]> 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
Don't complain about missing spares when reshaping a raid0.
[thirdparty/mdadm.git] / Grow.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 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
28 #if ! defined(__BIG_ENDIAN) && ! defined(__LITTLE_ENDIAN)
29 #error no endian defined
30 #endif
31 #include "md_u.h"
32 #include "md_p.h"
33
34 #ifndef offsetof
35 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
36 #endif
37
38 int Grow_Add_device(char *devname, int fd, char *newdev)
39 {
40 /* Add a device to an active array.
41 * Currently, just extend a linear array.
42 * This requires writing a new superblock on the
43 * new device, calling the kernel to add the device,
44 * and if that succeeds, update the superblock on
45 * all other devices.
46 * This means that we need to *find* all other devices.
47 */
48 struct mdinfo info;
49
50 struct stat stb;
51 int nfd, fd2;
52 int d, nd;
53 struct supertype *st = NULL;
54 char *subarray = NULL;
55
56 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
57 fprintf(stderr, Name ": cannot get array info for %s\n", devname);
58 return 1;
59 }
60
61 if (info.array.level != -1) {
62 fprintf(stderr, Name ": can only add devices to linear arrays\n");
63 return 1;
64 }
65
66 st = super_by_fd(fd, &subarray);
67 if (!st) {
68 fprintf(stderr, Name ": cannot handle arrays with superblock version %d\n", info.array.major_version);
69 return 1;
70 }
71
72 if (subarray) {
73 fprintf(stderr, Name ": Cannot grow linear sub-arrays yet\n");
74 free(subarray);
75 free(st);
76 }
77
78 nfd = open(newdev, O_RDWR|O_EXCL|O_DIRECT);
79 if (nfd < 0) {
80 fprintf(stderr, Name ": cannot open %s\n", newdev);
81 free(st);
82 return 1;
83 }
84 fstat(nfd, &stb);
85 if ((stb.st_mode & S_IFMT) != S_IFBLK) {
86 fprintf(stderr, Name ": %s is not a block device!\n", newdev);
87 close(nfd);
88 free(st);
89 return 1;
90 }
91 /* now check out all the devices and make sure we can read the superblock */
92 for (d=0 ; d < info.array.raid_disks ; d++) {
93 mdu_disk_info_t disk;
94 char *dv;
95
96 st->ss->free_super(st);
97
98 disk.number = d;
99 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
100 fprintf(stderr, Name ": cannot get device detail for device %d\n",
101 d);
102 close(nfd);
103 free(st);
104 return 1;
105 }
106 dv = map_dev(disk.major, disk.minor, 1);
107 if (!dv) {
108 fprintf(stderr, Name ": cannot find device file for device %d\n",
109 d);
110 close(nfd);
111 free(st);
112 return 1;
113 }
114 fd2 = dev_open(dv, O_RDWR);
115 if (!fd2) {
116 fprintf(stderr, Name ": cannot open device file %s\n", dv);
117 close(nfd);
118 free(st);
119 return 1;
120 }
121
122 if (st->ss->load_super(st, fd2, NULL)) {
123 fprintf(stderr, Name ": cannot find super block on %s\n", dv);
124 close(nfd);
125 close(fd2);
126 free(st);
127 return 1;
128 }
129 close(fd2);
130 }
131 /* Ok, looks good. Lets update the superblock and write it out to
132 * newdev.
133 */
134
135 info.disk.number = d;
136 info.disk.major = major(stb.st_rdev);
137 info.disk.minor = minor(stb.st_rdev);
138 info.disk.raid_disk = d;
139 info.disk.state = (1 << MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE);
140 st->ss->update_super(st, &info, "linear-grow-new", newdev,
141 0, 0, NULL);
142
143 if (st->ss->store_super(st, nfd)) {
144 fprintf(stderr, Name ": Cannot store new superblock on %s\n",
145 newdev);
146 close(nfd);
147 return 1;
148 }
149 close(nfd);
150
151 if (ioctl(fd, ADD_NEW_DISK, &info.disk) != 0) {
152 fprintf(stderr, Name ": Cannot add new disk to this array\n");
153 return 1;
154 }
155 /* Well, that seems to have worked.
156 * Now go through and update all superblocks
157 */
158
159 if (ioctl(fd, GET_ARRAY_INFO, &info.array) < 0) {
160 fprintf(stderr, Name ": cannot get array info for %s\n", devname);
161 return 1;
162 }
163
164 nd = d;
165 for (d=0 ; d < info.array.raid_disks ; d++) {
166 mdu_disk_info_t disk;
167 char *dv;
168
169 disk.number = d;
170 if (ioctl(fd, GET_DISK_INFO, &disk) < 0) {
171 fprintf(stderr, Name ": cannot get device detail for device %d\n",
172 d);
173 return 1;
174 }
175 dv = map_dev(disk.major, disk.minor, 1);
176 if (!dv) {
177 fprintf(stderr, Name ": cannot find device file for device %d\n",
178 d);
179 return 1;
180 }
181 fd2 = dev_open(dv, O_RDWR);
182 if (fd2 < 0) {
183 fprintf(stderr, Name ": cannot open device file %s\n", dv);
184 return 1;
185 }
186 if (st->ss->load_super(st, fd2, NULL)) {
187 fprintf(stderr, Name ": cannot find super block on %s\n", dv);
188 close(fd);
189 return 1;
190 }
191 info.array.raid_disks = nd+1;
192 info.array.nr_disks = nd+1;
193 info.array.active_disks = nd+1;
194 info.array.working_disks = nd+1;
195
196 st->ss->update_super(st, &info, "linear-grow-update", dv,
197 0, 0, NULL);
198
199 if (st->ss->store_super(st, fd2)) {
200 fprintf(stderr, Name ": Cannot store new superblock on %s\n", dv);
201 close(fd2);
202 return 1;
203 }
204 close(fd2);
205 }
206
207 return 0;
208 }
209
210 int Grow_addbitmap(char *devname, int fd, char *file, int chunk, int delay, int write_behind, int force)
211 {
212 /*
213 * First check that array doesn't have a bitmap
214 * Then create the bitmap
215 * Then add it
216 *
217 * For internal bitmaps, we need to check the version,
218 * find all the active devices, and write the bitmap block
219 * to all devices
220 */
221 mdu_bitmap_file_t bmf;
222 mdu_array_info_t array;
223 struct supertype *st;
224 char *subarray = NULL;
225 int major = BITMAP_MAJOR_HI;
226 int vers = md_get_version(fd);
227 unsigned long long bitmapsize, array_size;
228
229 if (vers < 9003) {
230 major = BITMAP_MAJOR_HOSTENDIAN;
231 fprintf(stderr, Name ": Warning - bitmaps created on this kernel"
232 " are not portable\n"
233 " between different architectures. Consider upgrading"
234 " the Linux kernel.\n");
235 }
236
237 if (ioctl(fd, GET_BITMAP_FILE, &bmf) != 0) {
238 if (errno == ENOMEM)
239 fprintf(stderr, Name ": Memory allocation failure.\n");
240 else
241 fprintf(stderr, Name ": bitmaps not supported by this kernel.\n");
242 return 1;
243 }
244 if (bmf.pathname[0]) {
245 if (strcmp(file,"none")==0) {
246 if (ioctl(fd, SET_BITMAP_FILE, -1)!= 0) {
247 fprintf(stderr, Name ": failed to remove bitmap %s\n",
248 bmf.pathname);
249 return 1;
250 }
251 return 0;
252 }
253 fprintf(stderr, Name ": %s already has a bitmap (%s)\n",
254 devname, bmf.pathname);
255 return 1;
256 }
257 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
258 fprintf(stderr, Name ": cannot get array status for %s\n", devname);
259 return 1;
260 }
261 if (array.state & (1<<MD_SB_BITMAP_PRESENT)) {
262 if (strcmp(file, "none")==0) {
263 array.state &= ~(1<<MD_SB_BITMAP_PRESENT);
264 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
265 fprintf(stderr, Name ": failed to remove internal bitmap.\n");
266 return 1;
267 }
268 return 0;
269 }
270 fprintf(stderr, Name ": Internal bitmap already present on %s\n",
271 devname);
272 return 1;
273 }
274
275 if (strcmp(file, "none") == 0) {
276 fprintf(stderr, Name ": no bitmap found on %s\n", devname);
277 return 1;
278 }
279 if (array.level <= 0) {
280 fprintf(stderr, Name ": Bitmaps not meaningful with level %s\n",
281 map_num(pers, array.level)?:"of this array");
282 return 1;
283 }
284 bitmapsize = array.size;
285 bitmapsize <<= 1;
286 if (get_dev_size(fd, NULL, &array_size) &&
287 array_size > (0x7fffffffULL<<9)) {
288 /* Array is big enough that we cannot trust array.size
289 * try other approaches
290 */
291 bitmapsize = get_component_size(fd);
292 }
293 if (bitmapsize == 0) {
294 fprintf(stderr, Name ": Cannot reliably determine size of array to create bitmap - sorry.\n");
295 return 1;
296 }
297
298 if (array.level == 10) {
299 int ncopies = (array.layout&255)*((array.layout>>8)&255);
300 bitmapsize = bitmapsize * array.raid_disks / ncopies;
301 }
302
303 st = super_by_fd(fd, &subarray);
304 if (!st) {
305 fprintf(stderr, Name ": Cannot understand version %d.%d\n",
306 array.major_version, array.minor_version);
307 return 1;
308 }
309 if (subarray) {
310 fprintf(stderr, Name ": Cannot add bitmaps to sub-arrays yet\n");
311 free(subarray);
312 free(st);
313 return 1;
314 }
315 if (strcmp(file, "internal") == 0) {
316 int d;
317 if (st->ss->add_internal_bitmap == NULL) {
318 fprintf(stderr, Name ": Internal bitmaps not supported "
319 "with %s metadata\n", st->ss->name);
320 return 1;
321 }
322 for (d=0; d< st->max_devs; d++) {
323 mdu_disk_info_t disk;
324 char *dv;
325 disk.number = d;
326 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
327 continue;
328 if (disk.major == 0 &&
329 disk.minor == 0)
330 continue;
331 if ((disk.state & (1<<MD_DISK_SYNC))==0)
332 continue;
333 dv = map_dev(disk.major, disk.minor, 1);
334 if (dv) {
335 int fd2 = dev_open(dv, O_RDWR);
336 if (fd2 < 0)
337 continue;
338 if (st->ss->load_super(st, fd2, NULL)==0) {
339 if (st->ss->add_internal_bitmap(
340 st,
341 &chunk, delay, write_behind,
342 bitmapsize, 0, major)
343 )
344 st->ss->write_bitmap(st, fd2);
345 else {
346 fprintf(stderr, Name ": failed to create internal bitmap - chunksize problem.\n");
347 close(fd2);
348 return 1;
349 }
350 }
351 close(fd2);
352 }
353 }
354 array.state |= (1<<MD_SB_BITMAP_PRESENT);
355 if (ioctl(fd, SET_ARRAY_INFO, &array)!= 0) {
356 if (errno == EBUSY)
357 fprintf(stderr, Name
358 ": Cannot add bitmap while array is"
359 " resyncing or reshaping etc.\n");
360 fprintf(stderr, Name ": failed to set internal bitmap.\n");
361 return 1;
362 }
363 } else {
364 int uuid[4];
365 int bitmap_fd;
366 int d;
367 int max_devs = st->max_devs;
368
369 /* try to load a superblock */
370 for (d=0; d<max_devs; d++) {
371 mdu_disk_info_t disk;
372 char *dv;
373 int fd2;
374 disk.number = d;
375 if (ioctl(fd, GET_DISK_INFO, &disk) < 0)
376 continue;
377 if ((disk.major==0 && disk.minor==0) ||
378 (disk.state & (1<<MD_DISK_REMOVED)))
379 continue;
380 dv = map_dev(disk.major, disk.minor, 1);
381 if (!dv) continue;
382 fd2 = dev_open(dv, O_RDONLY);
383 if (fd2 >= 0 &&
384 st->ss->load_super(st, fd2, NULL) == 0) {
385 close(fd2);
386 st->ss->uuid_from_super(st, uuid);
387 break;
388 }
389 close(fd2);
390 }
391 if (d == max_devs) {
392 fprintf(stderr, Name ": cannot find UUID for array!\n");
393 return 1;
394 }
395 if (CreateBitmap(file, force, (char*)uuid, chunk,
396 delay, write_behind, bitmapsize, major)) {
397 return 1;
398 }
399 bitmap_fd = open(file, O_RDWR);
400 if (bitmap_fd < 0) {
401 fprintf(stderr, Name ": weird: %s cannot be opened\n",
402 file);
403 return 1;
404 }
405 if (ioctl(fd, SET_BITMAP_FILE, bitmap_fd) < 0) {
406 int err = errno;
407 if (errno == EBUSY)
408 fprintf(stderr, Name
409 ": Cannot add bitmap while array is"
410 " resyncing or reshaping etc.\n");
411 fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
412 devname, strerror(err));
413 return 1;
414 }
415 }
416
417 return 0;
418 }
419
420
421 /*
422 * When reshaping an array we might need to backup some data.
