util.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2012 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
  25 #include        "mdadm.h"
  26 #include        "md_p.h"
  27 #include        <sys/socket.h>
  28 #include        <sys/utsname.h>
  29 #include        <sys/wait.h>
  30 #include        <sys/un.h>
  31 #include        <ctype.h>
  32 #include        <dirent.h>
  33 #include        <signal.h>
  34
  35 int __offroot;
  36
  37 /*
  38  * following taken from linux/blkpg.h because they aren't
  39  * anywhere else and it isn't safe to #include linux/ * stuff.
  40  */
  41
  42 #define BLKPG      _IO(0x12,105)
  43
  44 /* The argument structure */
  45 struct blkpg_ioctl_arg {
  46         int op;
  47         int flags;
  48         int datalen;
  49         void *data;
  50 };
  51
  52 /* The subfunctions (for the op field) */
  53 #define BLKPG_ADD_PARTITION     1
  54 #define BLKPG_DEL_PARTITION     2
  55
  56 /* Sizes of name fields. Unused at present. */
  57 #define BLKPG_DEVNAMELTH        64
  58 #define BLKPG_VOLNAMELTH        64
  59
  60 /* The data structure for ADD_PARTITION and DEL_PARTITION */
  61 struct blkpg_partition {
  62         long long start;                /* starting offset in bytes */
  63         long long length;               /* length in bytes */
  64         int pno;                        /* partition number */
  65         char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2,
  66                                            to be used in kernel messages */
  67         char volname[BLKPG_VOLNAMELTH]; /* volume label */
  68 };
  69
  70 #include "part.h"
  71
  72 /* Force a compilation error if condition is true */
  73 #define BUILD_BUG_ON(condition) ((void)BUILD_BUG_ON_ZERO(condition))
  74
  75 /* Force a compilation error if condition is true, but also produce a
  76    result (of value 0 and type size_t), so the expression can be used
  77    e.g. in a structure initializer (or where-ever else comma expressions
  78    aren't permitted). */
  79 #define BUILD_BUG_ON_ZERO(e) (sizeof(struct { int:-!!(e); }))
  80
  81 /*
  82  * Parse a 128 bit uuid in 4 integers
  83  * format is 32 hexx nibbles with options :.<space> separator
  84  * If not exactly 32 hex digits are found, return 0
  85  * else return 1
  86  */
  87 int parse_uuid(char *str, int uuid[4])
  88 {
  89         int hit = 0; /* number of Hex digIT */
  90         int i;
  91         char c;
  92         for (i = 0; i < 4; i++)
  93                 uuid[i] = 0;
  94
  95         while ((c = *str++) != 0) {
  96                 int n;
  97                 if (c >= '0' && c <= '9')
  98                         n = c-'0';
  99                 else if (c >= 'a' && c <= 'f')
 100                         n = 10 + c - 'a';
 101                 else if (c >= 'A' && c <= 'F')
 102                         n = 10 + c - 'A';
 103                 else if (strchr(":. -", c))
 104                         continue;
 105                 else return 0;
 106
 107                 if (hit<32) {
 108                         uuid[hit/8] <<= 4;
 109                         uuid[hit/8] += n;
 110                 }
 111                 hit++;
 112         }
 113         if (hit == 32)
 114                 return 1;
 115         return 0;
 116 }
 117
 118 /*
 119  * Get the md version number.
 120  * We use the RAID_VERSION ioctl if it is supported
 121  * If not, but we have a block device with major '9', we assume
 122  * 0.36.0
 123  *
 124  * Return version number as 24 but number - assume version parts
 125  * always < 255
 126  */
 127
 128 int md_get_version(int fd)
 129 {
 130     struct stat stb;
 131     mdu_version_t vers;
 132
 133     if (fstat(fd, &stb)<0)
 134         return -1;
 135     if ((S_IFMT&stb.st_mode) != S_IFBLK)
 136         return -1;
 137
 138     if (ioctl(fd, RAID_VERSION, &vers) == 0)
 139         return  (vers.major*10000) + (vers.minor*100) + vers.patchlevel;
 140     if (errno == EACCES)
 141             return -1;
 142     if (major(stb.st_rdev) == MD_MAJOR)
 143         return (3600);
 144     return -1;
 145 }
 146
 147 int get_linux_version()
 148 {
 149         struct utsname name;
 150         char *cp;
 151         int a = 0, b = 0,c = 0;
 152         if (uname(&name) <0)
 153                 return -1;
 154
 155         cp = name.release;
 156         a = strtoul(cp, &cp, 10);
 157         if (*cp == '.')
 158                 b = strtoul(cp+1, &cp, 10);
 159         if (*cp == '.')
 160                 c = strtoul(cp+1, &cp, 10);
 161
 162         return (a*1000000)+(b*1000)+c;
 163 }
 164
 165 #ifndef MDASSEMBLE
 166 int mdadm_version(char *version)
 167 {
 168         int a, b, c;
 169         char *cp;
 170
 171         if (!version)
 172                 version = Version;
 173
 174         cp = strchr(version, '-');
 175         if (!cp || *(cp+1) != ' ' || *(cp+2) != 'v')
 176                 return -1;
 177         cp += 3;
 178         a = strtoul(cp, &cp, 10);
 179         if (*cp != '.')
