#include "md_u.h"
#include "md_p.h"
-int Create(char *mddev, int mdfd,
- int chunk, int level, int layout, int size, int raiddisks, int sparedisks,
+int Create(struct supertype *st, char *mddev, int mdfd,
+ int chunk, int level, int layout, unsigned long size, int raiddisks, int sparedisks,
int subdevs, mddev_dev_t devlist,
- int runstop, int verbose, int force)
+ int runstop, int verbose, int force,
+ char *bitmap_file, int bitmap_chunk, int delay)
{
/*
* Create a new raid array.
* abort.
*
* SET_ARRAY_INFO and ADD_NEW_DISK, and
- * if runstop==run, or raiddisks diskswere used,
+ * if runstop==run, or raiddisks disks were used,
* RUN_ARRAY
*/
- int minsize, maxsize;
+ unsigned long long minsize=0, maxsize=0;
char *mindisc = NULL;
char *maxdisc = NULL;
- int i;
+ int dnum;
mddev_dev_t dv;
int fail=0, warn=0;
struct stat stb;
- int first_missing = MD_SB_DISKS*2;
+ int first_missing = subdevs * 2;
int missing_disks = 0;
- int insert_point = MD_SB_DISKS*2; /* where to insert a missing drive */
+ int insert_point = subdevs * 2; /* where to insert a missing drive */
+ void *super;
+ int pass;
+ int vers;
+ int rv;
+ int bitmap_fd;
mdu_array_info_t array;
-
- if (md_get_version(mdfd) < 9000) {
- fprintf(stderr, Name ": Create requires md driver verison 0.90.0 or later\n");
+ vers = md_get_version(mdfd);
+ if (vers < 9000) {
+ fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n");
return 1;
}
- if (level == -10) {
+ if (level == UnSet) {
fprintf(stderr,
Name ": a RAID level is needed to create an array.\n");
return 1;
}
if (raiddisks < 1) {
fprintf(stderr,
- Name ": a number of --raid-disks must be given to create an array\n");
+ Name ": a number of --raid-devices must be given to create an array\n");
return 1;
}
- if (raiddisks < 2 && level >= 4) {
+ if (raiddisks < 4 && level == 6) {
+ fprintf(stderr,
+ Name ": at least 4 raid-devices needed for level 6\n");
+ return 1;
+ }
+ if (raiddisks > 256 && level == 6) {
fprintf(stderr,
- Name ": atleast 2 raid-disks needed for level 4 or 5\n");
+ Name ": no more than 256 raid-devices supported for level 6\n");
return 1;
}
- if (raiddisks+sparedisks > MD_SB_DISKS) {
+ if (raiddisks < 2 && level >= 4) {
fprintf(stderr,
- Name ": too many discs requested: %d+%d > %d\n",
- raiddisks, sparedisks, MD_SB_DISKS);
+ Name ": at least 2 raid-devices needed for level 4 or 5\n");
return 1;
}
if (subdevs > raiddisks+sparedisks) {
- fprintf(stderr, Name ": You have listed more disks (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
+ fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
return 1;
}
if (subdevs < raiddisks+sparedisks) {
}
/* now set some defaults */
- if (layout == -1)
+ if (layout == UnSet)
switch(level) {
default: /* no layout */
layout = 0;
break;
+ case 10:
+ layout = 0x102; /* near=2, far=1 */
+ if (verbose)
+ fprintf(stderr,
+ Name ": layout defaults to n1\n");
+ break;
case 5:
+ case 6:
layout = map_name(r5layout, "default");
if (verbose)
fprintf(stderr,
Name ": layout defaults to %s\n", map_num(r5layout, layout));
break;
+ case LEVEL_FAULTY:
+ layout = map_name(faultylayout, "default");
+
+ if (verbose)
+ fprintf(stderr,
+ Name ": layout defaults to %s\n", map_num(faultylayout, layout));
+ break;
+ }
+