423 * This is written to all spares with a 'super_block' describing it.
424 * The superblock goes 4K from the end of the used space on the
425 * device.
426 * It if written after the backup is complete.
427 * It has the following structure.
428 */
429
430 static struct mdp_backup_super {
431 char magic[16]; /* md_backup_data-1 or -2 */
432 __u8 set_uuid[16];
433 __u64 mtime;
434 /* start/sizes in 512byte sectors */
435 __u64 devstart; /* address on backup device/file of data */
436 __u64 arraystart;
437 __u64 length;
438 __u32 sb_csum; /* csum of preceeding bytes. */
439 __u32 pad1;
440 __u64 devstart2; /* offset in to data of second section */
441 __u64 arraystart2;
442 __u64 length2;
443 __u32 sb_csum2; /* csum of preceeding bytes. */
444 __u8 pad[512-68-32];
445 } __attribute__((aligned(512))) bsb, bsb2;
446
447 static __u32 bsb_csum(char *buf, int len)
448 {
449 int i;
450 int csum = 0;
451 for (i=0; i<len; i++)
452 csum = (csum<<3) + buf[0];
453 return __cpu_to_le32(csum);
454 }
455
456 static int check_idle(struct supertype *st)
457 {
458 /* Check that all member arrays for this container, or the
459 * container of this array, are idle
460 */
461 int container_dev = (st->container_dev != NoMdDev
462 ? st->container_dev : st->devnum);
463 char container[40];
464 struct mdstat_ent *ent, *e;
465 int is_idle = 1;
466
467 fmt_devname(container, container_dev);
468 ent = mdstat_read(0, 0);
469 for (e = ent ; e; e = e->next) {
470 if (!is_container_member(e, container))
471 continue;
472 if (e->percent >= 0) {
473 is_idle = 0;
474 break;
475 }
476 }
477 free_mdstat(ent);
478 return is_idle;
479 }
480
481 static int freeze_container(struct supertype *st)
482 {
483 int container_dev = (st->container_dev != NoMdDev
484 ? st->container_dev : st->devnum);
485 char container[40];
486
487 if (!check_idle(st))
488 return -1;
489
490 fmt_devname(container, container_dev);
491
492 if (block_monitor(container, 1)) {
493 fprintf(stderr, Name ": failed to freeze container\n");
494 return -2;
495 }
496
497 return 1;
498 }
499
500 static void unfreeze_container(struct supertype *st)
501 {
502 int container_dev = (st->container_dev != NoMdDev
503 ? st->container_dev : st->devnum);
504 char container[40];
505
506 fmt_devname(container, container_dev);
507
508 unblock_monitor(container, 1);
509 }
510
511 static int freeze(struct supertype *st)
512 {
513 /* Try to freeze resync/rebuild on this array/container.
514 * Return -1 if the array is busy,
515 * return -2 container cannot be frozen,
516 * return 0 if this kernel doesn't support 'frozen'
517 * return 1 if it worked.
518 */
519 if (st->ss->external)
520 return freeze_container(st);
521 else {
522 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
523 int err;
524
525 if (!sra)
526 return -1;
527 err = sysfs_freeze_array(sra);
528 sysfs_free(sra);
529 return err;
530 }
531 }
532
533 static void unfreeze(struct supertype *st, int frozen)
534 {
535 /* If 'frozen' is 1, unfreeze the array */
536 if (frozen <= 0)
537 return;
538
539 if (st->ss->external)
540 return unfreeze_container(st);
541 else {
542 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
543
544 if (sra)
545 sysfs_set_str(sra, NULL, "sync_action", "idle");
546 else
547 fprintf(stderr, Name ": failed to unfreeze array\n");
548 sysfs_free(sra);
549 }
550 }
551
552 static void wait_reshape(struct mdinfo *sra)
553 {
554 int fd = sysfs_get_fd(sra, NULL, "sync_action");
555 char action[20];
556
557 if (fd < 0)
558 return;
559
560 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
561 strncmp(action, "reshape", 7) == 0) {
562 fd_set rfds;
563 FD_ZERO(&rfds);
564 FD_SET(fd, &rfds);
565 select(fd+1, NULL, NULL, &rfds, NULL);
566 }
567 close(fd);
568 }
569
570 static int reshape_super(struct supertype *st, long long size, int level,
571 int layout, int chunksize, int raid_disks,
572 char *backup_file, char *dev, int verbose)
573 {
574 /* nothing extra to check in the native case */
575 if (!st->ss->external)
576 return 0;
577 if (!st->ss->reshape_super ||
578 !st->ss->manage_reshape) {
579 fprintf(stderr, Name ": %s metadata does not support reshape\n",
580 st->ss->name);
581 return 1;
582 }
583
584 return st->ss->reshape_super(st, size, level, layout, chunksize,
585 raid_disks, backup_file, dev, verbose);
586 }
587
588 static void sync_metadata(struct supertype *st)
589 {
590 if (st->ss->external) {
591 if (st->update_tail) {
592 flush_metadata_updates(st);
593 st->update_tail = &st->updates;
594 } else
595 st->ss->sync_metadata(st);
596 }
597 }
598
599 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
600 {
601 /* when dealing with external metadata subarrays we need to be
602 * prepared to handle EAGAIN. The kernel may need to wait for
603 * mdmon to mark the array active so the kernel can handle
604 * allocations/writeback when preparing the reshape action
605 * (md_allow_write()). We temporarily disable safe_mode_delay
606 * to close a race with the array_state going clean before the
607 * next write to raid_disks / stripe_cache_size
608 */
609 char safe[50];
610 int rc;
611
612 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
613 if (!container ||
614 (strcmp(name, "raid_disks") != 0 &&
615 strcmp(name, "stripe_cache_size") != 0))
616 return sysfs_set_num(sra, NULL, name, n);
617
618 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
619 if (rc <= 0)
620 return -1;
621 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
622 rc = sysfs_set_num(sra, NULL, name, n);
623 if (rc < 0 && errno == EAGAIN) {
624 ping_monitor(container);
625 /* if we get EAGAIN here then the monitor is not active
626 * so stop trying
627 */
628 rc = sysfs_set_num(sra, NULL, name, n);
629 }
630 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
631 return rc;
632 }
633
634 int start_reshape(struct mdinfo *sra)
635 {
636 int err;
637 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
638 err = sysfs_set_num(sra, NULL, "suspend_hi", 0);
639 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", 0);
640 err = err ?: sysfs_set_num(sra, NULL, "sync_min", 0);
641 err = err ?: sysfs_set_num(sra, NULL, "sync_max", 0);
642 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
643
644 return err;
645 }
646
647 void abort_reshape(struct mdinfo *sra)
648 {
649 sysfs_set_str(sra, NULL, "sync_action", "idle");
650 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
651 sysfs_set_num(sra, NULL, "suspend_hi", 0);
652 sysfs_set_num(sra, NULL, "suspend_lo", 0);
653 sysfs_set_num(sra, NULL, "sync_min", 0);
654 sysfs_set_str(sra, NULL, "sync_max", "max");
655 }
656
657 int remove_disks_on_raid10_to_raid0_takeover(struct supertype *st,
658 struct mdinfo *sra,
659 int layout)
660 {
661 int nr_of_copies;
662 struct mdinfo *remaining;
663 int slot;
664
665 nr_of_copies = layout & 0xff;
666
667 remaining = sra->devs;
668 sra->devs = NULL;
669 /* for each 'copy', select one device and remove from the list. */
670 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
671 struct mdinfo **diskp;
672 int found = 0;
673
674 /* Find a working device to keep */
675 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
676 struct mdinfo *disk = *diskp;
677
678 if (disk->disk.raid_disk < slot)
679 continue;
680 if (disk->disk.raid_disk >= slot + nr_of_copies)
681 continue;
682 if (disk->disk.state & (1<<MD_DISK_REMOVED))
683 continue;
684 if (disk->disk.state & (1<<MD_DISK_FAULTY))
685 continue;
686 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
687 continue;
688
689 /* We have found a good disk to use! */
690 *diskp = disk->next;
691 disk->next = sra->devs;
692 sra->devs = disk;
693 found = 1;
694 break;
695 }
696 if (!found)
697 break;
698 }
699
700 if (slot < sra->array.raid_disks) {
701 /* didn't find all slots */
702 struct mdinfo **e;
703 e = &remaining;
704 while (*e)
705 e = &(*e)->next;
706 *e = sra->devs;
707 sra->devs = remaining;
708 return 1;
709 }
710
711 /* Remove all 'remaining' devices from the array */
712 while (remaining) {
713 struct mdinfo *sd = remaining;
714 remaining = sd->next;
715
716 sysfs_set_str(sra, sd, "state", "faulty");
717 sysfs_set_str(sra, sd, "slot", "none");
718 sysfs_set_str(sra, sd, "state", "remove");
719 sd->disk.state |= (1<<MD_DISK_REMOVED);
720 sd->disk.state &= ~(1<<MD_DISK_SYNC);
721 sd->next = sra->devs;
722 sra->devs = sd;
723 }
724 return 0;
725 }
726
727 void reshape_free_fdlist(int *fdlist,
728 unsigned long long *offsets,
729 int size)
730 {
731 int i;
732
733 for (i = 0; i < size; i++)
734 if (fdlist[i] >= 0)
735 close(fdlist[i]);
736
737 free(fdlist);
738 free(offsets);
739 }
740
741 int reshape_prepare_fdlist(char *devname,
742 struct mdinfo *sra,
743 int raid_disks,
744 int nrdisks,
745 unsigned long blocks,
746 char *backup_file,
747 int *fdlist,
748 unsigned long long *offsets)
749 {
750 int d = 0;
751 struct mdinfo *sd;
752
753 for (d = 0; d <= nrdisks; d++)
754 fdlist[d] = -1;
755 d = raid_disks;
756 for (sd = sra->devs; sd; sd = sd->next) {
757 if (sd->disk.state & (1<<MD_DISK_FAULTY))
758 continue;
759 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
760 char *dn = map_dev(sd->disk.major,
761 sd->disk.minor, 1);
762 fdlist[sd->disk.raid_disk]
763 = dev_open(dn, O_RDONLY);
764 offsets[sd->disk.raid_disk] = sd->data_offset*512;
765 if (fdlist[sd->disk.raid_disk] < 0) {
766 fprintf(stderr,
767 Name ": %s: cannot open component %s\n",
768 devname, dn ? dn : "-unknown-");
769 d = -1;
770 goto release;
771 }
772 } else if (backup_file == NULL) {
773 /* spare */
774 char *dn = map_dev(sd->disk.major,
775 sd->disk.minor, 1);
776 fdlist[d] = dev_open(dn, O_RDWR);
777 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
778 if (fdlist[d] < 0) {
779 fprintf(stderr, Name ": %s: cannot open component %s\n",
780 devname, dn ? dn : "-unknown-");
781 d = -1;
782 goto release;
783 }
784 d++;
785 }
786 }
787 release:
788 return d;
789 }
790
791 int reshape_open_backup_file(char *backup_file,
792 int fd,
793 char *devname,
794 long blocks,
795 int *fdlist,
796 unsigned long long *offsets)
797 {
798 /* Return 1 on success, 0 on any form of failure */
799 /* need to check backup file is large enough */
800 char buf[512];
801 struct stat stb;
802 unsigned int dev;
803 int i;
804
805 *fdlist = open(backup_file, O_RDWR|O_CREAT|O_EXCL,
806 S_IRUSR | S_IWUSR);
807 *offsets = 8 * 512;
808 if (*fdlist < 0) {
809 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
810 devname, backup_file, strerror(errno));
811 return 0;
812 }
813 /* Guard against backup file being on array device.
814 * If array is partitioned or if LVM etc is in the
815 * way this will not notice, but it is better than
816 * nothing.
817 */
818 fstat(*fdlist, &stb);
819 dev = stb.st_dev;
820 fstat(fd, &stb);
821 if (stb.st_rdev == dev) {
822 fprintf(stderr, Name ": backup file must NOT be"
823 " on the array being reshaped.\n");
824 close(*fdlist);
825 return 0;
826 }
827
828 memset(buf, 0, 512);
829 for (i=0; i < blocks + 1 ; i++) {
830 if (write(*fdlist, buf, 512) != 512) {
831 fprintf(stderr, Name ": %s: cannot create"
832 " backup file %s: %s\n",
833 devname, backup_file, strerror(errno));
834 return 0;
835 }
836 }
837 if (fsync(*fdlist) != 0) {
838 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
839 devname, backup_file, strerror(errno));
840 return 0;
841 }
842
843 return 1;
844 }
845
846 unsigned long compute_backup_blocks(int nchunk, int ochunk,
847 unsigned int ndata, unsigned int odata)
848 {
849 unsigned long a, b, blocks;
850 /* So how much do we need to backup.