 180                 return -1;
 181         b = strtoul(cp+1, &cp, 10);
 182         if (*cp == '.')
 183                 c = strtoul(cp+1, &cp, 10);
 184         else
 185                 c = 0;
 186         if (*cp != ' ' && *cp != '-')
 187                 return -1;
 188         return (a*1000000)+(b*1000)+c;
 189 }
 190
 191 long long parse_size(char *size)
 192 {
 193         /* parse 'size' which should be a number optionally
 194          * followed by 'K', 'M', or 'G'.
 195          * Without a suffix, K is assumed.
 196          * Number returned is in sectors (half-K)
 197          * -1 returned on error.
 198          */
 199         char *c;
 200         long long s = strtoll(size, &c, 10);
 201         if (s > 0) {
 202                 switch (*c) {
 203                 case 'K':
 204                         c++;
 205                 default:
 206                         s *= 2;
 207                         break;
 208                 case 'M':
 209                         c++;
 210                         s *= 1024 * 2;
 211                         break;
 212                 case 'G':
 213                         c++;
 214                         s *= 1024 * 1024 * 2;
 215                         break;
 216                 }
 217         }
 218         if (*c)
 219                 s = -1;
 220         return s;
 221 }
 222
 223 int parse_layout_10(char *layout)
 224 {
 225         int copies, rv;
 226         char *cp;
 227         /* Parse the layout string for raid10 */
 228         /* 'f', 'o' or 'n' followed by a number <= raid_disks */
 229         if ((layout[0] !=  'n' && layout[0] != 'f' && layout[0] != 'o') ||
 230             (copies = strtoul(layout+1, &cp, 10)) < 1 ||
 231             copies > 200 ||
 232             *cp)
 233                 return -1;
 234         if (layout[0] == 'n')
 235                 rv = 256 + copies;
 236         else if (layout[0] == 'o')
 237                 rv = 0x10000 + (copies<<8) + 1;
 238         else
 239                 rv = 1 + (copies<<8);
 240         return rv;
 241 }
 242
 243 int parse_layout_faulty(char *layout)
 244 {
 245         /* Parse the layout string for 'faulty' */
 246         int ln = strcspn(layout, "0123456789");
 247         char *m = xstrdup(layout);
 248         int mode;
 249         m[ln] = 0;
 250         mode = map_name(faultylayout, m);
 251         if (mode == UnSet)
 252                 return -1;
 253
 254         return mode | (atoi(layout+ln)<< ModeShift);
 255 }
 256
 257 long parse_num(char *num)
 258 {
 259         /* Either return a valid number, or -1 */
 260         char *c;
 261         long rv = strtol(num, &c, 10);
 262         if (rv < 0 || *c || !num[0])
 263                 return -1;
 264         else
 265                 return rv;
 266 }
 267 #endif
 268
 269 void remove_partitions(int fd)
 270 {
 271         /* remove partitions from this block devices.
 272          * This is used for components added to an array
 273          */
 274 #ifdef BLKPG_DEL_PARTITION
 275         struct blkpg_ioctl_arg a;
 276         struct blkpg_partition p;
 277
 278         a.op = BLKPG_DEL_PARTITION;
 279         a.data = (void*)&p;
 280         a.datalen = sizeof(p);
 281         a.flags = 0;
 282         memset(a.data, 0, a.datalen);
 283         for (p.pno = 0; p.pno < 16; p.pno++)
 284                 ioctl(fd, BLKPG, &a);
 285 #endif
 286 }
 287
 288 int test_partition(int fd)
 289 {
 290         /* Check if fd is a whole-disk or a partition.
 291          * BLKPG will return EINVAL on a partition, and BLKPG_DEL_PARTITION
 292          * will return ENXIO on an invalid partition number.
 293          */
 294         struct blkpg_ioctl_arg a;
 295         struct blkpg_partition p;
 296         a.op = BLKPG_DEL_PARTITION;
 297         a.data = (void*)&p;
 298         a.datalen = sizeof(p);
 299         a.flags = 0;
 300         memset(a.data, 0, a.datalen);
 301         p.pno = 1<<30;
 302         if (ioctl(fd, BLKPG, &a) == 0)
 303                 /* Very unlikely, but not a partition */
 304                 return 0;
 305         if (errno == ENXIO)
 306                 /* not a partition */
 307                 return 0;
 308
 309         return 1;
 310 }
 311
 312 int test_partition_from_id(dev_t id)
 313 {
 314         char buf[20];
 315         int fd, rv;
 316
 317         sprintf(buf, "%d:%d", major(id), minor(id));
 318         fd = dev_open(buf, O_RDONLY);
 319         if (fd < 0)
 320                 return -1;
 321         rv = test_partition(fd);
 322         close(fd);
 323         return rv;
 324 }
 325
 326 int enough(int level, int raid_disks, int layout, int clean, char *avail)
 327 {
 328         int copies, first;
 329         int i;
 330         int avail_disks = 0;
 331
 332         for (i = 0; i < raid_disks; i++)
 333                 avail_disks += !!avail[i];
 334
 335         switch (level) {
 336         case 10:
 337                 /* This is the tricky one - we need to check
 338                  * which actual disks are present.
