]> 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
Clean up exit paths from reshape_array.
[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)
534 {
535 if (st->ss->external)
536 return unfreeze_container(st);
537 else {
538 struct mdinfo *sra = sysfs_read(-1, st->devnum, GET_VERSION);
539
540 if (sra)
541 sysfs_set_str(sra, NULL, "sync_action", "idle");
542 else
543 fprintf(stderr, Name ": failed to unfreeze array\n");
544 sysfs_free(sra);
545 }
546 }
547
548 static void wait_reshape(struct mdinfo *sra)
549 {
550 int fd = sysfs_get_fd(sra, NULL, "sync_action");
551 char action[20];
552
553 if (fd < 0)
554 return;
555
556 while (sysfs_fd_get_str(fd, action, 20) > 0 &&
557 strncmp(action, "reshape", 7) == 0) {
558 fd_set rfds;
559 FD_ZERO(&rfds);
560 FD_SET(fd, &rfds);
561 select(fd+1, NULL, NULL, &rfds, NULL);
562 }
563 close(fd);
564 }
565
566 static int reshape_super(struct supertype *st, long long size, int level,
567 int layout, int chunksize, int raid_disks,
568 char *backup_file, char *dev, int verbose)
569 {
570 /* nothing extra to check in the native case */
571 if (!st->ss->external)
572 return 0;
573 if (!st->ss->reshape_super ||
574 !st->ss->manage_reshape) {
575 fprintf(stderr, Name ": %s metadata does not support reshape\n",
576 st->ss->name);
577 return 1;
578 }
579
580 return st->ss->reshape_super(st, size, level, layout, chunksize,
581 raid_disks, backup_file, dev, verbose);
582 }
583
584 static void sync_metadata(struct supertype *st)
585 {
586 if (st->ss->external) {
587 if (st->update_tail) {
588 flush_metadata_updates(st);
589 st->update_tail = &st->updates;
590 } else
591 st->ss->sync_metadata(st);
592 }
593 }
594
595 static int subarray_set_num(char *container, struct mdinfo *sra, char *name, int n)
596 {
597 /* when dealing with external metadata subarrays we need to be
598 * prepared to handle EAGAIN. The kernel may need to wait for
599 * mdmon to mark the array active so the kernel can handle
600 * allocations/writeback when preparing the reshape action
601 * (md_allow_write()). We temporarily disable safe_mode_delay
602 * to close a race with the array_state going clean before the
603 * next write to raid_disks / stripe_cache_size
604 */
605 char safe[50];
606 int rc;
607
608 /* only 'raid_disks' and 'stripe_cache_size' trigger md_allow_write */
609 if (!container ||
610 (strcmp(name, "raid_disks") != 0 &&
611 strcmp(name, "stripe_cache_size") != 0))
612 return sysfs_set_num(sra, NULL, name, n);
613
614 rc = sysfs_get_str(sra, NULL, "safe_mode_delay", safe, sizeof(safe));
615 if (rc <= 0)
616 return -1;
617 sysfs_set_num(sra, NULL, "safe_mode_delay", 0);
618 rc = sysfs_set_num(sra, NULL, name, n);
619 if (rc < 0 && errno == EAGAIN) {
620 ping_monitor(container);
621 /* if we get EAGAIN here then the monitor is not active
622 * so stop trying
623 */
624 rc = sysfs_set_num(sra, NULL, name, n);
625 }
626 sysfs_set_str(sra, NULL, "safe_mode_delay", safe);
627 return rc;
628 }
629
630 int start_reshape(struct mdinfo *sra)
631 {
632 int err;
633 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
634 err = sysfs_set_num(sra, NULL, "suspend_hi", 0);
635 err = err ?: sysfs_set_num(sra, NULL, "suspend_lo", 0);
636 err = err ?: sysfs_set_num(sra, NULL, "sync_min", 0);
637 err = err ?: sysfs_set_num(sra, NULL, "sync_max", 0);
638 err = err ?: sysfs_set_str(sra, NULL, "sync_action", "reshape");
639
640 return err;
641 }
642
643 void abort_reshape(struct mdinfo *sra)
644 {
645 sysfs_set_str(sra, NULL, "sync_action", "idle");
646 sysfs_set_num(sra, NULL, "suspend_lo", 0x7FFFFFFFFFFFFFFFULL);
647 sysfs_set_num(sra, NULL, "suspend_hi", 0);
648 sysfs_set_num(sra, NULL, "suspend_lo", 0);
649 sysfs_set_num(sra, NULL, "sync_min", 0);
650 sysfs_set_str(sra, NULL, "sync_max", "max");
651 }
652
653 int remove_disks_on_raid10_to_raid0_takeover(struct supertype *st,
654 struct mdinfo *sra,
655 int layout)
656 {
657 int nr_of_copies;
658 struct mdinfo *remaining;
659 int slot;
660
661 nr_of_copies = layout & 0xff;
662
663 remaining = sra->devs;
664 sra->devs = NULL;
665 /* for each 'copy', select one device and remove from the list. */
666 for (slot = 0; slot < sra->array.raid_disks; slot += nr_of_copies) {
667 struct mdinfo **diskp;
668 int found = 0;
669
670 /* Find a working device to keep */
671 for (diskp = &remaining; *diskp ; diskp = &(*diskp)->next) {
672 struct mdinfo *disk = *diskp;
673
674 if (disk->disk.raid_disk < slot)
675 continue;
676 if (disk->disk.raid_disk >= slot + nr_of_copies)
677 continue;
678 if (disk->disk.state & (1<<MD_DISK_REMOVED))
679 continue;
680 if (disk->disk.state & (1<<MD_DISK_FAULTY))
681 continue;
682 if (!(disk->disk.state & (1<<MD_DISK_SYNC)))
683 continue;
684
685 /* We have found a good disk to use! */
686 *diskp = disk->next;
687 disk->next = sra->devs;
688 sra->devs = disk;
689 found = 1;
690 break;
691 }
692 if (!found)
693 break;
694 }
695
696 if (slot < sra->array.raid_disks) {
697 /* didn't find all slots */
698 struct mdinfo **e;
699 e = &remaining;
700 while (*e)
701 e = &(*e)->next;
702 *e = sra->devs;
703 sra->devs = remaining;
704 return 1;
705 }
706
707 /* Remove all 'remaining' devices from the array */
708 while (remaining) {
709 struct mdinfo *sd = remaining;
710 remaining = sd->next;
711
712 sysfs_set_str(sra, sd, "state", "faulty");
713 sysfs_set_str(sra, sd, "slot", "none");
714 sysfs_set_str(sra, sd, "state", "remove");
715 sd->disk.state |= (1<<MD_DISK_REMOVED);
716 sd->disk.state &= ~(1<<MD_DISK_SYNC);
717 sd->next = sra->devs;
718 sra->devs = sd;
719 }
720 return 0;
721 }
722
723 void reshape_free_fdlist(int *fdlist,
724 unsigned long long *offsets,
725 int size)
726 {
727 int i;
728
729 for (i = 0; i < size; i++)
730 if (fdlist[i] >= 0)
731 close(fdlist[i]);
732
733 free(fdlist);
734 free(offsets);
735 }
736
737 int reshape_prepare_fdlist(char *devname,
738 struct mdinfo *sra,
739 int raid_disks,
740 int nrdisks,
741 unsigned long blocks,
742 char *backup_file,
743 int *fdlist,
744 unsigned long long *offsets)
745 {
746 int d = 0;
747 struct mdinfo *sd;
748
749 for (d = 0; d <= nrdisks; d++)
750 fdlist[d] = -1;
751 d = raid_disks;
752 for (sd = sra->devs; sd; sd = sd->next) {
753 if (sd->disk.state & (1<<MD_DISK_FAULTY))
754 continue;
755 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
756 char *dn = map_dev(sd->disk.major,
757 sd->disk.minor, 1);
758 fdlist[sd->disk.raid_disk]
759 = dev_open(dn, O_RDONLY);
760 offsets[sd->disk.raid_disk] = sd->data_offset*512;
761 if (fdlist[sd->disk.raid_disk] < 0) {
762 fprintf(stderr,
763 Name ": %s: cannot open component %s\n",
764 devname, dn ? dn : "-unknown-");
765 d = -1;
766 goto release;
767 }
768 } else if (backup_file == NULL) {
769 /* spare */
770 char *dn = map_dev(sd->disk.major,
771 sd->disk.minor, 1);
772 fdlist[d] = dev_open(dn, O_RDWR);
773 offsets[d] = (sd->data_offset + sra->component_size - blocks - 8)*512;
774 if (fdlist[d] < 0) {
775 fprintf(stderr, Name ": %s: cannot open component %s\n",
776 devname, dn ? dn : "-unknown-");
777 d = -1;
778 goto release;
779 }
780 d++;
781 }
782 }
783 release:
784 return d;
785 }
786
787 int reshape_open_backup_file(char *backup_file,
788 int fd,
789 char *devname,
790 long blocks,
791 int *fdlist,
792 unsigned long long *offsets)
793 {
794 /* Return 1 on success, 0 on any form of failure */
795 /* need to check backup file is large enough */
796 char buf[512];
797 struct stat stb;
798 unsigned int dev;
799 int i;
800
801 *fdlist = open(backup_file, O_RDWR|O_CREAT|O_EXCL,
802 S_IRUSR | S_IWUSR);
803 *offsets = 8 * 512;
804 if (*fdlist < 0) {
805 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
806 devname, backup_file, strerror(errno));
807 return 0;
808 }
809 /* Guard against backup file being on array device.
810 * If array is partitioned or if LVM etc is in the
811 * way this will not notice, but it is better than
812 * nothing.
813 */
814 fstat(*fdlist, &stb);
815 dev = stb.st_dev;
816 fstat(fd, &stb);
817 if (stb.st_rdev == dev) {
818 fprintf(stderr, Name ": backup file must NOT be"
819 " on the array being reshaped.\n");
820 close(*fdlist);
821 return 0;
822 }
823
824 memset(buf, 0, 512);
825 for (i=0; i < blocks + 1 ; i++) {
826 if (write(*fdlist, buf, 512) != 512) {
827 fprintf(stderr, Name ": %s: cannot create"
828 " backup file %s: %s\n",
829 devname, backup_file, strerror(errno));
830 return 0;
831 }
832 }
833 if (fsync(*fdlist) != 0) {
834 fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
835 devname, backup_file, strerror(errno));
836 return 0;
837 }
838
839 return 1;
840 }
841
842 unsigned long compute_backup_blocks(int nchunk, int ochunk,
843 unsigned int ndata, unsigned int odata)
844 {
845 unsigned long a, b, blocks;
846 /* So how much do we need to backup.
847 * We need an amount of data which is both a whole number of
848 * old stripes and a whole number of new stripes.
849 * So LCM for (chunksize*datadisks).
850 */
851 a = (ochunk/512) * odata;
852 b = (nchunk/512) * ndata;
853 /* Find GCD */
854 while (a != b) {
855 if (a < b)
856 b -= a;
857 if (b < a)
858 a -= b;
859 }
860 /* LCM == product / GCD */
861 blocks = (ochunk/512) * (nchunk/512) * odata * ndata / a;
862
863 return blocks;
864 }
865
866 char *analyse_change(struct mdinfo *info, struct reshape *re)
867 {
868 /* Based on the current array state in info->array and
869 * the changes in info->new_* etc, determine:
870 * - whether the change is possible
871 * - Intermediate level/raid_disks/layout
872 * - whether a restriping reshape is needed
873 * - number of sectors in minimum change unit. This
874 * will cover a whole number of stripes in 'before' and
875 * 'after'.