851 * We need an amount of data which is both a whole number of
852 * old stripes and a whole number of new stripes.
853 * So LCM for (chunksize*datadisks).
854 */
855 a = (ochunk/512) * odata;
856 b = (nchunk/512) * ndata;
857 /* Find GCD */
858 while (a != b) {
859 if (a < b)
860 b -= a;
861 if (b < a)
862 a -= b;
863 }
864 /* LCM == product / GCD */
865 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
866
867 return blocks;
868 }
869
870 char *analyse_change(struct mdinfo *info, struct reshape *re)
871 {
872 /* Based on the current array state in info->array and
873 * the changes in info->new_* etc, determine:
874 * - whether the change is possible
875 * - Intermediate level/raid_disks/layout
876 * - whether a restriping reshape is needed
877 * - number of sectors in minimum change unit. This
878 * will cover a whole number of stripes in 'before' and
879 * 'after'.
880 *
881 * Return message if the change should be rejected
882 * NULL if the change can be achieved
883 *
884 * This can be called as part of starting a reshape, or
885 * when assembling an array that is undergoing reshape.
886 */
887 int new_disks;
888
889 /* If a new level not explicitly given, we assume no-change */
890 if (info->new_level == UnSet)
891 info->new_level = info->array.level;
892
893 if (info->new_chunk)
894 switch (info->new_level) {
895 case 0:
896 case 4:
897 case 5:
898 case 6:
899 case 10:
900 /* chunk size is meaningful, must divide component_size
901 * evenly
902 */
903 if (info->component_size % (info->new_chunk/512))
904 return "New chunk size does not"
905 " divide component size";
906 break;
907 default:
908 return "chunk size not meaningful for this level";
909 }
910 else
911 info->new_chunk = info->array.chunk_size;
912
913 switch (info->array.level) {
914 case 1:
915 /* RAID1 can convert to RAID1 with different disks, or
916 * raid5 with 2 disks
917 */
918 if (info->new_level == 1) {
919 if (info->delta_disks == UnSet)
920 /* Don't know what to do */
921 return "no change requested for Growing RAID1";
922 re->level = 1;
923 re->before.data_disks = (info->array.raid_disks +
924 info->delta_disks);
925 re->before.layout = 0;
926 re->backup_blocks = 0;
927 re->parity = 0;
928 return NULL;
929 }
930 if (info->array.raid_disks == 2 &&
931 info->array.raid_disks == 5) {
932 /* simple in-place conversion */
933 re->level = 5;
934 re->parity = 1;
935 re->before.data_disks = 1;
936 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
937 re->backup_blocks = 0;
938 return NULL;
939 }
940 /* Could do some multi-stage conversions, but leave that to
941 * later.
942 */
943 return "Impossibly level change request for RAID1";
944
945 case 10:
946 /* RAID10 can only be converted from near mode to
947 * RAID0 by removing some devices
948 */
949 if ((info->array.layout & ~0xff) != 0x100)
950 return "Cannot Grow RAID10 with far/offset layout";
951 /* number of devices must be multiple of number of copies */
952 if (info->array.raid_disks % (info->array.layout & 0xff))
953 return "RAID10 layout too complex for Grow operation";
954
955 if (info->new_level != 0)
956 return "RAID10 can only be changed to RAID0";
957 new_disks = (info->array.raid_disks
958 / (info->array.layout & 0xff));
959 if (info->delta_disks != UnSet) {
960 info->delta_disks = (new_disks
961 - info->array.raid_disks);
962 }
963 if (info->delta_disks != new_disks - info->array.raid_disks)
964 return "New number of raid-devices impossible for RAID10";
965 if (info->new_chunk &&
966 info->new_chunk != info->array.chunk_size)
967 return "Cannot change chunk-size with RAID10 Grow";
968
969 /* looks good */
970 re->level = 0;
971 re->parity = 0;
972 re->before.data_disks = new_disks;
973 re->before.layout = 0;
974 re->backup_blocks = 0;
975 return NULL;
976
977 case 0:
978 /* RAID0 can be converted to RAID10, or to RAID456 */
979 if (info->new_level == 10) {
980 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
981 /* Assume near=2 layout */
982 info->new_layout = 0x102;
983 info->delta_disks = info->array.raid_disks;
984 }
985 if (info->new_layout == UnSet) {
986 int copies = 1 + (info->delta_disks
987 / info->array.raid_disks);
988 if (info->array.raid_disks * (copies-1)
989 != info->delta_disks)
990 return "Impossible number of devices"
991 " for RAID0->RAID10";
992 info->new_layout = 0x100 + copies;
993 }
994 if (info->delta_disks == UnSet) {
995 int copies = info->new_layout & 0xff;
996 if (info->new_layout != 0x100 + copies)
997 return "New layout impossible"
998 " for RAID0->RAID10";;
999 info->delta_disks = (copies - 1) *
1000 info->array.raid_disks;
1001 }
1002 if (info->new_chunk &&
1003 info->new_chunk != info->array.chunk_size)
1004 return "Cannot change chunk-size with RAID0->RAID10";
1005 /* looks good */
1006 re->level = 10;
1007 re->parity = 0;
1008 re->before.data_disks = (info->array.raid_disks +
1009 info->delta_disks);
1010 re->before.layout = info->new_layout;
1011 re->backup_blocks = 0;
1012 return NULL;
1013 }
1014
1015 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1016 * a raid4 style layout of the final level.
1017 */
1018 switch (info->new_level) {
1019 case 0:
1020 case 4:
1021 re->level = 4;
1022 re->before.layout = 0;
1023 break;
1024 case 5:
1025 re->level = 5;
1026 re->before.layout = ALGORITHM_PARITY_N;
1027 break;
1028 case 6:
1029 re->level = 6;
1030 re->before.layout = ALGORITHM_PARITY_N;
1031 break;
1032 default:
1033 return "Impossible level change requested";
1034 }
1035 re->before.data_disks = info->array.raid_disks;
1036 /* determining 'after' layout happens outside this 'switch' */
1037 break;
1038
1039 case 4:
1040 info->array.layout = ALGORITHM_PARITY_N;
1041 case 5:
1042 switch (info->new_level) {
1043 case 4:
1044 re->level = info->array.level;
1045 re->before.data_disks = info->array.raid_disks - 1;
1046 re->before.layout = info->array.layout;
1047 break;
1048 case 5:
1049 re->level = 5;
1050 re->before.data_disks = info->array.raid_disks - 1;
1051 re->before.layout = info->array.layout;
1052 break;
1053 case 6:
1054 re->level = 6;
1055 re->before.data_disks = info->array.raid_disks - 1;
1056 switch (info->array.layout) {
1057 case ALGORITHM_LEFT_ASYMMETRIC:
1058 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1059 break;
1060 case ALGORITHM_RIGHT_ASYMMETRIC:
1061 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1062 break;
1063 case ALGORITHM_LEFT_SYMMETRIC:
1064 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1065 break;
1066 case ALGORITHM_RIGHT_SYMMETRIC:
1067 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1068 break;
1069 case ALGORITHM_PARITY_0:
1070 re->before.layout = ALGORITHM_PARITY_0_6;
1071 break;
1072 case ALGORITHM_PARITY_N:
1073 re->before.layout = ALGORITHM_PARITY_N_6;
1074 break;
1075 default:
1076 return "Cannot convert an array with this layout";
1077 }
1078 break;
1079 case 1:
1080 if (info->array.raid_disks != 2)
1081 return "Can only convert a 2-device array to RAID1";
1082 re->level = 1;
1083 re->before.data_disks = 2;
1084 re->before.layout = 0;
1085 break;
1086 default:
1087 return "Impossible level change requested";
1088 }
1089 break;
1090 case 6:
1091 switch (info->new_level) {
1092 case 4:
1093 case 5:
1094 case 6:
1095 re->level = 6;
1096 re->before.data_disks = info->array.raid_disks - 2;
1097 re->before.layout = info->array.layout;
1098 break;
1099 default:
1100 return "Impossible level change requested";
1101 }
1102 break;
1103 }
1104
1105 /* If we reached here then it looks like a re-stripe is
1106 * happening. We have determined the intermediate level
1107 * and initial raid_disks/layout and stored these in 're'.
1108 *
1109 * We need to deduce the final layout that can be atomically
1110 * converted to the end state.
1111 */
1112 switch (info->new_level) {
1113 case 0:
1114 /* We can only get to RAID0 from RAID4 or RAID5
1115 * with appropriate layout and one extra device
1116 */
1117 if (re->level != 4 && re->level != 5)
1118 return "Cannot covert to RAID0 from this level";
1119 if (info->delta_disks == UnSet)
1120 re->after.data_disks = re->before.data_disks;
1121 else
1122 re->after.data_disks =
1123 info->array.raid_disks + info->delta_disks;
1124 switch (re->level) {
1125 case 4:
1126 re->after.layout = 0 ; break;
1127 case 5:
1128 re->after.layout = ALGORITHM_PARITY_N; break;
1129 }
1130 break;
1131
1132 case 4:
1133 /* We can only get to RAID4 from RAID5 */
1134 if (re->level != 4 && re->level != 5)
1135 return "Cannot convert to RAID4 from this level";
1136 if (info->delta_disks == UnSet)
1137 re->after.data_disks = re->before.data_disks;
1138 else
1139 re->after.data_disks =
1140 re->before.data_disks + info->delta_disks;
1141 switch (re->level) {
1142 case 4:
1143 re->after.layout = 0 ; break;
1144 case 5:
1145 re->after.layout = ALGORITHM_PARITY_N; break;
1146 }
1147 break;
1148
1149 case 5:
1150 /* We get to RAID5 for RAID5 or RAID6 */
1151 if (re->level != 5 && re->level != 6)
1152 return "Cannot convert to RAID5 from this level";
1153 if (info->delta_disks == UnSet)
1154 re->after.data_disks = re->before.data_disks;
1155 else if (re->level == 5)
1156 re->after.data_disks =
1157 re->before.data_disks + info->delta_disks;
1158 else
1159 re->after.data_disks =
1160 info->array.raid_disks + info->delta_disks - 1;
1161 switch (re->level) {
1162 case 5:
1163 if (info->new_layout == UnSet)
1164 re->after.layout = re->before.layout;
1165 else
1166 re->after.layout = info->new_layout;
1167 break;
1168 case 6:
1169 if (info->new_layout == UnSet)
1170 info->new_layout = re->before.layout;
1171
1172 /* after.layout needs to be raid6 version of new_layout */
1173 if (info->new_layout == ALGORITHM_PARITY_N)
1174 re->after.layout = ALGORITHM_PARITY_N;
1175 else {
1176 char layout[40];
1177 char *ls = map_num(r5layout, info->new_layout);
1178 int l;
1179 strcat(strcpy(layout, ls), "-6");
1180 l = map_name(r6layout, layout);
1181 if (l == UnSet)
1182 return "Cannot find RAID6 layout"
1183 " to convert to";
1184 re->after.layout = l;
1185 }
1186 }
1187 break;
1188
1189 case 6:
1190 /* We must already be at level 6 */
1191 if (re->level != 6)
1192 return "Impossible level change";
1193 if (info->delta_disks == UnSet)
1194 re->after.data_disks = re->before.data_disks;
1195 else
1196 re->after.data_disks = (info->array.raid_disks +
1197 info->delta_disks) - 2;
1198 if (info->new_layout == UnSet)
1199 re->after.layout = info->array.layout;
1200 else
1201 re->after.layout = info->new_layout;
1202 break;
1203 default:
1204 return "Impossible level change requested";
1205 }
1206 switch (re->level) {
1207 case 6: re->parity = 2; break;
1208 case 4:
1209 case 5: re->parity = 1; break;
1210 default: re->parity = 0; break;
1211 }
1212 /* So we have a restripe operation, we need to calculate the number
1213 * of blocks per reshape operation.
1214 */
1215 if (info->new_chunk == 0)
1216 info->new_chunk = info->array.chunk_size;
1217 if (re->after.data_disks == re->before.data_disks &&
1218 re->after.layout == re->before.layout &&
1219 info->new_chunk == info->array.chunk_size) {
1220 /* Nothing to change */
1221 re->backup_blocks = 0;
1222 return NULL;
1223 }
1224 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1225 /* chunks can layout changes make no difference */
1226 re->backup_blocks = 0;
1227 return NULL;
1228 }
1229
1230 if (re->after.data_disks == re->before.data_disks &&
1231 get_linux_version() < 2006032)
1232 return "in-place reshape is not safe before 2.6.32 - sorry.";
1233
1234 if (re->after.data_disks < re->before.data_disks &&
1235 get_linux_version() < 2006030)
1236 return "reshape to fewer devices is not supported before 2.6.32 - sorry.";
1237
1238 re->backup_blocks = compute_backup_blocks(
1239 info->new_chunk, info->array.chunk_size,
1240 re->after.data_disks,
1241 re->before.data_disks);
1242
1243 re->new_size = info->component_size * re->after.data_disks;
1244 return NULL;
1245 }
1246
1247 static int reshape_array(char *container, int fd, char *devname,
1248 struct supertype *st, struct mdinfo *info,
1249 int force, char *backup_file, int quiet, int forked);
1250 static int reshape_container(char *container, int cfd, char *devname,
1251 struct supertype *st,
1252 struct mdinfo *info,
1253 int force,
1254 char *backup_file,
1255 int quiet);
1256
1257 int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
1258 long long size,
1259 int level, char *layout_str, int chunksize, int raid_disks,
1260 int force)
1261 {
1262 /* Make some changes in the shape of an array.
