID5 from this level";
1360
1361 switch (re->level) {
1362 case 5:
1363 if (info->new_layout == UnSet)
1364 re->after.layout = re->before.layout;
1365 else
1366 re->after.layout = info->new_layout;
1367 break;
1368 case 6:
1369 if (info->new_layout == UnSet)
1370 info->new_layout = re->before.layout;
1371
1372 /* after.layout needs to be raid6 version of new_layout */
1373 if (info->new_layout == ALGORITHM_PARITY_N)
1374 re->after.layout = ALGORITHM_PARITY_N;
1375 else {
1376 char layout[40];
1377 char *ls = map_num(r5layout, info->new_layout);
1378 int l;
1379 if (ls) {
1380 /* Current RAID6 layout has a RAID5
1381 * equivalent - good
1382 */
1383 strcat(strcpy(layout, ls), "-6");
1384 l = map_name(r6layout, layout);
1385 if (l == UnSet)
1386 return "Cannot find RAID6 layout"
1387 " to convert to";
1388 } else {
1389 /* Current RAID6 has no equivalent.
1390 * If it is already a '-6' layout we
1391 * can leave it unchanged, else we must
1392 * fail
1393 */
1394 ls = map_num(r6layout, info->new_layout);
1395 if (!ls ||
1396 strcmp(ls+strlen(ls)-2, "-6") != 0)
1397 return "Please specify new layout";
1398 l = info->new_layout;
1399 }
1400 re->after.layout = l;
1401 }
1402 }
1403 break;
1404
1405 case 6:
1406 /* We must already be at level 6 */
1407 if (re->level != 6)
1408 return "Impossible level change";
1409 if (info->new_layout == UnSet)
1410 re->after.layout = info->array.layout;
1411 else
1412 re->after.layout = info->new_layout;
1413 break;
1414 default:
1415 return "Impossible level change requested";
1416 }
1417 if (info->delta_disks == UnSet)
1418 info->delta_disks = delta_parity;
1419
1420 re->after.data_disks = (re->before.data_disks
1421 + info->delta_disks
1422 - delta_parity);
1423 switch (re->level) {
1424 case 6: re->parity = 2;
1425 break;
1426 case 4:
1427 case 5: re->parity = 1;
1428 break;
1429 default: re->parity = 0;
1430 break;
1431 }
1432 /* So we have a restripe operation, we need to calculate the number
1433 * of blocks per reshape operation.
1434 */
1435 re->new_size = info->component_size * re->before.data_disks;
1436 if (info->new_chunk == 0)
1437 info->new_chunk = info->array.chunk_size;
1438 if (re->after.data_disks == re->before.data_disks &&
1439 re->after.layout == re->before.layout &&
1440 info->new_chunk == info->array.chunk_size) {
1441 /* Nothing to change, can change level immediately. */
1442 re->level = info->new_level;
1443 re->backup_blocks = 0;
1444 return NULL;
1445 }
1446 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1447 /* chunk and layout changes make no difference */
1448 re->level = info->new_level;
1449 re->backup_blocks = 0;
1450 return NULL;
1451 }
1452
1453 if (re->after.data_disks == re->before.data_disks &&
1454 get_linux_version() < 2006032)
1455 return "in-place reshape is not safe before 2.6.32 - sorry.";
1456
1457 if (re->after.data_disks < re->before.data_disks &&
1458 get_linux_version() < 2006030)
1459 return "reshape to fewer devices is not supported before 2.6.30 - sorry.";
1460
1461 re->backup_blocks = compute_backup_blocks(
1462 info->new_chunk, info->array.chunk_size,
1463 re->after.data_disks,
1464 re->before.data_disks);
1465 re->min_offset_change = re->backup_blocks / re->before.data_disks;
1466
1467 re->new_size = info->component_size * re->after.data_disks;
1468 return NULL;
1469 }
1470
1471 static int set_array_size(struct supertype *st, struct mdinfo *sra,
1472 char *text_version)
1473 {
1474 struct mdinfo *info;
1475 char *subarray;
1476 int ret_val = -1;
1477
1478 if ((st == NULL) || (sra == NULL))
1479 return ret_val;
1480
1481 if (text_version == NULL)
1482 text_version = sra->text_version;
1483 subarray = strchr(text_version+1, '/')+1;
1484 info = st->ss->container_content(st, subarray);
1485 if (info) {
1486 unsigned long long current_size = 0;
1487 unsigned long long new_size =
1488 info->custom_array_size/2;
1489
1490 if (sysfs_get_ll(sra, NULL, "array_size", &current_size) == 0 &&
1491 new_size > current_size) {
1492 if (sysfs_set_num(sra, NULL, "array_size", new_size)
1493 < 0)
1494 dprintf("Error: Cannot set array size");
1495 else {
1496 ret_val = 0;
1497 dprintf("Array size changed");
1498 }
1499 dprintf(" from %llu to %llu.\n",
1500 current_size, new_size);
1501 }
1502 sysfs_free(info);
1503 } else
1504 dprintf("Error: set_array_size(): info pointer in NULL\n");
1505
1506 return ret_val;
1507 }
1508
1509 static int reshape_array(char *container, int fd, char *devname,
1510 struct supertype *st, struct mdinfo *info,
1511 int force, struct mddev_dev *devlist,
1512 unsigned long long data_offset,
1513 char *backup_file, int verbose, int forked,
1514 int restart, int freeze_reshape);
1515 static int reshape_container(char *container, char *devname,
1516 int mdfd,
1517 struct supertype *st,
1518 struct mdinfo *info,
1519 int force,
1520 char *backup_file, int verbose,
1521 int forked, int restart, int freeze_reshape);
1522
1523 int Grow_reshape(char *devname, int fd,
1524 struct mddev_dev *devlist,
1525 unsigned long long data_offset,
1526 struct context *c, struct shape *s)
1527 {
1528 /* Make some changes in the shape of an array.
1529 * The kernel must support the change.
1530 *
1531 * There are three different changes. Each can trigger
1532 * a resync or recovery so we freeze that until we have
1533 * requested everything (if kernel supports freezing - 2.6.30).
1534 * The steps are:
1535 * - change size (i.e. component_size)
1536 * - change level
1537 * - change layout/chunksize/ndisks
1538 *
1539 * The last can require a reshape. It is different on different
1540 * levels so we need to check the level before actioning it.
1541 * Some times the level change needs to be requested after the
1542 * reshape (e.g. raid6->raid5, raid5->raid0)
1543 *
1544 */
1545 struct mdu_array_info_s array;
1546 int rv = 0;
1547 struct supertype *st;
1548 char *subarray = NULL;
1549
1550 int frozen;
1551 int changed = 0;
1552 char *container = NULL;
1553 int cfd = -1;
1554
1555 struct mddev_dev *dv;
1556 int added_disks;
1557
1558 struct mdinfo info;
1559 struct mdinfo *sra;
1560
1561 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1562 pr_err("%s is not an active md array - aborting\n",
1563 devname);
1564 return 1;
1565 }
1566 if (data_offset != INVALID_SECTORS && array.level != 10
1567 && (array.level < 4 || array.level > 6)) {
1568 pr_err("--grow --data-offset not yet supported\n");
1569 return 1;
1570 }
1571
1572 if (s->size > 0 &&
1573 (s->chunk || s->level!= UnSet || s->layout_str || s->raiddisks)) {
1574 pr_err("cannot change component size at the same time "
1575 "as other changes.\n"
1576 " Change size first, then check data is intact before "
1577 "making other changes.\n");
1578 return 1;
1579 }
1580
1581 if (s->raiddisks && s->raiddisks < array.raid_disks && array.level > 1 &&
1582 get_linux_version() < 2006032 &&
1583 !check_env("MDADM_FORCE_FEWER")) {
1584 pr_err("reducing the number of devices is not safe before Linux 2.6.32\n"
1585 " Please use a newer kernel\n");
1586 return 1;
1587 }
1588
1589 st = super_by_fd(fd, &subarray);
1590 if (!st) {
1591 pr_err("Unable to determine metadata format for %s\n", devname);
1592 return 1;
1593 }
1594 if (s->raiddisks > st->max_devs) {
1595 pr_err("Cannot increase raid-disks on this array"
1596 " beyond %d\n", st->max_devs);
1597 return 1;
1598 }
1599
1600 /* in the external case we need to check that the requested reshape is
1601 * supported, and perform an initial check that the container holds the
1602 * pre-requisite spare devices (mdmon owns final validation)
1603 */
1604 if (st->ss->external) {
1605 int rv;
1606
1607 if (subarray) {
1608 container = st->container_devnm;
1609 cfd = open_dev_excl(st->container_devnm);
1610 } else {
1611 container = st->devnm;
1612 close(fd);
1613 cfd = open_dev_excl(st->devnm);
1614 fd = cfd;
1615 }
1616 if (cfd < 0) {
1617 pr_err("Unable to open container for %s\n",
1618 devname);
1619 free(subarray);
1620 return 1;
1621 }
1622
1623 rv = st->ss->load_container(st, cfd, NULL);
1624
1625 if (rv) {
1626 pr_err("Cannot read superblock for %s\n",
1627 devname);
1628 free(subarray);
1629 return 1;
1630 }
1631
1632 /* check if operation is supported for metadata handler */
1633 if (st->ss->container_content) {
1634 struct mdinfo *cc = NULL;
1635 struct mdinfo *content = NULL;
1636
1637 cc = st->ss->container_content(st, subarray);
1638 for (content = cc; content ; content = content->next) {
1639 int allow_reshape = 1;
1640
1641 /* check if reshape is allowed based on metadata
1642 * indications stored in content.array.status
1643 */
1644 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
1645 allow_reshape = 0;
1646 if (content->array.state
1647 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE))
1648 allow_reshape = 0;
1649 if (!allow_reshape) {
1650 pr_err("cannot reshape arrays in"
1651 " container with unsupported"
1652 " metadata: %s(%s)\n",
1653 devname, container);
1654 sysfs_free(cc);
1655 free(subarray);
1656 return 1;
1657 }
1658 }
1659 sysfs_free(cc);
1660 }
1661 if (mdmon_running(container))
1662 st->update_tail = &st->updates;
1663 }
1664
1665 added_disks = 0;
1666 for (dv = devlist; dv; dv = dv->next)
1667 added_disks++;
1668 if (s->raiddisks > array.raid_disks &&
1669 array.spare_disks +added_disks < (s->raiddisks - array.raid_disks) &&
1670 !c->force) {
1671 pr_err("Need %d spare%s to avoid degraded array,"
1672 " and only have %d.\n"
1673 " Use --force to over-ride this check.\n",
1674 s->raiddisks - array.raid_disks,
1675 s->raiddisks - array.raid_disks == 1 ? "" : "s",
1676 array.spare_disks + added_disks);
1677 return 1;
1678 }
1679
1680 sra = sysfs_read(fd, NULL, GET_LEVEL | GET_DISKS | GET_DEVS
1681 | GET_STATE | GET_VERSION);
1682 if (sra) {
1683 if (st->ss->external && subarray == NULL) {
1684 array.level = LEVEL_CONTAINER;
1685 sra->array.level = LEVEL_CONTAINER;
1686 }
1687 } else {
1688 pr_err("failed to read sysfs parameters for %s\n",
1689 devname);
1690 return 1;
1691 }
1692 frozen = freeze(st);
1693 if (frozen < -1) {
1694 /* freeze() already spewed the reason */
1695 sysfs_free(sra);
1696 return 1;
1697 } else if (frozen < 0) {
1698 pr_err("%s is performing resync/recovery and cannot"
1699 " be reshaped\n", devname);
1700 sysfs_free(sra);
1701 return 1;
1702 }
1703
1704 /* ========= set size =============== */
1705 if (s->size > 0 && (s->size == MAX_SIZE || s->size != (unsigned)array.size)) {
1706 unsigned long long orig_size = get_component_size(fd)/2;
1707 unsigned long long min_csize;
1708 struct mdinfo *mdi;
1709 int raid0_takeover = 0;
1710
1711 if (orig_size == 0)
1712 orig_size = (unsigned) array.size;
1713
1714 if (orig_size == 0) {
1715 pr_err("Cannot set device size in this type of array.\n");
1716 rv = 1;
1717 goto release;
1718 }
1719
1720 if (reshape_super(st, s->size, UnSet, UnSet, 0, 0, UnSet, NULL,
1721 devname, APPLY_METADATA_CHANGES, c->verbose > 0)) {
1722 rv = 1;
1723 goto release;
1724 }
1725 sync_metadata(st);
1726 if (st->ss->external) {
1727 /* metadata can have size limitation
1728 * update size value according to metadata information
1729 */
1730 struct mdinfo *sizeinfo =
1731 st->ss->container_content(st, subarray);
1732 if (sizeinfo) {
1733 unsigned long long new_size =
1734 sizeinfo->custom_array_size/2;
1735 int data_disks = get_data_disks(
1736 sizeinfo->array.level,
1737 sizeinfo->array.layout,
1738 sizeinfo->array.raid_disks);
1739 new_size /= data_disks;
1740 dprintf("Metadata size correction from %llu to "
1741 "%llu (%llu)\n", orig_size, new_size,
1742 new_size * data_disks);
1743 s->size = new_size;
1744 sysfs_free(sizeinfo);
1745 }
1746 }
1747
1748 /* Update the size of each member device in case
1749 * they have been resized. This will never reduce
1750 * below the current used-size. The "size" attribute
1751 * understands '0' to mean 'max'.
1752 */
1753 min_csize = 0;
1754 rv = 0;
1755 for (mdi = sra->devs; mdi; mdi = mdi->next) {
1756 if (sysfs_set_num(sra, mdi, "size",
1757 s->size == MAX_SIZE ? 0 : s->size) < 0) {
1758 /* Probably kernel refusing to let us
1759 * reduce the size - not an error.
1760 */
1761 break;
1762 }
1763 if (array.not_persistent == 0 &&
1764 array.major_version == 0 &&
1765 get_linux_version() < 3001000) {
1766 /* Dangerous to allow size to exceed 2TB */
1767 unsigned long long csize;
1768 if (sysfs_get_ll(sra, mdi, "size", &csize) == 0) {
1769 if (csize >= 2ULL*1024*1024*1024)
1770 csize = 2ULL*1024*1024*1024;
1771 if ((min_csize == 0 || (min_csize
1772 > csize)))
1773 min_csize = csize;
1774 }
1775 }
1776 }
1777 if (rv) {
1778 pr_err("Cannot set size on "
1779 "array members.\n");
1780 goto size_change_error;
1781 }
1782 if (min_csize && s->size > min_csize) {
1783 pr_err("Cannot safely make this array "
1784 "use more than 2TB per device on this kernel.\n");
1785 rv = 1;
1786 goto size_change_error;
1787 }
1788 if (min_csize && s->size == MAX_SIZE) {
1789 /* Don't let the kernel choose a size - it will get
1790 * it wrong
1791 */
1792 pr_err("Limited v0.90 array to "
1793 "2TB per device\n");
1794 s->size = min_csize;
1795 }
1796 if (st->ss->external) {
1797 if (sra->array.level == 0) {
1798 rv = sysfs_set_str(sra, NULL, "level",
1799 "raid5");
1800 if (!rv) {
1801 raid0_takeover = 1;
1802 /* get array parametes after takeover
1803 * to chane one parameter at time only
1804 */
1805 rv = ioctl(fd, GET_ARRAY_INFO, &array);
1806 }
1807 }
1808 /* make sure mdmon is
1809 * aware of the new level */
1810 if (!mdmon_running(st->container_devnm))
1811 start_mdmon(st->container_devnm);
1812 ping_monitor(container);
1813 if (mdmon_running(st->container_devnm) &&
1814 st->update_tail == NULL)
1815 st->update_tail = &st->updates;
1816 }
1817
1818 if (s->size == MAX_SIZE)
1819 s->size = 0;
1820 array.size = s->size;
1821 if (s->size & ~INT32_MAX) {
1822 /* got truncated to 32bit, write to
1823 * component_size instead
1824 */
1825 if (sra)
1826 rv = sysfs_set_num(sra, NULL,
1827 "component_size", s->size);
1828 else
1829 rv = -1;
1830 } else {
1831 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1832
1833 /* manage array size when it is managed externally
1834 */
1835 if ((rv == 0) && st->ss->external)
1836 rv = set_array_size(st, sra, sra->text_version);
1837 }
1838
1839 if (raid0_takeover) {
1840 /* do not recync non-existing parity,
1841 * we will drop it anyway
1842 */
1843 sysfs_set_str(sra, NULL, "sync_action", "frozen");
1844 /* go back to raid0, drop parity disk
1845 */
1846 sysfs_set_str(sra, NULL, "level", "raid0");
1847 ioctl(fd, GET_ARRAY_INFO, &array);
1848 }
1849
1850 size_change_error:
1851 if (rv != 0) {
1852 int err = errno;
1853
1854 /* restore metadata */
1855 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1856 UnSet, NULL, devname,
1857 ROLLBACK_METADATA_CHANGES,
1858 c->verbose) == 0)
1859 sync_metadata(st);
1860 pr_err("Cannot set device size for %s: %s\n",
1861 devname, strerror(err));
1862 if (err == EBUSY &&
1863 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1864 cont_err("Bitmap must be removed before size can be changed\n");
1865 rv = 1;
1866 goto release;
1867 }
1868 if (s->assume_clean) {
1869 /* This will fail on kernels older than 3.0 unless
1870 * a backport has been arranged.