876 *
877 * Return message if the change should be rejected
878 * NULL if the change can be achieved
879 *
880 * This can be called as part of starting a reshape, or
881 * when assembling an array that is undergoing reshape.
882 */
883 int new_disks;
884
885 /* If a new level not explicitly given, we assume no-change */
886 if (info->new_level == UnSet)
887 info->new_level = info->array.level;
888
889 if (info->new_chunk)
890 switch (info->new_level) {
891 case 0:
892 case 4:
893 case 5:
894 case 6:
895 case 10:
896 /* chunk size is meaningful, must divide component_size
897 * evenly
898 */
899 if (info->component_size % (info->new_chunk/512))
900 return "New chunk size does not"
901 " divide component size";
902 break;
903 default:
904 return "chunk size not meaningful for this level";
905 }
906 else
907 info->new_chunk = info->array.chunk_size;
908
909 switch (info->array.level) {
910 case 1:
911 /* RAID1 can convert to RAID1 with different disks, or
912 * raid5 with 2 disks
913 */
914 if (info->new_level == 1) {
915 if (info->delta_disks == UnSet)
916 /* Don't know what to do */
917 return "no change requested for Growing RAID1";
918 re->level = 1;
919 re->before.data_disks = (info->array.raid_disks +
920 info->delta_disks);
921 re->before.layout = 0;
922 re->backup_blocks = 0;
923 re->parity = 0;
924 return NULL;
925 }
926 if (info->array.raid_disks == 2 &&
927 info->array.raid_disks == 5) {
928 /* simple in-place conversion */
929 re->level = 5;
930 re->parity = 1;
931 re->before.data_disks = 1;
932 re->before.layout = ALGORITHM_LEFT_SYMMETRIC;
933 re->backup_blocks = 0;
934 return NULL;
935 }
936 /* Could do some multi-stage conversions, but leave that to
937 * later.
938 */
939 return "Impossibly level change request for RAID1";
940
941 case 10:
942 /* RAID10 can only be converted from near mode to
943 * RAID0 by removing some devices
944 */
945 if ((info->array.layout & ~0xff) != 0x100)
946 return "Cannot Grow RAID10 with far/offset layout";
947 /* number of devices must be multiple of number of copies */
948 if (info->array.raid_disks % (info->array.layout & 0xff))
949 return "RAID10 layout too complex for Grow operation";
950
951 if (info->new_level != 0)
952 return "RAID10 can only be changed to RAID0";
953 new_disks = (info->array.raid_disks
954 / (info->array.layout & 0xff));
955 if (info->delta_disks != UnSet) {
956 info->delta_disks = (new_disks
957 - info->array.raid_disks);
958 }
959 if (info->delta_disks != new_disks - info->array.raid_disks)
960 return "New number of raid-devices impossible for RAID10";
961 if (info->new_chunk &&
962 info->new_chunk != info->array.chunk_size)
963 return "Cannot change chunk-size with RAID10 Grow";
964
965 /* looks good */
966 re->level = 0;
967 re->parity = 0;
968 re->before.data_disks = new_disks;
969 re->before.layout = 0;
970 re->backup_blocks = 0;
971 return NULL;
972
973 case 0:
974 /* RAID0 can be converted to RAID10, or to RAID456 */
975 if (info->new_level == 10) {
976 if (info->new_layout == UnSet && info->delta_disks == UnSet) {
977 /* Assume near=2 layout */
978 info->new_layout = 0x102;
979 info->delta_disks = info->array.raid_disks;
980 }
981 if (info->new_layout == UnSet) {
982 int copies = 1 + (info->delta_disks
983 / info->array.raid_disks);
984 if (info->array.raid_disks * (copies-1)
985 != info->delta_disks)
986 return "Impossible number of devices"
987 " for RAID0->RAID10";
988 info->new_layout = 0x100 + copies;
989 }
990 if (info->delta_disks == UnSet) {
991 int copies = info->new_layout & 0xff;
992 if (info->new_layout != 0x100 + copies)
993 return "New layout impossible"
994 " for RAID0->RAID10";;
995 info->delta_disks = (copies - 1) *
996 info->array.raid_disks;
997 }
998 if (info->new_chunk &&
999 info->new_chunk != info->array.chunk_size)
1000 return "Cannot change chunk-size with RAID0->RAID10";
1001 /* looks good */
1002 re->level = 10;
1003 re->parity = 0;
1004 re->before.data_disks = (info->array.raid_disks +
1005 info->delta_disks);
1006 re->before.layout = info->new_layout;
1007 re->backup_blocks = 0;
1008 return NULL;
1009 }
1010
1011 /* RAID0 can also covert to RAID0/4/5/6 by first converting to
1012 * a raid4 style layout of the final level.
1013 */
1014 switch (info->new_level) {
1015 case 0:
1016 case 4:
1017 re->level = 4;
1018 re->before.layout = 0;
1019 break;
1020 case 5:
1021 re->level = 5;
1022 re->before.layout = ALGORITHM_PARITY_N;
1023 break;
1024 case 6:
1025 re->level = 6;
1026 re->before.layout = ALGORITHM_PARITY_N;
1027 break;
1028 default:
1029 return "Impossible level change requested";
1030 }
1031 re->before.data_disks = info->array.raid_disks;
1032 /* determining 'after' layout happens outside this 'switch' */
1033 break;
1034
1035 case 4:
1036 info->array.layout = ALGORITHM_PARITY_N;
1037 case 5:
1038 switch (info->new_level) {
1039 case 4:
1040 re->level = info->array.level;
1041 re->before.data_disks = info->array.raid_disks - 1;
1042 re->before.layout = info->array.layout;
1043 break;
1044 case 5:
1045 re->level = 5;
1046 re->before.data_disks = info->array.raid_disks - 1;
1047 re->before.layout = info->array.layout;
1048 break;
1049 case 6:
1050 re->level = 6;
1051 re->before.data_disks = info->array.raid_disks - 1;
1052 switch (info->array.layout) {
1053 case ALGORITHM_LEFT_ASYMMETRIC:
1054 re->before.layout = ALGORITHM_LEFT_ASYMMETRIC_6;
1055 break;
1056 case ALGORITHM_RIGHT_ASYMMETRIC:
1057 re->before.layout = ALGORITHM_RIGHT_ASYMMETRIC_6;
1058 break;
1059 case ALGORITHM_LEFT_SYMMETRIC:
1060 re->before.layout = ALGORITHM_LEFT_SYMMETRIC_6;
1061 break;
1062 case ALGORITHM_RIGHT_SYMMETRIC:
1063 re->before.layout = ALGORITHM_RIGHT_SYMMETRIC_6;
1064 break;
1065 case ALGORITHM_PARITY_0:
1066 re->before.layout = ALGORITHM_PARITY_0_6;
1067 break;
1068 case ALGORITHM_PARITY_N:
1069 re->before.layout = ALGORITHM_PARITY_N_6;
1070 break;
1071 default:
1072 return "Cannot convert an array with this layout";
1073 }
1074 break;
1075 case 1:
1076 if (info->array.raid_disks != 2)
1077 return "Can only convert a 2-device array to RAID1";
1078 re->level = 1;
1079 re->before.data_disks = 2;
1080 re->before.layout = 0;
1081 break;
1082 default:
1083 return "Impossible level change requested";
1084 }
1085 break;
1086 case 6:
1087 switch (info->new_level) {
1088 case 4:
1089 case 5:
1090 case 6:
1091 re->level = 6;
1092 re->before.data_disks = info->array.raid_disks - 2;
1093 re->before.layout = info->array.layout;
1094 break;
1095 default:
1096 return "Impossible level change requested";
1097 }
1098 break;
1099 }
1100
1101 /* If we reached here then it looks like a re-stripe is
1102 * happening. We have determined the intermediate level
1103 * and initial raid_disks/layout and stored these in 're'.
1104 *
1105 * We need to deduce the final layout that can be atomically
1106 * converted to the end state.
1107 */
1108 switch (info->new_level) {
1109 case 0:
1110 /* We can only get to RAID0 from RAID4 or RAID5
1111 * with appropriate layout and one extra device
1112 */
1113 if (re->level != 4 && re->level != 5)
1114 return "Cannot covert to RAID0 from this level";
1115 if (info->delta_disks == UnSet)
1116 re->after.data_disks = re->before.data_disks;
1117 else
1118 re->after.data_disks =
1119 info->array.raid_disks + info->delta_disks;
1120 switch (re->level) {
1121 case 4:
1122 re->after.layout = 0 ; break;
1123 case 5:
1124 re->after.layout = ALGORITHM_PARITY_N; break;
1125 }
1126 break;
1127
1128 case 4:
1129 /* We can only get to RAID4 from RAID5 */
1130 if (re->level != 4 && re->level != 5)
1131 return "Cannot convert to RAID4 from this level";
1132 if (info->delta_disks == UnSet)
1133 re->after.data_disks = re->before.data_disks;
1134 else
1135 re->after.data_disks =
1136 re->before.data_disks + info->delta_disks;
1137 switch (re->level) {
1138 case 4:
1139 re->after.layout = 0 ; break;
1140 case 5:
1141 re->after.layout = ALGORITHM_PARITY_N; break;
1142 }
1143 break;
1144
1145 case 5:
1146 /* We get to RAID5 for RAID5 or RAID6 */
1147 if (re->level != 5 && re->level != 6)
1148 return "Cannot convert to RAID5 from this level";
1149 if (info->delta_disks == UnSet)
1150 re->after.data_disks = re->before.data_disks;
1151 else if (re->level == 5)
1152 re->after.data_disks =
1153 re->before.data_disks + info->delta_disks;
1154 else
1155 re->after.data_disks =
1156 info->array.raid_disks + info->delta_disks - 1;
1157 switch (re->level) {
1158 case 5:
1159 if (info->new_layout == UnSet)
1160 re->after.layout = re->before.layout;
1161 else
1162 re->after.layout = info->new_layout;
1163 break;
1164 case 6:
1165 if (info->new_layout == UnSet)
1166 info->new_layout = re->before.layout;
1167
1168 /* after.layout needs to be raid6 version of new_layout */
1169 if (info->new_layout == ALGORITHM_PARITY_N)
1170 re->after.layout = ALGORITHM_PARITY_N;
1171 else {
1172 char layout[40];
1173 char *ls = map_num(r5layout, info->new_layout);
1174 int l;
1175 strcat(strcpy(layout, ls), "-6");
1176 l = map_name(r6layout, layout);
1177 if (l == UnSet)
1178 return "Cannot find RAID6 layout"
1179 " to convert to";
1180 re->after.layout = l;
1181 }
1182 }
1183 break;
1184
1185 case 6:
1186 /* We must already be at level 6 */
1187 if (re->level != 6)
1188 return "Impossible level change";
1189 if (info->delta_disks == UnSet)
1190 re->after.data_disks = re->before.data_disks;
1191 else
1192 re->after.data_disks = (info->array.raid_disks +
1193 info->delta_disks) - 2;
1194 if (info->new_layout == UnSet)
1195 re->after.layout = info->array.layout;
1196 else
1197 re->after.layout = info->new_layout;
1198 break;
1199 default:
1200 return "Impossible level change requested";
1201 }
1202 switch (re->level) {
1203 case 6: re->parity = 2; break;
1204 case 4:
1205 case 5: re->parity = 1; break;
1206 default: re->parity = 0; break;
1207 }
1208 /* So we have a restripe operation, we need to calculate the number
1209 * of blocks per reshape operation.
