]>
git.ipfire.org Git - thirdparty/mdadm.git/blob - Create.c
1099dcbafab33199dfbc9c49f0864003e58105b8
2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
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.
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.
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
22 * Email: <neilb@cse.unsw.edu.au>
24 * School of Computer Science and Engineering
25 * The University of New South Wales
35 int Create(struct supertype
*st
, char *mddev
, int mdfd
,
36 int chunk
, int level
, int layout
, unsigned long long size
, int raiddisks
, int sparedisks
,
37 char *name
, char *homehost
, int *uuid
,
38 int subdevs
, mddev_dev_t devlist
,
39 int runstop
, int verbose
, int force
, int assume_clean
,
40 char *bitmap_file
, int bitmap_chunk
, int write_behind
, int delay
)
43 * Create a new raid array.
45 * First check that necessary details are available
46 * (i.e. level, raid-disks)
48 * Then check each disk to see what might be on it
49 * and report anything interesting.
51 * If anything looks odd, and runstop not set,
54 * SET_ARRAY_INFO and ADD_NEW_DISK, and
55 * if runstop==run, or raiddisks disks were used,
58 unsigned long long minsize
=0, maxsize
=0;
65 int first_missing
= subdevs
* 2;
66 int second_missing
= subdevs
* 2;
67 int missing_disks
= 0;
68 int insert_point
= subdevs
* 2; /* where to insert a missing drive */
73 unsigned long long bitmapsize
;
77 int major_num
= BITMAP_MAJOR_HI
;
79 memset(&info
, 0, sizeof(info
));
81 vers
= md_get_version(mdfd
);
83 fprintf(stderr
, Name
": Create requires md driver version 0.90.0 or later\n");
87 memset(&inf
, 0, sizeof(inf
));
88 ioctl(mdfd
, GET_ARRAY_INFO
, &inf
);
89 if (inf
.working_disks
!= 0) {
90 fprintf(stderr
, Name
": another array by this name"
91 " is already running.\n");
96 /* "ddf" metadata only supports one level - should possibly
97 * push this into metadata handler??
99 if (st
&& st
->ss
== &super_ddf
)
100 level
= LEVEL_CONTAINER
;
103 if (level
== UnSet
) {
105 Name
": a RAID level is needed to create an array.\n");
108 if (raiddisks
< 4 && level
== 6) {
110 Name
": at least 4 raid-devices needed for level 6\n");
113 if (raiddisks
> 256 && level
== 6) {
115 Name
": no more than 256 raid-devices supported for level 6\n");
118 if (raiddisks
< 2 && level
>= 4) {
120 Name
": at least 2 raid-devices needed for level 4 or 5\n");
123 if (level
<= 0 && sparedisks
) {
125 Name
": This level does not support spare devices\n");
128 if (subdevs
> raiddisks
+sparedisks
) {
129 fprintf(stderr
, Name
": You have listed more devices (%d) than are in the array(%d)!\n", subdevs
, raiddisks
+sparedisks
);
132 if (subdevs
< raiddisks
+sparedisks
) {
133 fprintf(stderr
, Name
": You haven't given enough devices (real or missing) to create this array\n");
136 if (bitmap_file
&& level
<= 0) {
137 fprintf(stderr
, Name
": bitmaps not meaningful with level %s\n",
138 map_num(pers
, level
)?:"given");
142 /* now set some defaults */
145 default: /* no layout */
149 layout
= 0x102; /* near=2, far=1 */
152 Name
": layout defaults to n1\n");
156 layout
= map_name(r5layout
, "default");
159 Name
": layout defaults to %s\n", map_num(r5layout
, layout
));
162 layout
= map_name(faultylayout
, "default");
166 Name
": layout defaults to %s\n", map_num(faultylayout
, layout
));
171 /* check layout fits in array*/
172 if ((layout
&255) * ((layout
>>8)&255) > raiddisks