1263 * The kernel must support the change.
1264 *
1265 * There are three different changes. Each can trigger
1266 * a resync or recovery so we freeze that until we have
1267 * requested everything (if kernel supports freezing - 2.6.30).
1268 * The steps are:
1269 * - change size (i.e. component_size)
1270 * - change level
1271 * - change layout/chunksize/ndisks
1272 *
1273 * The last can require a reshape. It is different on different
1274 * levels so we need to check the level before actioning it.
1275 * Some times the level change needs to be requested after the
1276 * reshape (e.g. raid6->raid5, raid5->raid0)
1277 *
1278 */
1279 struct mdu_array_info_s array;
1280 int rv = 0;
1281 struct supertype *st;
1282 char *subarray = NULL;
1283
1284 int frozen;
1285 int changed = 0;
1286 char *container = NULL;
1287 char container_buf[20];
1288 int cfd = -1;
1289
1290 struct mdinfo info;
1291 struct mdinfo *sra;
1292
1293 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1294 fprintf(stderr, Name ": %s is not an active md array - aborting\n",
1295 devname);
1296 return 1;
1297 }
1298
1299 if (size >= 0 &&
1300 (chunksize || level!= UnSet || layout_str || raid_disks)) {
1301 fprintf(stderr, Name ": cannot change component size at the same time "
1302 "as other changes.\n"
1303 " Change size first, then check data is intact before "
1304 "making other changes.\n");
1305 return 1;
1306 }
1307
1308 if (raid_disks && raid_disks < array.raid_disks && array.level > 1 &&
1309 get_linux_version() < 2006032 &&
1310 !check_env("MDADM_FORCE_FEWER")) {
1311 fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n"
1312 " Please use a newer kernel\n");
1313 return 1;
1314 }
1315
1316 st = super_by_fd(fd, &subarray);
1317 if (!st) {
1318 fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname);
1319 return 1;
1320 }
1321 if (raid_disks > st->max_devs) {
1322 fprintf(stderr, Name ": Cannot increase raid-disks on this array"
1323 " beyond %d\n", st->max_devs);
1324 return 1;
1325 }
1326
1327 /* in the external case we need to check that the requested reshape is
1328 * supported, and perform an initial check that the container holds the
1329 * pre-requisite spare devices (mdmon owns final validation)
1330 */
1331 if (st->ss->external) {
1332 int container_dev;
1333 int rv;
1334
1335 if (subarray) {
1336 container_dev = st->container_dev;
1337 cfd = open_dev_excl(st->container_dev);
1338 } else {
1339 container_dev = st->devnum;
1340 close(fd);
1341 cfd = open_dev_excl(st->devnum);
1342 fd = cfd;
1343 }
1344 if (cfd < 0) {
1345 fprintf(stderr, Name ": Unable to open container for %s\n",
1346 devname);
1347 free(subarray);
1348 return 1;
1349 }
1350
1351 fmt_devname(container_buf, container_dev);
1352 container = container_buf;
1353
1354 if (subarray)
1355 rv = st->ss->load_container(st, cfd, NULL);
1356 else
1357 rv = st->ss->load_super(st, cfd, NULL);
1358 if (rv) {
1359 fprintf(stderr, Name ": Cannot read superblock for %s\n",
1360 devname);
1361 free(subarray);
1362 return 1;
1363 }
1364
1365 if (mdmon_running(container_dev))
1366 st->update_tail = &st->updates;
1367 }
1368
1369 if (raid_disks > array.raid_disks &&
1370 array.spare_disks < (raid_disks - array.raid_disks) &&
1371 !force) {
1372 fprintf(stderr,
1373 Name ": Need %d spare%s to avoid degraded array,"
1374 " and only have %d.\n"
1375 " Use --force to over-ride this check.\n",
1376 raid_disks - array.raid_disks,
1377 raid_disks - array.raid_disks == 1 ? "" : "s",
1378 array.spare_disks);
1379 return 1;
1380 }
1381
1382 sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS | GET_STATE);
1383 if (sra) {
1384 if (st->ss->external && subarray == NULL) {
1385 array.level = LEVEL_CONTAINER;
1386 sra->array.level = LEVEL_CONTAINER;
1387 }
1388 } else {
1389 fprintf(stderr, Name ": failed to read sysfs parameters for %s\n",
1390 devname);
1391 return 1;
1392 }
1393 frozen = freeze(st);
1394 if (frozen < -1) {
1395 /* freeze() already spewed the reason */
1396 return 1;
1397 } else if (frozen < 0) {
1398 fprintf(stderr, Name ": %s is performing resync/recovery and cannot"
1399 " be reshaped\n", devname);
1400 return 1;
1401 }
1402
1403 /* ========= set size =============== */
1404 if (size >= 0 && (size == 0 || size != array.size)) {
1405 long long orig_size = array.size;
1406
1407 if (reshape_super(st, size, UnSet, UnSet, 0, 0, NULL, devname, !quiet)) {
1408 rv = 1;
1409 goto release;
1410 }
1411 sync_metadata(st);
1412 array.size = size;
1413 if (array.size != size) {
1414 /* got truncated to 32bit, write to
1415 * component_size instead
1416 */
1417 if (sra)
1418 rv = sysfs_set_num(sra, NULL,
1419 "component_size", size);
1420 else
1421 rv = -1;
1422 } else
1423 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1424 if (rv != 0) {
1425 int err = errno;
1426
1427 /* restore metadata */
1428 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1429 NULL, devname, !quiet) == 0)
1430 sync_metadata(st);
1431 fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
1432 devname, strerror(err));
1433 if (err == EBUSY &&
1434 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1435 fprintf(stderr, " Bitmap must be removed before size can be changed\n");
1436 rv = 1;
1437 goto release;
1438 }
1439 ioctl(fd, GET_ARRAY_INFO, &array);
1440 size = get_component_size(fd)/2;
1441 if (size == 0)
1442 size = array.size;
1443 if (!quiet)
1444 fprintf(stderr, Name ": component size of %s has been set to %lluK\n",
1445 devname, size);
1446 changed = 1;
1447 } else if (array.level != LEVEL_CONTAINER) {
1448 size = get_component_size(fd)/2;
1449 if (size == 0)
1450 size = array.size;
1451 }
1452
1453 /* ========= check for Raid10 -> Raid0 conversion ===============
1454 * current implementation assumes that following conditions must be met:
1455 * - far_copies == 1
1456 * - near_copies == 2
1457 */
1458 if (level == 0 && array.level == 10 && sra &&
1459 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) {
1460 int err;
1461 err = remove_disks_on_raid10_to_raid0_takeover(st, sra, array.layout);
1462 if (err) {
1463 dprintf(Name": Array cannot be reshaped\n");
1464 if (cfd > -1)
1465 close(cfd);
1466 rv = 1;
1467 goto release;
1468 }
1469 }
1470
1471 info.array = array;
1472 sysfs_init(&info, fd, NoMdDev);
1473 info.component_size = size*2;
1474 info.new_level = level;
1475 info.new_chunk = chunksize * 1024;
1476 if (raid_disks)
1477 info.delta_disks = raid_disks - info.array.raid_disks;
1478 else
1479 info.delta_disks = UnSet;
1480 if (layout_str == NULL) {
1481 info.new_layout = UnSet;
1482 if (info.array.level == 6 &&
1483 (info.new_level == 6 || info.new_level == UnSet) &&
1484 info.array.layout >= 16) {
1485 fprintf(stderr, Name
1486 ": %s has a non-standard layout. If you"
1487 " wish to preserve this\n"
1488 " during the reshape, please specify"
1489 " --layout=preserve\n"
1490 " If you want to change it, specify a"
1491 " layout or use --layout=normalise\n",
1492 devname);
1493 rv = 1;
1494 goto release;
1495 }
1496 } else if (strcmp(layout_str, "normalise") == 0 ||
1497 strcmp(layout_str, "normalize") == 0) {
1498 /* If we have a -6 RAID6 layout, remove the '-6'. */
1499 info.new_layout = UnSet;
1500 if (info.array.level == 6 && info.new_level == UnSet) {
1501 char l[40], *h;
1502 strcpy(l, map_num(r6layout, info.array.layout));
1503 h = strrchr(l, '-');
1504 if (h && strcmp(h, "-6") == 0) {
1505 *h = 0;
1506 info.new_layout = map_name(r6layout, l);
1507 }
1508 }
1509 } else if (strcmp(layout_str, "preserve") == 0) {
1510 info.new_layout = UnSet;
1511 } else {
1512 int l = info.new_level;
1513 if (l == UnSet)
1514 l = info.array.level;
1515 switch (l) {
1516 case 5:
1517 info.new_layout = map_name(r5layout, layout_str);
1518 break;
1519 case 6:
1520 info.new_layout = map_name(r6layout, layout_str);
1521 break;
1522 case 10:
1523 info.new_layout = parse_layout_10(layout_str);
1524 break;
1525 case LEVEL_FAULTY:
1526 info.new_layout = parse_layout_faulty(layout_str);
1527 break;
1528 default:
1529 fprintf(stderr, Name ": layout not meaningful"
1530 " with this level\n");
1531 rv = 1;
1532 goto release;
1533 }
1534 if (info.new_layout == UnSet) {
1535 fprintf(stderr, Name ": layout %s not understood"
1536 " for this level\n",
1537 layout_str);
1538 rv = 1;
1539 goto release;
1540 }
1541 }
1542
1543 if (array.level == LEVEL_CONTAINER) {
1544 /* This change is to be applied to every array in the
1545 * container. This is only needed when the metadata imposes
1546 * restraints of the various arrays in the container.
1547 * Currently we only know that IMSM requires all arrays
1548 * to have the same number of devices so changing the
1549 * number of devices (On-Line Capacity Expansion) must be
1550 * performed at the level of the container
1551 */
1552 rv = reshape_container(container, fd, devname, st, &info,
1553 force, backup_file, quiet);
1554 } else {
1555 /* Impose these changes on a single array. First
1556 * check that the metadata is OK with the change. */
1557
1558 if (reshape_super(st, info.component_size, info.new_level,
1559 info.new_layout, info.new_chunk,
1560 info.array.raid_disks + info.delta_disks,
1561 backup_file, devname, quiet)) {
1562 rv = 1;
1563 goto release;
1564 }
1565 sync_metadata(st);
1566 rv = reshape_array(container, fd, devname, st, &info, force,
1567 backup_file, quiet, 0);
1568 }
1569 /* reshape_* released the array */
1570 return rv;
1571 release:
1572 unfreeze(st, frozen);
1573 return rv;
1574 }
1575
1576 static int reshape_array(char *container, int fd, char *devname,
1577 struct supertype *st, struct mdinfo *info,
1578 int force,
1579 char *backup_file, int quiet, int forked)
1580 {
1581 struct reshape reshape;
1582 int spares_needed;
1583 char *msg;
1584 int orig_level = UnSet;
1585 int disks, odisks;
1586
1587 struct mdu_array_info_s array;
1588 char *c;
1589 int rv = 0;
1590
1591 int *fdlist;
1592 unsigned long long *offsets;
1593 int d;
1594 int nrdisks;
1595 int err;
1596 int frozen;
1597 unsigned long blocks;
1598 unsigned long cache;
1599 unsigned long long array_size;
1600 int done;
1601 struct mdinfo *sra;
1602
1603 msg = analyse_change(info, &reshape);
1604 if (msg) {
1605 fprintf(stderr, Name ": %s\n", msg);
1606 return 1;
1607 }
1608 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
1609 dprintf("Canot get array information.\n");
1610 return 1;
1611 }
1612 spares_needed = max(reshape.before.data_disks,
1613 reshape.after.data_disks)
1614 + reshape.parity - array.raid_disks;
1615
1616 if (!force &&
1617 info->new_level > 0 &&
1618 spares_needed > info->array.spare_disks) {
1619 fprintf(stderr,
1620 Name ": Need %d spare%s to avoid degraded array,"
1621 " and only have %d.\n"
1622 " Use --force to over-ride this check.\n",
1623 spares_needed,
1624 spares_needed == 1 ? "" : "s",
1625 info->array.spare_disks);
1626 return 1;
1627 }
1628
1629 if (reshape.level != info->array.level) {
1630 char *c = map_num(pers, reshape.level);
1631 int err;
1632 if (c == NULL)
1633 return 1; /* This should not be possible */
1634
1635 err = sysfs_set_str(info, NULL, "level", c);
1636 if (err) {
1637 err = errno;
1638 fprintf(stderr, Name ": %s: could not set level to %s\n",
1639 devname, c);
1640 if (err == EBUSY &&
1641 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
1642 fprintf(stderr, " Bitmap must be removed"
1643 " before level can be changed\n");
1644 return 1;
1645 }
1646 if (!quiet)
1647 fprintf(stderr, Name ": level of %s changed to %s\n",
1648 devname, c);
1649 orig_level = info->array.level;
1650 }
1651
1652 if (reshape.level > 0 && st->ss->external &&
1653 !mdmon_running(st->container_dev)) {
1654 start_mdmon(st->container_dev);
1655 ping_monitor(container);
1656 }
1657
1658 /* ->reshape_super might have chosen some spares from the
1659 * container that it wants to be part of the new array.