1210 */
1211 if (info->new_chunk == 0)
1212 info->new_chunk = info->array.chunk_size;
1213 if (re->after.data_disks == re->before.data_disks &&
1214 re->after.layout == re->before.layout &&
1215 info->new_chunk == info->array.chunk_size) {
1216 /* Nothing to change */
1217 re->backup_blocks = 0;
1218 return NULL;
1219 }
1220 if (re->after.data_disks == 1 && re->before.data_disks == 1) {
1221 /* chunks can layout changes make no difference */
1222 re->backup_blocks = 0;
1223 return NULL;
1224 }
1225
1226 if (re->after.data_disks == re->before.data_disks &&
1227 get_linux_version() < 2006032)
1228 return "in-place reshape is not safe before 2.6.32 - sorry.";
1229
1230 if (re->after.data_disks < re->before.data_disks &&
1231 get_linux_version() < 2006030)
1232 return "reshape to fewer devices is not supported before 2.6.32 - sorry.";
1233
1234 re->backup_blocks = compute_backup_blocks(
1235 info->new_chunk, info->array.chunk_size,
1236 re->after.data_disks,
1237 re->before.data_disks);
1238
1239 re->new_size = info->component_size * re->after.data_disks;
1240 return NULL;
1241 }
1242
1243 static int reshape_array(char *container, int fd, char *devname,
1244 struct supertype *st, struct mdinfo *info,
1245 int force, char *backup_file, int quiet, int forked);
1246 static int reshape_container(char *container, int cfd, char *devname,
1247 struct supertype *st,
1248 struct mdinfo *info,
1249 int force,
1250 char *backup_file,
1251 int quiet);
1252
1253 int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
1254 long long size,
1255 int level, char *layout_str, int chunksize, int raid_disks,
1256 int force)
1257 {
1258 /* Make some changes in the shape of an array.
1259 * The kernel must support the change.
1260 *
1261 * There are three different changes. Each can trigger
1262 * a resync or recovery so we freeze that until we have
1263 * requested everything (if kernel supports freezing - 2.6.30).
1264 * The steps are:
1265 * - change size (i.e. component_size)
1266 * - change level
1267 * - change layout/chunksize/ndisks
1268 *
1269 * The last can require a reshape. It is different on different
1270 * levels so we need to check the level before actioning it.
1271 * Some times the level change needs to be requested after the
1272 * reshape (e.g. raid6->raid5, raid5->raid0)
1273 *
1274 */
1275 struct mdu_array_info_s array;
1276 int rv = 0;
1277 struct supertype *st;
1278 char *subarray = NULL;
1279
1280 int frozen;
1281 int changed = 0;
1282 char *container = NULL;
1283 char container_buf[20];
1284 int cfd = -1;
1285
1286 struct mdinfo info;
1287 struct mdinfo *sra;
1288
1289 if (ioctl(fd, GET_ARRAY_INFO, &array) < 0) {
1290 fprintf(stderr, Name ": %s is not an active md array - aborting\n",
1291 devname);
1292 return 1;
1293 }
1294
1295 if (size >= 0 &&
1296 (chunksize || level!= UnSet || layout_str || raid_disks)) {
1297 fprintf(stderr, Name ": cannot change component size at the same time "
1298 "as other changes.\n"
1299 " Change size first, then check data is intact before "
1300 "making other changes.\n");
1301 return 1;
1302 }
1303
1304 if (raid_disks && raid_disks < array.raid_disks && array.level > 1 &&
1305 get_linux_version() < 2006032 &&
1306 !check_env("MDADM_FORCE_FEWER")) {
1307 fprintf(stderr, Name ": reducing the number of devices is not safe before Linux 2.6.32\n"
1308 " Please use a newer kernel\n");
1309 return 1;
1310 }
1311
1312 st = super_by_fd(fd, &subarray);
1313 if (!st) {
1314 fprintf(stderr, Name ": Unable to determine metadata format for %s\n", devname);
1315 return 1;
1316 }
1317 if (raid_disks > st->max_devs) {
1318 fprintf(stderr, Name ": Cannot increase raid-disks on this array"
1319 " beyond %d\n", st->max_devs);
1320 return 1;
1321 }
1322
1323 /* in the external case we need to check that the requested reshape is
1324 * supported, and perform an initial check that the container holds the
1325 * pre-requisite spare devices (mdmon owns final validation)
1326 */
1327 if (st->ss->external) {
1328 int container_dev;
1329 int rv;
1330
1331 if (subarray) {
1332 container_dev = st->container_dev;
1333 cfd = open_dev_excl(st->container_dev);
1334 } else {
1335 container_dev = st->devnum;
1336 close(fd);
1337 cfd = open_dev_excl(st->devnum);
1338 fd = cfd;
1339 }
1340 if (cfd < 0) {
1341 fprintf(stderr, Name ": Unable to open container for %s\n",
1342 devname);
1343 free(subarray);
1344 return 1;
1345 }
1346
1347 fmt_devname(container_buf, container_dev);
1348 container = container_buf;
1349
1350 rv = st->ss->load_container(st, cfd, NULL);
1351
1352 if (rv) {
1353 fprintf(stderr, Name ": Cannot read superblock for %s\n",
1354 devname);
1355 free(subarray);
1356 return 1;
1357 }
1358
1359 if (mdmon_running(container_dev))
1360 st->update_tail = &st->updates;
1361 }
1362
1363 if (raid_disks > array.raid_disks &&
1364 array.spare_disks < (raid_disks - array.raid_disks) &&
1365 !force) {
1366 fprintf(stderr,
1367 Name ": Need %d spare%s to avoid degraded array,"
1368 " and only have %d.\n"
1369 " Use --force to over-ride this check.\n",
1370 raid_disks - array.raid_disks,
1371 raid_disks - array.raid_disks == 1 ? "" : "s",
1372 array.spare_disks);
1373 return 1;
1374 }
1375
1376 sra = sysfs_read(fd, 0, GET_LEVEL | GET_DISKS | GET_DEVS
1377 | GET_STATE | GET_VERSION);
1378 if (sra) {
1379 if (st->ss->external && subarray == NULL) {
1380 array.level = LEVEL_CONTAINER;
1381 sra->array.level = LEVEL_CONTAINER;
1382 }
1383 } else {
1384 fprintf(stderr, Name ": failed to read sysfs parameters for %s\n",
1385 devname);
1386 return 1;
1387 }
1388 frozen = freeze(st);
1389 if (frozen < -1) {
1390 /* freeze() already spewed the reason */
1391 return 1;
1392 } else if (frozen < 0) {
1393 fprintf(stderr, Name ": %s is performing resync/recovery and cannot"
1394 " be reshaped\n", devname);
1395 return 1;
1396 }
1397
1398 /* ========= set size =============== */
1399 if (size >= 0 && (size == 0 || size != array.size)) {
1400 long long orig_size = array.size;
1401
1402 if (reshape_super(st, size, UnSet, UnSet, 0, 0, NULL, devname, !quiet)) {
1403 rv = 1;
1404 goto release;
1405 }
1406 sync_metadata(st);
1407 array.size = size;
1408 if (array.size != size) {
1409 /* got truncated to 32bit, write to
1410 * component_size instead
1411 */
1412 if (sra)
1413 rv = sysfs_set_num(sra, NULL,
1414 "component_size", size);
1415 else
1416 rv = -1;
1417 } else
1418 rv = ioctl(fd, SET_ARRAY_INFO, &array);
1419 if (rv != 0) {
1420 int err = errno;
1421
1422 /* restore metadata */
1423 if (reshape_super(st, orig_size, UnSet, UnSet, 0, 0,
1424 NULL, devname, !quiet) == 0)
1425 sync_metadata(st);
1426 fprintf(stderr, Name ": Cannot set device size for %s: %s\n",
1427 devname, strerror(err));
1428 if (err == EBUSY &&
1429 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1430 fprintf(stderr, " Bitmap must be removed before size can be changed\n");
1431 rv = 1;
1432 goto release;
1433 }
1434 ioctl(fd, GET_ARRAY_INFO, &array);
1435 size = get_component_size(fd)/2;
1436 if (size == 0)
1437 size = array.size;
1438 if (!quiet)
1439 fprintf(stderr, Name ": component size of %s has been set to %lluK\n",
1440 devname, size);
1441 changed = 1;
1442 } else if (array.level != LEVEL_CONTAINER) {
1443 size = get_component_size(fd)/2;
1444 if (size == 0)
1445 size = array.size;
1446 }
1447
1448 /* ========= check for Raid10 -> Raid0 conversion ===============
1449 * current implementation assumes that following conditions must be met:
1450 * - far_copies == 1
1451 * - near_copies == 2
1452 */
1453 if (level == 0 && array.level == 10 && sra &&
1454 array.layout == ((1 << 8) + 2) && !(array.raid_disks & 1)) {
1455 int err;
1456 err = remove_disks_on_raid10_to_raid0_takeover(st, sra, array.layout);
1457 if (err) {
1458 dprintf(Name": Array cannot be reshaped\n");
1459 if (cfd > -1)
1460 close(cfd);
1461 rv = 1;
1462 goto release;
1463 }
1464 }
1465
1466 info.array = array;
1467 sysfs_init(&info, fd, NoMdDev);
1468 strcpy(info.text_version, sra->text_version);
1469 info.component_size = size*2;
1470 info.new_level = level;
1471 info.new_chunk = chunksize * 1024;
1472 if (raid_disks)
1473 info.delta_disks = raid_disks - info.array.raid_disks;
1474 else
1475 info.delta_disks = UnSet;
1476 if (layout_str == NULL) {
1477 info.new_layout = UnSet;
1478 if (info.array.level == 6 &&
1479 (info.new_level == 6 || info.new_level == UnSet) &&
1480 info.array.layout >= 16) {
1481 fprintf(stderr, Name
1482 ": %s has a non-standard layout. If you"
1483 " wish to preserve this\n"
1484 " during the reshape, please specify"
1485 " --layout=preserve\n"
1486 " If you want to change it, specify a"
1487 " layout or use --layout=normalise\n",
1488 devname);
1489 rv = 1;
1490 goto release;
1491 }
1492 } else if (strcmp(layout_str, "normalise") == 0 ||
1493 strcmp(layout_str, "normalize") == 0) {
1494 /* If we have a -6 RAID6 layout, remove the '-6'. */
1495 info.new_layout = UnSet;
1496 if (info.array.level == 6 && info.new_level == UnSet) {
1497 char l[40], *h;
1498 strcpy(l, map_num(r6layout, info.array.layout));
1499 h = strrchr(l, '-');
1500 if (h && strcmp(h, "-6") == 0) {
1501 *h = 0;
1502 info.new_layout = map_name(r6layout, l);
1503 }
1504 }
1505 } else if (strcmp(layout_str, "preserve") == 0) {
1506 info.new_layout = UnSet;
1507 } else {
1508 int l = info.new_level;
1509 if (l == UnSet)
1510 l = info.array.level;
1511 switch (l) {
1512 case 5:
1513 info.new_layout = map_name(r5layout, layout_str);
1514 break;
1515 case 6:
1516 info.new_layout = map_name(r6layout, layout_str);
1517 break;
1518 case 10:
1519 info.new_layout = parse_layout_10(layout_str);
1520 break;
1521 case LEVEL_FAULTY:
1522 info.new_layout = parse_layout_faulty(layout_str);
1523 break;
1524 default:
1525 fprintf(stderr, Name ": layout not meaningful"
1526 " with this level\n");
1527 rv = 1;
1528 goto release;
1529 }
1530 if (info.new_layout == UnSet) {
1531 fprintf(stderr, Name ": layout %s not understood"
1532 " for this level\n",
1533 layout_str);
1534 rv = 1;
1535 goto release;
1536 }
1537 }
1538
1539 if (array.level == LEVEL_CONTAINER) {
1540 /* This change is to be applied to every array in the
1541 * container. This is only needed when the metadata imposes
1542 * restraints of the various arrays in the container.