) {
173 fprintf(stderr
, Name
": that layout requires at least %d devices\n",
174 (layout
&255) * ((layout
>>8)&255));
184 case LEVEL_LINEAR
: /* linear */
188 fprintf(stderr
, Name
": chunk size defaults to 64K\n");
193 case LEVEL_MULTIPATH
:
194 case LEVEL_CONTAINER
:
198 fprintf(stderr
, Name
": chunk size ignored for this level\n");
202 fprintf(stderr
, Name
": unknown level %d\n", level
);
206 if (st
&& ! st
->ss
->validate_geometry(st
, level
, layout
, raiddisks
,
207 chunk
, size
, NULL
, NULL
))
210 /* now look at the subdevs */
211 info
.array
.active_disks
= 0;
212 info
.array
.working_disks
= 0;
214 for (dv
=devlist
; dv
; dv
=dv
->next
, dnum
++) {
215 char *dname
= dv
->devname
;
216 unsigned long long freesize
;
217 if (strcasecmp(dname
, "missing")==0) {
218 if (first_missing
> dnum
)
219 first_missing
= dnum
;
220 if (second_missing
> dnum
&& dnum
> first_missing
)
221 second_missing
= dnum
;
225 info
.array
.working_disks
++;
226 if (dnum
< raiddisks
)
227 info
.array
.active_disks
++;
229 struct createinfo
*ci
= conf_get_create_info();
234 /* Need to choose a default metadata, which is different
235 * depending on geometry of array.
238 char *name
= "default";
239 for(i
=0; !st
&& superlist
[i
]; i
++) {
240 st
= superlist
[i
]->match_metadata_desc(name
);
241 if (st
&& !st
->ss
->validate_geometry
242 (st
, level
, layout
, raiddisks
,
243 chunk
, size
, dname
, &freesize
))
248 fprintf(stderr
, Name
": device %s not suitable "
249 "for any style of array\n",
253 if (st
->ss
->major
!= 0 ||
254 st
->minor_version
!= 90)
255 fprintf(stderr
, Name
": Defaulting to version"
260 if (!st
->ss
->validate_geometry(st
, level
, layout
,
266 Name
": %s is not suitable for "
274 freesize
/= 2; /* convert to K */
276 /* round to chunk size */
277 freesize
= freesize
& ~(chunk
-1);
280 if (size
&& freesize
< size
) {
281 fprintf(stderr
, Name
": %s is smaller that given size."
282 " %lluK < %lluK + metadata\n",
283 dname
, freesize
, size
);
287 if (maxdisc
== NULL
|| (maxdisc
&& freesize
> maxsize
)) {
291 if (mindisc
==NULL
|| (mindisc
&& freesize
< minsize
)) {
295 if (runstop
!= 1 || verbose
>= 0) {
296 int fd
= open(dname
, O_RDONLY
, 0);
298 fprintf(stderr
, Name
": Cannot open %s: %s\n",
299 dname
, strerror(errno
));
303 warn
|= check_ext2(fd
, dname
);
304 warn
|= check_reiser(fd
, dname
);
305 warn
|= check_raid(fd
, dname
);
310 fprintf(stderr
, Name
": create aborted\n");
314 if (mindisc
== NULL
) {
315 fprintf(stderr
, Name
": no size and no drives given - aborting create.\n");
318 if (level
> 0 || level
== LEVEL_MULTIPATH
|| level
== LEVEL_FAULTY
) {
319 /* size is meaningful */
320 if (minsize
> 0x100000000ULL
&& st
->ss
->major
== 0) {
321 fprintf(stderr
, Name
": devices too large for RAID level %d\n", level
);
326 fprintf(stderr
, Name
": size set to %lluK\n", size
);
329 if (level
> 0 && ((maxsize
-size
)*100 > maxsize
)) {
330 if (runstop
!= 1 || verbose
>= 0)
331 fprintf(stderr
, Name
": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
338 if (!ask("Continue creating array? ")) {
339 fprintf(stderr
, Name
": create aborted.\n");
344 fprintf(stderr
, Name
": creation continuing despite oddities due to --run\n");
348 /* If this is raid4/5, we want to configure the last active slot
349 * as missing, so that a reconstruct happens (faster than re-parity)
350 * FIX: Can we do this for raid6 as well?