1871 */
1872 if (sra == NULL ||
1873 sysfs_set_str(sra, NULL, "resync_start", "none") < 0)
1874 pr_err("--assume-clean not supported with --grow on this kernel\n");
1875 }
1876 ioctl(fd, GET_ARRAY_INFO, &array);
1877 s->size = get_component_size(fd)/2;
1878 if (s->size == 0)
1879 s->size = array.size;
1880 if (c->verbose >= 0) {
1881 if (s->size == orig_size)
1882 pr_err("component size of %s "
1883 "unchanged at %lluK\n",
1884 devname, s->size);
1885 else
1886 pr_err("component size of %s "
1887 "has been set to %lluK\n",
1888 devname, s->size);
1889 }
1890 changed = 1;
1891 } else if (array.level != LEVEL_CONTAINER) {
1892 s->size = get_component_size(fd)/2;
1893 if (s->size == 0)
1894 s->size = array.size;
1895 }
1896
1897 /* See if there is anything else to do */
1898 if ((s->level == UnSet || s->level == array.level) &&
1899 (s->layout_str == NULL) &&
1900 (s->chunk == 0 || s->chunk == array.chunk_size) &&
1901 data_offset == INVALID_SECTORS &&
1902 (s->raiddisks == 0 || s->raiddisks == array.raid_disks)) {
1903 /* Nothing more to do */
1904 if (!changed && c->verbose >= 0)
1905 pr_err("%s: no change requested\n",
1906 devname);
1907 goto release;
1908 }
1909
1910 /* ========= check for Raid10/Raid1 -> Raid0 conversion ===============
1911 * current implementation assumes that following conditions must be met:
1912 * - RAID10:
1913 * - far_copies == 1
1914 * - near_copies == 2
1915 */
1916 if ((s->level == 0 && array.level == 10 && sra &&
1917 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) ||
1918 (s->level == 0 && array.level == 1 && sra)) {
1919 int err;
1920 err = remove_disks_for_takeover(st, sra, array.layout);
1921 if (err) {
1922 dprintf("%s: Array cannot be reshaped\n", Name);
1923 if (cfd > -1)
1924 close(cfd);
1925 rv = 1;
1926 goto release;
1927 }
1928 /* Make sure mdmon has seen the device removal
1929 * and updated metadata before we continue with
1930 * level change
1931 */
1932 if (container)
1933 ping_monitor(container);
1934 }
1935
1936 memset(&info, 0, sizeof(info));
1937 info.array = array;
1938 sysfs_init(&info, fd, NULL);
1939 strcpy(info.text_version, sra->text_version);
1940 info.component_size = s->size*2;
1941 info.new_level = s->level;
1942 info.new_chunk = s->chunk * 1024;
1943 if (info.array.level == LEVEL_CONTAINER) {
1944 info.delta_disks = UnSet;
1945 info.array.raid_disks = s->raiddisks;
1946 } else if (s->raiddisks)
1947 info.delta_disks = s->raiddisks - info.array.raid_disks;
1948 else
1949 info.delta_disks = UnSet;
1950 if (s->layout_str == NULL) {
1951 info.new_layout = UnSet;
1952 if (info.array.level == 6 &&
1953 (info.new_level == 6 || info.new_level == UnSet) &&
1954 info.array.layout >= 16) {
1955 pr_err("%s has a non-standard layout. If you"
1956 " wish to preserve this\n", devname);
1957 cont_err("during the reshape, please specify"
1958 " --layout=preserve\n");
1959 cont_err("If you want to change it, specify a"
1960 " layout or use --layout=normalise\n");
1961 rv = 1;
1962 goto release;
1963 }
1964 } else if (strcmp(s->layout_str, "normalise") == 0 ||
1965 strcmp(s->layout_str, "normalize") == 0) {
1966 /* If we have a -6 RAID6 layout, remove the '-6'. */
1967 info.new_layout = UnSet;
1968 if (info.array.level == 6 && info.new_level == UnSet) {
1969 char l[40], *h;
1970 strcpy(l, map_num(r6layout, info.array.layout));
1971 h = strrchr(l, '-');
1972 if (h && strcmp(h, "-6") == 0) {
1973 *h = 0;
1974 info.new_layout = map_name(r6layout, l);
1975 }
1976 } else {
1977 pr_err("%s is only meaningful when reshaping"
1978 " a RAID6 array.\n", s->layout_str);
1979 rv = 1;
1980 goto release;
1981 }
1982 } else if (strcmp(s->layout_str, "preserve") == 0) {
1983 /* This means that a non-standard RAID6 layout
1984 * is OK.
1985 * In particular:
1986 * - When reshape a RAID6 (e.g. adding a device)
1987 * which is in a non-standard layout, it is OK
1988 * to preserve that layout.
1989 * - When converting a RAID5 to RAID6, leave it in
1990 * the XXX-6 layout, don't re-layout.
1991 */
1992 if (info.array.level == 6 && info.new_level == UnSet)
1993 info.new_layout = info.array.layout;
1994 else if (info.array.level == 5 && info.new_level == 6) {
1995 char l[40];
1996 strcpy(l, map_num(r5layout, info.array.layout));
1997 strcat(l, "-6");
1998 info.new_layout = map_name(r6layout, l);
1999 } else {
2000 pr_err("%s in only meaningful when reshaping"
2001 " to RAID6\n", s->layout_str);
2002 rv = 1;
2003 goto release;
2004 }
2005 } else {
2006 int l = info.new_level;
2007 if (l == UnSet)
2008 l = info.array.level;
2009 switch (l) {
2010 case 5:
2011 info.new_layout = map_name(r5layout, s->layout_str);
2012 break;
2013 case 6:
2014 info.new_layout = map_name(r6layout, s->layout_str);
2015 break;
2016 case 10:
2017 info.new_layout = parse_layout_10(s->layout_str);
2018 break;
2019 case LEVEL_FAULTY:
2020 info.new_layout = parse_layout_faulty(s->layout_str);
2021 break;
2022 default:
2023 pr_err("layout not meaningful"
2024 " with this level\n");
2025 rv = 1;
2026 goto release;
2027 }
2028 if (info.new_layout == UnSet) {
2029 pr_err("layout %s not understood"
2030 " for this level\n",
2031 s->layout_str);
2032 rv = 1;
2033 goto release;
2034 }
2035 }
2036
2037 if (array.level == LEVEL_FAULTY) {
2038 if (s->level != UnSet && s->level != array.level) {
2039 pr_err("cannot change level of Faulty device\n");
2040 rv =1 ;
2041 }
2042 if (s->chunk) {
2043 pr_err("cannot set chunksize of Faulty device\n");
2044 rv =1 ;
2045 }
2046 if (s->raiddisks && s->raiddisks != 1) {
2047 pr_err("cannot set raid_disks of Faulty device\n");
2048 rv =1 ;
2049 }
2050 if (s->layout_str) {
2051 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2052 dprintf("Cannot get array information.\n");
2053 goto release;
2054 }
2055 array.layout = info.new_layout;
2056 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2057 pr_err("failed to set new layout\n");
2058 rv = 1;
2059 } else if (c->verbose >= 0)
2060 printf("layout for %s set to %d\n",
2061 devname, array.layout);
2062 }
2063 } else if (array.level == LEVEL_CONTAINER) {
2064 /* This change is to be applied to every array in the
2065 * container. This is only needed when the metadata imposes
2066 * restraints of the various arrays in the container.
2067 * Currently we only know that IMSM requires all arrays
2068 * to have the same number of devices so changing the
2069 * number of devices (On-Line Capacity Expansion) must be
2070 * performed at the level of the container
2071 */
2072 rv = reshape_container(container, devname, -1, st, &info,
2073 c->force, c->backup_file, c->verbose, 0, 0, 0);
2074 frozen = 0;
2075 } else {
2076 /* get spare devices from external metadata
2077 */
2078 if (st->ss->external) {
2079 struct mdinfo *info2;
2080
2081 info2 = st->ss->container_content(st, subarray);
2082 if (info2) {
2083 info.array.spare_disks =
2084 info2->array.spare_disks;
2085 sysfs_free(info2);
2086 }
2087 }
2088
2089 /* Impose these changes on a single array. First
2090 * check that the metadata is OK with the change. */
2091
2092 if (reshape_super(st, 0, info.new_level,
2093 info.new_layout, info.new_chunk,
2094 info.array.raid_disks, info.delta_disks,
2095 c->backup_file, devname, APPLY_METADATA_CHANGES,
2096 c->verbose)) {
2097 rv = 1;
2098 goto release;
2099 }
2100 sync_metadata(st);
2101 rv = reshape_array(container, fd, devname, st, &info, c->force,
2102 devlist, data_offset, c->backup_file, c->verbose,
2103 0, 0, 0);
2104 frozen = 0;
2105 }
2106 release:
2107 sysfs_free(sra);
2108 if (frozen > 0)
2109 unfreeze(st);
2110 return rv;
2111 }
2112
2113 /* verify_reshape_position()
2114 * Function checks if reshape position in metadata is not farther
2115 * than position in md.
2116 * Return value:
2117 * 0 : not valid sysfs entry
2118 * it can be caused by not started reshape, it should be started
2119 * by reshape array or raid0 array is before takeover
2120 * -1 : error, reshape position is obviously wrong
2121 * 1 : success, reshape progress correct or updated
2122 */
2123 static int verify_reshape_position(struct mdinfo *info, int level)
2124 {
2125 int ret_val = 0;
2126 char buf[40];
2127 int rv;
2128
2129 /* read sync_max, failure can mean raid0 array */
2130 rv = sysfs_get_str(info, NULL, "sync_max", buf, 40);
2131
2132 if (rv > 0) {
2133 char *ep;
2134 unsigned long long position = strtoull(buf, &ep, 0);
2135
2136 dprintf("%s: Read sync_max sysfs entry is: %s\n", Name, buf);
2137 if (!(ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))) {
2138 position *= get_data_disks(level,
2139 info->new_layout,
2140 info->array.raid_disks);
2141 if (info->reshape_progress < position) {
2142 dprintf("Corrected reshape progress (%llu) to "
2143 "md position (%llu)\n",
2144 info->reshape_progress, position);
2145 info->reshape_progress = position;
2146 ret_val = 1;
2147 } else if (info->reshape_progress > position) {
2148 pr_err("Fatal error: array "
2149 "reshape was not properly frozen "
2150 "(expected reshape position is %llu, "
2151 "but reshape progress is %llu.\n",
2152 position, info->reshape_progress);
2153 ret_val = -1;
2154 } else {
2155 dprintf("Reshape position in md and metadata "
2156 "are the same;");
2157 ret_val = 1;
2158 }
2159 }
2160 } else if (rv == 0) {
2161 /* for valid sysfs entry, 0-length content
2162 * should be indicated as error
2163 */
2164 ret_val = -1;
2165 }
2166
2167 return ret_val;
2168 }
2169
2170 static unsigned long long choose_offset(unsigned long long lo,
2171 unsigned long long hi,
2172 unsigned long long min,
2173 unsigned long long max)
2174 {
2175 /* Choose a new offset between hi and lo.
2176 * It must be between min and max, but
2177 * we would prefer something near the middle of hi/lo, and also
2178 * prefer to be aligned to a big power of 2.
2179 *
2180 * So we start with the middle, then for each bit,
2181 * starting at '1' and increasing, if it is set, we either
2182 * add it or subtract it if possible, preferring the option
2183 * which is furthest from the boundary.
2184 *
2185 * We stop once we get a 1MB alignment. As units are in sectors,
2186 * 1MB = 2*1024 sectors.
2187 */
2188 unsigned long long choice = (lo + hi) / 2;
2189 unsigned long long bit = 1;
2190
2191 for (bit = 1; bit < 2*1024; bit = bit << 1) {
2192 unsigned long long bigger, smaller;
2193 if (! (bit & choice))
2194 continue;
2195 bigger = choice + bit;
2196 smaller = choice - bit;
2197 if (bigger > max && smaller < min)
2198 break;
2199 if (bigger > max)
2200 choice = smaller;
2201 else if (smaller < min)
2202 choice = bigger;
2203 else if (hi - bigger > smaller - lo)
2204 choice = bigger;
2205 else
2206 choice = smaller;
2207 }
2208 return choice;
2209 }
2210
2211 static int set_new_data_offset(struct mdinfo *sra, struct supertype *st,
2212 char *devname, int delta_disks,
2213 unsigned long long data_offset,
2214 unsigned long long min,
2215 int can_fallback)
2216 {
2217 struct mdinfo *sd;
2218 int dir = 0;
2219 int err = 0;
2220 unsigned long long before, after;
2221
2222 /* Need to find min space before and after so same is used
2223 * on all devices
2224 */
2225 before = UINT64_MAX;
2226 after = UINT64_MAX;
2227 for (sd = sra->devs; sd; sd = sd->next) {
2228 char *dn;
2229 int dfd;
2230 int rv;
2231 struct supertype *st2;
2232 struct mdinfo info2;
2233
2234 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2235 continue;
2236 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2237 dfd = dev_open(dn, O_RDONLY);
2238 if (dfd < 0) {
2239 pr_err("%s: cannot open component %s\n",
2240 devname, dn ? dn : "-unknown-");
2241 goto release;
2242 }
2243 st2 = dup_super(st);
2244 rv = st2->ss->load_super(st2,dfd, NULL);
2245 close(dfd);
2246 if (rv) {
2247 free(st2);
2248 pr_err("%s: cannot get superblock from %s\n",
2249 devname, dn);
2250 goto release;
2251 }
2252 st2->ss->getinfo_super(st2, &info2, NULL);
2253 st2->ss->free_super(st2);
2254 free(st2);
2255 if (info2.space_before == 0 &&
2256 info2.space_after == 0) {
2257 /* Metadata doesn't support data_offset changes */
2258 pr_err("%s: Metadata version doesn't support"
2259 " data_offset changes\n", devname);
2260 goto fallback;
2261 }
2262 if (before > info2.space_before)
2263 before = info2.space_before;
2264 if (after > info2.space_after)
2265 after = info2.space_after;
2266
2267 if (data_offset != INVALID_SECTORS) {
2268 if (dir == 0) {
2269 if (info2.data_offset == data_offset) {
2270 pr_err("%s: already has that data_offset\n",
2271 dn);
2272 goto release;
2273 }
2274 if (data_offset < info2.data_offset)
2275 dir = -1;
2276 else
2277 dir = 1;
2278 } else if ((data_offset <= info2.data_offset && dir == 1) ||
2279 (data_offset >= info2.data_offset && dir == -1)) {
2280 pr_err("%s: differing data offsets on devices make this --data-offset setting impossible\n",
2281 dn);
2282 goto release;
2283 }
2284 }
2285 }
2286 if (before == UINT64_MAX)
2287 /* impossible really, there must be no devices */
2288 return 1;
2289
2290 for (sd = sra->devs; sd; sd = sd->next) {
2291 char *dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2292 unsigned long long new_data_offset;
2293
2294 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2295 continue;
2296 if (delta_disks < 0) {
2297 /* Don't need any space as array is shrinking
2298 * just move data_offset up by min
2299 */
2300 if (data_offset == INVALID_SECTORS)
2301 new_data_offset = sd->data_offset + min;
2302 else {
2303 if (data_offset < sd->data_offset + min) {
2304 pr_err("--data-offset too small for %s\n",
2305 dn);
2306 goto release;
2307 }
2308 new_data_offset = data_offset;
2309 }
2310 } else if (delta_disks > 0) {
2311 /* need space before */
2312 if (before < min) {
2313 if (can_fallback)
2314 goto fallback;
2315 pr_err("Insufficient head-space for reshape on %s\n",
2316 dn);
2317 goto release;
2318 }
2319 if (data_offset == INVALID_SECTORS)
2320 new_data_offset = sd->data_offset - min;
2321 else {
2322 if (data_offset > sd->data_offset - min) {
2323 pr_err("--data-offset too large for %s\n",
2324 dn);
2325 goto release;
2326 }
2327 new_data_offset = data_offset;
2328 }
2329 } else {
2330 if (dir == 0) {
2331 /* can move up or down. If 'data_offset'
2332 * was set we would have already decided,
2333 * so just choose direction with most space.