1543 * Currently we only know that IMSM requires all arrays
1544 * to have the same number of devices so changing the
1545 * number of devices (On-Line Capacity Expansion) must be
1546 * performed at the level of the container
1547 */
1548 rv = reshape_container(container, fd, devname, st, &info,
1549 force, backup_file, quiet);
1550 frozen = 0;
1551 } else {
1552 /* Impose these changes on a single array. First
1553 * check that the metadata is OK with the change. */
1554
1555 if (reshape_super(st, info.component_size, info.new_level,
1556 info.new_layout, info.new_chunk,
1557 info.array.raid_disks + info.delta_disks,
1558 backup_file, devname, quiet)) {
1559 rv = 1;
1560 goto release;
1561 }
1562 sync_metadata(st);
1563 rv = reshape_array(container, fd, devname, st, &info, force,
1564 backup_file, quiet, 0);
1565 frozen = 0;
1566 }
1567 release:
1568 if (frozen > 0)
1569 unfreeze(st);
1570 return rv;
1571 }
1572
1573 static int reshape_array(char *container, int fd, char *devname,
1574 struct supertype *st, struct mdinfo *info,
1575 int force,
1576 char *backup_file, int quiet, int forked)
1577 {
1578 struct reshape reshape;
1579 int spares_needed;
1580 char *msg;
1581 int orig_level = UnSet;
1582 int disks, odisks;
1583
1584 struct mdu_array_info_s array;
1585 char *c;
1586 int rv = 0;
1587
1588 int *fdlist;
1589 unsigned long long *offsets;
1590 int d;
1591 int nrdisks;
1592 int err;
1593 unsigned long blocks;
1594 unsigned long cache;
1595 unsigned long long array_size;
1596 int done;
1597 struct mdinfo *sra;
1598
1599 msg = analyse_change(info, &reshape);
1600 if (msg) {
1601 fprintf(stderr, Name ": %s\n", msg);
1602 return 1;
1603 }
1604 if (ioctl(fd, GET_ARRAY_INFO, &array) != 0) {
1605 dprintf("Canot get array information.\n");
1606 return 1;
1607 }
1608 spares_needed = max(reshape.before.data_disks,
1609 reshape.after.data_disks)
1610 + reshape.parity - array.raid_disks;
1611
1612 if (!force &&
1613 info->new_level > 0 &&
1614 spares_needed > info->array.spare_disks) {
1615 fprintf(stderr,
1616 Name ": Need %d spare%s to avoid degraded array,"
1617 " and only have %d.\n"
1618 " Use --force to over-ride this check.\n",
1619 spares_needed,
1620 spares_needed == 1 ? "" : "s",
1621 info->array.spare_disks);
1622 return 1;
1623 }
1624
1625 if (reshape.level != info->array.level) {
1626 char *c = map_num(pers, reshape.level);
1627 int err;
1628 if (c == NULL)
1629 return 1; /* This should not be possible */
1630
1631 err = sysfs_set_str(info, NULL, "level", c);
1632 if (err) {
1633 err = errno;
1634 fprintf(stderr, Name ": %s: could not set level to %s\n",
1635 devname, c);
1636 if (err == EBUSY &&
1637 (info->array.state & (1<<MD_SB_BITMAP_PRESENT)))
1638 fprintf(stderr, " Bitmap must be removed"
1639 " before level can be changed\n");
1640 return 1;
1641 }
1642 if (!quiet)
1643 fprintf(stderr, Name ": level of %s changed to %s\n",
1644 devname, c);
1645 orig_level = info->array.level;
1646 }
1647
1648 if (reshape.level > 0 && st->ss->external &&
1649 !mdmon_running(st->container_dev)) {
1650 start_mdmon(st->container_dev);
1651 ping_monitor(container);
1652 }
1653
1654 /* ->reshape_super might have chosen some spares from the
1655 * container that it wants to be part of the new array.
1656 * We can collect them with ->container_content and give
1657 * them to the kernel.
1658 */
1659 if (st->ss->reshape_super && st->ss->container_content) {
1660 char *subarray = strchr(info->text_version+1, '/')+1;
1661 struct mdinfo *info2 =
1662 st->ss->container_content(st, subarray);
1663 struct mdinfo *d;
1664
1665 if (info2) {
1666 sysfs_init(info2, fd, st->devnum);
1667 for (d = info2->devs; d; d = d->next) {
1668 if (d->disk.state == 0 &&
1669 d->disk.raid_disk >= 0) {
1670 /* This is a spare that wants to
1671 * be part of the array.
1672 */
1673 add_disk(fd, st, info2, d);
1674 }
1675 }
1676 sysfs_free(info2);
1677 }
1678 }
1679
1680 if (reshape.backup_blocks == 0) {
1681 /* No restriping needed, but we might need to impose
1682 * some more changes: layout, raid_disks, chunk_size
1683 */
1684 if (info->new_layout != UnSet &&
1685 info->new_layout != info->array.layout) {
1686 info->array.layout = info->new_layout;
1687 if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) {
1688 fprintf(stderr, Name ": failed to set new layout\n");
1689 rv = 1;
1690 } else if (!quiet)
1691 printf("layout for %s set to %d\n",
1692 devname, info->array.layout);
1693 }
1694 if (info->delta_disks != UnSet &&
1695 info->delta_disks != 0) {
1696 info->array.raid_disks += info->delta_disks;
1697 if (ioctl(fd, SET_ARRAY_INFO, &info->array) != 0) {
1698 fprintf(stderr, Name ": failed to set raid disks\n");
1699 rv = 1;
1700 } else if (!quiet)
1701 printf("raid_disks for %s set to %d\n",
1702 devname, info->array.raid_disks);
1703 }
1704 if (info->new_chunk != 0 &&
1705 info->new_chunk != info->array.chunk_size) {
1706 if (sysfs_set_num(info, NULL,
1707 "chunk_size", info->new_chunk) != 0) {
1708 fprintf(stderr, Name ": failed to set chunk size\n");
1709 rv = 1;
1710 } else if (!quiet)
1711 printf("chunk size for %s set to %d\n",
1712 devname, info->array.chunk_size);
1713 }
1714
1715 return rv;
1716 }
1717
1718 /*
1719 * There are three possibilities.
1720 * 1/ The array will shrink.
1721 * We need to ensure the reshape will pause before reaching
1722 * the 'critical section'. We also need to fork and wait for
1723 * that to happen. When it does we
1724 * suspend/backup/complete/unfreeze
1725 *
1726 * 2/ The array will not change size.
1727 * This requires that we keep a backup of a sliding window
1728 * so that we can restore data after a crash. So we need
1729 * to fork and monitor progress.
1730 * In future we will allow the data_offset to change, so
1731 * a sliding backup becomes unnecessary.
1732 *
1733 * 3/ The array will grow. This is relatively easy.
1734 * However the kernel's restripe routines will cheerfully
1735 * overwrite some early data before it is safe. So we
1736 * need to make a backup of the early parts of the array
1737 * and be ready to restore it if rebuild aborts very early.
1738 * For externally managed metadata, we still need a forked
1739 * child to monitor the reshape and suspend IO over the region
1740 * that is being reshaped.
1741 *
1742 * We backup data by writing it to one spare, or to a
1743 * file which was given on command line.
1744 *
1745 * In each case, we first make sure that storage is available
1746 * for the required backup.
1747 * Then we:
1748 * - request the shape change.
1749 * - fork to handle backup etc.
1750 */
1751
1752 /* Check that we can hold all the data */
1753 get_dev_size(fd, NULL, &array_size);
1754 if (reshape.new_size < (array_size/512)) {
1755 fprintf(stderr,
1756 Name ": this change will reduce the size of the array.\n"
1757 " use --grow --array-size first to truncate array.\n"
1758 " e.g. mdadm --grow %s --array-size %llu\n",
1759 devname, reshape.new_size/2);
1760 rv = 1;
1761 goto release;
1762 }
1763
1764 sra = sysfs_read(fd, 0,
1765 GET_COMPONENT|GET_DEVS|GET_OFFSET|GET_STATE|GET_CHUNK|
1766 GET_CACHE);
1767
1768 if (!sra) {
1769 fprintf(stderr, Name ": %s: Cannot get array details from sysfs\n",
1770 devname);
1771 rv = 1;
1772 goto release;
1773 }
1774
1775 /* Decide how many blocks (sectors) for a reshape
1776 * unit. The number we have so far is just a minimum
1777 */
1778 blocks = reshape.backup_blocks;
1779 if (reshape.before.data_disks ==
1780 reshape.after.data_disks) {
1781 /* Make 'blocks' bigger for better throughput, but
1782 * not so big that we reject it below.
1783 * Try for 16 megabytes
1784 */
1785 while (blocks * 32 < sra->component_size &&
1786 blocks < 16*1024*2)
1787 blocks *= 2;
1788 } else
1789 fprintf(stderr, Name ": Need to backup %luK of critical "
1790 "section..\n", blocks/2);
1791
1792 if (blocks >= sra->component_size/2) {
1793 fprintf(stderr, Name ": %s: Something wrong"
1794 " - reshape aborted\n",
1795 devname);
1796 rv = 1;
1797 goto release;
1798 }
1799
1800 /* Now we need to open all these devices so we can read/write.
1801 */
1802 nrdisks = array.raid_disks + sra->array.spare_disks;
1803 fdlist = malloc((1+nrdisks) * sizeof(int));
1804 offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
1805 if (!fdlist || !offsets) {
1806 fprintf(stderr, Name ": malloc failed: grow aborted\n");
1807 rv = 1;
1808 goto release;
1809 }
1810
1811 d = reshape_prepare_fdlist(devname, sra, array.raid_disks,
1812 nrdisks, blocks, backup_file,
1813 fdlist, offsets);
1814 if (d < 0) {
1815 rv = 1;
1816 goto release;
1817 }
1818 if (backup_file == NULL) {
1819 if (reshape.after.data_disks <= reshape.before.data_disks) {
1820 fprintf(stderr,
1821 Name ": %s: Cannot grow - need backup-file\n",
1822 devname);
1823 rv = 1;
1824 goto release;
1825 } else if (sra->array.spare_disks == 0) {
1826 fprintf(stderr, Name ": %s: Cannot grow - need a spare or "
1827 "backup-file to backup critical section\n",
1828 devname);
1829 rv = 1;
1830 goto release;
1831 }
1832 } else {
1833 if (!reshape_open_backup_file(backup_file, fd, devname,
1834 (signed)blocks,
1835 fdlist+d, offsets+d)) {
1836 rv = 1;
1837 goto release;
1838 }
1839 d++;
1840 }
1841
1842 /* lastly, check that the internal stripe cache is
1843 * large enough, or it won't work.
1844 * It must hold at least 4 stripes of the larger
1845 * chunk size
1846 */
1847 cache = max(info->array.chunk_size, info->new_chunk);
1848 cache *= 4; /* 4 stripes minimum */
1849 cache /= 512; /* convert to sectors */
1850 disks = min(reshape.before.data_disks, reshape.after.data_disks);
1851 /* make sure there is room for 'blocks' with a bit to spare */
1852 if (cache < 16 + blocks / disks)
1853 cache = 16 + blocks / disks;
1854 cache /= (4096/512); /* Covert from sectors to pages */
1855
1856 if (sra->cache_size < cache)
1857 subarray_set_num(container, sra, "stripe_cache_size",
1858 cache+1);
1859
1860 /* Right, everything seems fine. Let's kick things off.
1861 * If only changing raid_disks, use ioctl, else use
1862 * sysfs.