+ if (level == 10)
+ /* check layout fits in array*/
+ if ((layout&255) * ((layout>>8)&255) > raiddisks) {
+ fprintf(stderr, Name ": that layout requires at least %d devices\n",
+ (layout&255) * ((layout>>8)&255));
+ return 1;
}
- if (chunk == 0) {
- chunk = 64;
- if (verbose)
- fprintf(stderr, Name ": chunk size defaults to 64K\n");
+ switch(level) {
+ case 4:
+ case 5:
+ case 10:
+ case 6:
+ case 0:
+ case -1: /* linear */
+ if (chunk == 0) {
+ chunk = 64;
+ if (verbose)
+ fprintf(stderr, Name ": chunk size defaults to 64K\n");
+ }
+ break;
+ default: /* raid1, multipath */
+ if (chunk) {
+ chunk = 0;
+ if (verbose)
+ fprintf(stderr, Name ": chunk size ignored for this level\n");
+ }
+ break;
}
/* now look at the subdevs */
array.active_disks = 0;
array.working_disks = 0;
- for (dv=devlist; dv; dv=dv->next) {
+ dnum = 0;
+ for (dv=devlist; dv; dv=dv->next, dnum++) {
char *dname = dv->devname;
- int dsize, freesize;
+ unsigned long dsize;
+ unsigned long long ldsize, freesize;
int fd;
if (strcasecmp(dname, "missing")==0) {
- if (first_missing > i)
- first_missing = i;
+ if (first_missing > dnum)
+ first_missing = dnum;
missing_disks ++;
continue;
}
array.working_disks++;
- if (i < raiddisks)
+ if (dnum < raiddisks)
array.active_disks++;
- fd = open(dname, O_RDONLY, 0);
+ fd = open(dname, O_RDONLY|O_EXCL, 0);
if (fd <0 ) {
fprintf(stderr, Name ": Cannot open %s: %s\n",
dname, strerror(errno));
fail=1;
continue;
}
+#ifdef BLKGETSIZE64
+ if (ioctl(fd, BLKGETSIZE64, &ldsize)==0)
+ ;
+ else
+#endif
if (ioctl(fd, BLKGETSIZE, &dsize)) {
fprintf(stderr, Name ": Cannot get size of %s: %s\n",
dname, strerror(errno));
close(fd);
continue;
}
- if (dsize < MD_RESERVED_SECTORS*2) {
- fprintf(stderr, Name ": %s is too small: %dK\n",
- dname, dsize/2);
+ else {
+ ldsize = dsize;
+ ldsize <<= 9;
+ }
+ freesize = st->ss->avail_size(ldsize >> 9);
+ if (freesize == 0) {
+ fprintf(stderr, Name ": %s is too small: %luK\n",
+ dname, (unsigned long)(ldsize>>10));
fail = 1;
close(fd);
continue;
}
- freesize = MD_NEW_SIZE_SECTORS(dsize);
- freesize /= 2;
+
+ freesize /= 2; /* convert to K */
if (size && freesize < size) {
fprintf(stderr, Name ": %s is smaller that given size."
- " %dK < %dK + superblock\n", dname, freesize, size);
+ " %lluK < %luK + superblock\n", dname, freesize, size);
fail = 1;
close(fd);
continue;
fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
return 1;
}
- size = minsize;
- if (verbose && level>0)
- fprintf(stderr, Name ": size set to %dK\n", size);
+ if (level > 0) {
+ /* size is meaningful */
+ if (minsize > 0x100000000ULL) {
+ fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
+ return 1;
+ }
+ size = minsize;
+ if (verbose)
+ fprintf(stderr, Name ": size set to %luK\n", size);
+ }
}
- if (level >= 1 && ((maxsize-size)*100 > maxsize)) {
- fprintf(stderr, Name ": largest drive (%s) exceed size (%dK) by more than 1%\n",
+ if (level > 0 && ((maxsize-size)*100 > maxsize)) {
+ fprintf(stderr, Name ": largest drive (%s) exceed size (%luK) by more than 1%%\n",
maxdisc, size);
warn = 1;
}
/* If this is raid5, we want to configure the last active slot
* as missing, so that a reconstruct happens (faster than re-parity)
+ * FIX: Can we do this for raid6 as well?