 339                  */
 340                 copies = (layout&255)* ((layout>>8) & 255);
 341                 first = 0;
 342                 do {
 343                         /* there must be one of the 'copies' form 'first' */
 344                         int n = copies;
 345                         int cnt = 0;
 346                         while (n--) {
 347                                 if (avail[first])
 348                                         cnt++;
 349                                 first = (first+1) % raid_disks;
 350                         }
 351                         if (cnt == 0)
 352                                 return 0;
 353
 354                 } while (first != 0);
 355                 return 1;
 356
 357         case LEVEL_MULTIPATH:
 358                 return avail_disks>= 1;
 359         case LEVEL_LINEAR:
 360         case 0:
 361                 return avail_disks == raid_disks;
 362         case 1:
 363                 return avail_disks >= 1;
 364         case 4:
 365         case 5:
 366                 if (clean)
 367                         return avail_disks >= raid_disks-1;
 368                 else
 369                         return avail_disks >= raid_disks;
 370         case 6:
 371                 if (clean)
 372                         return avail_disks >= raid_disks-2;
 373                 else
 374                         return avail_disks >= raid_disks;
 375         default:
 376                 return 0;
 377         }
 378 }
 379
 380 int enough_fd(int fd)
 381 {
 382         struct mdu_array_info_s array;
 383         struct mdu_disk_info_s disk;
 384         int avail_disks = 0;
 385         int i, rv;
 386         char *avail;
 387
 388         if (ioctl(fd, GET_ARRAY_INFO, &array) != 0 ||
 389             array.raid_disks <= 0)
 390                 return 0;
 391         avail = xcalloc(array.raid_disks, 1);
 392         for (i = 0; i < MAX_DISKS && array.nr_disks > 0; i++) {
 393                 disk.number = i;
 394                 if (ioctl(fd, GET_DISK_INFO, &disk) != 0)
 395                         continue;
 396                 if (disk.major == 0 && disk.minor == 0)
 397                         continue;
 398                 array.nr_disks--;
 399
 400                 if (! (disk.state & (1<<MD_DISK_SYNC)))
 401                         continue;
 402                 if (disk.raid_disk < 0 || disk.raid_disk >= array.raid_disks)
 403                         continue;
 404                 avail_disks++;
 405                 avail[disk.raid_disk] = 1;
 406         }
 407         /* This is used on an active array, so assume it is clean */
 408         rv = enough(array.level, array.raid_disks, array.layout,
 409                     1, avail);
 410         free(avail);
 411         return rv;
 412 }
 413
 414 const int uuid_zero[4] = { 0, 0, 0, 0 };
 415
 416 int same_uuid(int a[4], int b[4], int swapuuid)
 417 {
 418         if (swapuuid) {
 419                 /* parse uuids are hostendian.
 420                  * uuid's from some superblocks are big-ending
 421                  * if there is a difference, we need to swap..
 422                  */
 423                 unsigned char *ac = (unsigned char *)a;
 424                 unsigned char *bc = (unsigned char *)b;
 425                 int i;
 426                 for (i = 0; i < 16; i += 4) {
 427                         if (ac[i+0] != bc[i+3] ||
 428                             ac[i+1] != bc[i+2] ||
 429                             ac[i+2] != bc[i+1] ||
 430                             ac[i+3] != bc[i+0])
 431                                 return 0;
 432                 }
 433                 return 1;
 434         } else {
 435                 if (a[0]==b[0] &&
 436                     a[1]==b[1] &&
 437                     a[2]==b[2] &&
 438                     a[3]==b[3])
 439                         return 1;
 440                 return 0;
 441         }
 442 }
 443
 444 void copy_uuid(void *a, int b[4], int swapuuid)
 445 {
 446         if (swapuuid) {
 447                 /* parse uuids are hostendian.
 448                  * uuid's from some superblocks are big-ending
 449                  * if there is a difference, we need to swap..
 450                  */
 451                 unsigned char *ac = (unsigned char *)a;
 452                 unsigned char *bc = (unsigned char *)b;
 453                 int i;
 454                 for (i = 0; i < 16; i += 4) {
 455                         ac[i+0] = bc[i+3];
 456                         ac[i+1] = bc[i+2];
 457                         ac[i+2] = bc[i+1];
 458                         ac[i+3] = bc[i+0];
 459                 }
 460         } else
 461                 memcpy(a, b, 16);
 462 }
 463
 464 char *__fname_from_uuid(int id[4], int swap, char *buf, char sep)
 465 {
 466         int i, j;
 467         char uuid[16];
 468         char *c = buf;
 469         strcpy(c, "UUID-");
 470         c += strlen(c);
 471         copy_uuid(uuid, id, swap);
 472         for (i = 0; i < 4; i++) {
 473                 if (i)
 474                         *c++ = sep;
 475                 for (j = 3; j >= 0; j--) {
 476                         sprintf(c,"%02x", (unsigned char) uuid[j+4*i]);
 477                         c+= 2;
 478                 }
 479         }
 480         return buf;
 481
 482 }
 483
 484 char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep)
 485 {
 486         // dirty hack to work around an issue with super1 superblocks...