2334 */
2335 if (before > after)
2336 dir = -1;
2337 else
2338 dir = 1;
2339 }
2340 sysfs_set_str(sra, NULL, "reshape_direction",
2341 dir == 1 ? "backwards" : "forwards");
2342 if (dir > 0) {
2343 /* Increase data offset */
2344 if (after < min) {
2345 if (can_fallback)
2346 goto fallback;
2347 pr_err("Insufficient tail-space for reshape on %s\n",
2348 dn);
2349 goto release;
2350 }
2351 if (data_offset != INVALID_SECTORS &&
2352 data_offset < sd->data_offset + min) {
2353 pr_err("--data-offset too small on %s\n",
2354 dn);
2355 goto release;
2356 }
2357 if (data_offset != INVALID_SECTORS)
2358 new_data_offset = data_offset;
2359 else
2360 new_data_offset = choose_offset(sd->data_offset,
2361 sd->data_offset + after,
2362 sd->data_offset + min,
2363 sd->data_offset + after);
2364 } else {
2365 /* Decrease data offset */
2366 if (before < min) {
2367 if (can_fallback)
2368 goto fallback;
2369 pr_err("insufficient head-room on %s\n",
2370 dn);
2371 goto release;
2372 }
2373 if (data_offset != INVALID_SECTORS &&
2374 data_offset < sd->data_offset - min) {
2375 pr_err("--data-offset too small on %s\n",
2376 dn);
2377 goto release;
2378 }
2379 if (data_offset != INVALID_SECTORS)
2380 new_data_offset = data_offset;
2381 else
2382 new_data_offset = choose_offset(sd->data_offset - before,
2383 sd->data_offset,
2384 sd->data_offset - before,
2385 sd->data_offset - min);
2386 }
2387 }
2388 err = sysfs_set_num(sra, sd, "new_offset", new_data_offset);
2389 if (err < 0 && errno == E2BIG) {
2390 /* try again after increasing data size to max */
2391 err = sysfs_set_num(sra, sd, "size", 0);
2392 if (err < 0 && errno == EINVAL &&
2393 !(sd->disk.state & (1<<MD_DISK_SYNC))) {
2394 /* some kernels have a bug where you cannot
2395 * use '0' on spare devices. */
2396 sysfs_set_num(sra, sd, "size",
2397 (sra->component_size + after)/2);
2398 }
2399 err = sysfs_set_num(sra, sd, "new_offset",
2400 new_data_offset);
2401 }
2402 if (err < 0) {
2403 if (errno == E2BIG && data_offset != INVALID_SECTORS) {
2404 pr_err("data-offset is too big for %s\n",
2405 dn);
2406 goto release;
2407 }
2408 if (sd == sra->devs &&
2409 (errno == ENOENT || errno == E2BIG))
2410 /* Early kernel, no 'new_offset' file,
2411 * or kernel doesn't like us.
2412 * For RAID5/6 this is not fatal
2413 */
2414 return 1;
2415 pr_err("Cannot set new_offset for %s\n",
2416 dn);
2417 break;
2418 }
2419 }
2420 return err;
2421 release:
2422 return -1;
2423 fallback:
2424 /* Just use a backup file */
2425 return 1;
2426 }
2427
2428 static int raid10_reshape(char *container, int fd, char *devname,
2429 struct supertype *st, struct mdinfo *info,
2430 struct reshape *reshape,
2431 unsigned long long data_offset,
2432 int force, int verbose)
2433 {
2434 /* Changing raid_disks, layout, chunksize or possibly
2435 * just data_offset for a RAID10.
2436 * We must always change data_offset. We change by at least
2437 * ->min_offset_change which is the largest of the old and new
2438 * chunk sizes.
2439 * If raid_disks is increasing, then data_offset must decrease
2440 * by at least this copy size.
2441 * If raid_disks is unchanged, data_offset must increase or
2442 * decrease by at least min_offset_change but preferably by much more.
2443 * We choose half of the available space.
2444 * If raid_disks is decreasing, data_offset must increase by
2445 * at least min_offset_change. To allow of this, component_size
2446 * must be decreased by the same amount.
2447 *
2448 * So we calculate the required minimum and direction, possibly
2449 * reduce the component_size, then iterate through the devices
2450 * and set the new_data_offset.
2451 * If that all works, we set chunk_size, layout, raid_disks, and start
2452 * 'reshape'
2453 */
2454 struct mdinfo *sra;
2455 unsigned long long min;
2456 int err = 0;
2457
2458 sra = sysfs_read(fd, NULL,
2459 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK
2460 );
2461 if (!sra) {
2462 pr_err("%s: Cannot get array details from sysfs\n",
2463 devname);
2464 goto release;
2465 }
2466 min = reshape->min_offset_change;
2467
2468 if (info->delta_disks)
2469 sysfs_set_str(sra, NULL, "reshape_direction",
2470 info->delta_disks < 0 ? "backwards" : "forwards");
2471 if (info->delta_disks < 0 &&
2472 info->space_after < min) {
2473 int rv = sysfs_set_num(sra, NULL, "component_size",
2474 (sra->component_size -
2475 min)/2);
2476 if (rv) {
2477 pr_err("cannot reduce component size\n");
2478 goto release;
2479 }
2480 }
2481 err = set_new_data_offset(sra, st, devname, info->delta_disks, data_offset,
2482 min, 0);
2483 if (err == 1) {
2484 pr_err("Cannot set new_data_offset: RAID10 reshape not\n");
2485 cont_err("supported on this kernel\n");
2486 err = -1;
2487 }
2488 if (err < 0)
2489 goto release;
2490
2491 if (!err && sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2492 err = errno;
2493 if (!err && sysfs_set_num(sra, NULL, "layout", reshape->after.layout) < 0)
2494 err = errno;
2495 if (!err && sysfs_set_num(sra, NULL, "raid_disks",
2496 info->array.raid_disks + info->delta_disks) < 0)
2497 err = errno;
2498 if (!err && sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0)
2499 err = errno;
2500 if (err) {
2501 pr_err("Cannot set array shape for %s\n",
2502 devname);
2503 if (err == EBUSY &&
2504 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
2505 cont_err(" Bitmap must be removed before"
2506 " shape can be changed\n");
2507 goto release;
2508 }
2509 sysfs_free(sra);
2510 return 0;
2511 release:
2512 sysfs_free(sra);
2513 return 1;
2514 }
2515
2516 static void get_space_after(int fd, struct supertype *st, struct mdinfo *info)
2517 {
2518 struct mdinfo *sra, *sd;
2519 /* Initialisation to silence compiler warning */
2520 unsigned long long min_space_before = 0, min_space_after = 0;
2521 int first = 1;
2522
2523 sra = sysfs_read(fd, NULL, GET_DEVS);
2524 if (!sra)
2525 return;
2526 for (sd = sra->devs; sd; sd = sd->next) {
2527 char *dn;
2528 int dfd;
2529 struct supertype *st2;
2530 struct mdinfo info2;
2531
2532 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2533 continue;
2534 dn = map_dev(sd->disk.major, sd->disk.minor, 0);
2535 dfd = dev_open(dn, O_RDONLY);
2536 if (dfd < 0)
2537 break;
2538 st2 = dup_super(st);
2539 if (st2->ss->load_super(st2,dfd, NULL)) {
2540 close(dfd);
2541 free(st2);
2542 break;
2543 }
2544 close(dfd);
2545 st2->ss->getinfo_super(st2, &info2, NULL);
2546 st2->ss->free_super(st2);
2547 free(st2);
2548 if (first ||
2549 min_space_before > info2.space_before)
2550 min_space_before = info2.space_before;
2551 if (first ||
2552 min_space_after > info2.space_after)
2553 min_space_after = info2.space_after;
2554 first = 0;
2555 }
2556 if (sd == NULL && !first) {
2557 info->space_after = min_space_after;
2558 info->space_before = min_space_before;
2559 }
2560 sysfs_free(sra);
2561 }
2562
2563 static void update_cache_size(char *container, struct mdinfo *sra,
2564 struct mdinfo *info,
2565 int disks, unsigned long long blocks)
2566 {
2567 /* Check that the internal stripe cache is
2568 * large enough, or it won't work.
2569 * It must hold at least 4 stripes of the larger
2570 * chunk size
2571 */
2572 unsigned long cache;
2573 cache = max(info->array.chunk_size, info->new_chunk);
2574 cache *= 4; /* 4 stripes minimum */
2575 cache /= 512; /* convert to sectors */
2576 /* make sure there is room for 'blocks' with a bit to spare */
2577 if (cache < 16 + blocks / disks)
2578 cache = 16 + blocks / disks;
2579 cache /= (4096/512); /* Covert from sectors to pages */
2580
2581 if (sra->cache_size < cache)
2582 subarray_set_num(container, sra, "stripe_cache_size",
2583 cache+1);
2584 }
2585
2586 static int impose_reshape(struct mdinfo *sra,
2587 struct mdinfo *info,
2588 struct supertype *st,
2589 int fd,
2590 int restart,
2591 char *devname, char *container,
2592 struct reshape *reshape)
2593 {
2594 struct mdu_array_info_s array;
2595
2596 sra->new_chunk = info->new_chunk;
2597
2598 if (restart) {
2599 /* for external metadata checkpoint saved by mdmon can be lost
2600 * or missed /due to e.g. crash/. Check if md is not during
2601 * restart farther than metadata points to.
2602 * If so, this means metadata information is obsolete.
2603 */
2604 if (st->ss->external)
2605 verify_reshape_position(info, reshape->level);
2606 sra->reshape_progress = info->reshape_progress;
2607 } else {
2608 sra->reshape_progress = 0;
2609 if (reshape->after.data_disks < reshape->before.data_disks)
2610 /* start from the end of the new array */
2611 sra->reshape_progress = (sra->component_size
2612 * reshape->after.data_disks);
2613 }
2614
2615 ioctl(fd, GET_ARRAY_INFO, &array);
2616 if (info->array.chunk_size == info->new_chunk &&
2617 reshape->before.layout == reshape->after.layout &&
2618 st->ss->external == 0) {
2619 /* use SET_ARRAY_INFO but only if reshape hasn't started */
2620 array.raid_disks = reshape->after.data_disks + reshape->parity;
2621 if (!restart &&
2622 ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
2623 int err = errno;
2624
2625 pr_err("Cannot set device shape for %s: %s\n",
2626 devname, strerror(errno));
2627
2628 if (err == EBUSY &&
2629 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2630 cont_err("Bitmap must be removed before"
2631 " shape can be changed\n");
2632
2633 goto release;
2634 }
2635 } else if (!restart) {
2636 /* set them all just in case some old 'new_*' value
2637 * persists from some earlier problem.
2638 */
2639 int err = 0;
2640 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
2641 err = errno;
2642 if (!err && sysfs_set_num(sra, NULL, "layout",
2643 reshape->after.layout) < 0)
2644 err = errno;
2645 if (!err && subarray_set_num(container, sra, "raid_disks",
2646 reshape->after.data_disks +
2647 reshape->parity) < 0)
2648 err = errno;
2649 if (err) {
2650 pr_err("Cannot set device shape for %s\n",
2651 devname);
2652
2653 if (err == EBUSY &&
2654 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2655 cont_err("Bitmap must be removed before"
2656 " shape can be changed\n");
2657 goto release;
2658 }
2659 }
2660 return 0;
2661 release:
2662 return -1;
2663 }
2664
2665 static int impose_level(int fd, int level, char *devname, int verbose)
2666 {
2667 char *c;
2668 struct mdu_array_info_s array;
2669 struct mdinfo info;
2670 sysfs_init(&info, fd, NULL);
2671
2672 ioctl(fd, GET_ARRAY_INFO, &array);
2673 if (level == 0 &&
2674 (array.level >= 4 && array.level <= 6)) {
2675 /* To convert to RAID0 we need to fail and
2676 * remove any non-data devices. */
2677 int found = 0;
2678 int d;
2679 int data_disks = array.raid_disks - 1;
2680 if (array.level == 6)
2681 data_disks -= 1;
2682 if (array.level == 5 &&
2683 array.layout != ALGORITHM_PARITY_N)
2684 return -1;
2685 if (array.level == 6 &&
2686 array.layout != ALGORITHM_PARITY_N_6)
2687 return -1;
2688 sysfs_set_str(&info, NULL,"sync_action", "idle");
2689 /* First remove any spares so no recovery starts */
2690 for (d = 0, found = 0;
2691 d < MAX_DISKS && found < array.nr_disks;
2692 d++) {
2693 mdu_disk_info_t disk;
2694 disk.number = d;
2695 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2696 continue;
2697 if (disk.major == 0 && disk.minor == 0)
2698 continue;
2699 found++;
2700 if ((disk.state & (1 << MD_DISK_ACTIVE))
2701 && disk.raid_disk < data_disks)
2702 /* keep this */
2703 continue;
2704 ioctl(fd, HOT_REMOVE_DISK,
2705 makedev(disk.major, disk.minor));
2706 }
2707 /* Now fail anything left */
2708 ioctl(fd, GET_ARRAY_INFO, &array);
2709 for (d = 0, found = 0;
2710 d < MAX_DISKS && found < array.nr_disks;
2711 d++) {
2712 int cnt;
2713 mdu_disk_info_t disk;
2714 disk.number = d;
2715 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
2716 continue;
2717 if (disk.major == 0 && disk.minor == 0)
2718 continue;
2719 found++;
2720 if ((disk.state & (1 << MD_DISK_ACTIVE))
2721 && disk.raid_disk < data_disks)
2722 /* keep this */
2723 continue;
2724 ioctl(fd, SET_DISK_FAULTY,
2725 makedev(disk.major, disk.minor));
2726 cnt = 5;
2727 while (ioctl(fd, HOT_REMOVE_DISK,
2728 makedev(disk.major, disk.minor)) < 0
2729 && errno == EBUSY
2730 && cnt--) {
2731 usleep(10000);
2732 }
2733 }
2734 }
2735 c = map_num(pers, level);
2736 if (c) {
2737 int err = sysfs_set_str(&info, NULL, "level", c);
2738 if (err) {
2739 err = errno;
2740 pr_err("%s: could not set level to %s\n",
2741 devname, c);
2742 if (err == EBUSY &&
2743 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
2744 cont_err("Bitmap must be removed"
2745 " before level can be changed\n");
2746 return err;
2747 }
2748 if (verbose >= 0)
2749 pr_err("level of %s changed to %s\n",
2750 devname, c);
2751 }
2752 return 0;
2753 }
2754
2755 int sigterm = 0;
2756 static void catch_term(int sig)
2757 {
2758 sigterm = 1;
2759 }
2760
2761 static int continue_via_systemd(char *devnm)
2762 {
2763 int skipped, i, pid, status;
2764 char pathbuf[1024];
2765 /* In a systemd/udev world, it is best to get systemd to
2766 * run "mdadm --grow --continue" rather than running in the
2767 * background.
