Create.c

   1 /*
   2  * mdctl - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001 Neil Brown <neilb@cse.unsw.edu.au>
   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@cse.unsw.edu.au>
  23  *    Paper: Neil Brown
  24  *           School of Computer Science and Engineering
  25  *           The University of New South Wales
  26  *           Sydney, 2052
  27  *           Australia
  28  */
  29
  30 #include "mdctl.h"
  31 #include        "md_u.h"
  32 #include        "md_p.h"
  33
  34 int Create(char *mddev, int mdfd,
  35            int chunk, int level, int layout, int size, int raiddisks, int sparedisks,
  36            int subdevs, char *subdev[],
  37            int runstop, int verbose, int force)
  38 {
  39         /*
  40          * Create a new raid array.
  41          *
  42          * First check that necessary details are available
  43          * (i.e. level, raid-disks)
  44          *
  45          * Then check each disk to see what might be on it
  46          * and report anything interesting.
  47          *
  48          * If anything looks odd, and runstop not set,
  49          * abort.
  50          *
  51          * SET_ARRAY_INFO and ADD_NEW_DISK, and
  52          * if runstop==run, or raiddisks diskswere used,
  53          * RUN_ARRAY
  54          */
  55         int minsize, maxsize;
  56         int maxdisc= -1, mindisc = -1;
  57         int i;
  58         int fail=0, warn=0;
  59         struct stat stb;
  60         int first_missing = MD_SB_DISKS*2;
  61         int missing_disks = 0;
  62         int insert_point = MD_SB_DISKS*2; /* where to insert a missing drive */
  63
  64         mdu_array_info_t array;
  65
  66
  67         if (md_get_version(mdfd) < 9000) {
  68                 fprintf(stderr, Name ": Create requires md driver verison 0.90.0 or later\n");
  69                 return 1;
  70         }
  71         if (level == -10) {
  72                 fprintf(stderr,
  73                         Name ": a RAID level is needed to create an array.\n");
  74                 return 1;
  75         }
  76         if (raiddisks < 1) {
  77                 fprintf(stderr,
  78                         Name ": a number of --raid-disks must be given to create an array\n");
  79                 return 1;
  80         }
  81         if (raiddisks < 2 && level >= 4) {
  82                 fprintf(stderr,
  83                         Name ": atleast 2 raid-disks needed for level 4 or 5\n");
  84                 return 1;
  85         }
  86         if (raiddisks+sparedisks > MD_SB_DISKS) {
  87                 fprintf(stderr,
  88                         Name ": too many discs requested: %d+%d > %d\n",
  89                         raiddisks, sparedisks, MD_SB_DISKS);
  90                 return 1;
  91         }
  92         if (subdevs > raiddisks+sparedisks) {
  93                 fprintf(stderr, Name ": You have listed more disks (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
  94                 return 1;
  95         }
  96         if (subdevs < raiddisks) {
  97                 fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
  98                 return 1;
  99         }
 100
 101         /* now set some defaults */
 102         if (layout == -1)
 103                 switch(level) {
 104                 default: /* no layout */
 105                         layout = 0;
 106                         break;
 107                 case 5:
 108                         layout = map_name(r5layout, "default");
 109                         if (verbose)
 110                                 fprintf(stderr,
 111                                         Name ": layout defaults to %s\n", map_num(r5layout, layout));
 112                         break;
 113                 }
 114
 115         if (chunk == 0) {
 116                 chunk = 64;
 117                 if (verbose)
 118                         fprintf(stderr, Name ": chunk size defaults to 64K\n");
 119         }
 120
 121         /* now look at the subdevs */
 122         array.active_disks = 0;
 123         array.working_disks = 0;
 124         for (i=0; i<subdevs; i++) {
 125                 char *dname = subdev[i];
 126                 int dsize, freesize;
 127                 int fd;
 128                 if (strcasecmp(subdev[i], "missing")==0) {
 129                         if (first_missing > i)
 130                                 first_missing = i;
 131                         missing_disks ++;
 132                         continue;
 133                 }
 134                 array.working_disks++;
 135                 if (i < raiddisks)
 136                         array.active_disks++;
 137                 fd = open(dname, O_RDONLY, 0);
 138                 if (fd <0 ) {
 139                         fprintf(stderr, Name ": Cannot open %s: %s\n",
 140                                 dname, strerror(errno));
 141                         fail=1;
 142                         continue;
 143                 }
 144                 if (ioctl(fd, BLKGETSIZE, &dsize)) {
 145                         fprintf(stderr, Name ": Cannot get size of %s: %s\n",
 146                                 dname, strerror(errno));
 147                         fail = 1;
 148                         close(fd);
 149                         continue;
 150                 }
 151                 if (dsize < MD_RESERVED_SECTORS*2) {
 152                         fprintf(stderr, Name ": %s is too small: %dK\n",
 153                                 dname, dsize/2);
 154                         fail = 1;
 155                         close(fd);
 156                         continue;
 157                 }
 158                 freesize = MD_NEW_SIZE_SECTORS(dsize);
 159                 freesize /= 2;
 160
 161                 if (size && freesize < size) {
 162                         fprintf(stderr, Name ": %s is smaller that given size."