352 if (assume_clean
==0 && force
== 0 && first_missing
>= raiddisks
) {
356 insert_point
= raiddisks
-1;
358 info
.array
.active_disks
--;
365 /* For raid6, if creating with 1 missing drive, make a good drive
366 * into a spare, else the create will fail
368 if (assume_clean
== 0 && force
== 0 && first_missing
< raiddisks
&&
369 second_missing
>= raiddisks
&& level
== 6) {
370 insert_point
= raiddisks
- 1;
371 if (insert_point
== first_missing
)
374 info
.array
.active_disks
--;
378 if (level
<= 0 && first_missing
!= subdevs
* 2) {
380 Name
": This level does not support missing devices\n");
384 /* Ok, lets try some ioctls */
386 info
.array
.level
= level
;
387 info
.array
.size
= size
;
388 info
.array
.raid_disks
= raiddisks
;
389 /* The kernel should *know* what md_minor we are dealing
390 * with, but it chooses to trust me instead. Sigh
392 info
.array
.md_minor
= 0;
393 if (fstat(mdfd
, &stb
)==0)
394 info
.array
.md_minor
= minor(stb
.st_rdev
);
395 info
.array
.not_persistent
= 0;
397 if ( ( (level
== 4 || level
== 5) &&
398 (insert_point
< raiddisks
|| first_missing
< raiddisks
) )
400 ( level
== 6 && (insert_point
< raiddisks
401 || second_missing
< raiddisks
))
405 info
.array
.state
= 1; /* clean, but one+ drive will be missing*/
407 info
.array
.state
= 0; /* not clean, but no errors */
410 /* for raid10, the bitmap size is the capacity of the array,
411 * which is array.size * raid_disks / ncopies;
412 * .. but convert to sectors.
414 int ncopies
= ((layout
>>8) & 255) * (layout
& 255);
415 bitmapsize
= (unsigned long long)size
* raiddisks
/ ncopies
* 2;
416 /* printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
418 bitmapsize
= (unsigned long long)size
* 2;
420 /* There is lots of redundancy in these disk counts,
421 * raid_disks is the most meaningful value
422 * it describes the geometry of the array
424 * nr_disks is total number of used slots.
425 * it should be raid_disks+spare_disks
426 * spare_disks is the number of extra disks present
428 * active_disks is the number of working disks in
429 * active slots. (With raid_disks)
430 * working_disks is the total number of working disks,
432 * failed_disks is the number of disks marked failed
434 * Ideally, the kernel would keep these (except raid_disks)
435 * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
436 * So for now, we assume that all raid and spare
437 * devices will be given.
439 info
.array
.spare_disks
=sparedisks
;
440 info
.array
.failed_disks
=missing_disks
;
441 info
.array
.nr_disks
= info
.array
.working_disks
442 + info
.array
.failed_disks
;
443 info
.array
.layout
= layout
;
444 info
.array
.chunk_size
= chunk
*1024;
445 info
.array
.major_version
= st
->ss
->major
;
447 if (name
== NULL
|| *name
== 0) {
448 /* base name on mddev */
453 * /dev/md/home -> home
454 * /dev/mdhome -> home
456 name
= strrchr(mddev
, '/');
459 if (strncmp(name
, "md_d", 4)==0 &&
462 (name
-mddev
) == 5 /* /dev/ */)
464 else if (strncmp(name
, "md", 2)==0 &&
467 (name
-mddev
) == 5 /* /dev/ */)
471 if (!st
->ss
->init_super(st
, &info
.array
, size
, name
, homehost
, uuid
))
474 if (bitmap_file
&& vers
< 9003) {
475 major_num
= BITMAP_MAJOR_HOSTENDIAN
;
477 fprintf(stderr
, Name
": Warning - bitmaps created on this kernel are not portable\n"