2768 */
2769 switch(fork()) {
2770 case 0:
2771 /* FIXME yuk. CLOSE_EXEC?? */
2772 skipped = 0;
2773 for (i = 3; skipped < 20; i++)
2774 if (close(i) < 0)
2775 skipped++;
2776 else
2777 skipped = 0;
2778
2779 /* Don't want to see error messages from
2780 * systemctl. If the service doesn't exist,
2781 * we fork ourselves.
2782 */
2783 close(2);
2784 open("/dev/null", O_WRONLY);
2785 snprintf(pathbuf, sizeof(pathbuf), "mdadm-grow-continue@%s.service",
2786 devnm);
2787 status = execl("/usr/bin/systemctl", "systemctl",
2788 "start",
2789 pathbuf, NULL);
2790 status = execl("/bin/systemctl", "systemctl", "start",
2791 pathbuf, NULL);
2792 exit(1);
2793 case -1: /* Just do it ourselves. */
2794 break;
2795 default: /* parent - good */
2796 pid = wait(&status);
2797 if (pid >= 0 && status == 0)
2798 return 1;
2799 }
2800 return 0;
2801 }
2802
2803 static int reshape_array(char *container, int fd, char *devname,
2804 struct supertype *st, struct mdinfo *info,
2805 int force, struct mddev_dev *devlist,
2806 unsigned long long data_offset,
2807 char *backup_file, int verbose, int forked,
2808 int restart, int freeze_reshape)
2809 {
2810 struct reshape reshape;
2811 int spares_needed;
2812 char *msg;
2813 int orig_level = UnSet;
2814 int odisks;
2815 int delayed;
2816
2817 struct mdu_array_info_s array;
2818 char *c;
2819
2820 struct mddev_dev *dv;
2821 int added_disks;
2822
2823 int *fdlist = NULL;
2824 unsigned long long *offsets = NULL;
2825 int d;
2826 int nrdisks;
2827 int err;
2828 unsigned long blocks;
2829 unsigned long long array_size;
2830 int done;
2831 struct mdinfo *sra = NULL;
2832 char buf[20];
2833
2834 /* when reshaping a RAID0, the component_size might be zero.
2835 * So try to fix that up.
2836 */
2837 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
2838 dprintf("Cannot get array information.\n");
2839 goto release;
2840 }
2841 if (array.level == 0 && info->component_size == 0) {
2842 get_dev_size(fd, NULL, &array_size);
2843 info->component_size = array_size / array.raid_disks;
2844 }
2845
2846 if (array.level == 10)
2847 /* Need space_after info */
2848 get_space_after(fd, st, info);
2849
2850 if (info->reshape_active) {
2851 int new_level = info->new_level;
2852 info->new_level = UnSet;
2853 if (info->delta_disks > 0)
2854 info->array.raid_disks -= info->delta_disks;
2855 msg = analyse_change(devname, info, &reshape);
2856 info->new_level = new_level;
2857 if (info->delta_disks > 0)
2858 info->array.raid_disks += info->delta_disks;
2859 if (!restart)
2860 /* Make sure the array isn't read-only */
2861 ioctl(fd, RESTART_ARRAY_RW, 0);
2862 } else
2863 msg = analyse_change(devname, info, &reshape);
2864 if (msg) {
2865 /* if msg == "", error has already been printed */
2866 if (msg[0])
2867 pr_err("%s\n", msg);
2868 goto release;
2869 }
2870 if (restart &&
2871 (reshape.level != info->array.level ||
2872 reshape.before.layout != info->array.layout ||
2873 reshape.before.data_disks + reshape.parity
2874 != info->array.raid_disks - max(0, info->delta_disks))) {
2875 pr_err("reshape info is not in native format -"
2876 " cannot continue.\n");
2877 goto release;
2878 }
2879
2880 if (st->ss->external && restart && (info->reshape_progress == 0) &&
2881 !((sysfs_get_str(info, NULL, "sync_action", buf, sizeof(buf)) > 0) &&
2882 (strncmp(buf, "reshape", 7) == 0))) {
2883 /* When reshape is restarted from '0', very begin of array
2884 * it is possible that for external metadata reshape and array
2885 * configuration doesn't happen.
2886 * Check if md has the same opinion, and reshape is restarted
2887 * from 0. If so, this is regular reshape start after reshape
2888 * switch in metadata to next array only.
2889 */
2890 if ((verify_reshape_position(info, reshape.level) >= 0) &&
2891 (info->reshape_progress == 0))
2892 restart = 0;
2893 }
2894 if (restart) {
2895 /* reshape already started. just skip to monitoring the reshape */
2896 if (reshape.backup_blocks == 0)
2897 return 0;
2898 if (restart & RESHAPE_NO_BACKUP)
2899 return 0;
2900
2901 /* Need 'sra' down at 'started:' */
2902 sra = sysfs_read(fd, NULL,
2903 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
2904 GET_CACHE);
2905 if (!sra) {
2906 pr_err("%s: Cannot get array details from sysfs\n",
2907 devname);
2908 goto release;
2909 }
2910
2911 if (!backup_file)
2912 backup_file = locate_backup(sra->sys_name);
2913
2914 goto started;
2915 }
2916 /* The container is frozen but the array may not be.
2917 * So freeze the array so spares don't get put to the wrong use
2918 * FIXME there should probably be a cleaner separation between
2919 * freeze_array and freeze_container.
2920 */
2921 sysfs_freeze_array(info);
2922 /* Check we have enough spares to not be degraded */
2923 added_disks = 0;
2924 for (dv = devlist; dv ; dv=dv->next)
2925 added_disks++;
2926 spares_needed = max(reshape.before.data_disks,
2927 reshape.after.data_disks)
2928 + reshape.parity - array.raid_disks;
2929
2930 if (!force &&
2931 info->new_level > 1 && info->array.level > 1 &&
2932 spares_needed > info->array.spare_disks + added_disks) {
2933 pr_err("Need %d spare%s to avoid degraded array,"
2934 " and only have %d.\n"
2935 " Use --force to over-ride this check.\n",
2936 spares_needed,
2937 spares_needed == 1 ? "" : "s",
2938 info->array.spare_disks + added_disks);
2939 goto release;
2940 }
2941 /* Check we have enough spares to not fail */
2942 spares_needed = max(reshape.before.data_disks,
2943 reshape.after.data_disks)
2944 - array.raid_disks;
2945 if ((info->new_level > 1 || info->new_level == 0) &&
2946 spares_needed > info->array.spare_disks +added_disks) {
2947 pr_err("Need %d spare%s to create working array,"
2948 " and only have %d.\n",
2949 spares_needed,
2950 spares_needed == 1 ? "" : "s",
2951 info->array.spare_disks + added_disks);
2952 goto release;
2953 }
2954
2955 if (reshape.level != array.level) {
2956 int err = impose_level(fd, reshape.level, devname, verbose);
2957 if (err)
2958 goto release;
2959 info->new_layout = UnSet; /* after level change,
2960 * layout is meaningless */
2961 orig_level = array.level;
2962 sysfs_freeze_array(info);
2963
2964 if (reshape.level > 0 && st->ss->external) {
2965 /* make sure mdmon is aware of the new level */
2966 if (mdmon_running(container))
2967 flush_mdmon(container);
2968
2969 if (!mdmon_running(container))
2970 start_mdmon(container);
2971 ping_monitor(container);
2972 if (mdmon_running(container) &&
2973 st->update_tail == NULL)
2974 st->update_tail = &st->updates;
2975 }
2976 }
2977 /* ->reshape_super might have chosen some spares from the
2978 * container that it wants to be part of the new array.
2979 * We can collect them with ->container_content and give
2980 * them to the kernel.
2981 */
2982 if (st->ss->reshape_super && st->ss->container_content) {
2983 char *subarray = strchr(info->text_version+1, '/')+1;
2984 struct mdinfo *info2 =
2985 st->ss->container_content(st, subarray);
2986 struct mdinfo *d;
2987
2988 if (info2) {
2989 sysfs_init(info2, fd, st->devnm);
2990 /* When increasing number of devices, we need to set
2991 * new raid_disks before adding these, or they might
2992 * be rejected.
2993 */
2994 if (reshape.backup_blocks &&
2995 reshape.after.data_disks > reshape.before.data_disks)
2996 subarray_set_num(container, info2, "raid_disks",
2997 reshape.after.data_disks +
2998 reshape.parity);
2999 for (d = info2->devs; d; d = d->next) {
3000 if (d->disk.state == 0 &&
3001 d->disk.raid_disk >= 0) {
3002 /* This is a spare that wants to
3003 * be part of the array.
3004 */
3005 add_disk(fd, st, info2, d);
3006 }
3007 }
3008 sysfs_free(info2);
3009 }
3010 }
3011 /* We might have been given some devices to add to the
3012 * array. Now that the array has been changed to the right
3013 * level and frozen, we can safely add them.
3014 */
3015 if (devlist)
3016 Manage_subdevs(devname, fd, devlist, verbose,
3017 0,NULL, 0);
3018
3019 if (reshape.backup_blocks == 0 && data_offset != INVALID_SECTORS)
3020 reshape.backup_blocks = reshape.before.data_disks * info->array.chunk_size/512;
3021 if (reshape.backup_blocks == 0) {
3022 /* No restriping needed, but we might need to impose
3023 * some more changes: layout, raid_disks, chunk_size
3024 */
3025 /* read current array info */
3026 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
3027 dprintf("Cannot get array information.\n");
3028 goto release;
3029 }
3030 /* compare current array info with new values and if
3031 * it is different update them to new */
3032 if (info->new_layout != UnSet &&
3033 info->new_layout != array.layout) {
3034 array.layout = info->new_layout;
3035 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3036 pr_err("failed to set new layout\n");
3037 goto release;
3038 } else if (verbose >= 0)
3039 printf("layout for %s set to %d\n",
3040 devname, array.layout);
3041 }
3042 if (info->delta_disks != UnSet &&
3043 info->delta_disks != 0 &&
3044 array.raid_disks != (info->array.raid_disks + info->delta_disks)) {
3045 array.raid_disks += info->delta_disks;
3046 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
3047 pr_err("failed to set raid disks\n");
3048 goto release;
3049 } else if (verbose >= 0) {
3050 printf("raid_disks for %s set to %d\n",
3051 devname, array.raid_disks);
3052 }
3053 }
3054 if (info->new_chunk != 0 &&
3055 info->new_chunk != array.chunk_size) {
3056 if (sysfs_set_num(info, NULL,
3057 "chunk_size", info->new_chunk) != 0) {
3058 pr_err("failed to set chunk size\n");
3059 goto release;
3060 } else if (verbose >= 0)
3061 printf("chunk size for %s set to %d\n",
3062 devname, array.chunk_size);
3063 }
3064 unfreeze(st);
3065 return 0;
3066 }
3067
3068 /*
3069 * There are three possibilities.
3070 * 1/ The array will shrink.
3071 * We need to ensure the reshape will pause before reaching
3072 * the 'critical section'. We also need to fork and wait for
3073 * that to happen. When it does we
3074 * suspend/backup/complete/unfreeze
3075 *
3076 * 2/ The array will not change size.
3077 * This requires that we keep a backup of a sliding window
3078 * so that we can restore data after a crash. So we need
3079 * to fork and monitor progress.
3080 * In future we will allow the data_offset to change, so
3081 * a sliding backup becomes unnecessary.
3082 *
3083 * 3/ The array will grow. This is relatively easy.
3084 * However the kernel's restripe routines will cheerfully
3085 * overwrite some early data before it is safe. So we
3086 * need to make a backup of the early parts of the array
3087 * and be ready to restore it if rebuild aborts very early.
3088 * For externally managed metadata, we still need a forked
3089 * child to monitor the reshape and suspend IO over the region
3090 * that is being reshaped.
3091 *
3092 * We backup data by writing it to one spare, or to a
3093 * file which was given on command line.
3094 *
3095 * In each case, we first make sure that storage is available
3096 * for the required backup.
3097 * Then we:
3098 * - request the shape change.
3099 * - fork to handle backup etc.
3100 */
3101 /* Check that we can hold all the data */
3102 get_dev_size(fd, NULL, &array_size);
3103 if (reshape.new_size < (array_size/512)) {
3104 pr_err("this change will reduce the size of the array.\n"
3105 " use --grow --array-size first to truncate array.\n"
3106 " e.g. mdadm --grow %s --array-size %llu\n",
3107 devname, reshape.new_size/2);
3108 goto release;
3109 }
3110
3111 if (array.level == 10) {
3112 /* Reshaping RAID10 does not require any data backup by
3113 * user-space. Instead it requires that the data_offset
3114 * is changed to avoid the need for backup.
3115 * So this is handled very separately
3116 */
3117 if (restart)
3118 /* Nothing to do. */
3119 return 0;
3120 return raid10_reshape(container, fd, devname, st, info,
3121 &reshape, data_offset,
3122 force, verbose);
3123 }
3124 sra = sysfs_read(fd, NULL,
3125 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
3126 GET_CACHE);
3127 if (!sra) {
3128 pr_err("%s: Cannot get array details from sysfs\n",
3129 devname);
3130 goto release;
3131 }
3132
3133 if (!backup_file)
3134 switch(set_new_data_offset(sra, st, devname,
3135 reshape.after.data_disks - reshape.before.data_disks,
3136 data_offset,
3137 reshape.min_offset_change, 1)) {
3138 case -1:
3139 goto release;
3140 case 0:
3141 /* Updated data_offset, so it's easy now */
3142 update_cache_size(container, sra, info,
3143 min(reshape.before.data_disks,
3144 reshape.after.data_disks),
3145 reshape.backup_blocks);
3146
3147 /* Right, everything seems fine. Let's kick things off.