 487         // super1 superblocks need swapuuid set in order for assembly to
 488         // work, but can't have it set if we want this printout to match
 489         // all the other uuid printouts in super1.c, so we force swapuuid
 490         // to 1 to make our printout match the rest of super1
 491         return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep);
 492 }
 493
 494 #ifndef MDASSEMBLE
 495 int check_ext2(int fd, char *name)
 496 {
 497         /*
 498          * Check for an ext2fs file system.
 499          * Superblock is always 1K at 1K offset
 500          *
 501          * s_magic is le16 at 56 == 0xEF53
 502          * report mtime - le32 at 44
 503          * blocks - le32 at 4
 504          * logblksize - le32 at 24
 505          */
 506         unsigned char sb[1024];
 507         time_t mtime;
 508         int size, bsize;
 509         if (lseek(fd, 1024,0)!= 1024)
 510                 return 0;
 511         if (read(fd, sb, 1024)!= 1024)
 512                 return 0;
 513         if (sb[56] != 0x53 || sb[57] != 0xef)
 514                 return 0;
 515
 516         mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8;
 517         bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8;
 518         size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8;
 519         pr_err("%s appears to contain an ext2fs file system\n",
 520                 name);
 521         fprintf(stderr,"    size=%dK  mtime=%s",
 522                 size*(1<<bsize), ctime(&mtime));
 523         return 1;
 524 }
 525
 526 int check_reiser(int fd, char *name)
 527 {
 528         /*
 529          * superblock is at 64K
 530          * size is 1024;
 531          * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
 532          *
 533          */
 534         unsigned char sb[1024];
 535         unsigned long size;
 536         if (lseek(fd, 64*1024, 0) != 64*1024)
 537                 return 0;
 538         if (read(fd, sb, 1024) != 1024)
 539                 return 0;
 540         if (strncmp((char*)sb+52, "ReIsErFs",8) != 0 &&
 541             strncmp((char*)sb+52, "ReIsEr2Fs",9) != 0)
 542                 return 0;
 543         pr_err("%s appears to contain a reiserfs file system\n",name);
 544         size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8;
 545         fprintf(stderr, "    size = %luK\n", size*4);
 546
 547         return 1;
 548 }
 549
 550 int check_raid(int fd, char *name)
 551 {
 552         struct mdinfo info;
 553         time_t crtime;
 554         char *level;
 555         struct supertype *st = guess_super(fd);
 556
 557         if (!st) return 0;
 558         st->ignore_hw_compat = 1;
 559         st->ss->load_super(st, fd, name);
 560         /* Looks like a raid array .. */
 561         pr_err("%s appears to be part of a raid array:\n",
 562                 name);
 563         st->ss->getinfo_super(st, &info, NULL);
 564         st->ss->free_super(st);
 565         crtime = info.array.ctime;
 566         level = map_num(pers, info.array.level);
 567         if (!level) level = "-unknown-";
 568         fprintf(stderr, "    level=%s devices=%d ctime=%s",
 569                 level, info.array.raid_disks, ctime(&crtime));
 570         return 1;
 571 }
 572
 573 int ask(char *mesg)
 574 {
 575         char *add = "";
 576         int i;
 577         for (i = 0; i < 5; i++) {
 578                 char buf[100];
 579                 fprintf(stderr, "%s%s", mesg, add);
 580                 fflush(stderr);
 581                 if (fgets(buf, 100, stdin)==NULL)
 582                         return 0;
 583                 if (buf[0]=='y' || buf[0]=='Y')
 584                         return 1;
 585                 if (buf[0]=='n' || buf[0]=='N')
 586                         return 0;
 587                 add = "(y/n) ";
 588         }
 589         pr_err("assuming 'no'\n");
 590         return 0;
 591 }
 592 #endif /* MDASSEMBLE */
 593
 594 int is_standard(char *dev, int *nump)
 595 {
 596         /* tests if dev is a "standard" md dev name.
 597          * i.e if the last component is "/dNN" or "/mdNN",
 598          * where NN is a string of digits
 599          * Returns 1 if a partitionable standard,
 600          *   -1 if non-partitonable,
 601          *   0 if not a standard name.