1660 * We can collect them with ->container_content and give
1661 * them to the kernel.
1662 */
1663 if (st->ss->reshape_super && st->ss->container_content) {
1664 char *subarray = strchr(info->text_version+1, '/')+1;
1665 struct mdinfo *info2 =
1666 st->ss->container_content(st, subarray);
1667 struct mdinfo *d;
1668
1669 if (info2)
1670 for (d = info2->devs; d; d = d->next) {
1671 if (d->disk.state == 0 &&
1672 d->disk.raid_disk >= 0) {
1673 /* This is a spare that wants to
1674 * be part of the array.
1675 */
1676 add_disk(fd, st, info2, d);
1677 }
1678 }
1679 sysfs_free(info2);
1680 }
1681
1682 if (reshape.backup_blocks == 0) {
1683 /* No restriping needed, but we might need to impose
1684 * some more changes: layout, raid_disks, chunk_size
1685 */
1686 if (info->new_layout != UnSet &&
1687 info->new_layout != info->array.layout) {
1688 info->array.layout = info->new_layout;
1689 if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) {
1690 fprintf(stderr, Name ": failed to set new layout\n");
1691 rv = 1;
1692 } else if (!quiet)
1693 printf("layout for %s set to %d\n",
1694 devname, info->array.layout);
1695 }
1696 if (info->delta_disks != UnSet &&
1697 info->delta_disks != 0) {
1698 info->array.raid_disks += info->delta_disks;
1699 if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) {
1700 fprintf(stderr, Name ": failed to set raid disks\n");
1701 rv = 1;
1702 } else if (!quiet)
1703 printf("raid_disks for %s set to %d\n",
1704 devname, info->array.raid_disks);
1705 }
1706 if (info->new_chunk != 0 &&
1707 info->new_chunk != info->array.chunk_size) {
1708 if (sysfs_set_num(info, NULL,
1709 "chunk_size", info->new_chunk) != 0) {
1710 fprintf(stderr, Name ": failed to set chunk size\n");
1711 rv = 1;
1712 } else if (!quiet)
1713 printf("chunk size for %s set to %d\n",
1714 devname, info->array.chunk_size);
1715 }
1716
1717 return rv;
1718 }
1719
1720 /*
1721 * There are three possibilities.
1722 * 1/ The array will shrink.
1723 * We need to ensure the reshape will pause before reaching
1724 * the 'critical section'. We also need to fork and wait for
1725 * that to happen. When it does we
1726 * suspend/backup/complete/unfreeze
1727 *
1728 * 2/ The array will not change size.
1729 * This requires that we keep a backup of a sliding window
1730 * so that we can restore data after a crash. So we need
1731 * to fork and monitor progress.
1732 * In future we will allow the data_offset to change, so
1733 * a sliding backup becomes unnecessary.
1734 *
1735 * 3/ The array will grow. This is relatively easy.
1736 * However the kernel's restripe routines will cheerfully
1737 * overwrite some early data before it is safe. So we
1738 * need to make a backup of the early parts of the array
1739 * and be ready to restore it if rebuild aborts very early.
1740 * For externally managed metadata, we still need a forked
1741 * child to monitor the reshape and suspend IO over the region
1742 * that is being reshaped.
1743 *
1744 * We backup data by writing it to one spare, or to a
1745 * file which was given on command line.
1746 *
1747 * In each case, we first make sure that storage is available
1748 * for the required backup.
1749 * Then we:
1750 * - request the shape change.
1751 * - fork to handle backup etc.
1752 */
1753
1754 /* Check that we can hold all the data */
1755 get_dev_size(fd, NULL, &array_size);
1756 if (reshape.new_size < (array_size/512)) {
1757 fprintf(stderr,
1758 Name ": this change will reduce the size of the array.\n"
1759 " use --grow --array-size first to truncate array.\n"
1760 " e.g. mdadm --grow %s --array-size %llu\n",
1761 devname, reshape.new_size/2);
1762 rv = 1;
1763 goto release;
1764 }
1765
1766 sra = sysfs_read(fd, 0,
1767 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
1768 GET_CACHE);
1769
1770 if (!sra) {
1771 fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
1772 devname);
1773 rv = 1;
1774 goto release;
1775 }
1776
1777 /* Decide how many blocks (sectors) for a reshape
1778 * unit. The number we have so far is just a minimum
1779 */
1780 blocks = reshape.backup_blocks;
1781 if (reshape.before.data_disks ==
1782 reshape.after.data_disks) {
1783 /* Make 'blocks' bigger for better throughput, but
1784 * not so big that we reject it below.
1785 * Try for 16 megabytes
1786 */
1787 while (blocks * 32 < sra->component_size &&
1788 blocks < 16*1024*2)
1789 blocks *= 2;
1790 } else
1791 fprintf(stderr, Name ": Need to backup %luK of critical "
1792 "section..\n", blocks/2);
1793
1794 if (blocks >= sra->component_size/2) {
1795 fprintf(stderr, Name ": %s: Something wrong"
1796 " - reshape aborted\n",
1797 devname);
1798 rv = 1;
1799 goto release;
1800 }
1801
1802 /* Now we need to open all these devices so we can read/write.
1803 */
1804 nrdisks = array.raid_disks + sra->array.spare_disks;
1805 fdlist = malloc((1+nrdisks) * sizeof(int));
1806 offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
1807 if (!fdlist || !offsets) {
1808 fprintf(stderr, Name ": malloc failed: grow aborted\n");
1809 rv = 1;
1810 goto release;
1811 }
1812
1813 d = reshape_prepare_fdlist(devname, sra, array.raid_disks,
1814 nrdisks, blocks, backup_file,
1815 fdlist, offsets);
1816 if (d < 0) {
1817 rv = 1;
1818 goto release;
1819 }
1820 if (backup_file == NULL) {
1821 if (reshape.after.data_disks <= reshape.before.data_disks) {
1822 fprintf(stderr,
1823 Name ": %s: Cannot grow - need backup-file\n",
1824 devname);
1825 rv = 1;
1826 goto release;
1827 } else if (sra->array.spare_disks == 0) {
1828 fprintf(stderr, Name ": %s: Cannot grow - need a spare or "
1829 "backup-file to backup critical section\n",
1830 devname);
1831 rv = 1;
1832 goto release;
1833 }
1834 } else {
1835 if (!reshape_open_backup_file(backup_file, fd, devname,
1836 (signed)blocks,
1837 fdlist+d, offsets+d)) {
1838 rv = 1;
1839 goto release;
1840 }
1841 d++;
1842 }
1843
1844 /* lastly, check that the internal stripe cache is
1845 * large enough, or it won't work.
1846 * It must hold at least 4 stripes of the larger
1847 * chunk size
1848 */
1849 cache = max(info->array.chunk_size, info->new_chunk);
1850 cache *= 4; /* 4 stripes minimum */
1851 cache /= 512; /* convert to sectors */
1852 disks = min(reshape.before.data_disks, reshape.after.data_disks);
1853 /* make sure there is room for 'blocks' with a bit to spare */
1854 if (cache < 16 + blocks / disks)
1855 cache = 16 + blocks / disks;
1856 cache /= (4096/512); /* Covert from sectors to pages */
1857
1858 if (sra->cache_size < cache)
1859 subarray_set_num(container, sra, "stripe_cache_size",
1860 cache+1);
1861
1862 /* Right, everything seems fine. Let's kick things off.
1863 * If only changing raid_disks, use ioctl, else use
1864 * sysfs.
1865 */
1866 sync_metadata(st);
1867
1868 sra->new_chunk = info->new_chunk;
1869
1870 if (info->array.chunk_size == info->new_chunk &&
1871 reshape.before.layout == reshape.after.layout &&
1872 st->ss->external == 0) {
1873 array.raid_disks = reshape.after.data_disks + reshape.parity;
1874 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1875 int err = errno;
1876 rv = 1;
1877 fprintf(stderr,
1878 Name ": Cannot set device shape for %s: %s\n",
1879 devname, strerror(errno));
1880
1881 if (err == EBUSY &&
1882 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1883 fprintf(stderr,
1884 " Bitmap must be removed before"
1885 " shape can be changed\n");
1886
1887 goto release;
1888 }
1889 } else {
1890 /* set them all just in case some old 'new_*' value
1891 * persists from some earlier problem
1892 */
1893 int err = err; /* only used if rv==1, and always set if
1894 * rv==1, so initialisation not needed,
1895 * despite gcc warning
1896 */
1897 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
1898 rv = 1, err = errno;
1899 if (!rv && sysfs_set_num(sra, NULL, "layout",
1900 reshape.after.layout) < 0)
1901 rv = 1, err = errno;
1902 if (!rv && subarray_set_num(container, sra, "raid_disks",
1903 reshape.after.data_disks +
1904 reshape.parity) < 0)
1905 rv = 1, err = errno;
1906 if (rv) {
1907 fprintf(stderr, Name ": Cannot set device shape for %s\n",
1908 devname);
1909
1910 if (err == EBUSY &&
1911 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1912 fprintf(stderr,
1913 " Bitmap must be removed before"
1914 " shape can be changed\n");
1915 goto release;
1916 }
1917 }
1918
1919 start_reshape(sra);
1920
1921 /* Now we just need to kick off the reshape and watch, while
1922 * handling backups of the data...
1923 * This is all done by a forked background process.
1924 */
1925 switch(forked ? 0 : fork()) {
1926 case 0:
1927 close(fd);
1928 if (check_env("MDADM_GROW_VERIFY"))
1929 fd = open(devname, O_RDONLY | O_DIRECT);
1930 else
1931 fd = -1;
1932 mlockall(MCL_FUTURE);
1933
1934 odisks = reshape.before.data_disks + reshape.parity;
1935
1936 if (st->ss->external) {
1937 /* metadata handler takes it from here */
1938 done = st->ss->manage_reshape(
1939 fd, sra, &reshape, st, blocks,
1940 fdlist, offsets,
1941 d - odisks, fdlist+odisks,
1942 offsets+odisks);
1943 } else
1944 done = child_monitor(
1945 fd, sra, &reshape, st, blocks,
1946 fdlist, offsets,
1947 d - odisks, fdlist+odisks,
1948 offsets+odisks);
1949
1950 if (backup_file && done)
1951 unlink(backup_file);
1952 if (!done) {
1953 abort_reshape(sra);
1954 goto out;
1955 }
1956 /* set new array size if required customer_array_size is used
1957 * by this metadata.