3148 */
3149 sync_metadata(st);
3150
3151 if (impose_reshape(sra, info, st, fd, restart,
3152 devname, container, &reshape) < 0)
3153 goto release;
3154 if (sysfs_set_str(sra, NULL, "sync_action", "reshape") < 0) {
3155 pr_err("Failed to initiate reshape!\n");
3156 goto release;
3157 }
3158 if (info->new_level == reshape.level)
3159 return 0;
3160 /* need to adjust level when reshape completes */
3161 switch(fork()) {
3162 case -1: /* ignore error, but don't wait */
3163 return 0;
3164 default: /* parent */
3165 return 0;
3166 case 0:
3167 map_fork();
3168 break;
3169 }
3170 close(fd);
3171 wait_reshape(sra);
3172 fd = open_dev(sra->sys_name);
3173 if (fd >= 0)
3174 impose_level(fd, info->new_level, devname, verbose);
3175 return 0;
3176 case 1: /* Couldn't set data_offset, try the old way */
3177 if (data_offset != INVALID_SECTORS) {
3178 pr_err("Cannot update data_offset on this array\n");
3179 goto release;
3180 }
3181 break;
3182 }
3183
3184 started:
3185 /* Decide how many blocks (sectors) for a reshape
3186 * unit. The number we have so far is just a minimum
3187 */
3188 blocks = reshape.backup_blocks;
3189 if (reshape.before.data_disks ==
3190 reshape.after.data_disks) {
3191 /* Make 'blocks' bigger for better throughput, but
3192 * not so big that we reject it below.
3193 * Try for 16 megabytes
3194 */
3195 while (blocks * 32 < sra->component_size &&
3196 blocks < 16*1024*2)
3197 blocks *= 2;
3198 } else
3199 pr_err("Need to backup %luK of critical "
3200 "section..\n", blocks/2);
3201
3202 if (blocks >= sra->component_size/2) {
3203 pr_err("%s: Something wrong"
3204 " - reshape aborted\n",
3205 devname);
3206 goto release;
3207 }
3208
3209 /* Now we need to open all these devices so we can read/write.
3210 */
3211 nrdisks = max(reshape.before.data_disks,
3212 reshape.after.data_disks) + reshape.parity
3213 + sra->array.spare_disks;
3214 fdlist = xcalloc((1+nrdisks), sizeof(int));
3215 offsets = xcalloc((1+nrdisks), sizeof(offsets[0]));
3216
3217 odisks = reshape.before.data_disks + reshape.parity;
3218 d = reshape_prepare_fdlist(devname, sra, odisks,
3219 nrdisks, blocks, backup_file,
3220 fdlist, offsets);
3221 if (d < 0) {
3222 goto release;
3223 }
3224 if ((st->ss->manage_reshape == NULL) ||
3225 (st->ss->recover_backup == NULL)) {
3226 if (backup_file == NULL) {
3227 if (reshape.after.data_disks <=
3228 reshape.before.data_disks) {
3229 pr_err("%s: Cannot grow - need backup-file\n",
3230 devname);
3231 pr_err(" Please provide one with \"--backup=...\"\n");
3232 goto release;
3233 } else if (sra->array.spare_disks == 0) {
3234 pr_err("%s: Cannot grow - "
3235 "need a spare or backup-file to backup "
3236 "critical section\n", devname);
3237 goto release;
3238 }
3239 } else {
3240 if (!reshape_open_backup_file(backup_file, fd, devname,
3241 (signed)blocks,
3242 fdlist+d, offsets+d,
3243 sra->sys_name,
3244 restart)) {
3245 goto release;
3246 }
3247 d++;
3248 }
3249 }
3250
3251 update_cache_size(container, sra, info,
3252 min(reshape.before.data_disks, reshape.after.data_disks),
3253 blocks);
3254
3255 /* Right, everything seems fine. Let's kick things off.
3256 * If only changing raid_disks, use ioctl, else use
3257 * sysfs.
3258 */
3259 sync_metadata(st);
3260
3261 if (impose_reshape(sra, info, st, fd, restart,
3262 devname, container, &reshape) < 0)
3263 goto release;
3264
3265 err = start_reshape(sra, restart, reshape.before.data_disks,
3266 reshape.after.data_disks);
3267 if (err) {
3268 pr_err("Cannot %s reshape for %s\n",
3269 restart ? "continue" : "start",
3270 devname);
3271 goto release;
3272 }
3273 if (restart)
3274 sysfs_set_str(sra, NULL, "array_state", "active");
3275 if (freeze_reshape) {
3276 free(fdlist);
3277 free(offsets);
3278 sysfs_free(sra);
3279 pr_err("Reshape has to be continued from"
3280 " location %llu when root filesystem has been mounted.\n",
3281 sra->reshape_progress);
3282 return 1;
3283 }
3284
3285 if (!forked && !check_env("MDADM_NO_SYSTEMCTL"))
3286 if (continue_via_systemd(container ?: sra->sys_name)) {
3287 free(fdlist);
3288 free(offsets);
3289 sysfs_free(sra);
3290 return 0;
3291 }
3292
3293 /* Now we just need to kick off the reshape and watch, while
3294 * handling backups of the data...
3295 * This is all done by a forked background process.
3296 */
3297 switch(forked ? 0 : fork()) {
3298 case -1:
3299 pr_err("Cannot run child to monitor reshape: %s\n",
3300 strerror(errno));
3301 abort_reshape(sra);
3302 goto release;
3303 default:
3304 free(fdlist);
3305 free(offsets);
3306 sysfs_free(sra);
3307 return 0;
3308 case 0:
3309 map_fork();
3310 break;
3311 }
3312
3313 /* If another array on the same devices is busy, the
3314 * reshape will wait for them. This would mean that
3315 * the first section that we suspend will stay suspended
3316 * for a long time. So check on that possibility
3317 * by looking for "DELAYED" in /proc/mdstat, and if found,
3318 * wait a while
3319 */
3320 do {
3321 struct mdstat_ent *mds, *m;
3322 delayed = 0;
3323 mds = mdstat_read(1, 0);
3324 for (m = mds; m; m = m->next)
3325 if (strcmp(m->devnm, sra->sys_name) == 0) {
3326 if (m->resync &&
3327 m->percent == RESYNC_DELAYED)
3328 delayed = 1;
3329 if (m->resync == 0)
3330 /* Haven't started the reshape thread
3331 * yet, wait a bit
3332 */
3333 delayed = 2;
3334 break;
3335 }
3336 free_mdstat(mds);
3337 if (delayed == 1 && get_linux_version() < 3007000) {
3338 pr_err("Reshape is delayed, but cannot wait carefully with this kernel.\n"
3339 " You might experience problems until other reshapes complete.\n");
3340 delayed = 0;
3341 }
3342 if (delayed)
3343 mdstat_wait(30 - (delayed-1) * 25);
3344 } while (delayed);
3345 mdstat_close();
3346 close(fd);
3347 if (check_env("MDADM_GROW_VERIFY"))
3348 fd = open(devname, O_RDONLY | O_DIRECT);
3349 else
3350 fd = -1;
3351 mlockall(MCL_FUTURE);
3352
3353 signal(SIGTERM, catch_term);
3354
3355 if (st->ss->external) {
3356 /* metadata handler takes it from here */
3357 done = st->ss->manage_reshape(
3358 fd, sra, &reshape, st, blocks,
3359 fdlist, offsets,
3360 d - odisks, fdlist+odisks,
3361 offsets+odisks);
3362 } else
3363 done = child_monitor(
3364 fd, sra, &reshape, st, blocks,
3365 fdlist, offsets,
3366 d - odisks, fdlist+odisks,
3367 offsets+odisks);
3368
3369 free(fdlist);
3370 free(offsets);
3371
3372 if (backup_file && done) {
3373 char *bul;
3374 bul = make_backup(sra->sys_name);
3375 if (bul) {
3376 char buf[1024];
3377 int l = readlink(bul, buf, sizeof(buf));
3378 if (l > 0) {
3379 buf[l]=0;
3380 unlink(buf);
3381 }
3382 unlink(bul);
3383 free(bul);
3384 }
3385 unlink(backup_file);
3386 }
3387 if (!done) {
3388 abort_reshape(sra);
3389 goto out;
3390 }
3391
3392 if (!st->ss->external &&
3393 !(reshape.before.data_disks != reshape.after.data_disks
3394 && info->custom_array_size) &&
3395 info->new_level == reshape.level &&
3396 !forked) {
3397 /* no need to wait for the reshape to finish as
3398 * there is nothing more to do.
3399 */
3400 sysfs_free(sra);
3401 exit(0);
3402 }
3403 wait_reshape(sra);
3404
3405 if (st->ss->external) {
3406 /* Re-load the metadata as much could have changed */
3407 int cfd = open_dev(st->container_devnm);
3408 if (cfd >= 0) {
3409 flush_mdmon(container);
3410 st->ss->free_super(st);
3411 st->ss->load_container(st, cfd, container);
3412 close(cfd);
3413 }
3414 }
3415
3416 /* set new array size if required customer_array_size is used
3417 * by this metadata.
3418 */
3419 if (reshape.before.data_disks !=
3420 reshape.after.data_disks &&
3421 info->custom_array_size)
3422 set_array_size(st, info, info->text_version);
3423
3424 if (info->new_level != reshape.level) {
3425 if (fd < 0)
3426 fd = open(devname, O_RDONLY);
3427 impose_level(fd, info->new_level, devname, verbose);
3428 close(fd);
3429 if (info->new_level == 0)
3430 st->update_tail = NULL;
3431 }
3432 out:
3433 sysfs_free(sra);
3434 if (forked)
3435 return 0;
3436 unfreeze(st);
3437 exit(0);
3438
3439 release:
3440 free(fdlist);
3441 free(offsets);
3442 if (orig_level != UnSet && sra) {
3443 c = map_num(pers, orig_level);
3444 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
3445 pr_err("aborting level change\n");
3446 }
3447 sysfs_free(sra);
3448 if (!forked)
3449 unfreeze(st);
3450 return 1;
3451 }
3452
3453 /* mdfd handle is passed to be closed in child process (after fork).
3454 */
3455 int reshape_container(char *container, char *devname,
3456 int mdfd,
3457 struct supertype *st,
3458 struct mdinfo *info,
3459 int force,
3460 char *backup_file, int verbose,
3461 int forked, int restart, int freeze_reshape)
3462 {
3463 struct mdinfo *cc = NULL;
3464 int rv = restart;
3465 char last_devnm[32] = "";
3466
3467 /* component_size is not meaningful for a container,
3468 * so pass '0' meaning 'no change'
3469 */
3470 if (!restart &&
3471 reshape_super(st, 0, info->new_level,
3472 info->new_layout, info->new_chunk,
3473 info->array.raid_disks, info->delta_disks,
3474 backup_file, devname, APPLY_METADATA_CHANGES,
3475 verbose)) {
3476 unfreeze(st);
3477 return 1;
3478 }
3479
3480 sync_metadata(st);
3481
3482 /* ping monitor to be sure that update is on disk
3483 */
3484 ping_monitor(container);
3485
3486 if (!forked && !freeze_reshape && !check_env("MDADM_NO_SYSTEMCTL"))
3487 if (continue_via_systemd(container))
3488 return 0;
3489
3490 switch (forked ? 0 : fork()) {
3491 case -1: /* error */
3492 perror("Cannot fork to complete reshape\n");
3493 unfreeze(st);
3494 return 1;
3495 default: /* parent */
3496 if (!freeze_reshape)
3497 printf("%s: multi-array reshape continues"
3498 " in background\n", Name);
3499 return 0;
3500 case 0: /* child */
3501 map_fork();
3502 break;
3503 }
3504
3505 /* close unused handle in child process
3506 */
3507 if (mdfd > -1)
3508 close(mdfd);
3509
3510 while(1) {
3511 /* For each member array with reshape_active,
3512 * we need to perform the reshape.
3513 * We pick the first array that needs reshaping and
3514 * reshape it. reshape_array() will re-read the metadata
3515 * so the next time through a different array should be
3516 * ready for reshape.
3517 * It is possible that the 'different' array will not
3518 * be assembled yet. In that case we simple exit.
3519 * When it is assembled, the mdadm which assembles it
3520 * will take over the reshape.
3521 */
3522 struct mdinfo *content;
3523 int fd;
3524 struct mdstat_ent *mdstat;
3525 char *adev;
3526 int devid;
3527
3528 sysfs_free(cc);
3529
3530 cc = st->ss->container_content(st, NULL);
3531
3532 for (content = cc; content ; content = content->next) {
3533 char *subarray;
3534 if (!content->reshape_active)
3535 continue;
3536
3537 subarray = strchr(content->text_version+1, '/')+1;
3538 mdstat = mdstat_by_subdev(subarray, container);
3539 if (!mdstat)
3540 continue;
3541 if (mdstat->active == 0) {
3542 pr_err("Skipping inactive array %s.\n",
3543 mdstat->devnm);
3544 free_mdstat(mdstat);
3545 mdstat = NULL;
3546 continue;
3547 }
3548 break;
3549 }
3550 if (!content)
3551 break;
3552
3553 devid = devnm2devid(mdstat->devnm);
3554 adev = map_dev(major(devid), minor(devid), 0);
3555 if (!adev)
3556 adev = content->text_version;
3557
3558 fd = open_dev(mdstat->devnm);
3559 if (fd < 0) {
3560 printf("%s: Device %s cannot be opened for reshape.",
3561 Name, adev);
3562 break;
3563 }
3564
3565 if (strcmp(last_devnm, mdstat->devnm) == 0) {
3566 /* Do not allow for multiple reshape_array() calls for
3567 * the same array.
3568 * It can happen when reshape_array() returns without
3569 * error, when reshape is not finished (wrong reshape
3570 * starting/continuation conditions). Mdmon doesn't
3571 * switch to next array in container and reentry
3572 * conditions for the same array occur.
3573 * This is possibly interim until the behaviour of
3574 * reshape_array is resolved().
3575 */
3576 printf("%s: Multiple reshape execution detected for "
3577 "device %s.", Name, adev);
3578 close(fd);
3579 break;
3580 }
3581 strcpy(last_devnm, mdstat->devnm);
3582
3583 sysfs_init(content, fd, mdstat->devnm);
3584
3585 if (mdmon_running(container))
3586 flush_mdmon(container);
3587
3588 rv = reshape_array(container, fd, adev, st,
3589 content, force, NULL, INVALID_SECTORS,
3590 backup_file, verbose, 1, restart,
3591 freeze_reshape);
3592 close(fd);
3593
3594 if (freeze_reshape) {
3595 sysfs_free(cc);
3596 exit(0);
3597 }
3598
3599 restart = 0;
3600 if (rv)
3601 break;
3602
3603 if (mdmon_running(container))
3604 flush_mdmon(container);
3605 }
3606 if (!rv)
3607 unfreeze(st);
3608 sysfs_free(cc);
3609 exit(0);
3610 }
3611
3612 /*
3613 * We run a child process in the background which performs the following
3614 * steps:
3615 * - wait for resync to reach a certain point
3616 * - suspend io to the following section
3617 * - backup that section
3618 * - allow resync to proceed further
3619 * - resume io
3620 * - discard the backup.
3621 *
3622 * When are combined in slightly different ways in the three cases.
3623 * Grow:
3624 * - suspend/backup/allow/wait/resume/discard
3625 * Shrink:
3626 * - allow/wait/suspend/backup/allow/wait/resume/discard
3627 * same-size:
3628 * - wait/resume/discard/suspend/backup/allow
3629 *
3630 * suspend/backup/allow always come together
3631 * wait/resume/discard do too.
3632 * For the same-size case we have two backups to improve flow.
3633 *
3634 */
3635
3636 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
3637 unsigned long long backup_point,
3638 unsigned long long wait_point,
3639 unsigned long long *suspend_point,
3640 unsigned long long *reshape_completed, int *frozen)
3641 {
3642 /* This function is called repeatedly by the reshape manager.
3643 * It determines how much progress can safely be made and allows
3644 * that progress.
3645 * - 'info' identifies the array and particularly records in
3646 * ->reshape_progress the metadata's knowledge of progress
3647 * This is a sector offset from the start of the array
3648 * of the next array block to be relocated. This number
3649 * may increase from 0 or decrease from array_size, depending
3650 * on the type of reshape that is happening.
3651 * Note that in contrast, 'sync_completed' is a block count of the
3652 * reshape so far. It gives the distance between the start point
3653 * (head or tail of device) and the next place that data will be
3654 * written. It always increases.