1863 */
1864 sync_metadata(st);
1865
1866 sra->new_chunk = info->new_chunk;
1867
1868 if (info->array.chunk_size == info->new_chunk &&
1869 reshape.before.layout == reshape.after.layout &&
1870 st->ss->external == 0) {
1871 array.raid_disks = reshape.after.data_disks + reshape.parity;
1872 if (ioctl(fd, SET_ARRAY_INFO, &array) != 0) {
1873 int err = errno;
1874 rv = 1;
1875 fprintf(stderr,
1876 Name ": Cannot set device shape for %s: %s\n",
1877 devname, strerror(errno));
1878
1879 if (err == EBUSY &&
1880 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1881 fprintf(stderr,
1882 " Bitmap must be removed before"
1883 " shape can be changed\n");
1884
1885 goto release;
1886 }
1887 } else {
1888 /* set them all just in case some old 'new_*' value
1889 * persists from some earlier problem
1890 */
1891 int err = err; /* only used if rv==1, and always set if
1892 * rv==1, so initialisation not needed,
1893 * despite gcc warning
1894 */
1895 if (sysfs_set_num(sra, NULL, "chunk_size", info->new_chunk) < 0)
1896 rv = 1, err = errno;
1897 if (!rv && sysfs_set_num(sra, NULL, "layout",
1898 reshape.after.layout) < 0)
1899 rv = 1, err = errno;
1900 if (!rv && subarray_set_num(container, sra, "raid_disks",
1901 reshape.after.data_disks +
1902 reshape.parity) < 0)
1903 rv = 1, err = errno;
1904 if (rv) {
1905 fprintf(stderr, Name ": Cannot set device shape for %s\n",
1906 devname);
1907
1908 if (err == EBUSY &&
1909 (array.state & (1<<MD_SB_BITMAP_PRESENT)))
1910 fprintf(stderr,
1911 " Bitmap must be removed before"
1912 " shape can be changed\n");
1913 goto release;
1914 }
1915 }
1916
1917 start_reshape(sra);
1918
1919 /* Now we just need to kick off the reshape and watch, while
1920 * handling backups of the data...
1921 * This is all done by a forked background process.
1922 */
1923 switch(forked ? 0 : fork()) {
1924 case -1:
1925 fprintf(stderr, Name ": Cannot run child to monitor reshape: %s\n",
1926 strerror(errno));
1927 rv = 1;
1928 abort_reshape(sra);
1929 break;
1930 default:
1931 return 0;
1932 case 0:
1933 break;
1934 }
1935
1936 close(fd);
1937 if (check_env("MDADM_GROW_VERIFY"))
1938 fd = open(devname, O_RDONLY | O_DIRECT);
1939 else
1940 fd = -1;
1941 mlockall(MCL_FUTURE);
1942
1943 odisks = reshape.before.data_disks + reshape.parity;
1944
1945 if (st->ss->external) {
1946 /* metadata handler takes it from here */
1947 done = st->ss->manage_reshape(
1948 fd, sra, &reshape, st, blocks,
1949 fdlist, offsets,
1950 d - odisks, fdlist+odisks,
1951 offsets+odisks);
1952 } else
1953 done = child_monitor(
1954 fd, sra, &reshape, st, blocks,
1955 fdlist, offsets,
1956 d - odisks, fdlist+odisks,
1957 offsets+odisks);
1958
1959 if (backup_file && done)
1960 unlink(backup_file);
1961 if (!done) {
1962 abort_reshape(sra);
1963 goto out;
1964 }
1965
1966 if (!st->ss->external &&
1967 !(reshape.before.data_disks != reshape.after.data_disks
1968 && info->custom_array_size) &&
1969 info->new_level == reshape.level &&
1970 !forked) {
1971 /* no need to wait for the reshape to finish as
1972 * there is nothing more to do.
1973 */
1974 exit(0);
1975 }
1976 wait_reshape(sra);
1977
1978 if (st->ss->external) {
1979 /* Re-load the metadata as much could have changed */
1980 int cfd = open_dev(st->container_dev);
1981 if (cfd >= 0) {
1982 ping_monitor(container);
1983 st->ss->free_super(st);
1984 st->ss->load_container(st, cfd, container);
1985 close(cfd);
1986 }
1987 }
1988
1989 /* set new array size if required customer_array_size is used
1990 * by this metadata.
1991 */
1992 if (reshape.before.data_disks !=
1993 reshape.after.data_disks &&
1994 info->custom_array_size) {
1995 struct mdinfo *info2;
1996 char *subarray = strchr(info->text_version+1, '/')+1;
1997
1998 info2 = st->ss->container_content(st, subarray);
1999 if (info2) {
2000 unsigned long long current_size = 0;
2001 unsigned long long new_size =
2002 info2->custom_array_size/2;
2003
2004 if (sysfs_get_ll(sra,
2005 NULL,
2006 "array_size",
2007 &current_size) == 0 &&
2008 new_size > current_size) {
2009 if (sysfs_set_num(sra, NULL,
2010 "array_size", new_size)
2011 < 0)
2012 dprintf("Error: Cannot"
2013 " set array size");
2014 else
2015 dprintf("Array size "
2016 "changed");
2017 dprintf(" from %llu to %llu.\n",
2018 current_size, new_size);
2019 }
2020 sysfs_free(info2);
2021 }
2022 }
2023
2024 if (info->new_level != reshape.level) {
2025
2026 c = map_num(pers, info->new_level);
2027 if (c) {
2028 err = sysfs_set_str(sra, NULL, "level", c);
2029 if (err)
2030 fprintf(stderr, Name\
2031 ": %s: could not set level "
2032 "to %s\n", devname, c);
2033 }
2034 }
2035 out:
2036 if (forked)
2037 return 0;
2038 exit(0);
2039
2040 release:
2041 if (orig_level != UnSet && sra) {
2042 c = map_num(pers, orig_level);
2043 if (c && sysfs_set_str(sra, NULL, "level", c) == 0)
2044 fprintf(stderr, Name ": aborting level change\n");
2045 }
2046 if (!forked)
2047 unfreeze(st);
2048 return rv;
2049 }
2050
2051 int reshape_container(char *container, int cfd, char *devname,
2052 struct supertype *st,
2053 struct mdinfo *info,
2054 int force,
2055 char *backup_file,
2056 int quiet)
2057 {
2058 struct mdinfo *cc = NULL;
2059
2060 /* component_size is not meaningful for a container,
2061 * so pass '-1' meaning 'no change'
2062 */
2063 if (reshape_super(st, -1, 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 unfreeze(st);
2138 sysfs_free(cc);
2139 exit(0);
2140 }
2141
2142 /*
2143 * We run a child process in the background which performs the following
2144 * steps:
2145 * - wait for resync to reach a certain point
2146 * - suspend io to the following section
2147 * - backup that section
2148 * - allow resync to proceed further
2149 * - resume io
2150 * - discard the backup.
2151 *
2152 * When are combined in slightly different ways in the three cases.
2153 * Grow:
2154 * - suspend/backup/allow/wait/resume/discard
2155 * Shrink:
2156 * - allow/wait/suspend/backup/allow/wait/resume/discard
2157 * same-size:
2158 * - wait/resume/discard/suspend/backup/allow
2159 *
2160 * suspend/backup/allow always come together
2161 * wait/resume/discard do too.
2162 * For the same-size case we have two backups to improve flow.
2163 *
2164 */
2165
2166 int progress_reshape(struct mdinfo *info, struct reshape *reshape,
2167 unsigned long long backup_point,
2168 unsigned long long wait_point,
2169 unsigned long long *suspend_point,
2170 unsigned long long *reshape_completed)
2171 {
2172 /* This function is called repeatedly by the reshape manager.
2173 * It determines how much progress can safely be made and allows
2174 * that progress.
2175 * - 'info' identifies the array and particularly records in
2176 * ->reshape_progress the metadata's knowledge of progress
2177 * This is a sector offset from the start of the array
2178 * of the next array block to be relocated. This number
2179 * may increase from 0 or decrease from array_size, depending
2180 * on the type of reshape that is happening.
2181 * Note that in contrast, 'sync_completed' is a block count of the
2182 * reshape so far. It gives the distance between the start point
2183 * (head or tail of device) and the next place that data will be
2184 * written. It always increases.
2185 * - 'reshape' is the structure created by analyse_change
2186 * - 'backup_point' shows how much the metadata manager has backed-up
2187 * data. For reshapes with increasing progress, it is the next address
2188 * to be backed up, previous addresses have been backed-up. For
2189 * decreasing progress, it is the earliest address that has been
2190 * backed up - later address are also backed up.
2191 * So addresses between reshape_progress and backup_point are
2192 * backed up providing those are in the 'correct' order.
2193 * - 'wait_point' is an array address. When reshape_completed
2194 * passes this point, progress_reshape should return. It might
2195 * return earlier if it determines that ->reshape_progress needs
2196 * to be updated or further backup is needed.
2197 * - suspend_point is maintained by progress_reshape and the caller
2198 * should not touch it except to initialise to zero.
2199 * It is an array address and it only increases in 2.6.37 and earlier.
2200 * This makes it difficult to handle reducing reshapes with
2201 * external metadata.
2202 * However: it is similar to backup_point in that it records the
2203 * other end of a suspended region from reshape_progress.
2204 * it is moved to extend the region that is safe to backup and/or
2205 * reshape
2206 * - reshape_completed is read from sysfs and returned. The caller
2207 * should copy this into ->reshape_progress when it has reason to
2208 * believe that the metadata knows this, and any backup outside this
2209 * has been erased.
2210 *
2211 * Return value is:
2212 * 1 if more data from backup_point - but only as far as suspend_point,
2213 * should be backed up
2214 * 0 if things are progressing smoothly
2215 * -1 if the reshape is finished, either because it is all done,
2216 * or due to an error.
2217 */
2218
2219 int advancing = (reshape->after.data_disks
2220 >= reshape->before.data_disks);
2221 unsigned long long need_backup; /* need to eventually backup all the way
2222 * to here
2223 */
2224 unsigned long long read_offset, write_offset;
2225 unsigned long long write_range;
2226 unsigned long long max_progress, target, completed;
2227 unsigned long long array_size = (info->component_size
2228 * reshape->before.data_disks);
2229 int fd;
2230
2231 /* First, we unsuspend any region that is now known to be safe.
2232 * If suspend_point is on the 'wrong' side of reshape_progress, then
2233 * we don't have or need suspension at the moment. This is true for
2234 * native metadata when we don't need to back-up.
2235 */
2236 if (advancing) {
2237 if (info->reshape_progress <= *suspend_point)
2238 sysfs_set_num(info, NULL, "suspend_lo",
2239 info->reshape_progress);
2240 } else {
2241 /* Note: this won't work in 2.6.37 and before.
2242 * Something somewhere should make sure we don't need it!
2243 */
2244 if (info->reshape_progress >= *suspend_point)
2245 sysfs_set_num(info, NULL, "suspend_hi",
2246 info->reshape_progress);
2247 }
2248
2249 /* Now work out how far it is safe to progress.
2250 * If the read_offset for ->reshape_progress is less than
2251 * 'blocks' beyond the write_offset, we can only progress as far
2252 * as a backup.
2253 * Otherwise we can progress until the write_offset for the new location
2254 * reaches (within 'blocks' of) the read_offset at the current location.
2255 * However that region must be suspended unless we are using native
2256 * metadata.
2257 * If we need to suspend more, we limit it to 128M per device, which is
2258 * rather arbitrary and should be some time-based calculation.