1958 */
1959 if (reshape.before.data_disks !=
1960 reshape.after.data_disks &&
1961 info->custom_array_size) {
1962 struct mdinfo *info2;
1963 char *subarray = strchr(info->text_version+1, '/')+1;
1964
1965 wait_reshape(sra);
1966 ping_monitor(container);
1967
1968 info2 = st->ss->container_content(st, subarray);
1969 if (info2) {
1970 unsigned long long current_size = 0;
1971 unsigned long long new_size =
1972 info2->custom_array_size/2;
1973
1974 if (sysfs_get_ll(sra,
1975 NULL,
1976 "array_size",
1977 &current_size) == 0 &&
1978 new_size > current_size) {
1979 if (sysfs_set_num(sra, NULL,
1980 "array_size", new_size)
1981 < 0)
1982 dprintf("Error: Cannot"
1983 " set array size");
1984 else
1985 dprintf("Array size "
1986 "changed");
1987 dprintf(" from %llu to %llu.\n",
1988 current_size, new_size);
1989 }
1990 sysfs_free(info2);
1991 }
1992 }
1993
1994 if (info->new_level != reshape.level) {
1995 /* We need to wait for the reshape to finish
1996 * (which will have happened unless
1997 * odata < ndata) and then set the level
1998 */
1999
2000 if (reshape.before.data_disks <
2001 reshape.after.data_disks)
2002 wait_reshape(sra);
2003
2004 c = map_num(pers, info->new_level);
2005 if (c == NULL)
2006 goto out;/* not possible */
2007
2008 err = sysfs_set_str(sra, NULL, "level", c);
2009 if (err)
2010 fprintf(stderr, Name\
2011 ": %s: could not set level "
2012 "to %s\n", devname, c);
2013 }
2014 out:
2015 if (forked)
2016 return 0;
2017 exit(0);
2018 case -1:
2019 fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n",
2020 strerror(errno));
2021 rv = 1;
2022 abort_reshape(sra);
2023 break;
2024 default:
2025 /* The child will take care of unfreezing the array */
2026 frozen = 0;
2027 break;
2028 }
2029
2030
2031 release:
2032 if (!rv) {
2033 if (container)
2034 ping_monitor(container);
2035 if (st->ss->external) {
2036 /* Re-load the metadata as much could have changed */
2037 int cfd = open_dev(st->container_dev);
2038 if (cfd >= 0) {
2039 st->ss->free_super(st);
2040 st->ss->load_container(st, cfd, container);
2041 close(cfd);
2042 }
2043 }
2044 }
2045 if (rv && orig_level != UnSet && sra) {
2046 c = map_num(pers, orig_level);
2047 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
2048 fprintf(stderr, Name ": aborting level change\n");
2049 }
2050 unfreeze(st, frozen);
2051 return rv;
2052 }
2053
2054 int reshape_container(char *container, int cfd, char *devname,
2055 struct supertype *st,
2056 struct mdinfo *info,
2057 int force,
2058 char *backup_file,
2059 int quiet)
2060 {
2061 struct mdinfo *cc = NULL;
2062
2063 if (reshape_super(st, info->component_size, info->new_level,
2064 info->new_layout, info->new_chunk,
2065 info->array.raid_disks + info->delta_disks,
2066 backup_file, devname, quiet))
2067 return 1;
2068
2069 sync_metadata(st);
2070
2071 /* ping monitor to be sure that update is on disk
2072 */
2073 ping_monitor(container);
2074
2075 switch (fork()) {
2076 case -1: /* error */
2077 perror("Cannot fork to complete reshape\n");
2078 return 1;
2079 default: /* parent */
2080 printf(Name ": multi-array reshape continues in background\n");
2081 return 0;
2082 case 0: /* child */
2083 break;
2084 }
2085
2086 while(1) {
2087 /* For each member array with reshape_active,
2088 * we need to perform the reshape.
2089 * We pick the first array that needs reshaping and
2090 * reshape it. reshape_array() will re-read the metadata
2091 * so the next time through a different array should be
2092 * ready for reshape.
2093 */
2094 struct mdinfo *content;
2095 int rv;
2096 int fd;
2097 struct mdstat_ent *mdstat;
2098 char *adev;
2099
2100 sysfs_free(cc);
2101
2102 cc = st->ss->container_content(st, NULL);
2103
2104 for (content = cc; content ; content = content->next) {
2105 char *subarray;
2106 if (!content->reshape_active)
2107 continue;
2108
2109 subarray = strchr(content->text_version+1, '/')+1;
2110 mdstat = mdstat_by_subdev(subarray,
2111 devname2devnum(container));
2112 if (!mdstat)
2113 continue;
2114 break;
2115 }
2116 if (!content)
2117 break;
2118
2119 fd = open_dev_excl(mdstat->devnum);
2120 if (fd < 0)
2121 break;
2122 adev = map_dev(dev2major(mdstat->devnum),
2123 dev2minor(mdstat->devnum),
2124 0);
2125 if (!adev)
2126 adev = content->text_version;
2127
2128 sysfs_init(content, fd, mdstat->devnum);
2129
2130 rv = reshape_array(container, fd, adev, st,
2131 content, force,
2132 backup_file, quiet, 1);
2133 close(fd);
2134 if (rv)
2135 break;
2136 }
2137 sysfs_free(cc);
2138 exit(0);
2139 }
2140
2141 /*
2142 * We run a child process in the background which performs the following
2143 * steps:
2144 * - wait for resync to reach a certain point
2145 * - suspend io to the following section
2146 * - backup that section
2147 * - allow resync to proceed further
2148 * - resume io
2149 * - discard the backup.
2150 *
2151 * When are combined in slightly different ways in the three cases.
2152 * Grow:
2153 * - suspend/backup/allow/wait/resume/discard
2154 * Shrink:
2155 * - allow/wait/suspend/backup/allow/wait/resume/discard
2156 * same-size:
2157 * - wait/resume/discard/suspend/backup/allow
2158 *
2159 * suspend/backup/allow always come together
2160 * wait/resume/discard do too.
2161 * For the same-size case we have two backups to improve flow.
2162 *
2163 */
2164
2165 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
2166 unsigned long long backup_point,
2167 unsigned long long wait_point,
2168 unsigned long long *suspend_point,
2169 unsigned long long *reshape_completed)
2170 {
2171 /* This function is called repeatedly by the reshape manager.
2172 * It determines how much progress can safely be made and allows
2173 * that progress.
2174 * - 'info' identifies the array and particularly records in
2175 * ->reshape_progress the metadata's knowledge of progress
2176 * This is a sector offset from the start of the array
2177 * of the next array block to be relocated. This number
2178 * may increase from 0 or decrease from array_size, depending
2179 * on the type of reshape that is happening.
2180 * Note that in contrast, 'sync_completed' is a block count of the
2181 * reshape so far. It gives the distance between the start point
2182 * (head or tail of device) and the next place that data will be
2183 * written. It always increases.
2184 * - 'reshape' is the structure created by analyse_change
2185 * - 'backup_point' shows how much the metadata manager has backed-up
2186 * data. For reshapes with increasing progress, it is the next address
2187 * to be backed up, previous addresses have been backed-up. For
2188 * decreasing progress, it is the earliest address that has been
2189 * backed up - later address are also backed up.
2190 * So addresses between reshape_progress and backup_point are
2191 * backed up providing those are in the 'correct' order.
2192 * - 'wait_point' is an array address. When reshape_completed
2193 * passes this point, progress_reshape should return. It might
2194 * return earlier if it determines that ->reshape_progress needs
2195 * to be updated or further backup is needed.
2196 * - suspend_point is maintained by progress_reshape and the caller
2197 * should not touch it except to initialise to zero.
2198 * It is an array address and it only increases in 2.6.37 and earlier.
2199 * This makes it difficult to handle reducing reshapes with
2200 * external metadata.
2201 * However: it is similar to backup_point in that it records the
2202 * other end of a suspended region from reshape_progress.
2203 * it is moved to extend the region that is safe to backup and/or
2204 * reshape
2205 * - reshape_completed is read from sysfs and returned. The caller
2206 * should copy this into ->reshape_progress when it has reason to
2207 * believe that the metadata knows this, and any backup outside this
2208 * has been erased.
2209 *
2210 * Return value is:
2211 * 1 if more data from backup_point - but only as far as suspend_point,
2212 * should be backed up
2213 * 0 if things are progressing smoothly
2214 * -1 if the reshape is finished, either because it is all done,
2215 * or due to an error.
2216 */
2217
2218 int advancing = (reshape->after.data_disks
2219 >= reshape->before.data_disks);
2220 unsigned long long need_backup; /* need to eventually backup all the way
2221 * to here
2222 */
2223 unsigned long long read_offset, write_offset;
2224 unsigned long long write_range;
2225 unsigned long long max_progress, target, completed;
2226 unsigned long long array_size = (info->component_size
2227 * reshape->before.data_disks);
2228 int fd;
2229
2230 /* First, we unsuspend any region that is now known to be safe.
2231 * If suspend_point is on the 'wrong' side of reshape_progress, then
2232 * we don't have or need suspension at the moment. This is true for
2233 * native metadata when we don't need to back-up.
2234 */
2235 if (advancing) {
2236 if (info->reshape_progress <= *suspend_point)
2237 sysfs_set_num(info, NULL, "suspend_lo",
2238 info->reshape_progress);
2239 } else {
2240 /* Note: this won't work in 2.6.37 and before.
2241 * Something somewhere should make sure we don't need it!
2242 */
2243 if (info->reshape_progress >= *suspend_point)
2244 sysfs_set_num(info, NULL, "suspend_hi",
2245 info->reshape_progress);
2246 }
2247
2248 /* Now work out how far it is safe to progress.
2249 * If the read_offset for ->reshape_progress is less than
2250 * 'blocks' beyond the write_offset, we can only progress as far
2251 * as a backup.
2252 * Otherwise we can progress until the write_offset for the new location
2253 * reaches (within 'blocks' of) the read_offset at the current location.
2254 * However that region must be suspended unless we are using native
2255 * metadata.
2256 * If we need to suspend more, we limit it to 128M per device, which is
2257 * rather arbitrary and should be some time-based calculation.
2258 */
2259 read_offset = info->reshape_progress / reshape->before.data_disks;
2260 write_offset = info->reshape_progress / reshape->after.data_disks;
2261 write_range = info->new_chunk/512;
2262 if (advancing) {
2263 need_backup = 0;
2264 if (read_offset < write_offset + write_range) {
2265 max_progress = backup_point;
2266 if (reshape->before.data_disks == reshape->after.data_disks)
2267 need_backup = array_size;
2268 else
2269 need_backup = reshape->backup_blocks;
2270 } else {
2271 max_progress =
2272 read_offset *
2273 reshape->after.data_disks;
2274 }
2275 } else {
2276 need_backup = array_size;
2277 if (read_offset > write_offset - write_range) {
2278 max_progress = backup_point;
2279 if (max_progress >= info->reshape_progress)
2280 need_backup = 0;
2281 } else {
2282 max_progress =
2283 read_offset *
2284 reshape->after.data_disks;
2285 /* If we are using internal metadata, then we can
2286 * progress all the way to the suspend_point without
2287 * worrying about backing-up/suspending along the
2288 * way.
2289 */
2290 if (max_progress < *suspend_point &&
2291 info->array.major_version >= 0)
2292 max_progress = *suspend_point;
2293 }
2294 }
2295
2296 /* We know it is safe to progress to 'max_progress' providing
2297 * it is suspended or we are using native metadata.
2298 * Consider extending suspend_point 128M per device if it
2299 * is less than 64M per device beyond reshape_progress.
2300 * But always do a multiple of 'blocks'
2301 * FIXME this is too big - it takes to long to complete
2302 * this much.
2303 */
2304 target = 64*1024*2 * min(reshape->before.data_disks,
2305 reshape->after.data_disks);
2306 target /= reshape->backup_blocks;
2307 if (target < 2)
2308 target = 2;
2309 target *= reshape->backup_blocks;
2310
2311 /* For externally managed metadata we always need to suspend IO to
2312 * the area being reshaped so we regularly push suspend_point forward.
2313 * For native metadata we only need the suspend if we are going to do
2314 * a backup.
2315 */
2316 if (advancing) {
2317 if ((need_backup > info->reshape_progress
2318 || info->array.major_version < 0) &&
2319 *suspend_point < info->reshape_progress + target) {
2320 if (need_backup < *suspend_point + 2 * target)
2321 *suspend_point = need_backup;
2322 else if (*suspend_point + 2 * target < array_size)
2323 *suspend_point += 2 * target;
2324 else
2325 *suspend_point = array_size;
2326 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
2327 if (max_progress > *suspend_point)
2328 max_progress = *suspend_point;
2329 }
2330 } else {
2331 if ((need_backup < info->reshape_progress
2332 || info->array.major_version < 0) &&
2333 *suspend_point > info->reshape_progress - target) {
2334 if (need_backup > *suspend_point - 2 * target)
2335 *suspend_point = need_backup;
2336 else if (*suspend_point >= 2 * target)
2337 *suspend_point -= 2 * target;
2338 else
2339 *suspend_point = 0;
2340 sysfs_set_num(info, NULL, "suspend_lo", *suspend_point);
2341 if (max_progress < *suspend_point)
2342 max_progress = *suspend_point;
2343 }
2344 }
2345
2346 /* now set sync_max to allow that progress. sync_max, like
2347 * sync_completed is a count of sectors written per device, so
2348 * we find the difference between max_progress and the start point,
2349 * and divide that by after.data_disks to get a sync_max
2350 * number.
2351 * At the same time we convert wait_point to a similar number
2352 * for comparing against sync_completed.
2353 */
2354 /* scale down max_progress to per_disk */
2355 max_progress /= reshape->after.data_disks;
2356 /* Round to chunk size as some kernels give an erroneously high number */
2357 max_progress /= info->new_chunk/512;
2358 max_progress *= info->new_chunk/512;
2359 /* Limit progress to the whole device */
2360 if (max_progress > info->component_size)
2361 max_progress = info->component_size;
2362 wait_point /= reshape->after.data_disks;
2363 if (!advancing) {
2364 /* switch from 'device offset' to 'processed block count' */
2365 max_progress = info->component_size - max_progress;
2366 wait_point = info->component_size - wait_point;
2367 }
2368
2369 sysfs_set_num(info, NULL, "sync_max", max_progress);
2370
2371 /* Now wait. If we have already reached the point that we were
2372 * asked to wait to, don't wait at all, else wait for any change.