3655 * - 'reshape' is the structure created by analyse_change
3656 * - 'backup_point' shows how much the metadata manager has backed-up
3657 * data. For reshapes with increasing progress, it is the next address
3658 * to be backed up, previous addresses have been backed-up. For
3659 * decreasing progress, it is the earliest address that has been
3660 * backed up - later address are also backed up.
3661 * So addresses between reshape_progress and backup_point are
3662 * backed up providing those are in the 'correct' order.
3663 * - 'wait_point' is an array address. When reshape_completed
3664 * passes this point, progress_reshape should return. It might
3665 * return earlier if it determines that ->reshape_progress needs
3666 * to be updated or further backup is needed.
3667 * - suspend_point is maintained by progress_reshape and the caller
3668 * should not touch it except to initialise to zero.
3669 * It is an array address and it only increases in 2.6.37 and earlier.
3670 * This makes it difficult to handle reducing reshapes with
3671 * external metadata.
3672 * However: it is similar to backup_point in that it records the
3673 * other end of a suspended region from reshape_progress.
3674 * it is moved to extend the region that is safe to backup and/or
3675 * reshape
3676 * - reshape_completed is read from sysfs and returned. The caller
3677 * should copy this into ->reshape_progress when it has reason to
3678 * believe that the metadata knows this, and any backup outside this
3679 * has been erased.
3680 *
3681 * Return value is:
3682 * 1 if more data from backup_point - but only as far as suspend_point,
3683 * should be backed up
3684 * 0 if things are progressing smoothly
3685 * -1 if the reshape is finished because it is all done,
3686 * -2 if the reshape is finished due to an error.
3687 */
3688
3689 int advancing = (reshape->after.data_disks
3690 >= reshape->before.data_disks);
3691 unsigned long long need_backup; /* All data between start of array and
3692 * here will at some point need to
3693 * be backed up.
3694 */
3695 unsigned long long read_offset, write_offset;
3696 unsigned long long write_range;
3697 unsigned long long max_progress, target, completed;
3698 unsigned long long array_size = (info->component_size
3699 * reshape->before.data_disks);
3700 int fd;
3701 char buf[20];
3702
3703 /* First, we unsuspend any region that is now known to be safe.
3704 * If suspend_point is on the 'wrong' side of reshape_progress, then
3705 * we don't have or need suspension at the moment. This is true for
3706 * native metadata when we don't need to back-up.
3707 */
3708 if (advancing) {
3709 if (info->reshape_progress <= *suspend_point)
3710 sysfs_set_num(info, NULL, "suspend_lo",
3711 info->reshape_progress);
3712 } else {
3713 /* Note: this won't work in 2.6.37 and before.
3714 * Something somewhere should make sure we don't need it!
3715 */
3716 if (info->reshape_progress >= *suspend_point)
3717 sysfs_set_num(info, NULL, "suspend_hi",
3718 info->reshape_progress);
3719 }
3720
3721 /* Now work out how far it is safe to progress.
3722 * If the read_offset for ->reshape_progress is less than
3723 * 'blocks' beyond the write_offset, we can only progress as far
3724 * as a backup.
3725 * Otherwise we can progress until the write_offset for the new location
3726 * reaches (within 'blocks' of) the read_offset at the current location.
3727 * However that region must be suspended unless we are using native
3728 * metadata.
3729 * If we need to suspend more, we limit it to 128M per device, which is
3730 * rather arbitrary and should be some time-based calculation.
3731 */
3732 read_offset = info->reshape_progress / reshape->before.data_disks;
3733 write_offset = info->reshape_progress / reshape->after.data_disks;
3734 write_range = info->new_chunk/512;
3735 if (reshape->before.data_disks == reshape->after.data_disks)
3736 need_backup = array_size;
3737 else
3738 need_backup = reshape->backup_blocks;
3739 if (advancing) {
3740 if (read_offset < write_offset + write_range)
3741 max_progress = backup_point;
3742 else
3743 max_progress =
3744 read_offset *
3745 reshape->after.data_disks;
3746 } else {
3747 if (read_offset > write_offset - write_range)
3748 /* Can only progress as far as has been backed up,
3749 * which must be suspended */
3750 max_progress = backup_point;
3751 else if (info->reshape_progress <= need_backup)
3752 max_progress = backup_point;
3753 else {
3754 if (info->array.major_version >= 0)
3755 /* Can progress until backup is needed */
3756 max_progress = need_backup;
3757 else {
3758 /* Can progress until metadata update is required */
3759 max_progress =
3760 read_offset *
3761 reshape->after.data_disks;
3762 /* but data must be suspended */
3763 if (max_progress < *suspend_point)
3764 max_progress = *suspend_point;
3765 }
3766 }
3767 }
3768
3769 /* We know it is safe to progress to 'max_progress' providing
3770 * it is suspended or we are using native metadata.
3771 * Consider extending suspend_point 128M per device if it
3772 * is less than 64M per device beyond reshape_progress.
3773 * But always do a multiple of 'blocks'
3774 * FIXME this is too big - it takes to long to complete
3775 * this much.
3776 */
3777 target = 64*1024*2 * min(reshape->before.data_disks,
3778 reshape->after.data_disks);
3779 target /= reshape->backup_blocks;
3780 if (target < 2)
3781 target = 2;
3782 target *= reshape->backup_blocks;
3783
3784 /* For externally managed metadata we always need to suspend IO to
3785 * the area being reshaped so we regularly push suspend_point forward.
3786 * For native metadata we only need the suspend if we are going to do
3787 * a backup.
3788 */
3789 if (advancing) {
3790 if ((need_backup > info->reshape_progress
3791 || info->array.major_version < 0) &&
3792 *suspend_point < info->reshape_progress + target) {
3793 if (need_backup < *suspend_point + 2 * target)
3794 *suspend_point = need_backup;
3795 else if (*suspend_point + 2 * target < array_size)
3796 *suspend_point += 2 * target;
3797 else
3798 *suspend_point = array_size;
3799 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
3800 if (max_progress > *suspend_point)
3801 max_progress = *suspend_point;
3802 }
3803 } else {
3804 if (info->array.major_version >= 0) {
3805 /* Only need to suspend when about to backup */
3806 if (info->reshape_progress < need_backup * 2 &&
3807 *suspend_point > 0) {
3808 *suspend_point = 0;
3809 sysfs_set_num(info, NULL, "suspend_lo", 0);
3810 sysfs_set_num(info, NULL, "suspend_hi", need_backup);
3811 }
3812 } else {
3813 /* Need to suspend continually */
3814 if (info->reshape_progress < *suspend_point)
3815 *suspend_point = info->reshape_progress;
3816 if (*suspend_point + target < info->reshape_progress)
3817 /* No need to move suspend region yet */;
3818 else {
3819 if (*suspend_point >= 2 * target)
3820 *suspend_point -= 2 * target;
3821 else
3822 *suspend_point = 0;
3823 sysfs_set_num(info, NULL, "suspend_lo",
3824 *suspend_point);
3825 }
3826 if (max_progress < *suspend_point)
3827 max_progress = *suspend_point;
3828 }
3829 }
3830
3831 /* now set sync_max to allow that progress. sync_max, like
3832 * sync_completed is a count of sectors written per device, so
3833 * we find the difference between max_progress and the start point,
3834 * and divide that by after.data_disks to get a sync_max
3835 * number.
3836 * At the same time we convert wait_point to a similar number
3837 * for comparing against sync_completed.
3838 */
3839 /* scale down max_progress to per_disk */
3840 max_progress /= reshape->after.data_disks;
3841 /* Round to chunk size as some kernels give an erroneously high number */
3842 max_progress /= info->new_chunk/512;
3843 max_progress *= info->new_chunk/512;
3844 /* And round to old chunk size as the kernel wants that */
3845 max_progress /= info->array.chunk_size/512;
3846 max_progress *= info->array.chunk_size/512;
3847 /* Limit progress to the whole device */
3848 if (max_progress > info->component_size)
3849 max_progress = info->component_size;
3850 wait_point /= reshape->after.data_disks;
3851 if (!advancing) {
3852 /* switch from 'device offset' to 'processed block count' */
3853 max_progress = info->component_size - max_progress;
3854 wait_point = info->component_size - wait_point;
3855 }
3856
3857 if (!*frozen)
3858 sysfs_set_num(info, NULL, "sync_max", max_progress);
3859
3860 /* Now wait. If we have already reached the point that we were
3861 * asked to wait to, don't wait at all, else wait for any change.
3862 * We need to select on 'sync_completed' as that is the place that
3863 * notifications happen, but we are really interested in
3864 * 'reshape_position'
3865 */
3866 fd = sysfs_get_fd(info, NULL, "sync_completed");
3867 if (fd < 0)
3868 goto check_progress;
3869
3870 if (sysfs_fd_get_ll(fd, &completed) < 0)
3871 goto check_progress;
3872
3873 while (completed < max_progress && completed < wait_point) {
3874 /* Check that sync_action is still 'reshape' to avoid
3875 * waiting forever on a dead array
3876 */
3877 char action[20];
3878 if (sysfs_get_str(info, NULL, "sync_action",
3879 action, 20) <= 0 ||
3880 strncmp(action, "reshape", 7) != 0)
3881 break;
3882 /* Some kernels reset 'sync_completed' to zero
3883 * before setting 'sync_action' to 'idle'.
3884 * So we need these extra tests.
3885 */
3886 if (completed == 0 && advancing
3887 && info->reshape_progress > 0)
3888 break;
3889 if (completed == 0 && !advancing
3890 && info->reshape_progress < (info->component_size
3891 * reshape->after.data_disks))
3892 break;
3893 sysfs_wait(fd, NULL);
3894 if (sysfs_fd_get_ll(fd, &completed) < 0)
3895 goto check_progress;
3896 }
3897 /* Some kernels reset 'sync_completed' to zero,
3898 * we need to have real point we are in md
3899 */
3900 if (completed == 0)
3901 completed = max_progress;
3902
3903 /* some kernels can give an incorrectly high 'completed' number */
3904 completed /= (info->new_chunk/512);
3905 completed *= (info->new_chunk/512);
3906 /* Convert 'completed' back in to a 'progress' number */
3907 completed *= reshape->after.data_disks;
3908 if (!advancing) {
3909 completed = info->component_size * reshape->after.data_disks
3910 - completed;
3911 }
3912 *reshape_completed = completed;
3913
3914 close(fd);
3915
3916 /* We return the need_backup flag. Caller will decide
3917 * how much - a multiple of ->backup_blocks up to *suspend_point
3918 */
3919 if (advancing)
3920 return need_backup > info->reshape_progress;
3921 else
3922 return need_backup >= info->reshape_progress;
3923
3924 check_progress:
3925 /* if we couldn't read a number from sync_completed, then
3926 * either the reshape did complete, or it aborted.
3927 * We can tell which by checking for 'none' in reshape_position.
3928 * If it did abort, then it might immediately restart if it
3929 * it was just a device failure that leaves us degraded but
3930 * functioning.
3931 */
3932 strcpy(buf, "hi");
3933 if (sysfs_get_str(info, NULL, "reshape_position", buf, sizeof(buf)) < 0
3934 || strncmp(buf, "none", 4) != 0) {
3935 /* The abort might only be temporary. Wait up to 10
3936 * seconds for fd to contain a valid number again.
3937 */
3938 int wait = 10000;
3939 int rv = -2;
3940 unsigned long long new_sync_max;
3941 while (fd >= 0 && rv < 0 && wait > 0) {
3942 if (sysfs_wait(fd, &wait) != 1)
3943 break;
3944 switch (sysfs_fd_get_ll(fd, &completed)) {
3945 case 0:
3946 /* all good again */
3947 rv = 1;
3948 /* If "sync_max" is no longer max_progress
3949 * we need to freeze things
3950 */
3951 sysfs_get_ll(info, NULL, "sync_max", &new_sync_max);
3952 *frozen = (new_sync_max != max_progress);
3953 break;
3954 case -2: /* read error - abort */
3955 wait = 0;
3956 break;
3957 }
3958 }
3959 if (fd >= 0)
3960 close(fd);
3961 return rv; /* abort */
3962 } else {
3963 /* Maybe racing with array shutdown - check state */
3964 if (fd >= 0)
3965 close(fd);
3966 if (sysfs_get_str(info, NULL, "array_state", buf, sizeof(buf)) < 0
3967 || strncmp(buf, "inactive", 8) == 0
3968 || strncmp(buf, "clear",5) == 0)
3969 return -2; /* abort */
3970 return -1; /* complete */
3971 }
3972 }
3973
3974 /* FIXME return status is never checked */
3975 static int grow_backup(struct mdinfo *sra,
3976 unsigned long long offset, /* per device */
3977 unsigned long stripes, /* per device, in old chunks */
3978 int *sources, unsigned long long *offsets,
3979 int disks, int chunk, int level, int layout,
3980 int dests, int *destfd, unsigned long long *destoffsets,
3981 int part, int *degraded,
3982 char *buf)
3983 {
3984 /* Backup 'blocks' sectors at 'offset' on each device of the array,
3985 * to storage 'destfd' (offset 'destoffsets'), after first
3986 * suspending IO. Then allow resync to continue
3987 * over the suspended section.
3988 * Use part 'part' of the backup-super-block.
3989 */
3990 int odata = disks;
3991 int rv = 0;
3992 int i;
3993 unsigned long long ll;
3994 int new_degraded;
3995 //printf("offset %llu\n", offset);
3996 if (level >= 4)
3997 odata--;
3998 if (level == 6)
3999 odata--;
4000
4001 /* Check that array hasn't become degraded, else we might backup the wrong data */
4002 if (sysfs_get_ll(sra, NULL, "degraded", &ll) < 0)
4003 return -1; /* FIXME this error is ignored */
4004 new_degraded = (int)ll;
4005 if (new_degraded != *degraded) {
4006 /* check each device to ensure it is still working */
4007 struct mdinfo *sd;
4008 for (sd = sra->devs ; sd ; sd = sd->next) {
4009 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4010 continue;
4011 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
4012 char sbuf[20];
4013 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
4014 strstr(sbuf, "faulty") ||
4015 strstr(sbuf, "in_sync") == NULL) {
4016 /* this device is dead */
4017 sd->disk.state = (1<<MD_DISK_FAULTY);
4018 if (sd->disk.raid_disk >= 0 &&
4019 sources[sd->disk.raid_disk] >= 0) {
4020 close(sources[sd->disk.raid_disk]);
4021 sources[sd->disk.raid_disk] = -1;
4022 }
4023 }
4024 }
4025 }
4026 *degraded = new_degraded;
4027 }
4028 if (part) {
4029 bsb.arraystart2 = __cpu_to_le64(offset * odata);
4030 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
4031 } else {
4032 bsb.arraystart = __cpu_to_le64(offset * odata);
4033 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
4034 }
4035 if (part)
4036 bsb.magic[15] = '2';
4037 for (i = 0; i < dests; i++)
4038 if (part)
4039 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
4040 else
4041 lseek64(destfd[i], destoffsets[i], 0);
4042
4043 rv = save_stripes(sources, offsets,
4044 disks, chunk, level, layout,
4045 dests, destfd,
4046 offset*512*odata, stripes * chunk * odata,
4047 buf);
4048
4049 if (rv)
4050 return rv;
4051 bsb.mtime = __cpu_to_le64(time(0));
4052 for (i = 0; i < dests; i++) {
4053 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4054
4055 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4056 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4057 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4058 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4059
4060 rv = -1;
4061 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
4062 != destoffsets[i] - 4096)
4063 break;
4064 if (write(destfd[i], &bsb, 512) != 512)
4065 break;
4066 if (destoffsets[i] > 4096) {
4067 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
4068 destoffsets[i]+stripes*chunk*odata)
4069 break;
4070 if (write(destfd[i], &bsb, 512) != 512)
4071 break;
4072 }
4073 fsync(destfd[i]);
4074 rv = 0;
4075 }
4076
4077 return rv;
4078 }
4079
4080 /* in 2.6.30, the value reported by sync_completed can be
4081 * less that it should be by one stripe.