 602          */
 603         char *d = strrchr(dev, '/');
 604         int type = 0;
 605         int num;
 606         if (!d)
 607                 return 0;
 608         if (strncmp(d, "/d",2) == 0)
 609                 d += 2, type = 1; /* /dev/md/dN{pM} */
 610         else if (strncmp(d, "/md_d", 5) == 0)
 611                 d += 5, type = 1; /* /dev/md_dN{pM} */
 612         else if (strncmp(d, "/md", 3) == 0)
 613                 d += 3, type = -1; /* /dev/mdN */
 614         else if (d-dev > 3 && strncmp(d-2, "md/", 3) == 0)
 615                 d += 1, type = -1; /* /dev/md/N */
 616         else
 617                 return 0;
 618         if (!*d)
 619                 return 0;
 620         num = atoi(d);
 621         while (isdigit(*d))
 622                 d++;
 623         if (*d)
 624                 return 0;
 625         if (nump) *nump = num;
 626
 627         return type;
 628 }
 629
 630 unsigned long calc_csum(void *super, int bytes)
 631 {
 632         unsigned long long newcsum = 0;
 633         int i;
 634         unsigned int csum;
 635         unsigned int *superc = (unsigned int*) super;
 636
 637         for(i = 0; i < bytes/4; i++)
 638                 newcsum += superc[i];
 639         csum = (newcsum& 0xffffffff) + (newcsum>>32);
 640 #ifdef __alpha__
 641 /* The in-kernel checksum calculation is always 16bit on
 642  * the alpha, though it is 32 bit on i386...
 643  * I wonder what it is elsewhere... (it uses an API in
 644  * a way that it shouldn't).
 645  */
 646         csum = (csum & 0xffff) + (csum >> 16);
 647         csum = (csum & 0xffff) + (csum >> 16);
 648 #endif
 649         return csum;
 650 }
 651
 652 #ifndef MDASSEMBLE
 653 char *human_size(long long bytes)
 654 {
 655         static char buf[30];
 656
 657         /* We convert bytes to either centi-M{ega,ibi}bytes or
 658          * centi-G{igi,ibi}bytes, with appropriate rounding,
 659          * and then print 1/100th of those as a decimal.
 660          * We allow upto 2048Megabytes before converting to
 661          * gigabytes, as that shows more precision and isn't
 662          * too large a number.
 663          * Terabytes are not yet handled.
 664          */
 665
 666         if (bytes < 5000*1024)
 667                 buf[0] = 0;
 668         else if (bytes < 2*1024LL*1024LL*1024LL) {
 669                 long cMiB = (bytes / ( (1LL<<20) / 200LL ) +1) /2;
 670                 long cMB  = (bytes / ( 1000000LL / 200LL ) +1) /2;
 671                 snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)",
 672                         cMiB/100 , cMiB % 100,
 673                         cMB/100, cMB % 100);
 674         } else {
 675                 long cGiB = (bytes / ( (1LL<<30) / 200LL ) +1) /2;
 676                 long cGB  = (bytes / (1000000000LL/200LL ) +1) /2;
 677                 snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)",
 678                         cGiB/100 , cGiB % 100,
 679                         cGB/100, cGB % 100);
 680         }
 681         return buf;
 682 }
 683
 684 char *human_size_brief(long long bytes)
 685 {
 686         static char buf[30];
 687
 688         if (bytes < 5000*1024)
 689                 snprintf(buf, sizeof(buf), "%ld.%02ldKiB",
 690                         (long)(bytes>>10), (long)(((bytes&1023)*100+512)/1024)
 691                         );
 692         else if (bytes < 2*1024LL*1024LL*1024LL)
 693                 snprintf(buf, sizeof(buf), "%ld.%02ldMiB",
 694                         (long)(bytes>>20),
 695                         (long)((bytes&0xfffff)+0x100000/200)/(0x100000/100)
 696                         );
 697         else
 698                 snprintf(buf, sizeof(buf), "%ld.%02ldGiB",
 699                         (long)(bytes>>30),
 700                         (long)(((bytes>>10)&0xfffff)+0x100000/200)/(0x100000/100)
 701                         );
 702         return buf;
 703 }
 704
 705 void print_r10_layout(int layout)
 706 {
 707         int near = layout & 255;
 708         int far = (layout >> 8) & 255;
 709         int offset = (layout&0x10000);
 710         char *sep = "";
 711
 712         if (near != 1) {
 713                 printf("%s near=%d", sep, near);
 714                 sep = ",";
 715         }
 716         if (far != 1)
 717                 printf("%s %s=%d", sep, offset?"offset":"far", far);
 718         if (near*far == 1)
 719                 printf("NO REDUNDANCY");
 720 }
 721 #endif
 722
 723 unsigned long long calc_array_size(int level, int raid_disks, int layout,
 724                                    int chunksize, unsigned long long devsize)
 725 {
 726         if (level == 1)
 727                 return devsize;
 728         devsize &= ~(unsigned long long)((chunksize>>9)-1);
 729         return get_data_disks(level, layout, raid_disks) * devsize;
 730 }
 731
 732 int get_data_disks(int level, int layout, int raid_disks)
 733 {
 734         int data_disks = 0;
 735         switch (level) {
 736         case 0: data_disks = raid_disks;
 737                 break;
 738         case 1: data_disks = 1;
 739                 break;
 740         case 4:
 741         case 5: data_disks = raid_disks - 1;
 742                 break;
 743         case 6: data_disks = raid_disks - 2;
 744                 break;
 745         case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255);
 746                 break;
 747         }
 748