2373 * We need to select on 'sync_completed' as that is the place that
2374 * notifications happen, but we are really interested in
2375 * 'reshape_position'
2376 */
2377 fd = sysfs_get_fd(info, NULL, "sync_completed");
2378 if (fd < 0)
2379 return -1;
2380
2381 if (sysfs_fd_get_ll(fd, &completed) < 0) {
2382 close(fd);
2383 return -1;
2384 }
2385 while (completed < max_progress && completed < wait_point) {
2386 /* Check that sync_action is still 'reshape' to avoid
2387 * waiting forever on a dead array
2388 */
2389 char action[20];
2390 fd_set rfds;
2391 if (sysfs_get_str(info, NULL, "sync_action",
2392 action, 20) <= 0 ||
2393 strncmp(action, "reshape", 7) != 0)
2394 break;
2395 FD_ZERO(&rfds);
2396 FD_SET(fd, &rfds);
2397 select(fd+1, NULL, NULL, &rfds, NULL);
2398 if (sysfs_fd_get_ll(fd, &completed) < 0) {
2399 close(fd);
2400 return -1;
2401 }
2402 }
2403 /* some kernels can give an incorrectly high 'completed' number */
2404 completed /= (info->new_chunk/512);
2405 completed *= (info->new_chunk/512);
2406 /* Convert 'completed' back in to a 'progress' number */
2407 completed *= reshape->after.data_disks;
2408 if (!advancing) {
2409 completed = info->component_size * reshape->after.data_disks
2410 - completed;
2411 }
2412 *reshape_completed = completed;
2413
2414 close(fd);
2415
2416 /* We return the need_backup flag. Caller will decide
2417 * how much - a multiple of ->backup_blocks up to *suspend_point
2418 */
2419 return advancing
2420 ? (need_backup > info->reshape_progress)
2421 : (need_backup < info->reshape_progress);
2422 }
2423
2424
2425 /* FIXME return status is never checked */
2426 static int grow_backup(struct mdinfo *sra,
2427 unsigned long long offset, /* per device */
2428 unsigned long stripes, /* per device, in old chunks */
2429 int *sources, unsigned long long *offsets,
2430 int disks, int chunk, int level, int layout,
2431 int dests, int *destfd, unsigned long long *destoffsets,
2432 int part, int *degraded,
2433 char *buf)
2434 {
2435 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2436 * to storage 'destfd' (offset 'destoffsets'), after first
2437 * suspending IO. Then allow resync to continue
2438 * over the suspended section.
2439 * Use part 'part' of the backup-super-block.
2440 */
2441 int odata = disks;
2442 int rv = 0;
2443 int i;
2444 unsigned long long ll;
2445 int new_degraded;
2446 //printf("offset %llu\n", offset);
2447 if (level >= 4)
2448 odata--;
2449 if (level == 6)
2450 odata--;
2451
2452 /* Check that array hasn't become degraded, else we might backup the wrong data */
2453 sysfs_get_ll(sra, NULL, "degraded", &ll);
2454 new_degraded = (int)ll;
2455 if (new_degraded != *degraded) {
2456 /* check each device to ensure it is still working */
2457 struct mdinfo *sd;
2458 for (sd = sra->devs ; sd ; sd = sd->next) {
2459 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2460 continue;
2461 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
2462 char sbuf[20];
2463 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
2464 strstr(sbuf, "faulty") ||
2465 strstr(sbuf, "in_sync") == NULL) {
2466 /* this device is dead */
2467 sd->disk.state = (1<<MD_DISK_FAULTY);
2468 if (sd->disk.raid_disk >= 0 &&
2469 sources[sd->disk.raid_disk] >= 0) {
2470 close(sources[sd->disk.raid_disk]);
2471 sources[sd->disk.raid_disk] = -1;
2472 }
2473 }
2474 }
2475 }
2476 *degraded = new_degraded;
2477 }
2478 if (part) {
2479 bsb.arraystart2 = __cpu_to_le64(offset * odata);
2480 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
2481 } else {
2482 bsb.arraystart = __cpu_to_le64(offset * odata);
2483 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
2484 }
2485 if (part)
2486 bsb.magic[15] = '2';
2487 for (i = 0; i < dests; i++)
2488 if (part)
2489 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
2490 else
2491 lseek64(destfd[i], destoffsets[i], 0);
2492
2493 rv = save_stripes(sources, offsets,
2494 disks, chunk, level, layout,
2495 dests, destfd,
2496 offset*512*odata, stripes * chunk * odata,
2497 buf);
2498
2499 if (rv)
2500 return rv;
2501 bsb.mtime = __cpu_to_le64(time(0));
2502 for (i = 0; i < dests; i++) {
2503 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
2504
2505 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
2506 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
2507 bsb.sb_csum2 = bsb_csum((char*)&bsb,
2508 ((char*)&bsb.sb_csum2)-((char*)&bsb));
2509
2510 rv = -1;
2511 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
2512 != destoffsets[i] - 4096)
2513 break;
2514 if (write(destfd[i], &bsb, 512) != 512)
2515 break;
2516 if (destoffsets[i] > 4096) {
2517 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
2518 destoffsets[i]+stripes*chunk*odata)
2519 break;
2520 if (write(destfd[i], &bsb, 512) != 512)
2521 break;
2522 }
2523 fsync(destfd[i]);
2524 rv = 0;
2525 }
2526
2527 return rv;
2528 }
2529
2530 /* in 2.6.30, the value reported by sync_completed can be
2531 * less that it should be by one stripe.
2532 * This only happens when reshape hits sync_max and pauses.
2533 * So allow wait_backup to either extent sync_max further
2534 * than strictly necessary, or return before the
2535 * sync has got quite as far as we would really like.
2536 * This is what 'blocks2' is for.
2537 * The various caller give appropriate values so that
2538 * every works.
2539 */
2540 /* FIXME return value is often ignored */
2541 static int forget_backup(
2542 int dests, int *destfd, unsigned long long *destoffsets,
2543 int part)
2544 {
2545 /*
2546 * Erase backup 'part' (which is 0 or 1)
2547 */
2548 int i;
2549 int rv;
2550
2551 if (part) {
2552 bsb.arraystart2 = __cpu_to_le64(0);
2553 bsb.length2 = __cpu_to_le64(0);
2554 } else {
2555 bsb.arraystart = __cpu_to_le64(0);
2556 bsb.length = __cpu_to_le64(0);
2557 }
2558 bsb.mtime = __cpu_to_le64(time(0));
2559 rv = 0;
2560 for (i = 0; i < dests; i++) {
2561 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
2562 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
2563 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
2564 bsb.sb_csum2 = bsb_csum((char*)&bsb,
2565 ((char*)&bsb.sb_csum2)-((char*)&bsb));
2566 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
2567 destoffsets[i]-4096)
2568 rv = -1;
2569 if (rv == 0 &&
2570 write(destfd[i], &bsb, 512) != 512)
2571 rv = -1;
2572 fsync(destfd[i]);
2573 }
2574 return rv;
2575 }
2576
2577 static void fail(char *msg)
2578 {
2579 int rv;
2580 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
2581 rv |= (write(2, "\n", 1) != 1);
2582 exit(rv ? 1 : 2);
2583 }
2584
2585 static char *abuf, *bbuf;
2586 static unsigned long long abuflen;
2587 static void validate(int afd, int bfd, unsigned long long offset)
2588 {
2589 /* check that the data in the backup against the array.
2590 * This is only used for regression testing and should not
2591 * be used while the array is active
2592 */
2593 if (afd < 0)
2594 return;
2595 lseek64(bfd, offset - 4096, 0);
2596 if (read(bfd, &bsb2, 512) != 512)
2597 fail("cannot read bsb");
2598 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
2599 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
2600 fail("first csum bad");
2601 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
2602 fail("magic is bad");
2603 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
2604 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
2605 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
2606 fail("second csum bad");
2607
2608 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
2609 fail("devstart is wrong");
2610
2611 if (bsb2.length) {
2612 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
2613
2614 if (abuflen < len) {
2615 free(abuf);
2616 free(bbuf);
2617 abuflen = len;
2618 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
2619 posix_memalign((void**)&bbuf, 4096, abuflen)) {
2620 abuflen = 0;
2621 /* just stop validating on mem-alloc failure */
2622 return;
2623 }
2624 }
2625
2626 lseek64(bfd, offset, 0);
2627 if ((unsigned long long)read(bfd, bbuf, len) != len) {
2628 //printf("len %llu\n", len);
2629 fail("read first backup failed");
2630 }
2631 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
2632 if ((unsigned long long)read(afd, abuf, len) != len)
2633 fail("read first from array failed");
2634 if (memcmp(bbuf, abuf, len) != 0) {
2635 #if 0
2636 int i;
2637 printf("offset=%llu len=%llu\n",
2638 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
2639 for (i=0; i<len; i++)
2640 if (bbuf[i] != abuf[i]) {
2641 printf("first diff byte %d\n", i);
2642 break;
2643 }
2644 #endif
2645 fail("data1 compare failed");
2646 }
2647 }
2648 if (bsb2.length2) {
2649 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
2650
2651 if (abuflen < len) {
2652 free(abuf);
2653 free(bbuf);
2654 abuflen = len;
2655 abuf = malloc(abuflen);
2656 bbuf = malloc(abuflen);
2657 }
2658
2659 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
2660 if ((unsigned long long)read(bfd, bbuf, len) != len)
2661 fail("read second backup failed");
2662 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
2663 if ((unsigned long long)read(afd, abuf, len) != len)
2664 fail("read second from array failed");
2665 if (memcmp(bbuf, abuf, len) != 0)
2666 fail("data2 compare failed");
2667 }
2668 }
2669
2670 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
2671 struct supertype *st, unsigned long blocks,
2672 int *fds, unsigned long long *offsets,
2673 int dests, int *destfd, unsigned long long *destoffsets)
2674 {
2675 /* Monitor a reshape where backup is being performed using
2676 * 'native' mechanism - either to a backup file, or
2677 * to some space in a spare.
2678 */
2679 char *buf;
2680 int degraded = -1;
2681 unsigned long long speed;
2682 unsigned long long suspend_point, array_size;
2683 unsigned long long backup_point, wait_point;
2684 unsigned long long reshape_completed;
2685 int done = 0;
2686 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
2687 int part = 0; /* The next part of the backup area to fill. It may already
2688 * be full, so we need to check */
2689 int level = reshape->level;
2690 int layout = reshape->before.layout;
2691 int data = reshape->before.data_disks;
2692 int disks = reshape->before.data_disks + reshape->parity;
2693 int chunk = sra->array.chunk_size;
2694 struct mdinfo *sd;
2695 unsigned long stripes;
2696
2697 /* set up the backup-super-block. This requires the
2698 * uuid from the array.
2699 */
2700 /* Find a superblock */
2701 for (sd = sra->devs; sd; sd = sd->next) {
2702 char *dn;
2703 int devfd;
2704 int ok;
2705 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2706 continue;
2707 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
2708 devfd = dev_open(dn, O_RDONLY);
2709 if (devfd < 0)
2710 continue;
2711 ok = st->ss->load_super(st, devfd, NULL);
2712 close(devfd);
2713 if (ok >= 0)
2714 break;
2715 }
2716 if (!sd) {
2717 fprintf(stderr, Name ": Cannot find a superblock\n");
2718 return 0;
2719 }
2720
2721 memset(&bsb, 0, 512);
2722 memcpy(bsb.magic, "md_backup_data-1", 16);
2723 st->ss->uuid_from_super(st, (int*)&bsb.set_uuid);
2724 bsb.mtime = __cpu_to_le64(time(0));
2725 bsb.devstart2 = blocks;
2726
2727 stripes = blocks / (sra->array.chunk_size/512) /
2728 reshape->before.data_disks;
2729
2730 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2731 /* Don't start the 'reshape' */
2732 return 0;
2733 if (reshape->before.data_disks == reshape->after.data_disks) {
2734 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
2735 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
2736 }
2737
2738 if (increasing) {
2739 array_size = sra->component_size * reshape->after.data_disks;
2740 backup_point = sra->reshape_progress;
2741 suspend_point = 0;
2742 } else {
2743 array_size = sra->component_size * reshape->before.data_disks;
2744 backup_point = array_size;
2745 suspend_point = array_size;
2746 }
2747
2748 while (!done) {
2749 int rv;
2750
2751 /* Want to return as soon the oldest backup slot can
2752 * be released as that allows us to start backing up
2753 * some more, providing suspend_point has been
2754 * advanced, which it should have.