4082 * This only happens when reshape hits sync_max and pauses.
4083 * So allow wait_backup to either extent sync_max further
4084 * than strictly necessary, or return before the
4085 * sync has got quite as far as we would really like.
4086 * This is what 'blocks2' is for.
4087 * The various caller give appropriate values so that
4088 * every works.
4089 */
4090 /* FIXME return value is often ignored */
4091 static int forget_backup(int dests, int *destfd,
4092 unsigned long long *destoffsets,
4093 int part)
4094 {
4095 /*
4096 * Erase backup 'part' (which is 0 or 1)
4097 */
4098 int i;
4099 int rv;
4100
4101 if (part) {
4102 bsb.arraystart2 = __cpu_to_le64(0);
4103 bsb.length2 = __cpu_to_le64(0);
4104 } else {
4105 bsb.arraystart = __cpu_to_le64(0);
4106 bsb.length = __cpu_to_le64(0);
4107 }
4108 bsb.mtime = __cpu_to_le64(time(0));
4109 rv = 0;
4110 for (i = 0; i < dests; i++) {
4111 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
4112 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
4113 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
4114 bsb.sb_csum2 = bsb_csum((char*)&bsb,
4115 ((char*)&bsb.sb_csum2)-((char*)&bsb));
4116 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
4117 destoffsets[i]-4096)
4118 rv = -1;
4119 if (rv == 0 &&
4120 write(destfd[i], &bsb, 512) != 512)
4121 rv = -1;
4122 fsync(destfd[i]);
4123 }
4124 return rv;
4125 }
4126
4127 static void fail(char *msg)
4128 {
4129 int rv;
4130 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
4131 rv |= (write(2, "\n", 1) != 1);
4132 exit(rv ? 1 : 2);
4133 }
4134
4135 static char *abuf, *bbuf;
4136 static unsigned long long abuflen;
4137 static void validate(int afd, int bfd, unsigned long long offset)
4138 {
4139 /* check that the data in the backup against the array.
4140 * This is only used for regression testing and should not
4141 * be used while the array is active
4142 */
4143 if (afd < 0)
4144 return;
4145 lseek64(bfd, offset - 4096, 0);
4146 if (read(bfd, &bsb2, 512) != 512)
4147 fail("cannot read bsb");
4148 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
4149 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
4150 fail("first csum bad");
4151 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
4152 fail("magic is bad");
4153 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
4154 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
4155 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
4156 fail("second csum bad");
4157
4158 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
4159 fail("devstart is wrong");
4160
4161 if (bsb2.length) {
4162 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
4163
4164 if (abuflen < len) {
4165 free(abuf);
4166 free(bbuf);
4167 abuflen = len;
4168 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
4169 posix_memalign((void**)&bbuf, 4096, abuflen)) {
4170 abuflen = 0;
4171 /* just stop validating on mem-alloc failure */
4172 return;
4173 }
4174 }
4175
4176 lseek64(bfd, offset, 0);
4177 if ((unsigned long long)read(bfd, bbuf, len) != len) {
4178 //printf("len %llu\n", len);
4179 fail("read first backup failed");
4180 }
4181 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
4182 if ((unsigned long long)read(afd, abuf, len) != len)
4183 fail("read first from array failed");
4184 if (memcmp(bbuf, abuf, len) != 0) {
4185 #if 0
4186 int i;
4187 printf("offset=%llu len=%llu\n",
4188 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
4189 for (i=0; i<len; i++)
4190 if (bbuf[i] != abuf[i]) {
4191 printf("first diff byte %d\n", i);
4192 break;
4193 }
4194 #endif
4195 fail("data1 compare failed");
4196 }
4197 }
4198 if (bsb2.length2) {
4199 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
4200
4201 if (abuflen < len) {
4202 free(abuf);
4203 free(bbuf);
4204 abuflen = len;
4205 abuf = xmalloc(abuflen);
4206 bbuf = xmalloc(abuflen);
4207 }
4208
4209 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
4210 if ((unsigned long long)read(bfd, bbuf, len) != len)
4211 fail("read second backup failed");
4212 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
4213 if ((unsigned long long)read(afd, abuf, len) != len)
4214 fail("read second from array failed");
4215 if (memcmp(bbuf, abuf, len) != 0)
4216 fail("data2 compare failed");
4217 }
4218 }
4219
4220 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
4221 struct supertype *st, unsigned long blocks,
4222 int *fds, unsigned long long *offsets,
4223 int dests, int *destfd, unsigned long long *destoffsets)
4224 {
4225 /* Monitor a reshape where backup is being performed using
4226 * 'native' mechanism - either to a backup file, or
4227 * to some space in a spare.
4228 */
4229 char *buf;
4230 int degraded = -1;
4231 unsigned long long speed;
4232 unsigned long long suspend_point, array_size;
4233 unsigned long long backup_point, wait_point;
4234 unsigned long long reshape_completed;
4235 int done = 0;
4236 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
4237 int part = 0; /* The next part of the backup area to fill. It may already
4238 * be full, so we need to check */
4239 int level = reshape->level;
4240 int layout = reshape->before.layout;
4241 int data = reshape->before.data_disks;
4242 int disks = reshape->before.data_disks + reshape->parity;
4243 int chunk = sra->array.chunk_size;
4244 struct mdinfo *sd;
4245 unsigned long stripes;
4246 int uuid[4];
4247 int frozen = 0;
4248
4249 /* set up the backup-super-block. This requires the
4250 * uuid from the array.
4251 */
4252 /* Find a superblock */
4253 for (sd = sra->devs; sd; sd = sd->next) {
4254 char *dn;
4255 int devfd;
4256 int ok;
4257 if (sd->disk.state & (1<<MD_DISK_FAULTY))
4258 continue;
4259 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
4260 devfd = dev_open(dn, O_RDONLY);
4261 if (devfd < 0)
4262 continue;
4263 ok = st->ss->load_super(st, devfd, NULL);
4264 close(devfd);
4265 if (ok == 0)
4266 break;
4267 }
4268 if (!sd) {
4269 pr_err("Cannot find a superblock\n");
4270 return 0;
4271 }
4272
4273 memset(&bsb, 0, 512);
4274 memcpy(bsb.magic, "md_backup_data-1", 16);
4275 st->ss->uuid_from_super(st, uuid);
4276 memcpy(bsb.set_uuid, uuid, 16);
4277 bsb.mtime = __cpu_to_le64(time(0));
4278 bsb.devstart2 = blocks;
4279
4280 stripes = blocks / (sra->array.chunk_size/512) /
4281 reshape->before.data_disks;
4282
4283 if (posix_memalign((void**)&buf, 4096, disks * chunk))
4284 /* Don't start the 'reshape' */
4285 return 0;
4286 if (reshape->before.data_disks == reshape->after.data_disks) {
4287 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
4288 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
4289 }
4290
4291 if (increasing) {
4292 array_size = sra->component_size * reshape->after.data_disks;
4293 backup_point = sra->reshape_progress;
4294 suspend_point = 0;
4295 } else {
4296 array_size = sra->component_size * reshape->before.data_disks;
4297 backup_point = reshape->backup_blocks;
4298 suspend_point = array_size;
4299 }
4300
4301 while (!done) {
4302 int rv;
4303
4304 /* Want to return as soon the oldest backup slot can
4305 * be released as that allows us to start backing up
4306 * some more, providing suspend_point has been
4307 * advanced, which it should have.
4308 */
4309 if (increasing) {
4310 wait_point = array_size;
4311 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4312 wait_point = (__le64_to_cpu(bsb.arraystart) +
4313 __le64_to_cpu(bsb.length));
4314 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4315 wait_point = (__le64_to_cpu(bsb.arraystart2) +
4316 __le64_to_cpu(bsb.length2));
4317 } else {
4318 wait_point = 0;
4319 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
4320 wait_point = __le64_to_cpu(bsb.arraystart);
4321 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
4322 wait_point = __le64_to_cpu(bsb.arraystart2);
4323 }
4324
4325 reshape_completed = sra->reshape_progress;
4326 rv = progress_reshape(sra, reshape,
4327 backup_point, wait_point,
4328 &suspend_point, &reshape_completed,
4329 &frozen);
4330 /* external metadata would need to ping_monitor here */
4331 sra->reshape_progress = reshape_completed;
4332
4333 /* Clear any backup region that is before 'here' */
4334 if (increasing) {
4335 if (__le64_to_cpu(bsb.length) > 0 &&
4336 reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
4337 __le64_to_cpu(bsb.length)))
4338 forget_backup(dests, destfd,
4339 destoffsets, 0);
4340 if (__le64_to_cpu(bsb.length2) > 0 &&
4341 reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
4342 __le64_to_cpu(bsb.length2)))
4343 forget_backup(dests, destfd,
4344 destoffsets, 1);
4345 } else {
4346 if (__le64_to_cpu(bsb.length) > 0 &&
4347 reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
4348 forget_backup(dests, destfd,
4349 destoffsets, 0);
4350 if (__le64_to_cpu(bsb.length2) > 0 &&
4351 reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
4352 forget_backup(dests, destfd,
4353 destoffsets, 1);
4354 }
4355 if (sigterm)
4356 rv = -2;
4357 if (rv < 0) {
4358 if (rv == -1)
4359 done = 1;
4360 break;
4361 }
4362 if (rv == 0 && increasing && !st->ss->external) {
4363 /* No longer need to monitor this reshape */
4364 sysfs_set_str(sra, NULL, "sync_max", "max");
4365 done = 1;
4366 break;
4367 }
4368
4369 while (rv) {
4370 unsigned long long offset;
4371 unsigned long actual_stripes;
4372 /* Need to backup some data.
4373 * If 'part' is not used and the desired
4374 * backup size is suspended, do a backup,
4375 * then consider the next part.
4376 */
4377 /* Check that 'part' is unused */
4378 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
4379 break;
4380 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
4381 break;
4382
4383 offset = backup_point / data;
4384 actual_stripes = stripes;
4385 if (increasing) {
4386 if (offset + actual_stripes * (chunk/512) >
4387 sra->component_size)
4388 actual_stripes = ((sra->component_size - offset)
4389 / (chunk/512));
4390 if (offset + actual_stripes * (chunk/512) >
4391 suspend_point/data)
4392 break;
4393 } else {
4394 if (offset < actual_stripes * (chunk/512))
4395 actual_stripes = offset / (chunk/512);
4396 offset -= actual_stripes * (chunk/512);
4397 if (offset < suspend_point/data)
4398 break;
4399 }
4400 if (actual_stripes == 0)
4401 break;
4402 grow_backup(sra, offset, actual_stripes,
4403 fds, offsets,
4404 disks, chunk, level, layout,
4405 dests, destfd, destoffsets,
4406 part, &degraded, buf);
4407 validate(afd, destfd[0], destoffsets[0]);
4408 /* record where 'part' is up to */
4409 part = !part;
4410 if (increasing)
4411 backup_point += actual_stripes * (chunk/512) * data;
4412 else
4413 backup_point -= actual_stripes * (chunk/512) * data;
4414 }
4415 }
4416
4417 /* FIXME maybe call progress_reshape one more time instead */
4418 /* remove any remaining suspension */
4419 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
4420 sysfs_set_num(sra, NULL, "suspend_hi", 0);
4421 sysfs_set_num(sra, NULL, "suspend_lo", 0);
4422 sysfs_set_num(sra, NULL, "sync_min", 0);
4423
4424 if (reshape->before.data_disks == reshape->after.data_disks)
4425 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
4426 free(buf);
4427 return done;
4428 }
4429
4430 /*
4431 * If any spare contains md_back_data-1 which is recent wrt mtime,
4432 * write that data into the array and update the super blocks with
4433 * the new reshape_progress
4434 */
4435 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
4436 char *backup_file, int verbose)
4437 {
4438 int i, j;
4439 int old_disks;
4440 unsigned long long *offsets;
4441 unsigned long long nstripe, ostripe;
4442 int ndata, odata;
4443
4444 odata = info->array.raid_disks - info->delta_disks - 1;
4445 if (info->array.level == 6) odata--; /* number of data disks */
4446 ndata = info->array.raid_disks - 1;
4447 if (info->new_level == 6) ndata--;
4448
4449 old_disks = info->array.raid_disks - info->delta_disks;
4450
4451 if (info->delta_disks <= 0)
4452 /* Didn't grow, so the backup file must have
4453 * been used
4454 */
4455 old_disks = cnt;
4456 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
4457 struct mdinfo dinfo;
4458 int fd;
4459 int bsbsize;
4460 char *devname, namebuf[20];
4461 unsigned long long lo, hi;
4462
4463 /* This was a spare and may have some saved data on it.
4464 * Load the superblock, find and load the
4465 * backup_super_block.
4466 * If either fail, go on to next device.
4467 * If the backup contains no new info, just return
4468 * else restore data and update all superblocks
4469 */
4470 if (i == old_disks-1) {
4471 fd = open(backup_file, O_RDONLY);
4472 if (fd<0) {
4473 pr_err("backup file %s inaccessible: %s\n",
4474 backup_file, strerror(errno));
4475 continue;
4476 }
4477 devname = backup_file;
4478 } else {
4479 fd = fdlist[i];
4480 if (fd < 0)
4481 continue;
4482 if (st->ss->load_super(st, fd, NULL))
4483 continue;
4484
4485 st->ss->getinfo_super(st, &dinfo, NULL);
4486 st->ss->free_super(st);
4487
4488 if (lseek64(fd,
4489 (dinfo.data_offset + dinfo.component_size - 8) <<9,
4490 0) < 0) {
4491 pr_err("Cannot seek on device %d\n", i);
4492 continue; /* Cannot seek */
4493 }
4494 sprintf(namebuf, "device-%d", i);
4495 devname = namebuf;
4496 }
4497 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
4498 if (verbose)
4499 pr_err("Cannot read from %s\n", devname);
4500 continue; /* Cannot read */
4501 }
4502 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
4503 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
4504 if (verbose)
4505 pr_err("No backup metadata on %s\n", devname);
4506 continue;
4507 }
4508 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
4509 if (verbose)
4510 pr_err("Bad backup-metadata checksum on %s\n", devname);
4511 continue; /* bad checksum */
4512 }
4513 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
4514 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
4515 if (verbose)
4516 pr_err("Bad backup-metadata checksum2 on %s\n", devname);
4517 continue; /* Bad second checksum */
4518 }
4519 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
4520 if (verbose)
4521 pr_err("Wrong uuid on backup-metadata on %s\n", devname);
4522 continue; /* Wrong uuid */
4523 }
4524
4525 /* array utime and backup-mtime should be updated at much the same time, but it seems that
4526 * sometimes they aren't... So allow considerable flexability in matching, and allow
4527 * this test to be overridden by an environment variable.