 749         return data_disks;
 750 }
 751
 752 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
 753 char *get_md_name(int dev)
 754 {
 755         /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
 756         /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
 757         static char devname[50];
 758         struct stat stb;
 759         dev_t rdev;
 760         char *dn;
 761
 762         if (dev < 0) {
 763                 int mdp =  get_mdp_major();
 764                 if (mdp < 0) return NULL;
 765                 rdev = makedev(mdp, (-1-dev)<<6);
 766                 snprintf(devname, sizeof(devname), "/dev/md/d%d", -1-dev);
 767                 if (stat(devname, &stb) == 0
 768                     && (S_IFMT&stb.st_mode) == S_IFBLK
 769                     && (stb.st_rdev == rdev))
 770                         return devname;
 771         } else {
 772                 rdev = makedev(MD_MAJOR, dev);
 773                 snprintf(devname, sizeof(devname), "/dev/md%d", dev);
 774                 if (stat(devname, &stb) == 0
 775                     && (S_IFMT&stb.st_mode) == S_IFBLK
 776                     && (stb.st_rdev == rdev))
 777                         return devname;
 778
 779                 snprintf(devname, sizeof(devname), "/dev/md/%d", dev);
 780                 if (stat(devname, &stb) == 0
 781                     && (S_IFMT&stb.st_mode) == S_IFBLK
 782                     && (stb.st_rdev == rdev))
 783                         return devname;
 784         }
 785         dn = map_dev(major(rdev), minor(rdev), 0);
 786         if (dn)
 787                 return dn;
 788         snprintf(devname, sizeof(devname), "/dev/.tmp.md%d", dev);
 789         if (mknod(devname, S_IFBLK | 0600, rdev) == -1)
 790                 if (errno != EEXIST)
 791                         return NULL;
 792
 793         if (stat(devname, &stb) == 0
 794             && (S_IFMT&stb.st_mode) == S_IFBLK
 795             && (stb.st_rdev == rdev))
 796                 return devname;
 797         unlink(devname);
 798         return NULL;
 799 }
 800
 801 void put_md_name(char *name)
 802 {
 803         if (strncmp(name, "/dev/.tmp.md", 12) == 0)
 804                 unlink(name);
 805 }
 806
 807 int find_free_devnum(int use_partitions)
 808 {
 809         int devnum;
 810         for (devnum = 127; devnum != 128;
 811              devnum = devnum ? devnum-1 : (1<<20)-1) {
 812                 char *dn;
 813                 int _devnum;
 814                 char nbuf[50];
 815
 816                 _devnum = use_partitions ? (-1-devnum) : devnum;
 817                 if (mddev_busy(_devnum))
 818                         continue;
 819                 sprintf(nbuf, "%s%d", use_partitions?"mdp":"md", devnum);
 820                 if (!conf_name_is_free(nbuf))
 821                         continue;
 822                 /* make sure it is new to /dev too, at least as a
 823                  * non-standard */
 824                 dn = map_dev(dev2major(_devnum), dev2minor(_devnum), 0);
 825                 if (dn && ! is_standard(dn, NULL))
 826                         continue;
 827                 break;
 828         }
 829         if (devnum == 128)
 830                 return NoMdDev;
 831         return use_partitions ? (-1-devnum) : devnum;
 832 }
 833 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
 834
 835 int dev_open(char *dev, int flags)
 836 {
 837         /* like 'open', but if 'dev' matches %d:%d, create a temp
 838          * block device and open that
 839          */
 840         char *e;
 841         int fd = -1;
 842         char devname[32];
 843         int major;
 844         int minor;
 845
 846         if (!dev) return -1;
 847         flags |= O_DIRECT;
 848
 849         major = strtoul(dev, &e, 0);
 850         if (e > dev && *e == ':' && e[1] &&
 851             (minor = strtoul(e+1, &e, 0)) >= 0 &&
 852             *e == 0) {
 853                 char *path = map_dev(major, minor, 0);
 854                 if (path)
 855                         fd = open(path, flags);
 856                 if (fd < 0) {
 857                         snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d",
 858                                  (int)getpid(), major, minor);
 859                         if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) {
 860                                 fd = open(devname, flags);
 861                                 unlink(devname);
 862                         }
 863                 }
 864                 if (fd < 0) {
 865                         snprintf(devname, sizeof(devname), "/tmp/.tmp.md.%d:%d:%d",
 866                                  (int)getpid(), major, minor);
 867                         if (mknod(devname, S_IFBLK|0600, makedev(major, minor)) == 0) {
 868                                 fd = open(devname, flags);
 869                                 unlink(devname);
 870                         }
 871                 }
 872         } else
 873                 fd = open(dev, flags);
 874         return fd;
 875 }
 876
 877 int open_dev_flags(int devnum, int flags)
 878 {
 879         char buf[20];
 880
 881         sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
 882         return dev_open(buf, flags);
 883 }
 884
 885 int open_dev(int devnum)
 886 {
 887         return open_dev_flags(devnum, O_RDONLY);
 888 }
 889
 890 int open_dev_excl(int devnum)