2755 */
2756 if (increasing) {
2757 wait_point = array_size;
2758 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
2759 wait_point = (__le64_to_cpu(bsb.arraystart) +
2760 __le64_to_cpu(bsb.length));
2761 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
2762 wait_point = (__le64_to_cpu(bsb.arraystart2) +
2763 __le64_to_cpu(bsb.length2));
2764 } else {
2765 wait_point = 0;
2766 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
2767 wait_point = __le64_to_cpu(bsb.arraystart);
2768 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
2769 wait_point = __le64_to_cpu(bsb.arraystart2);
2770 }
2771
2772 rv = progress_reshape(sra, reshape,
2773 backup_point, wait_point,
2774 &suspend_point, &reshape_completed);
2775 /* external metadata would need to ping_monitor here */
2776 sra->reshape_progress = reshape_completed;
2777
2778 /* Clear any backup region that is before 'here' */
2779 if (increasing) {
2780 if (reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
2781 __le64_to_cpu(bsb.length)))
2782 forget_backup(dests, destfd,
2783 destoffsets, 0);
2784 if (reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
2785 __le64_to_cpu(bsb.length2)))
2786 forget_backup(dests, destfd,
2787 destoffsets, 1);
2788 } else {
2789 if (reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
2790 forget_backup(dests, destfd,
2791 destoffsets, 0);
2792 if (reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
2793 forget_backup(dests, destfd,
2794 destoffsets, 1);
2795 }
2796
2797 if (rv < 0) {
2798 done = 1;
2799 break;
2800 }
2801
2802 while (rv) {
2803 unsigned long long offset;
2804 unsigned long actual_stripes;
2805 /* Need to backup some data.
2806 * If 'part' is not used and the desired
2807 * backup size is suspended, do a backup,
2808 * then consider the next part.
2809 */
2810 /* Check that 'part' is unused */
2811 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
2812 break;
2813 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
2814 break;
2815
2816 offset = backup_point / data;
2817 actual_stripes = stripes;
2818 if (increasing) {
2819 if (offset + actual_stripes * (chunk/512) >
2820 sra->component_size)
2821 actual_stripes = ((sra->component_size - offset)
2822 / (chunk/512));
2823 if (offset + actual_stripes * (chunk/512) >
2824 suspend_point/data)
2825 break;
2826 } else {
2827 if (offset < actual_stripes * (chunk/512))
2828 actual_stripes = offset / (chunk/512);
2829 offset -= actual_stripes * (chunk/512);
2830 if (offset < suspend_point/data)
2831 break;
2832 }
2833 grow_backup(sra, offset, actual_stripes,
2834 fds, offsets,
2835 disks, chunk, level, layout,
2836 dests, destfd, destoffsets,
2837 part, &degraded, buf);
2838 validate(afd, destfd[0], destoffsets[0]);
2839 /* record where 'part' is up to */
2840 part = !part;
2841 if (increasing)
2842 backup_point += actual_stripes * (chunk/512) * data;
2843 else
2844 backup_point -= actual_stripes * (chunk/512) * data;
2845 }
2846 }
2847
2848 /* FIXME maybe call progress_reshape one more time instead */
2849 abort_reshape(sra); /* remove any remaining suspension */
2850 if (reshape->before.data_disks == reshape->after.data_disks)
2851 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
2852 free(buf);
2853 return 1; /* FIXME what does this mean? */
2854 }
2855
2856 /*
2857 * If any spare contains md_back_data-1 which is recent wrt mtime,
2858 * write that data into the array and update the super blocks with
2859 * the new reshape_progress
2860 */
2861 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
2862 char *backup_file, int verbose)
2863 {
2864 int i, j;
2865 int old_disks;
2866 unsigned long long *offsets;
2867 unsigned long long nstripe, ostripe;
2868 int ndata, odata;
2869
2870 if (info->new_level != info->array.level)
2871 return 1; /* Cannot handle level changes (they are instantaneous) */
2872
2873 odata = info->array.raid_disks - info->delta_disks - 1;
2874 if (info->array.level == 6) odata--; /* number of data disks */
2875 ndata = info->array.raid_disks - 1;
2876 if (info->new_level == 6) ndata--;
2877
2878 old_disks = info->array.raid_disks - info->delta_disks;
2879
2880 if (info->delta_disks <= 0)
2881 /* Didn't grow, so the backup file must have
2882 * been used
2883 */
2884 old_disks = cnt;
2885 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
2886 struct mdinfo dinfo;
2887 int fd;
2888 int bsbsize;
2889 char *devname, namebuf[20];
2890
2891 /* This was a spare and may have some saved data on it.
2892 * Load the superblock, find and load the
2893 * backup_super_block.
2894 * If either fail, go on to next device.
2895 * If the backup contains no new info, just return
2896 * else restore data and update all superblocks
2897 */
2898 if (i == old_disks-1) {
2899 fd = open(backup_file, O_RDONLY);
2900 if (fd<0) {
2901 fprintf(stderr, Name ": backup file %s inaccessible: %s\n",
2902 backup_file, strerror(errno));
2903 continue;
2904 }
2905 devname = backup_file;
2906 } else {
2907 fd = fdlist[i];
2908 if (fd < 0)
2909 continue;
2910 if (st->ss->load_super(st, fd, NULL))
2911 continue;
2912
2913 st->ss->getinfo_super(st, &dinfo, NULL);
2914 st->ss->free_super(st);
2915
2916 if (lseek64(fd,
2917 (dinfo.data_offset + dinfo.component_size - 8) <<9,
2918 0) < 0) {
2919 fprintf(stderr, Name ": Cannot seek on device %d\n", i);
2920 continue; /* Cannot seek */
2921 }
2922 sprintf(namebuf, "device-%d", i);
2923 devname = namebuf;
2924 }
2925 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
2926 if (verbose)
2927 fprintf(stderr, Name ": Cannot read from %s\n", devname);
2928 continue; /* Cannot read */
2929 }
2930 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
2931 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
2932 if (verbose)
2933 fprintf(stderr, Name ": No backup metadata on %s\n", devname);
2934 continue;
2935 }
2936 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
2937 if (verbose)
2938 fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname);
2939 continue; /* bad checksum */
2940 }
2941 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
2942 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
2943 if (verbose)
2944 fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname);
2945 continue; /* Bad second checksum */
2946 }
2947 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
2948 if (verbose)
2949 fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname);
2950 continue; /* Wrong uuid */
2951 }
2952
2953 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2954 * sometimes they aren't... So allow considerable flexability in matching, and allow
2955 * this test to be overridden by an environment variable.
2956 */
2957 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
2958 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
2959 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2960 fprintf(stderr, Name ": accepting backup with timestamp %lu "
2961 "for array with timestamp %lu\n",
2962 (unsigned long)__le64_to_cpu(bsb.mtime),
2963 (unsigned long)info->array.utime);
2964 } else {
2965 if (verbose)
2966 fprintf(stderr, Name ": too-old timestamp on "
2967 "backup-metadata on %s\n", devname);
2968 continue; /* time stamp is too bad */
2969 }
2970 }
2971
2972 if (bsb.magic[15] == '1') {
2973 if (info->delta_disks >= 0) {
2974 /* reshape_progress is increasing */
2975 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2976 info->reshape_progress) {
2977 nonew:
2978 if (verbose)
2979 fprintf(stderr, Name ": backup-metadata found on %s but is not needed\n", devname);
2980 continue; /* No new data here */
2981 }
2982 } else {
2983 /* reshape_progress is decreasing */
2984 if (__le64_to_cpu(bsb.arraystart) >=
2985 info->reshape_progress)
2986 goto nonew; /* No new data here */
2987 }
2988 } else {
2989 if (info->delta_disks >= 0) {
2990 /* reshape_progress is increasing */
2991 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2992 info->reshape_progress &&
2993 __le64_to_cpu(bsb.arraystart2) + __le64_to_cpu(bsb.length2) <
2994 info->reshape_progress)
2995 goto nonew; /* No new data here */
2996 } else {
2997 /* reshape_progress is decreasing */
2998 if (__le64_to_cpu(bsb.arraystart) >=
2999 info->reshape_progress &&
3000 __le64_to_cpu(bsb.arraystart2) >=
3001 info->reshape_progress)
3002 goto nonew; /* No new data here */
3003 }
3004 }
3005 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
3006 second_fail:
3007 if (verbose)
3008 fprintf(stderr, Name ": Failed to verify secondary backup-metadata block on %s\n",
3009 devname);
3010 continue; /* Cannot seek */
3011 }
3012 /* There should be a duplicate backup superblock 4k before here */
3013 if (lseek64(fd, -4096, 1) < 0 ||
3014 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
3015 goto second_fail; /* Cannot find leading superblock */
3016 if (bsb.magic[15] == '1')
3017 bsbsize = offsetof(struct mdp_backup_super, pad1);
3018 else
3019 bsbsize = offsetof(struct mdp_backup_super, pad);
3020 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
3021 goto second_fail; /* Cannot find leading superblock */
3022
3023 /* Now need the data offsets for all devices. */
3024 offsets = malloc(sizeof(*offsets)*info->array.raid_disks);
3025 for(j=0; j<info->array.raid_disks; j++) {
3026 if (fdlist[j] < 0)
3027 continue;
3028 if (st->ss->load_super(st, fdlist[j], NULL))
3029 /* FIXME should be this be an error */
3030 continue;
3031 st->ss->getinfo_super(st, &dinfo, NULL);
3032 st->ss->free_super(st);
3033 offsets[j] = dinfo.data_offset * 512;
3034 }
3035 printf(Name ": restoring critical section\n");
3036
3037 if (restore_stripes(fdlist, offsets,
3038 info->array.raid_disks,
3039 info->new_chunk,
3040 info->new_level,
3041 info->new_layout,
3042 fd, __le64_to_cpu(bsb.devstart)*512,
3043 __le64_to_cpu(bsb.arraystart)*512,
3044 __le64_to_cpu(bsb.length)*512)) {
3045 /* didn't succeed, so giveup */
3046 if (verbose)
3047 fprintf(stderr, Name ": Error restoring backup from %s\n",
3048 devname);
3049 return 1;
3050 }
3051
3052 if (bsb.magic[15] == '2' &&
3053 restore_stripes(fdlist, offsets,
3054 info->array.raid_disks,
3055 info->new_chunk,
3056 info->new_level,
3057 info->new_layout,
3058 fd, __le64_to_cpu(bsb.devstart)*512 +
3059 __le64_to_cpu(bsb.devstart2)*512,
3060 __le64_to_cpu(bsb.arraystart2)*512,
3061 __le64_to_cpu(bsb.length2)*512)) {
3062 /* didn't succeed, so giveup */
3063 if (verbose)
3064 fprintf(stderr, Name ": Error restoring second backup from %s\n",
3065 devname);
3066 return 1;
3067 }
3068
3069
3070 /* Ok, so the data is restored. Let's update those superblocks. */
3071
3072 if (info->delta_disks >= 0) {
3073 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
3074 __le64_to_cpu(bsb.length);
3075 if (bsb.magic[15] == '2') {
3076 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
3077 __le64_to_cpu(bsb.length2);
3078 if (p2 > info->reshape_progress)
3079 info->reshape_progress = p2;
3080 }
3081 } else {
3082 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
3083 if (bsb.magic[15] == '2') {
3084 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
3085 if (p2 < info->reshape_progress)
3086 info->reshape_progress = p2;
3087 }
3088 }
3089 for (j=0; j<info->array.raid_disks; j++) {
3090 if (fdlist[j] < 0) continue;
3091 if (st->ss->load_super(st, fdlist[j], NULL))
3092 continue;
3093 st->ss->getinfo_super(st, &dinfo, NULL);
3094 dinfo.reshape_progress = info->reshape_progress;
3095 st->ss->update_super(st, &dinfo,
3096 "_reshape_progress",
3097 NULL,0, 0, NULL);
3098 st->ss->store_super(st, fdlist[j]);
3099 st->ss->free_super(st);
3100 }
3101 return 0;
3102 }
3103 /* Didn't find any backup data, try to see if any
3104 * was needed.
3105 */
3106 if (info->delta_disks < 0) {
3107 /* When shrinking, the critical section is at the end.
3108 * So see if we are before the critical section.
3109 */
3110 unsigned long long first_block;
3111 nstripe = ostripe = 0;
3112 first_block = 0;
3113 while (ostripe >= nstripe) {
3114 ostripe += info->array.chunk_size / 512;
3115 first_block = ostripe * odata;
3116 nstripe = first_block / ndata / (info->new_chunk/512) *
3117 (info->new_chunk/512);
3118 }
3119
3120 if (info->reshape_progress >= first_block)
3121 return 0;
3122 }
3123 if (info->delta_disks > 0) {
3124 /* See if we are beyond the critical section. */
3125 unsigned long long last_block;
3126 nstripe = ostripe = 0;
3127 last_block = 0;
3128 while (nstripe >= ostripe) {
3129 nstripe += info->new_chunk / 512;
3130 last_block = nstripe * ndata;
3131 ostripe = last_block / odata / (info->array.chunk_size/512) *
3132 (info->array.chunk_size/512);
3133 }
3134
3135 if (info->reshape_progress >= last_block)
3136 return 0;
3137 }
3138 /* needed to recover critical section! */
3139 if (verbose)
3140 fprintf(stderr, Name ": Failed to find backup of critical section\n");
3141 return 1;
3142 }
3143
3144 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
3145 char *backup_file)
3146 {
3147 int err = sysfs_set_str(info, NULL, "array_state", "readonly");
3148 if (err)
3149 return err;
3150 return reshape_array(NULL, mdfd, "array", st, info, 1, backup_file, 0, 0);
3151 }
3152
3153
Unnamed repository; edit this file 'description' to name the repository.