4528 */
4529 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
4530 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
4531 if (check_env("MDADM_GROW_ALLOW_OLD")) {
4532 pr_err("accepting backup with timestamp %lu "
4533 "for array with timestamp %lu\n",
4534 (unsigned long)__le64_to_cpu(bsb.mtime),
4535 (unsigned long)info->array.utime);
4536 } else {
4537 pr_err("too-old timestamp on backup-metadata on %s\n", devname);
4538 pr_err("If you think it is should be safe, try 'export MDADM_GROW_ALLOW_OLD=1'\n");
4539 continue; /* time stamp is too bad */
4540 }
4541 }
4542
4543 if (bsb.magic[15] == '1') {
4544 if (bsb.length == 0)
4545 continue;
4546 if (info->delta_disks >= 0) {
4547 /* reshape_progress is increasing */
4548 if (__le64_to_cpu(bsb.arraystart)
4549 + __le64_to_cpu(bsb.length)
4550 < info->reshape_progress) {
4551 nonew:
4552 if (verbose)
4553 pr_err("backup-metadata found on %s but is not needed\n", devname);
4554 continue; /* No new data here */
4555 }
4556 } else {
4557 /* reshape_progress is decreasing */
4558 if (__le64_to_cpu(bsb.arraystart) >=
4559 info->reshape_progress)
4560 goto nonew; /* No new data here */
4561 }
4562 } else {
4563 if (bsb.length == 0 && bsb.length2 == 0)
4564 continue;
4565 if (info->delta_disks >= 0) {
4566 /* reshape_progress is increasing */
4567 if ((__le64_to_cpu(bsb.arraystart)
4568 + __le64_to_cpu(bsb.length)
4569 < info->reshape_progress)
4570 &&
4571 (__le64_to_cpu(bsb.arraystart2)
4572 + __le64_to_cpu(bsb.length2)
4573 < info->reshape_progress))
4574 goto nonew; /* No new data here */
4575 } else {
4576 /* reshape_progress is decreasing */
4577 if (__le64_to_cpu(bsb.arraystart) >=
4578 info->reshape_progress &&
4579 __le64_to_cpu(bsb.arraystart2) >=
4580 info->reshape_progress)
4581 goto nonew; /* No new data here */
4582 }
4583 }
4584 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
4585 second_fail:
4586 if (verbose)
4587 pr_err("Failed to verify secondary backup-metadata block on %s\n",
4588 devname);
4589 continue; /* Cannot seek */
4590 }
4591 /* There should be a duplicate backup superblock 4k before here */
4592 if (lseek64(fd, -4096, 1) < 0 ||
4593 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
4594 goto second_fail; /* Cannot find leading superblock */
4595 if (bsb.magic[15] == '1')
4596 bsbsize = offsetof(struct mdp_backup_super, pad1);
4597 else
4598 bsbsize = offsetof(struct mdp_backup_super, pad);
4599 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
4600 goto second_fail; /* Cannot find leading superblock */
4601
4602 /* Now need the data offsets for all devices. */
4603 offsets = xmalloc(sizeof(*offsets)*info->array.raid_disks);
4604 for(j=0; j<info->array.raid_disks; j++) {
4605 if (fdlist[j] < 0)
4606 continue;
4607 if (st->ss->load_super(st, fdlist[j], NULL))
4608 /* FIXME should be this be an error */
4609 continue;
4610 st->ss->getinfo_super(st, &dinfo, NULL);
4611 st->ss->free_super(st);
4612 offsets[j] = dinfo.data_offset * 512;
4613 }
4614 printf("%s: restoring critical section\n", Name);
4615
4616 if (restore_stripes(fdlist, offsets,
4617 info->array.raid_disks,
4618 info->new_chunk,
4619 info->new_level,
4620 info->new_layout,
4621 fd, __le64_to_cpu(bsb.devstart)*512,
4622 __le64_to_cpu(bsb.arraystart)*512,
4623 __le64_to_cpu(bsb.length)*512, NULL)) {
4624 /* didn't succeed, so giveup */
4625 if (verbose)
4626 pr_err("Error restoring backup from %s\n",
4627 devname);
4628 free(offsets);
4629 return 1;
4630 }
4631
4632 if (bsb.magic[15] == '2' &&
4633 restore_stripes(fdlist, offsets,
4634 info->array.raid_disks,
4635 info->new_chunk,
4636 info->new_level,
4637 info->new_layout,
4638 fd, __le64_to_cpu(bsb.devstart)*512 +
4639 __le64_to_cpu(bsb.devstart2)*512,
4640 __le64_to_cpu(bsb.arraystart2)*512,
4641 __le64_to_cpu(bsb.length2)*512, NULL)) {
4642 /* didn't succeed, so giveup */
4643 if (verbose)
4644 pr_err("Error restoring second backup from %s\n",
4645 devname);
4646 free(offsets);
4647 return 1;
4648 }
4649
4650 free(offsets);
4651
4652 /* Ok, so the data is restored. Let's update those superblocks. */
4653
4654 lo = hi = 0;
4655 if (bsb.length) {
4656 lo = __le64_to_cpu(bsb.arraystart);
4657 hi = lo + __le64_to_cpu(bsb.length);
4658 }
4659 if (bsb.magic[15] == '2' && bsb.length2) {
4660 unsigned long long lo1, hi1;
4661 lo1 = __le64_to_cpu(bsb.arraystart2);
4662 hi1 = lo1 + __le64_to_cpu(bsb.length2);
4663 if (lo == hi) {
4664 lo = lo1;
4665 hi = hi1;
4666 } else if (lo < lo1)
4667 hi = hi1;
4668 else
4669 lo = lo1;
4670 }
4671 if (lo < hi &&
4672 (info->reshape_progress < lo ||
4673 info->reshape_progress > hi))
4674 /* backup does not affect reshape_progress*/ ;
4675 else if (info->delta_disks >= 0) {
4676 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
4677 __le64_to_cpu(bsb.length);
4678 if (bsb.magic[15] == '2') {
4679 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
4680 __le64_to_cpu(bsb.length2);
4681 if (p2 > info->reshape_progress)
4682 info->reshape_progress = p2;
4683 }
4684 } else {
4685 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
4686 if (bsb.magic[15] == '2') {
4687 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
4688 if (p2 < info->reshape_progress)
4689 info->reshape_progress = p2;
4690 }
4691 }
4692 for (j=0; j<info->array.raid_disks; j++) {
4693 if (fdlist[j] < 0)
4694 continue;
4695 if (st->ss->load_super(st, fdlist[j], NULL))
4696 continue;
4697 st->ss->getinfo_super(st, &dinfo, NULL);
4698 dinfo.reshape_progress = info->reshape_progress;
4699 st->ss->update_super(st, &dinfo,
4700 "_reshape_progress",
4701 NULL,0, 0, NULL);
4702 st->ss->store_super(st, fdlist[j]);
4703 st->ss->free_super(st);
4704 }
4705 return 0;
4706 }
4707 /* Didn't find any backup data, try to see if any
4708 * was needed.
4709 */
4710 if (info->delta_disks < 0) {
4711 /* When shrinking, the critical section is at the end.
4712 * So see if we are before the critical section.
4713 */
4714 unsigned long long first_block;
4715 nstripe = ostripe = 0;
4716 first_block = 0;
4717 while (ostripe >= nstripe) {
4718 ostripe += info->array.chunk_size / 512;
4719 first_block = ostripe * odata;
4720 nstripe = first_block / ndata / (info->new_chunk/512) *
4721 (info->new_chunk/512);
4722 }
4723
4724 if (info->reshape_progress >= first_block)
4725 return 0;
4726 }
4727 if (info->delta_disks > 0) {
4728 /* See if we are beyond the critical section. */
4729 unsigned long long last_block;
4730 nstripe = ostripe = 0;
4731 last_block = 0;
4732 while (nstripe >= ostripe) {
4733 nstripe += info->new_chunk / 512;
4734 last_block = nstripe * ndata;
4735 ostripe = last_block / odata / (info->array.chunk_size/512) *
4736 (info->array.chunk_size/512);
4737 }
4738
4739 if (info->reshape_progress >= last_block)
4740 return 0;
4741 }
4742 /* needed to recover critical section! */
4743 if (verbose)
4744 pr_err("Failed to find backup of critical section\n");
4745 return 1;
4746 }
4747
4748 int Grow_continue_command(char *devname, int fd,
4749 char *backup_file, int verbose)
4750 {
4751 int ret_val = 0;
4752 struct supertype *st = NULL;
4753 struct mdinfo *content = NULL;
4754 struct mdinfo array;
4755 char *subarray = NULL;
4756 struct mdinfo *cc = NULL;
4757 struct mdstat_ent *mdstat = NULL;
4758 int cfd = -1;
4759 int fd2 = -1;
4760
4761 dprintf("Grow continue from command line called for %s\n",
4762 devname);
4763
4764 st = super_by_fd(fd, &subarray);
4765 if (!st || !st->ss) {
4766 pr_err("Unable to determine metadata format for %s\n",
4767 devname);
4768 return 1;
4769 }
4770 dprintf("Grow continue is run for ");
4771 if (st->ss->external == 0) {
4772 int d;
4773 dprintf("native array (%s)\n", devname);
4774 if (ioctl(fd, GET_ARRAY_INFO, &array.array) < 0) {
4775 pr_err("%s is not an active md array -"
4776 " aborting\n", devname);
4777 ret_val = 1;
4778 goto Grow_continue_command_exit;
4779 }
4780 content = &array;
4781 /* Need to load a superblock.
4782 * FIXME we should really get what we need from
4783 * sysfs
4784 */
4785 for (d = 0; d < MAX_DISKS; d++) {
4786 mdu_disk_info_t disk;
4787 char *dv;
4788 int err;
4789 disk.number = d;
4790 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
4791 continue;
4792 if (disk.major == 0 && disk.minor == 0)
4793 continue;
4794 if ((disk.state & (1 << MD_DISK_ACTIVE)) == 0)
4795 continue;
4796 dv = map_dev(disk.major, disk.minor, 1);
4797 if (!dv)
4798 continue;
4799 fd2 = dev_open(dv, O_RDONLY);
4800 if (fd2 < 0)
4801 continue;
4802 err = st->ss->load_super(st, fd2, NULL);
4803 close(fd2);
4804 /* invalidate fd2 to avoid possible double close() */
4805 fd2 = -1;
4806 if (err)
4807 continue;
4808 break;
4809 }
4810 if (d == MAX_DISKS) {
4811 pr_err("Unable to load metadata for %s\n",
4812 devname);
4813 ret_val = 1;
4814 goto Grow_continue_command_exit;
4815 }
4816 st->ss->getinfo_super(st, content, NULL);
4817 } else {
4818 char *container;
4819
4820 if (subarray) {
4821 dprintf("subarray (%s)\n", subarray);
4822 container = st->container_devnm;
4823 cfd = open_dev_excl(st->container_devnm);
4824 } else {
4825 container = st->devnm;
4826 close(fd);
4827 cfd = open_dev_excl(st->devnm);
4828 dprintf("container (%s)\n", container);
4829 fd = cfd;
4830 }
4831 if (cfd < 0) {
4832 pr_err("Unable to open container "
4833 "for %s\n", devname);
4834 ret_val = 1;
4835 goto Grow_continue_command_exit;
4836 }
4837
4838 /* find in container array under reshape
4839 */
4840 ret_val = st->ss->load_container(st, cfd, NULL);
4841 if (ret_val) {
4842 pr_err("Cannot read superblock for %s\n",
4843 devname);
4844 ret_val = 1;
4845 goto Grow_continue_command_exit;
4846 }
4847
4848 cc = st->ss->container_content(st, subarray);
4849 for (content = cc; content ; content = content->next) {
4850 char *array;
4851 int allow_reshape = 1;
4852
4853 if (content->reshape_active == 0)
4854 continue;
4855 /* The decision about array or container wide
4856 * reshape is taken in Grow_continue based
4857 * content->reshape_active state, therefore we
4858 * need to check_reshape based on
4859 * reshape_active and subarray name
4860 */
4861 if (content->array.state & (1<<MD_SB_BLOCK_VOLUME))
4862 allow_reshape = 0;
4863 if (content->reshape_active == CONTAINER_RESHAPE &&
4864 (content->array.state
4865 & (1<<MD_SB_BLOCK_CONTAINER_RESHAPE)))
4866 allow_reshape = 0;
4867
4868 if (!allow_reshape) {
4869 pr_err("cannot continue reshape of an array"
4870 " in container with unsupported"
4871 " metadata: %s(%s)\n",
4872 devname, container);
4873 ret_val = 1;
4874 goto Grow_continue_command_exit;
4875 }
4876
4877 array = strchr(content->text_version+1, '/')+1;
4878 mdstat = mdstat_by_subdev(array, container);
4879 if (!mdstat)
4880 continue;
4881 if (mdstat->active == 0) {
4882 pr_err("Skipping inactive array %s.\n",
4883 mdstat->devnm);
4884 free_mdstat(mdstat);
4885 mdstat = NULL;
4886 continue;
4887 }
4888 break;
4889 }
4890 if (!content) {
4891 pr_err("Unable to determine reshaped "
4892 "array for %s\n", devname);
4893 ret_val = 1;
4894 goto Grow_continue_command_exit;
4895 }
4896 fd2 = open_dev(mdstat->devnm);
4897 if (fd2 < 0) {
4898 pr_err("cannot open (%s)\n", mdstat->devnm);
4899 ret_val = 1;
4900 goto Grow_continue_command_exit;
4901 }
4902
4903 sysfs_init(content, fd2, mdstat->devnm);
4904
4905 /* start mdmon in case it is not running
4906 */
4907 if (!mdmon_running(container))
4908 start_mdmon(container);
4909 ping_monitor(container);
4910
4911 if (mdmon_running(container))
4912 st->update_tail = &st->updates;
4913 else {
4914 pr_err("No mdmon found. "
4915 "Grow cannot continue.\n");
4916 ret_val = 1;
4917 goto Grow_continue_command_exit;
4918 }
4919 }
4920
4921 /* verify that array under reshape is started from
4922 * correct position
4923 */
4924 if (verify_reshape_position(content, content->array.level) < 0) {
4925 ret_val = 1;
4926 goto Grow_continue_command_exit;
4927 }
4928
4929 /* continue reshape
4930 */
4931 ret_val = Grow_continue(fd, st, content, backup_file, 1, 0);
4932
4933 Grow_continue_command_exit:
4934 if (fd2 > -1)
4935 close(fd2);
4936 if (cfd > -1)
4937 close(cfd);
4938 st->ss->free_super(st);
4939 free_mdstat(mdstat);
4940 sysfs_free(cc);
4941 free(subarray);
4942
4943 return ret_val;
4944 }
4945
4946 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
4947 char *backup_file, int forked, int freeze_reshape)
4948 {
4949 int ret_val = 2;
4950
4951 if (!info->reshape_active)
4952 return ret_val;
4953
4954 if (st->ss->external) {
4955 int cfd = open_dev(st->container_devnm);
4956
4957 if (cfd < 0)
4958 return 1;
4959
4960 st->ss->load_container(st, cfd, st->container_devnm);
4961 close(cfd);
4962 ret_val = reshape_container(st->container_devnm, NULL, mdfd,
4963 st, info, 0, backup_file,
4964 0, forked,
4965 1 | info->reshape_active,
4966 freeze_reshape);
4967 } else
4968 ret_val = reshape_array(NULL, mdfd, "array", st, info, 1,
4969 NULL, INVALID_SECTORS,
4970 backup_file, 0, forked,
4971 1 | info->reshape_active,
4972 freeze_reshape);
4973
4974 return ret_val;
4975 }
4976
4977 char *make_backup(char *name)
4978 {
4979 char *base = "backup_file-";
4980 int len;
4981 char *fname;
4982
4983 len = strlen(MAP_DIR) + 1 + strlen(base) + strlen(name)+1;
4984 fname = xmalloc(len);
4985 sprintf(fname, "%s/%s%s", MAP_DIR, base, name);
4986 return fname;
4987 }
4988
4989 char *locate_backup(char *name)
4990 {
4991 char *fl = make_backup(name);
4992 struct stat stb;
4993
4994 if (stat(fl, &stb) == 0 &&
4995 S_ISREG(stb.st_mode))
4996 return fl;
4997
4998 free(fl);
4999 return NULL;
5000 }
Unnamed repository; edit this file 'description' to name the repository.