 891 {
 892         char buf[20];
 893         int i;
 894         int flags = O_RDWR;
 895
 896         sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
 897         for (i = 0 ; i < 25 ; i++) {
 898                 int fd = dev_open(buf, flags|O_EXCL);
 899                 if (fd >= 0)
 900                         return fd;
 901                 if (errno == EACCES && flags == O_RDWR) {
 902                         flags = O_RDONLY;
 903                         continue;
 904                 }
 905                 if (errno != EBUSY)
 906                         return fd;
 907                 usleep(200000);
 908         }
 909         return -1;
 910 }
 911
 912 int same_dev(char *one, char *two)
 913 {
 914         struct stat st1, st2;
 915         if (stat(one, &st1) != 0)
 916                 return 0;
 917         if (stat(two, &st2) != 0)
 918                 return 0;
 919         if ((st1.st_mode & S_IFMT) != S_IFBLK)
 920                 return 0;
 921         if ((st2.st_mode & S_IFMT) != S_IFBLK)
 922                 return 0;
 923         return st1.st_rdev == st2.st_rdev;
 924 }
 925
 926 void wait_for(char *dev, int fd)
 927 {
 928         int i;
 929         struct stat stb_want;
 930
 931         if (fstat(fd, &stb_want) != 0 ||
 932             (stb_want.st_mode & S_IFMT) != S_IFBLK)
 933                 return;
 934
 935         for (i = 0 ; i < 25 ; i++) {
 936                 struct stat stb;
 937                 if (stat(dev, &stb) == 0 &&
 938                     (stb.st_mode & S_IFMT) == S_IFBLK &&
 939                     (stb.st_rdev == stb_want.st_rdev))
 940                         return;
 941                 usleep(200000);
 942         }
 943         if (i == 25)
 944                 dprintf("%s: timeout waiting for %s\n", __func__, dev);
 945 }
 946
 947 struct superswitch *superlist[] =
 948 {
 949         &super0, &super1,
 950         &super_ddf, &super_imsm,
 951         &mbr, &gpt,
 952         NULL };
 953
 954 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
 955
 956 struct supertype *super_by_fd(int fd, char **subarrayp)
 957 {
 958         mdu_array_info_t array;
 959         int vers;
 960         int minor;
 961         struct supertype *st = NULL;
 962         struct mdinfo *sra;
 963         char *verstr;
 964         char version[20];
 965         int i;
 966         char *subarray = NULL;
 967         int container = NoMdDev;
 968
 969         sra = sysfs_read(fd, 0, GET_VERSION);
 970
 971         if (sra) {
 972                 vers = sra->array.major_version;
 973                 minor = sra->array.minor_version;
 974                 verstr = sra->text_version;
 975         } else {
 976                 if (ioctl(fd, GET_ARRAY_INFO, &array))
 977                         array.major_version = array.minor_version = 0;
 978                 vers = array.major_version;
 979                 minor = array.minor_version;
 980                 verstr = "";
 981         }
 982
 983         if (vers != -1) {
 984                 sprintf(version, "%d.%d", vers, minor);
 985                 verstr = version;
 986         }
 987         if (minor == -2 && is_subarray(verstr)) {
 988                 char *dev = verstr+1;
 989
 990                 subarray = strchr(dev, '/');
 991                 if (subarray) {
 992                         *subarray++ = '\0';
 993                         subarray = xstrdup(subarray);
 994                 }
 995                 container = devname2devnum(dev);
 996                 if (sra)
 997                         sysfs_free(sra);
 998                 sra = sysfs_read(-1, container, GET_VERSION);
 999                 if (sra && sra->text_version[0])
1000                         verstr = sra->text_version;
1001                 else
1002                         verstr = "-no-metadata-";
1003         }
1004
1005         for (i = 0; st == NULL && superlist[i] ; i++)
1006                 st = superlist[i]->match_metadata_desc(verstr);
1007
1008         if (sra)
1009                 sysfs_free(sra);
1010         if (st) {
1011                 st->sb = NULL;
1012                 if (subarrayp)
1013                         *subarrayp = subarray;
1014                 st->container_dev = container;
1015                 st->devnum = fd2devnum(fd);
1016         } else
1017                 free(subarray);
1018
1019         return st;
1020 }
1021 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
1022
1023 int dev_size_from_id(dev_t id, unsigned long long *size)
1024 {
1025         char buf[20];
1026         int fd;
1027
1028         sprintf(buf, "%d:%d", major(id), minor(id));
1029         fd = dev_open(buf, O_RDONLY);
1030         if (fd < 0)
1031                 return 0;
1032         if (get_dev_size(fd, NULL, size)) {
1033                 close(fd);
1034                 return 1;
1035         }
1036         close(fd);
1037         return 0;
1038 }
1039
1040 struct supertype *dup_super(struct supertype *orig)
1041 {
1042         struct supertype *st;
1043
1044         if (!orig)
1045                 return orig;
1046         st = xcalloc(1, sizeof(*st));
1047         st->ss = orig->ss;
1048         st->max_devs = orig->max_devs;
1049         st->minor_version = orig->minor_version;
1050         st->sb = NULL;
1051         st->info = NULL;
1052         return st;
1053 }
1054
1055 struct supertype *guess_super_type(int fd, enum guess_types guess_type)
1056 {