2259 */
2260 read_offset = info->reshape_progress / reshape->before.data_disks;
2261 write_offset = info->reshape_progress / reshape->after.data_disks;
2262 write_range = info->new_chunk/512;
2263 if (advancing) {
2264 need_backup = 0;
2265 if (read_offset < write_offset + write_range) {
2266 max_progress = backup_point;
2267 if (reshape->before.data_disks == reshape->after.data_disks)
2268 need_backup = array_size;
2269 else
2270 need_backup = reshape->backup_blocks;
2271 } else {
2272 max_progress =
2273 read_offset *
2274 reshape->after.data_disks;
2275 }
2276 } else {
2277 need_backup = array_size;
2278 if (read_offset > write_offset - write_range) {
2279 max_progress = backup_point;
2280 if (max_progress >= info->reshape_progress)
2281 need_backup = 0;
2282 } else {
2283 max_progress =
2284 read_offset *
2285 reshape->after.data_disks;
2286 /* If we are using internal metadata, then we can
2287 * progress all the way to the suspend_point without
2288 * worrying about backing-up/suspending along the
2289 * way.
2290 */
2291 if (max_progress < *suspend_point &&
2292 info->array.major_version >= 0)
2293 max_progress = *suspend_point;
2294 }
2295 }
2296
2297 /* We know it is safe to progress to 'max_progress' providing
2298 * it is suspended or we are using native metadata.
2299 * Consider extending suspend_point 128M per device if it
2300 * is less than 64M per device beyond reshape_progress.
2301 * But always do a multiple of 'blocks'
2302 * FIXME this is too big - it takes to long to complete
2303 * this much.
2304 */
2305 target = 64*1024*2 * min(reshape->before.data_disks,
2306 reshape->after.data_disks);
2307 target /= reshape->backup_blocks;
2308 if (target < 2)
2309 target = 2;
2310 target *= reshape->backup_blocks;
2311
2312 /* For externally managed metadata we always need to suspend IO to
2313 * the area being reshaped so we regularly push suspend_point forward.
2314 * For native metadata we only need the suspend if we are going to do
2315 * a backup.
2316 */
2317 if (advancing) {
2318 if ((need_backup > info->reshape_progress
2319 || info->array.major_version < 0) &&
2320 *suspend_point < info->reshape_progress + target) {
2321 if (need_backup < *suspend_point + 2 * target)
2322 *suspend_point = need_backup;
2323 else if (*suspend_point + 2 * target < array_size)
2324 *suspend_point += 2 * target;
2325 else
2326 *suspend_point = array_size;
2327 sysfs_set_num(info, NULL, "suspend_hi", *suspend_point);
2328 if (max_progress > *suspend_point)
2329 max_progress = *suspend_point;
2330 }
2331 } else {
2332 if ((need_backup < info->reshape_progress
2333 || info->array.major_version < 0) &&
2334 *suspend_point > info->reshape_progress - target) {
2335 if (need_backup > *suspend_point - 2 * target)
2336 *suspend_point = need_backup;
2337 else if (*suspend_point >= 2 * target)
2338 *suspend_point -= 2 * target;
2339 else
2340 *suspend_point = 0;
2341 sysfs_set_num(info, NULL, "suspend_lo", *suspend_point);
2342 if (max_progress < *suspend_point)
2343 max_progress = *suspend_point;
2344 }
2345 }
2346
2347 /* now set sync_max to allow that progress. sync_max, like
2348 * sync_completed is a count of sectors written per device, so
2349 * we find the difference between max_progress and the start point,
2350 * and divide that by after.data_disks to get a sync_max
2351 * number.
2352 * At the same time we convert wait_point to a similar number
2353 * for comparing against sync_completed.
2354 */
2355 /* scale down max_progress to per_disk */
2356 max_progress /= reshape->after.data_disks;
2357 /* Round to chunk size as some kernels give an erroneously high number */
2358 max_progress /= info->new_chunk/512;
2359 max_progress *= info->new_chunk/512;
2360 /* Limit progress to the whole device */
2361 if (max_progress > info->component_size)
2362 max_progress = info->component_size;
2363 wait_point /= reshape->after.data_disks;
2364 if (!advancing) {
2365 /* switch from 'device offset' to 'processed block count' */
2366 max_progress = info->component_size - max_progress;
2367 wait_point = info->component_size - wait_point;
2368 }
2369
2370 sysfs_set_num(info, NULL, "sync_max", max_progress);
2371
2372 /* Now wait. If we have already reached the point that we were
2373 * asked to wait to, don't wait at all, else wait for any change.
2374 * We need to select on 'sync_completed' as that is the place that
2375 * notifications happen, but we are really interested in
2376 * 'reshape_position'
2377 */
2378 fd = sysfs_get_fd(info, NULL, "sync_completed");
2379 if (fd < 0)
2380 return -1;
2381
2382 if (sysfs_fd_get_ll(fd, &completed) < 0) {
2383 close(fd);
2384 return -1;
2385 }
2386 while (completed < max_progress && completed < wait_point) {
2387 /* Check that sync_action is still 'reshape' to avoid
2388 * waiting forever on a dead array
2389 */
2390 char action[20];
2391 fd_set rfds;
2392 if (sysfs_get_str(info, NULL, "sync_action",
2393 action, 20) <= 0 ||
2394 strncmp(action, "reshape", 7) != 0)
2395 break;
2396 FD_ZERO(&rfds);
2397 FD_SET(fd, &rfds);
2398 select(fd+1, NULL, NULL, &rfds, NULL);
2399 if (sysfs_fd_get_ll(fd, &completed) < 0) {
2400 close(fd);
2401 return -1;
2402 }
2403 }
2404 /* some kernels can give an incorrectly high 'completed' number */
2405 completed /= (info->new_chunk/512);
2406 completed *= (info->new_chunk/512);
2407 /* Convert 'completed' back in to a 'progress' number */
2408 completed *= reshape->after.data_disks;
2409 if (!advancing) {
2410 completed = info->component_size * reshape->after.data_disks
2411 - completed;
2412 }
2413 *reshape_completed = completed;
2414
2415 close(fd);
2416
2417 /* We return the need_backup flag. Caller will decide
2418 * how much - a multiple of ->backup_blocks up to *suspend_point
2419 */
2420 return advancing
2421 ? (need_backup > info->reshape_progress)
2422 : (need_backup < info->reshape_progress);
2423 }
2424
2425
2426 /* FIXME return status is never checked */
2427 static int grow_backup(struct mdinfo *sra,
2428 unsigned long long offset, /* per device */
2429 unsigned long stripes, /* per device, in old chunks */
2430 int *sources, unsigned long long *offsets,
2431 int disks, int chunk, int level, int layout,
2432 int dests, int *destfd, unsigned long long *destoffsets,
2433 int part, int *degraded,
2434 char *buf)
2435 {
2436 /* Backup 'blocks' sectors at 'offset' on each device of the array,
2437 * to storage 'destfd' (offset 'destoffsets'), after first
2438 * suspending IO. Then allow resync to continue
2439 * over the suspended section.
2440 * Use part 'part' of the backup-super-block.
2441 */
2442 int odata = disks;
2443 int rv = 0;
2444 int i;
2445 unsigned long long ll;
2446 int new_degraded;
2447 //printf("offset %llu\n", offset);
2448 if (level >= 4)
2449 odata--;
2450 if (level == 6)
2451 odata--;
2452
2453 /* Check that array hasn't become degraded, else we might backup the wrong data */
2454 sysfs_get_ll(sra, NULL, "degraded", &ll);
2455 new_degraded = (int)ll;
2456 if (new_degraded != *degraded) {
2457 /* check each device to ensure it is still working */
2458 struct mdinfo *sd;
2459 for (sd = sra->devs ; sd ; sd = sd->next) {
2460 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2461 continue;
2462 if (sd->disk.state & (1<<MD_DISK_SYNC)) {
2463 char sbuf[20];
2464 if (sysfs_get_str(sra, sd, "state", sbuf, 20) < 0 ||
2465 strstr(sbuf, "faulty") ||
2466 strstr(sbuf, "in_sync") == NULL) {
2467 /* this device is dead */
2468 sd->disk.state = (1<<MD_DISK_FAULTY);
2469 if (sd->disk.raid_disk >= 0 &&
2470 sources[sd->disk.raid_disk] >= 0) {
2471 close(sources[sd->disk.raid_disk]);
2472 sources[sd->disk.raid_disk] = -1;
2473 }
2474 }
2475 }
2476 }
2477 *degraded = new_degraded;
2478 }
2479 if (part) {
2480 bsb.arraystart2 = __cpu_to_le64(offset * odata);
2481 bsb.length2 = __cpu_to_le64(stripes * (chunk/512) * odata);
2482 } else {
2483 bsb.arraystart = __cpu_to_le64(offset * odata);
2484 bsb.length = __cpu_to_le64(stripes * (chunk/512) * odata);
2485 }
2486 if (part)
2487 bsb.magic[15] = '2';
2488 for (i = 0; i < dests; i++)
2489 if (part)
2490 lseek64(destfd[i], destoffsets[i] + __le64_to_cpu(bsb.devstart2)*512, 0);
2491 else
2492 lseek64(destfd[i], destoffsets[i], 0);
2493
2494 rv = save_stripes(sources, offsets,
2495 disks, chunk, level, layout,
2496 dests, destfd,
2497 offset*512*odata, stripes * chunk * odata,
2498 buf);
2499
2500 if (rv)
2501 return rv;
2502 bsb.mtime = __cpu_to_le64(time(0));
2503 for (i = 0; i < dests; i++) {
2504 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
2505
2506 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
2507 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
2508 bsb.sb_csum2 = bsb_csum((char*)&bsb,
2509 ((char*)&bsb.sb_csum2)-((char*)&bsb));
2510
2511 rv = -1;
2512 if ((unsigned long long)lseek64(destfd[i], destoffsets[i] - 4096, 0)
2513 != destoffsets[i] - 4096)
2514 break;
2515 if (write(destfd[i], &bsb, 512) != 512)
2516 break;
2517 if (destoffsets[i] > 4096) {
2518 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]+stripes*chunk*odata, 0) !=
2519 destoffsets[i]+stripes*chunk*odata)
2520 break;
2521 if (write(destfd[i], &bsb, 512) != 512)
2522 break;
2523 }
2524 fsync(destfd[i]);
2525 rv = 0;
2526 }
2527
2528 return rv;
2529 }
2530
2531 /* in 2.6.30, the value reported by sync_completed can be
2532 * less that it should be by one stripe.
2533 * This only happens when reshape hits sync_max and pauses.
2534 * So allow wait_backup to either extent sync_max further
2535 * than strictly necessary, or return before the
2536 * sync has got quite as far as we would really like.
2537 * This is what 'blocks2' is for.
2538 * The various caller give appropriate values so that
2539 * every works.
2540 */
2541 /* FIXME return value is often ignored */
2542 static int forget_backup(
2543 int dests, int *destfd, unsigned long long *destoffsets,
2544 int part)
2545 {
2546 /*
2547 * Erase backup 'part' (which is 0 or 1)
2548 */
2549 int i;
2550 int rv;
2551
2552 if (part) {
2553 bsb.arraystart2 = __cpu_to_le64(0);
2554 bsb.length2 = __cpu_to_le64(0);
2555 } else {
2556 bsb.arraystart = __cpu_to_le64(0);
2557 bsb.length = __cpu_to_le64(0);
2558 }
2559 bsb.mtime = __cpu_to_le64(time(0));
2560 rv = 0;
2561 for (i = 0; i < dests; i++) {
2562 bsb.devstart = __cpu_to_le64(destoffsets[i]/512);
2563 bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
2564 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0)
2565 bsb.sb_csum2 = bsb_csum((char*)&bsb,
2566 ((char*)&bsb.sb_csum2)-((char*)&bsb));
2567 if ((unsigned long long)lseek64(destfd[i], destoffsets[i]-4096, 0) !=
2568 destoffsets[i]-4096)
2569 rv = -1;
2570 if (rv == 0 &&
2571 write(destfd[i], &bsb, 512) != 512)
2572 rv = -1;
2573 fsync(destfd[i]);
2574 }
2575 return rv;
2576 }
2577
2578 static void fail(char *msg)
2579 {
2580 int rv;
2581 rv = (write(2, msg, strlen(msg)) != (int)strlen(msg));
2582 rv |= (write(2, "\n", 1) != 1);
2583 exit(rv ? 1 : 2);
2584 }
2585
2586 static char *abuf, *bbuf;
2587 static unsigned long long abuflen;
2588 static void validate(int afd, int bfd, unsigned long long offset)
2589 {
2590 /* check that the data in the backup against the array.