478 " between different architectured. Consider upgrading the Linux kernel.\n");
482 if (bitmap_file
&& strcmp(bitmap_file
, "internal")==0) {
483 if ((vers
%100) < 2) {
484 fprintf(stderr
, Name
": internal bitmaps not supported by this kernel.\n");
487 if (!st
->ss
->add_internal_bitmap(st
, &bitmap_chunk
,
489 bitmapsize
, 1, major_num
)) {
490 fprintf(stderr
, Name
": Given bitmap chunk size not supported.\n");
497 sra
= sysfs_read(mdfd
, 0, 0);
499 if (st
->ss
->external
) {
501 strcat(strcpy(ver
, "external:"), st
->ss
->text_version
);
502 if ((vers
% 100) < 2 ||
504 sysfs_set_str(sra
, NULL
, "metadata_version",
506 fprintf(stderr
, Name
": This kernel does not "
507 "support external metadata.\n");
511 } else if ((vers
% 100) >= 1) { /* can use different versions */
512 mdu_array_info_t inf
;
513 memset(&inf
, 0, sizeof(inf
));
514 inf
.major_version
= st
->ss
->major
;
515 inf
.minor_version
= st
->minor_version
;
516 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, &inf
);
518 rv
= ioctl(mdfd
, SET_ARRAY_INFO
, NULL
);
520 fprintf(stderr
, Name
": SET_ARRAY_INFO failed for %s: %s\n",
521 mddev
, strerror(errno
));
528 st
->ss
->uuid_from_super(st
, uuid
);
529 if (CreateBitmap(bitmap_file
, force
, (char*)uuid
, bitmap_chunk
,
535 bitmap_fd
= open(bitmap_file
, O_RDWR
);
537 fprintf(stderr
, Name
": weird: %s cannot be openned\n",
541 if (ioctl(mdfd
, SET_BITMAP_FILE
, bitmap_fd
) < 0) {
542 fprintf(stderr
, Name
": Cannot set bitmap file for %s: %s\n",
543 mddev
, strerror(errno
));
549 for (pass
=1; pass
<=2 ; pass
++) {
550 mddev_dev_t moved_disk
= NULL
; /* the disk that was moved out of the insert point */
552 for (dnum
=0, dv
= devlist
; dv
;
553 dv
=(dv
->next
)?(dv
->next
):moved_disk
, dnum
++) {
557 info
.disk
.number
= dnum
;
558 if (dnum
== insert_point
) {
561 info
.disk
.raid_disk
= info
.disk
.number
;
562 if (info
.disk
.raid_disk
< raiddisks
)
563 info
.disk
.state
= (1<<MD_DISK_ACTIVE
) |
568 info
.disk
.state
|= (1<<MD_DISK_WRITEMOSTLY
);
570 if (dnum
== insert_point
||
571 strcasecmp(dv
->devname
, "missing")==0)
574 fd
= open(dv
->devname
, O_RDWR
|O_EXCL
, 0);
576 fprintf(stderr
, Name
": failed to open %s "
577 "after earlier success - aborting\n",
582 info
.disk
.major
= major(stb
.st_rdev
);
583 info
.disk
.minor
= minor(stb
.st_rdev
);
587 remove_partitions(fd
);
588 st
->ss
->add_to_super(st
, &info
.disk
,
594 if (st
->ss
->external
) {
596 sprintf(dv
, "%d:%d\n",
599 sysfs_set_str(sra
, NULL
, "new_dev", dv
);
600 /* FIXME check error */
601 /*FIXME find that device and set it up*/
602 } else if (ioctl(mdfd
, ADD_NEW_DISK
,
605 Name
": ADD_NEW_DISK for %s "
607 dv
->devname
, strerror(errno
));
608 st
->ss
->free_super(st
);
613 if (dv
== moved_disk
&& dnum
!= insert_point
) break;
616 st
->ss
->write_init_super(st
);
618 st
->ss
->free_super(st
);
620 /* param is not actually used */
621 if (level
== LEVEL_CONTAINER
)
622 /* No need to start */
624 else if (runstop
== 1 || subdevs
>= raiddisks
) {
626 if (ioctl(mdfd
, RUN_ARRAY
, ¶m
)) {
627 fprintf(stderr
, Name
": RUN_ARRAY failed: %s\n",
629 Manage_runstop(mddev
, mdfd
, -1, 0);
633 fprintf(stderr
, Name
": array %s started.\n", mddev
);
635 fprintf(stderr
, Name
": not starting array - not enough devices.\n");