 163                                 " %dK < %dK + superblock\n", dname, freesize, size);
 164                         fail = 1;
 165                         close(fd);
 166                         continue;
 167                 }
 168                 if (maxdisc< 0 || (maxdisc>=0 && freesize > maxsize)) {
 169                         maxdisc = i;
 170                         maxsize = freesize;
 171                 }
 172                 if (mindisc < 0 || (mindisc >=0 && freesize < minsize)) {
 173                         mindisc = i;
 174                         minsize = freesize;
 175                 }
 176                 warn |= check_ext2(fd, dname);
 177                 warn |= check_reiser(fd, dname);
 178                 warn |= check_raid(fd, dname);
 179                 close(fd);
 180         }
 181         if (fail) {
 182                 fprintf(stderr, Name ": create aborted\n");
 183                 return 1;
 184         }
 185         if (size == 0) {
 186                 if (mindisc == -1) {
 187                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 188                         return 1;
 189                 }
 190                 size = minsize;
 191                 if (verbose && level>0)
 192                         fprintf(stderr, Name ": size set to %dK\n", size);
 193         }
 194         if ((maxsize-size)*100 > maxsize) {
 195                 fprintf(stderr, Name ": largest drive (%s) exceed size (%dK) by more than 1%\n",
 196                         subdev[maxdisc], size);
 197                 warn = 1;
 198         }
 199
 200         if (warn) {
 201                 if (runstop!= 1) {
 202                         if (!ask("Continue creating array? ")) {
 203                                 fprintf(stderr, Name ": create aborted.\n");
 204                                 return 1;
 205                         }
 206                 } else {
 207                         if (verbose)
 208                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 209                 }
 210         }
 211
 212         /* If this is  raid5, we want to configure the last active slot
 213          * as missing, so that a reconstruct happens (faster than re-parity)
 214          */
 215         if (force == 0 && level == 5 && first_missing >= raiddisks) {
 216                 insert_point = raiddisks-1;
 217                 sparedisks++;
 218                 array.active_disks--;
 219                 missing_disks++;
 220         }
 221
 222         /* Ok, lets try some ioctls */
 223
 224         array.level = level;
 225         array.size = size;
 226         array.raid_disks = raiddisks;
 227         /* The kernel should *know* what md_minor we are dealing
 228          * with, but it chooses to trust me instead. Sigh
 229          */
 230         array.md_minor = 0;
 231         if (fstat(mdfd, &stb)==0)
 232                 array.md_minor = MINOR(stb.st_rdev);
 233         array.not_persistent = 0;
 234         if (level == 5 && (insert_point < raiddisks || first_missing < raiddisks))
 235                 array.state = 1; /* clean, but one drive will be missing */
 236         else
 237                 array.state = 0; /* not clean, but no errors */
 238
 239         /* There is lots of redundancy in these disk counts,
 240          * raid_disks is the most meaningful value
 241          *          it describes the geometry of the array
 242          *          it is constant
 243          * nr_disks is total number of used slots.
 244          *          it should be raid_disks+spare_disks
 245          * spare_disks is the number of extra disks present
 246          *          see above
 247          * active_disks is the number of working disks in
 248          *          active slots. (With raid_disks)
 249          * working_disks is the total number of working disks,
 250          *          including spares
 251          * failed_disks is the number of disks marked failed
 252          *
 253          * Ideally, the kernel would keep these (except raid_disks)
 254          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
 255          * So for now, we assume that all raid and spare
 256          * devices will be given.
 257          */
 258         array.spare_disks=sparedisks;
 259         array.failed_disks=missing_disks;
 260         array.nr_disks = array.working_disks + array.failed_disks;
 261         array.layout = layout;
 262         array.chunk_size = chunk*1024;
 263
 264         if (ioctl(mdfd, SET_ARRAY_INFO, &array)) {
 265                 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
 266                         mddev, strerror(errno));
 267                 return 1;
 268         }
 269
 270         for (i=0; i<subdevs; i++) {
 271                 int fd;
 272                 struct stat stb;
 273                 mdu_disk_info_t disk;
 274
 275                 disk.number = i;
 276                 if (i >= insert_point)
 277                         disk.number++;
 278                 disk.raid_disk = disk.number;
 279                 if (disk.raid_disk < raiddisks)
 280                         disk.state = 6; /* active and in sync */
 281                 else
 282                         disk.state = 0;
 283                 if (strcasecmp(subdev[i], "missing")==0) {
 284                         disk.major = 0;
 285                         disk.minor = 0;
 286                         disk.state = 1; /* faulty */
 287                 } else {
 288                         fd = open(subdev[i], O_RDONLY, 0);
 289                         if (fd < 0) {
 290                                 fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
 291                                         subdev[i]);
 292                                 return 1;
 293                         }
 294                         fstat(fd, &stb);
 295                         disk.major = MAJOR(stb.st_rdev);
 296                         disk.minor = MINOR(stb.st_rdev);
 297                         close(fd);
 298                 }
 299                 if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
 300                         fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
 301                                 subdev[i], strerror(errno));
 302                         return 1;
 303                 }
 304         }
 305
 306         if (insert_point < MD_SB_DISKS) {
 307                 mdu_disk_info_t disk;
 308                 disk.number = insert_point;
 309                 disk.raid_disk = disk.number;
 310                 disk.state = 1; /* faulty */
 311                 disk.major = disk.minor = 0;
 312                 ioctl(mdfd,ADD_NEW_DISK, &disk);
 313         }
 314
 315         /* param is not actually used */
 316         if (runstop == 1 || subdevs >= raiddisks) {
 317                 mdu_param_t param;
 318                 if (ioctl(mdfd, RUN_ARRAY, &param)) {
 319                         fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
 320                                 strerror(errno));
 321                         return 1;
 322                 }
 323                 fprintf(stderr, Name ": array %s started.\n", mddev);
 324         } else {
 325                 fprintf(stderr, Name ": not starting array - not enough discs.\n");
 326         }
 327         return 0;
 328 }