2591 * This is only used for regression testing and should not
2592 * be used while the array is active
2593 */
2594 if (afd < 0)
2595 return;
2596 lseek64(bfd, offset - 4096, 0);
2597 if (read(bfd, &bsb2, 512) != 512)
2598 fail("cannot read bsb");
2599 if (bsb2.sb_csum != bsb_csum((char*)&bsb2,
2600 ((char*)&bsb2.sb_csum)-((char*)&bsb2)))
2601 fail("first csum bad");
2602 if (memcmp(bsb2.magic, "md_backup_data", 14) != 0)
2603 fail("magic is bad");
2604 if (memcmp(bsb2.magic, "md_backup_data-2", 16) == 0 &&
2605 bsb2.sb_csum2 != bsb_csum((char*)&bsb2,
2606 ((char*)&bsb2.sb_csum2)-((char*)&bsb2)))
2607 fail("second csum bad");
2608
2609 if (__le64_to_cpu(bsb2.devstart)*512 != offset)
2610 fail("devstart is wrong");
2611
2612 if (bsb2.length) {
2613 unsigned long long len = __le64_to_cpu(bsb2.length)*512;
2614
2615 if (abuflen < len) {
2616 free(abuf);
2617 free(bbuf);
2618 abuflen = len;
2619 if (posix_memalign((void**)&abuf, 4096, abuflen) ||
2620 posix_memalign((void**)&bbuf, 4096, abuflen)) {
2621 abuflen = 0;
2622 /* just stop validating on mem-alloc failure */
2623 return;
2624 }
2625 }
2626
2627 lseek64(bfd, offset, 0);
2628 if ((unsigned long long)read(bfd, bbuf, len) != len) {
2629 //printf("len %llu\n", len);
2630 fail("read first backup failed");
2631 }
2632 lseek64(afd, __le64_to_cpu(bsb2.arraystart)*512, 0);
2633 if ((unsigned long long)read(afd, abuf, len) != len)
2634 fail("read first from array failed");
2635 if (memcmp(bbuf, abuf, len) != 0) {
2636 #if 0
2637 int i;
2638 printf("offset=%llu len=%llu\n",
2639 (unsigned long long)__le64_to_cpu(bsb2.arraystart)*512, len);
2640 for (i=0; i<len; i++)
2641 if (bbuf[i] != abuf[i]) {
2642 printf("first diff byte %d\n", i);
2643 break;
2644 }
2645 #endif
2646 fail("data1 compare failed");
2647 }
2648 }
2649 if (bsb2.length2) {
2650 unsigned long long len = __le64_to_cpu(bsb2.length2)*512;
2651
2652 if (abuflen < len) {
2653 free(abuf);
2654 free(bbuf);
2655 abuflen = len;
2656 abuf = malloc(abuflen);
2657 bbuf = malloc(abuflen);
2658 }
2659
2660 lseek64(bfd, offset+__le64_to_cpu(bsb2.devstart2)*512, 0);
2661 if ((unsigned long long)read(bfd, bbuf, len) != len)
2662 fail("read second backup failed");
2663 lseek64(afd, __le64_to_cpu(bsb2.arraystart2)*512, 0);
2664 if ((unsigned long long)read(afd, abuf, len) != len)
2665 fail("read second from array failed");
2666 if (memcmp(bbuf, abuf, len) != 0)
2667 fail("data2 compare failed");
2668 }
2669 }
2670
2671 int child_monitor(int afd, struct mdinfo *sra, struct reshape *reshape,
2672 struct supertype *st, unsigned long blocks,
2673 int *fds, unsigned long long *offsets,
2674 int dests, int *destfd, unsigned long long *destoffsets)
2675 {
2676 /* Monitor a reshape where backup is being performed using
2677 * 'native' mechanism - either to a backup file, or
2678 * to some space in a spare.
2679 */
2680 char *buf;
2681 int degraded = -1;
2682 unsigned long long speed;
2683 unsigned long long suspend_point, array_size;
2684 unsigned long long backup_point, wait_point;
2685 unsigned long long reshape_completed;
2686 int done = 0;
2687 int increasing = reshape->after.data_disks >= reshape->before.data_disks;
2688 int part = 0; /* The next part of the backup area to fill. It may already
2689 * be full, so we need to check */
2690 int level = reshape->level;
2691 int layout = reshape->before.layout;
2692 int data = reshape->before.data_disks;
2693 int disks = reshape->before.data_disks + reshape->parity;
2694 int chunk = sra->array.chunk_size;
2695 struct mdinfo *sd;
2696 unsigned long stripes;
2697
2698 /* set up the backup-super-block. This requires the
2699 * uuid from the array.
2700 */
2701 /* Find a superblock */
2702 for (sd = sra->devs; sd; sd = sd->next) {
2703 char *dn;
2704 int devfd;
2705 int ok;
2706 if (sd->disk.state & (1<<MD_DISK_FAULTY))
2707 continue;
2708 dn = map_dev(sd->disk.major, sd->disk.minor, 1);
2709 devfd = dev_open(dn, O_RDONLY);
2710 if (devfd < 0)
2711 continue;
2712 ok = st->ss->load_super(st, devfd, NULL);
2713 close(devfd);
2714 if (ok >= 0)
2715 break;
2716 }
2717 if (!sd) {
2718 fprintf(stderr, Name ": Cannot find a superblock\n");
2719 return 0;
2720 }
2721
2722 memset(&bsb, 0, 512);
2723 memcpy(bsb.magic, "md_backup_data-1", 16);
2724 st->ss->uuid_from_super(st, (int*)&bsb.set_uuid);
2725 bsb.mtime = __cpu_to_le64(time(0));
2726 bsb.devstart2 = blocks;
2727
2728 stripes = blocks / (sra->array.chunk_size/512) /
2729 reshape->before.data_disks;
2730
2731 if (posix_memalign((void**)&buf, 4096, disks * chunk))
2732 /* Don't start the 'reshape' */
2733 return 0;
2734 if (reshape->before.data_disks == reshape->after.data_disks) {
2735 sysfs_get_ll(sra, NULL, "sync_speed_min", &speed);
2736 sysfs_set_num(sra, NULL, "sync_speed_min", 200000);
2737 }
2738
2739 if (increasing) {
2740 array_size = sra->component_size * reshape->after.data_disks;
2741 backup_point = sra->reshape_progress;
2742 suspend_point = 0;
2743 } else {
2744 array_size = sra->component_size * reshape->before.data_disks;
2745 backup_point = array_size;
2746 suspend_point = array_size;
2747 }
2748
2749 while (!done) {
2750 int rv;
2751
2752 /* Want to return as soon the oldest backup slot can
2753 * be released as that allows us to start backing up
2754 * some more, providing suspend_point has been
2755 * advanced, which it should have.
2756 */
2757 if (increasing) {
2758 wait_point = array_size;
2759 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
2760 wait_point = (__le64_to_cpu(bsb.arraystart) +
2761 __le64_to_cpu(bsb.length));
2762 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
2763 wait_point = (__le64_to_cpu(bsb.arraystart2) +
2764 __le64_to_cpu(bsb.length2));
2765 } else {
2766 wait_point = 0;
2767 if (part == 0 && __le64_to_cpu(bsb.length) > 0)
2768 wait_point = __le64_to_cpu(bsb.arraystart);
2769 if (part == 1 && __le64_to_cpu(bsb.length2) > 0)
2770 wait_point = __le64_to_cpu(bsb.arraystart2);
2771 }
2772
2773 rv = progress_reshape(sra, reshape,
2774 backup_point, wait_point,
2775 &suspend_point, &reshape_completed);
2776 /* external metadata would need to ping_monitor here */
2777 sra->reshape_progress = reshape_completed;
2778
2779 /* Clear any backup region that is before 'here' */
2780 if (increasing) {
2781 if (reshape_completed >= (__le64_to_cpu(bsb.arraystart) +
2782 __le64_to_cpu(bsb.length)))
2783 forget_backup(dests, destfd,
2784 destoffsets, 0);
2785 if (reshape_completed >= (__le64_to_cpu(bsb.arraystart2) +
2786 __le64_to_cpu(bsb.length2)))
2787 forget_backup(dests, destfd,
2788 destoffsets, 1);
2789 } else {
2790 if (reshape_completed <= (__le64_to_cpu(bsb.arraystart)))
2791 forget_backup(dests, destfd,
2792 destoffsets, 0);
2793 if (reshape_completed <= (__le64_to_cpu(bsb.arraystart2)))
2794 forget_backup(dests, destfd,
2795 destoffsets, 1);
2796 }
2797
2798 if (rv < 0) {
2799 done = 1;
2800 break;
2801 }
2802
2803 while (rv) {
2804 unsigned long long offset;
2805 unsigned long actual_stripes;
2806 /* Need to backup some data.
2807 * If 'part' is not used and the desired
2808 * backup size is suspended, do a backup,
2809 * then consider the next part.
2810 */
2811 /* Check that 'part' is unused */
2812 if (part == 0 && __le64_to_cpu(bsb.length) != 0)
2813 break;
2814 if (part == 1 && __le64_to_cpu(bsb.length2) != 0)
2815 break;
2816
2817 offset = backup_point / data;
2818 actual_stripes = stripes;
2819 if (increasing) {
2820 if (offset + actual_stripes * (chunk/512) >
2821 sra->component_size)
2822 actual_stripes = ((sra->component_size - offset)
2823 / (chunk/512));
2824 if (offset + actual_stripes * (chunk/512) >
2825 suspend_point/data)
2826 break;
2827 } else {
2828 if (offset < actual_stripes * (chunk/512))
2829 actual_stripes = offset / (chunk/512);
2830 offset -= actual_stripes * (chunk/512);
2831 if (offset < suspend_point/data)
2832 break;
2833 }
2834 grow_backup(sra, offset, actual_stripes,
2835 fds, offsets,
2836 disks, chunk, level, layout,
2837 dests, destfd, destoffsets,
2838 part, &degraded, buf);
2839 validate(afd, destfd[0], destoffsets[0]);
2840 /* record where 'part' is up to */
2841 part = !part;
2842 if (increasing)
2843 backup_point += actual_stripes * (chunk/512) * data;
2844 else
2845 backup_point -= actual_stripes * (chunk/512) * data;
2846 }
2847 }
2848
2849 /* FIXME maybe call progress_reshape one more time instead */
2850 abort_reshape(sra); /* remove any remaining suspension */
2851 if (reshape->before.data_disks == reshape->after.data_disks)
2852 sysfs_set_num(sra, NULL, "sync_speed_min", speed);
2853 free(buf);
2854 return 1; /* FIXME what does this mean? */
2855 }
2856
2857 /*
2858 * If any spare contains md_back_data-1 which is recent wrt mtime,
2859 * write that data into the array and update the super blocks with
2860 * the new reshape_progress
2861 */
2862 int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt,
2863 char *backup_file, int verbose)
2864 {
2865 int i, j;
2866 int old_disks;
2867 unsigned long long *offsets;
2868 unsigned long long nstripe, ostripe;
2869 int ndata, odata;
2870
2871 if (info->new_level != info->array.level)
2872 return 1; /* Cannot handle level changes (they are instantaneous) */
2873
2874 odata = info->array.raid_disks - info->delta_disks - 1;
2875 if (info->array.level == 6) odata--; /* number of data disks */
2876 ndata = info->array.raid_disks - 1;
2877 if (info->new_level == 6) ndata--;
2878
2879 old_disks = info->array.raid_disks - info->delta_disks;
2880
2881 if (info->delta_disks <= 0)
2882 /* Didn't grow, so the backup file must have
2883 * been used
2884 */
2885 old_disks = cnt;
2886 for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
2887 struct mdinfo dinfo;
2888 int fd;
2889 int bsbsize;
2890 char *devname, namebuf[20];
2891
2892 /* This was a spare and may have some saved data on it.