*/
- if (force == 0 && level == 5 && first_missing >= raiddisks) {
- insert_point = raiddisks-1;
- sparedisks++;
- array.active_disks--;
- missing_disks++;
+ if (force == 0 && first_missing >= raiddisks) {
+ switch ( level ) {
+ case 5:
+ insert_point = raiddisks-1;
+ sparedisks++;
+ array.active_disks--;
+ missing_disks++;
+ break;
+ default:
+ break;
+ }
}
/* Ok, lets try some ioctls */
*/
array.md_minor = 0;
if (fstat(mdfd, &stb)==0)
- array.md_minor = MINOR(stb.st_rdev);
+ array.md_minor = minor(stb.st_rdev);
array.not_persistent = 0;
- if (level == 5 && (insert_point < raiddisks || first_missing < raiddisks))
- array.state = 1; /* clean, but one drive will be missing */
+ /*** FIX: Need to do something about RAID-6 here ***/
+ if ( ( (level == 5) &&
+ (insert_point < raiddisks || first_missing < raiddisks) )
+ ||
+ ( level == 6 && missing_disks == 2)
+ )
+ array.state = 1; /* clean, but one+ drive will be missing */
else
array.state = 0; /* not clean, but no errors */
array.layout = layout;
array.chunk_size = chunk*1024;
- if (ioctl(mdfd, SET_ARRAY_INFO, &array)) {
+
+ if (!st->ss->init_super(&super, &array))
+ return 1;
+
+ if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
+ if ((vers%100) < 2) {
+ fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
+ return 1;
+ }
+ if (!st->ss->add_internal_bitmap(super, bitmap_chunk, delay,
+ size ? size : maxsize)) {
+ fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
+ return 1;
+ }
+ bitmap_file = NULL;
+ }
+
+
+
+ if ((vers % 100) >= 1) { /* can use different versions */
+ mdu_array_info_t inf;
+ memset(&inf, 0, sizeof(inf));
+ inf.major_version = st->ss->major;
+ inf.minor_version = st->minor_version;
+ rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
+ } else
+ rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
+ if (rv) {
fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
mddev, strerror(errno));
return 1;
}
-
- for (i=0, dv = devlist ; dv ; dv=dv->next, i++) {
- int fd;
- struct stat stb;
- mdu_disk_info_t disk;
- disk.number = i;
- if (i >= insert_point)
- disk.number++;
- disk.raid_disk = disk.number;
- if (disk.raid_disk < raiddisks)
- disk.state = 6; /* active and in sync */
- else
- disk.state = 0;
- if (strcasecmp(dv->devname, "missing")==0) {
- disk.major = 0;
- disk.minor = 0;
- disk.state = 1; /* faulty */
- } else {
- fd = open(dv->devname, O_RDONLY, 0);
- if (fd < 0) {
- fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
- dv->devname);
- return 1;
- }
- fstat(fd, &stb);
- disk.major = MAJOR(stb.st_rdev);
- disk.minor = MINOR(stb.st_rdev);
- close(fd);
+ if (bitmap_file) {
+ int uuid[4];
+
+ if (bitmap_chunk == UnSet)
+ bitmap_chunk = DEFAULT_BITMAP_CHUNK;
+
+ st->ss->uuid_from_super(uuid, super);
+ if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk, delay,
+ array.size*2ULL /* FIXME wrong for raid10 */)) {
+ return 1;
}
- if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
- fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
- dv->devname, strerror(errno));
+ bitmap_fd = open(bitmap_file, O_RDWR);
+ if (bitmap_fd < 0) {
+ fprintf(stderr, Name ": weird: %s cannot be openned\n",
+ bitmap_file);
+ return 1;
+ }
+ if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
+ fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
+ mddev, strerror(errno));
return 1;
}
}
- if (insert_point < MD_SB_DISKS) {
- mdu_disk_info_t disk;
- disk.number = insert_point;
- disk.raid_disk = disk.number;
- disk.state = 1; /* faulty */
- disk.major = disk.minor = 0;
- ioctl(mdfd,ADD_NEW_DISK, &disk);
+
+
+ for (pass=1; pass <=2 ; pass++) {
+ mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
+
+ for (dnum=0, dv = devlist ; dv ;
+ dv=(dv->next)?(dv->next):moved_disk, dnum++) {
+ int fd;
+ struct stat stb;
+ mdu_disk_info_t disk;
+
+ disk.number = dnum;
+ if (dnum == insert_point) {
+ moved_disk = dv;
+ }
+ disk.raid_disk = disk.number;
+ if (disk.raid_disk < raiddisks)
+ disk.state = 6; /* active and in sync */
+ else
+ disk.state = 0;
+ if (dnum == insert_point ||
+ strcasecmp(dv->devname, "missing")==0) {
+ disk.major = 0;
+ disk.minor = 0;
+ disk.state = 1; /* faulty */
+ } else {
+ fd = open(dv->devname, O_RDONLY|O_EXCL, 0);
+ if (fd < 0) {
+ fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
+ dv->devname);
+ return 1;
+ }
+ fstat(fd, &stb);
+ disk.major = major(stb.st_rdev);
+ disk.minor = minor(stb.st_rdev);
+ close(fd);
+ }
+ switch(pass){
+ case 1:
+ st->ss->add_to_super(super, &disk);
+ break;
+ case 2:
+ if (disk.state == 1) break;
+ st->ss->write_init_super(st, super, &disk, dv->devname);
+
+ if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
+ fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
+ dv->devname, strerror(errno));
+ free(super);
+ return 1;
+ }
+
+ break;
+ }
+ if (dv == moved_disk && dnum != insert_point) break;
+ }
}
+ free(super);
/* param is not actually used */
if (runstop == 1 || subdevs >= raiddisks) {
if (ioctl(mdfd, RUN_ARRAY, ¶m)) {
fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
strerror(errno));
+ Manage_runstop(mddev, mdfd, -1);
return 1;
}
fprintf(stderr, Name ": array %s started.\n", mddev);
} else {
- fprintf(stderr, Name ": not starting array - not enough discs.\n");
+ fprintf(stderr, Name ": not starting array - not enough devices.\n");
}
return 0;
}