2893 * Load the superblock, find and load the
2894 * backup_super_block.
2895 * If either fail, go on to next device.
2896 * If the backup contains no new info, just return
2897 * else restore data and update all superblocks
2898 */
2899 if (i == old_disks-1) {
2900 fd = open(backup_file, O_RDONLY);
2901 if (fd<0) {
2902 fprintf(stderr, Name ": backup file %s inaccessible: %s\n",
2903 backup_file, strerror(errno));
2904 continue;
2905 }
2906 devname = backup_file;
2907 } else {
2908 fd = fdlist[i];
2909 if (fd < 0)
2910 continue;
2911 if (st->ss->load_super(st, fd, NULL))
2912 continue;
2913
2914 st->ss->getinfo_super(st, &dinfo, NULL);
2915 st->ss->free_super(st);
2916
2917 if (lseek64(fd,
2918 (dinfo.data_offset + dinfo.component_size - 8) <<9,
2919 0) < 0) {
2920 fprintf(stderr, Name ": Cannot seek on device %d\n", i);
2921 continue; /* Cannot seek */
2922 }
2923 sprintf(namebuf, "device-%d", i);
2924 devname = namebuf;
2925 }
2926 if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb)) {
2927 if (verbose)
2928 fprintf(stderr, Name ": Cannot read from %s\n", devname);
2929 continue; /* Cannot read */
2930 }
2931 if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0 &&
2932 memcmp(bsb.magic, "md_backup_data-2", 16) != 0) {
2933 if (verbose)
2934 fprintf(stderr, Name ": No backup metadata on %s\n", devname);
2935 continue;
2936 }
2937 if (bsb.sb_csum != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb))) {
2938 if (verbose)
2939 fprintf(stderr, Name ": Bad backup-metadata checksum on %s\n", devname);
2940 continue; /* bad checksum */
2941 }
2942 if (memcmp(bsb.magic, "md_backup_data-2", 16) == 0 &&
2943 bsb.sb_csum2 != bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum2)-((char*)&bsb))) {
2944 if (verbose)
2945 fprintf(stderr, Name ": Bad backup-metadata checksum2 on %s\n", devname);
2946 continue; /* Bad second checksum */
2947 }
2948 if (memcmp(bsb.set_uuid,info->uuid, 16) != 0) {
2949 if (verbose)
2950 fprintf(stderr, Name ": Wrong uuid on backup-metadata on %s\n", devname);
2951 continue; /* Wrong uuid */
2952 }
2953
2954 /* array utime and backup-mtime should be updated at much the same time, but it seems that
2955 * sometimes they aren't... So allow considerable flexability in matching, and allow
2956 * this test to be overridden by an environment variable.
2957 */
2958 if (info->array.utime > (int)__le64_to_cpu(bsb.mtime) + 2*60*60 ||
2959 info->array.utime < (int)__le64_to_cpu(bsb.mtime) - 10*60) {
2960 if (check_env("MDADM_GROW_ALLOW_OLD")) {
2961 fprintf(stderr, Name ": accepting backup with timestamp %lu "
2962 "for array with timestamp %lu\n",
2963 (unsigned long)__le64_to_cpu(bsb.mtime),
2964 (unsigned long)info->array.utime);
2965 } else {
2966 if (verbose)
2967 fprintf(stderr, Name ": too-old timestamp on "
2968 "backup-metadata on %s\n", devname);
2969 continue; /* time stamp is too bad */
2970 }
2971 }
2972
2973 if (bsb.magic[15] == '1') {
2974 if (info->delta_disks >= 0) {
2975 /* reshape_progress is increasing */
2976 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2977 info->reshape_progress) {
2978 nonew:
2979 if (verbose)
2980 fprintf(stderr, Name ": backup-metadata found on %s but is not needed\n", devname);
2981 continue; /* No new data here */
2982 }
2983 } else {
2984 /* reshape_progress is decreasing */
2985 if (__le64_to_cpu(bsb.arraystart) >=
2986 info->reshape_progress)
2987 goto nonew; /* No new data here */
2988 }
2989 } else {
2990 if (info->delta_disks >= 0) {
2991 /* reshape_progress is increasing */
2992 if (__le64_to_cpu(bsb.arraystart) + __le64_to_cpu(bsb.length) <
2993 info->reshape_progress &&
2994 __le64_to_cpu(bsb.arraystart2) + __le64_to_cpu(bsb.length2) <
2995 info->reshape_progress)
2996 goto nonew; /* No new data here */
2997 } else {
2998 /* reshape_progress is decreasing */
2999 if (__le64_to_cpu(bsb.arraystart) >=
3000 info->reshape_progress &&
3001 __le64_to_cpu(bsb.arraystart2) >=
3002 info->reshape_progress)
3003 goto nonew; /* No new data here */
3004 }
3005 }
3006 if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0) {
3007 second_fail:
3008 if (verbose)
3009 fprintf(stderr, Name ": Failed to verify secondary backup-metadata block on %s\n",
3010 devname);
3011 continue; /* Cannot seek */
3012 }
3013 /* There should be a duplicate backup superblock 4k before here */
3014 if (lseek64(fd, -4096, 1) < 0 ||
3015 read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
3016 goto second_fail; /* Cannot find leading superblock */
3017 if (bsb.magic[15] == '1')
3018 bsbsize = offsetof(struct mdp_backup_super, pad1);
3019 else
3020 bsbsize = offsetof(struct mdp_backup_super, pad);
3021 if (memcmp(&bsb2, &bsb, bsbsize) != 0)
3022 goto second_fail; /* Cannot find leading superblock */
3023
3024 /* Now need the data offsets for all devices. */
3025 offsets = malloc(sizeof(*offsets)*info->array.raid_disks);
3026 for(j=0; j<info->array.raid_disks; j++) {
3027 if (fdlist[j] < 0)
3028 continue;
3029 if (st->ss->load_super(st, fdlist[j], NULL))
3030 /* FIXME should be this be an error */
3031 continue;
3032 st->ss->getinfo_super(st, &dinfo, NULL);
3033 st->ss->free_super(st);
3034 offsets[j] = dinfo.data_offset * 512;
3035 }
3036 printf(Name ": restoring critical section\n");
3037
3038 if (restore_stripes(fdlist, offsets,
3039 info->array.raid_disks,
3040 info->new_chunk,
3041 info->new_level,
3042 info->new_layout,
3043 fd, __le64_to_cpu(bsb.devstart)*512,
3044 __le64_to_cpu(bsb.arraystart)*512,
3045 __le64_to_cpu(bsb.length)*512)) {
3046 /* didn't succeed, so giveup */
3047 if (verbose)
3048 fprintf(stderr, Name ": Error restoring backup from %s\n",
3049 devname);
3050 return 1;
3051 }
3052
3053 if (bsb.magic[15] == '2' &&
3054 restore_stripes(fdlist, offsets,
3055 info->array.raid_disks,
3056 info->new_chunk,
3057 info->new_level,
3058 info->new_layout,
3059 fd, __le64_to_cpu(bsb.devstart)*512 +
3060 __le64_to_cpu(bsb.devstart2)*512,
3061 __le64_to_cpu(bsb.arraystart2)*512,
3062 __le64_to_cpu(bsb.length2)*512)) {
3063 /* didn't succeed, so giveup */
3064 if (verbose)
3065 fprintf(stderr, Name ": Error restoring second backup from %s\n",
3066 devname);
3067 return 1;
3068 }
3069
3070
3071 /* Ok, so the data is restored. Let's update those superblocks. */
3072
3073 if (info->delta_disks >= 0) {
3074 info->reshape_progress = __le64_to_cpu(bsb.arraystart) +
3075 __le64_to_cpu(bsb.length);
3076 if (bsb.magic[15] == '2') {
3077 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2) +
3078 __le64_to_cpu(bsb.length2);
3079 if (p2 > info->reshape_progress)
3080 info->reshape_progress = p2;
3081 }
3082 } else {
3083 info->reshape_progress = __le64_to_cpu(bsb.arraystart);
3084 if (bsb.magic[15] == '2') {
3085 unsigned long long p2 = __le64_to_cpu(bsb.arraystart2);
3086 if (p2 < info->reshape_progress)
3087 info->reshape_progress = p2;
3088 }
3089 }
3090 for (j=0; j<info->array.raid_disks; j++) {
3091 if (fdlist[j] < 0) continue;
3092 if (st->ss->load_super(st, fdlist[j], NULL))
3093 continue;
3094 st->ss->getinfo_super(st, &dinfo, NULL);
3095 dinfo.reshape_progress = info->reshape_progress;
3096 st->ss->update_super(st, &dinfo,
3097 "_reshape_progress",
3098 NULL,0, 0, NULL);
3099 st->ss->store_super(st, fdlist[j]);
3100 st->ss->free_super(st);
3101 }
3102 return 0;
3103 }
3104 /* Didn't find any backup data, try to see if any
3105 * was needed.
3106 */
3107 if (info->delta_disks < 0) {
3108 /* When shrinking, the critical section is at the end.
3109 * So see if we are before the critical section.
3110 */
3111 unsigned long long first_block;
3112 nstripe = ostripe = 0;
3113 first_block = 0;
3114 while (ostripe >= nstripe) {
3115 ostripe += info->array.chunk_size / 512;
3116 first_block = ostripe * odata;
3117 nstripe = first_block / ndata / (info->new_chunk/512) *
3118 (info->new_chunk/512);
3119 }
3120
3121 if (info->reshape_progress >= first_block)
3122 return 0;
3123 }
3124 if (info->delta_disks > 0) {
3125 /* See if we are beyond the critical section. */
3126 unsigned long long last_block;
3127 nstripe = ostripe = 0;
3128 last_block = 0;
3129 while (nstripe >= ostripe) {
3130 nstripe += info->new_chunk / 512;
3131 last_block = nstripe * ndata;
3132 ostripe = last_block / odata / (info->array.chunk_size/512) *
3133 (info->array.chunk_size/512);
3134 }
3135
3136 if (info->reshape_progress >= last_block)
3137 return 0;
3138 }
3139 /* needed to recover critical section! */
3140 if (verbose)
3141 fprintf(stderr, Name ": Failed to find backup of critical section\n");
3142 return 1;
3143 }
3144
3145 int Grow_continue(int mdfd, struct supertype *st, struct mdinfo *info,
3146 char *backup_file)
3147 {
3148 int err = sysfs_set_str(info, NULL, "array_state", "readonly");
3149 if (err)
3150 return err;
3151 return reshape_array(NULL, mdfd, "array", st, info, 1, backup_file, 0, 0);
3152 }
3153
3154
Unnamed repository; edit this file 'description' to name the repository.