1057         /* try each load_super to find the best match,
1058          * and return the best superswitch
1059          */
1060         struct superswitch  *ss;
1061         struct supertype *st;
1062         time_t besttime = 0;
1063         int bestsuper = -1;
1064         int i;
1065
1066         st = xcalloc(1, sizeof(*st));
1067         st->container_dev = NoMdDev;
1068
1069         for (i = 0 ; superlist[i]; i++) {
1070                 int rv;
1071                 ss = superlist[i];
1072                 if (guess_type == guess_array && ss->add_to_super == NULL)
1073                         continue;
1074                 if (guess_type == guess_partitions && ss->add_to_super != NULL)
1075                         continue;
1076                 memset(st, 0, sizeof(*st));
1077                 st->ignore_hw_compat = 1;
1078                 rv = ss->load_super(st, fd, NULL);
1079                 if (rv == 0) {
1080                         struct mdinfo info;
1081                         st->ss->getinfo_super(st, &info, NULL);
1082                         if (bestsuper == -1 ||
1083                             besttime < info.array.ctime) {
1084                                 bestsuper = i;
1085                                 besttime = info.array.ctime;
1086                         }
1087                         ss->free_super(st);
1088                 }
1089         }
1090         if (bestsuper != -1) {
1091                 int rv;
1092                 memset(st, 0, sizeof(*st));
1093                 st->ignore_hw_compat = 1;
1094                 rv = superlist[bestsuper]->load_super(st, fd, NULL);
1095                 if (rv == 0) {
1096                         superlist[bestsuper]->free_super(st);
1097                         st->ignore_hw_compat = 0;
1098                         return st;
1099                 }
1100         }
1101         free(st);
1102         return NULL;
1103 }
1104
1105 /* Return size of device in bytes */
1106 int get_dev_size(int fd, char *dname, unsigned long long *sizep)
1107 {
1108         unsigned long long ldsize;
1109         struct stat st;
1110
1111         if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
1112                 ldsize = (unsigned long long)st.st_size;
1113         else
1114 #ifdef BLKGETSIZE64
1115         if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0)
1116 #endif
1117         {
1118                 unsigned long dsize;
1119                 if (ioctl(fd, BLKGETSIZE, &dsize) == 0) {
1120                         ldsize = dsize;
1121                         ldsize <<= 9;
1122                 } else {
1123                         if (dname)
1124                                 pr_err("Cannot get size of %s: %s\b",
1125                                         dname, strerror(errno));
1126                         return 0;
1127                 }
1128         }
1129         *sizep = ldsize;
1130         return 1;
1131 }
1132
1133 /* Return true if this can only be a container, not a member device.
1134  * i.e. is and md device and size is zero
1135  */
1136 int must_be_container(int fd)
1137 {
1138         unsigned long long size;
1139         if (md_get_version(fd) < 0)
1140                 return 0;
1141         if (get_dev_size(fd, NULL, &size) == 0)
1142                 return 1;
1143         if (size == 0)
1144                 return 1;
1145         return 0;
1146 }
1147
1148 /* Sets endofpart parameter to the last block used by the last GPT partition on the device.
1149  * Returns: 1 if successful
1150  *         -1 for unknown partition type
1151  *          0 for other errors
1152  */
1153 static int get_gpt_last_partition_end(int fd, unsigned long long *endofpart)
1154 {
1155         struct GPT gpt;
1156         unsigned char empty_gpt_entry[16]= {0};
1157         struct GPT_part_entry *part;
1158         char buf[512];
1159         unsigned long long curr_part_end;
1160         unsigned all_partitions, entry_size;
1161         unsigned part_nr;
1162
1163         *endofpart = 0;
1164
1165         BUILD_BUG_ON(sizeof(gpt) != 512);
1166         /* skip protective MBR */
1167         lseek(fd, 512, SEEK_SET);
1168         /* read GPT header */
1169         if (read(fd, &gpt, 512) != 512)
1170                 return 0;
1171
1172         /* get the number of partition entries and the entry size */
1173         all_partitions = __le32_to_cpu(gpt.part_cnt);
1174         entry_size = __le32_to_cpu(gpt.part_size);
1175
1176         /* Check GPT signature*/
1177         if (gpt.magic != GPT_SIGNATURE_MAGIC)
1178                 return -1;
1179
1180         /* sanity checks */
1181         if (all_partitions > 1024 ||
1182             entry_size > sizeof(buf))
1183                 return -1;
1184
1185         part = (struct GPT_part_entry *)buf;
1186
1187         for (part_nr = 0; part_nr < all_partitions; part_nr++) {
1188                 /* read partition entry */
1189                 if (read(fd, buf, entry_size) != (ssize_t)entry_size)
1190                         return 0;
1191
1192                 /* is this valid partition? */
1193                 if (memcmp(part->type_guid, empty_gpt_entry, 16) != 0) {
1194                         /* check the last lba for the current partition */
1195                         curr_part_end = __le64_to_cpu(part->ending_lba);
1196                         if (curr_part_end > *endofpart)
1197                                 *endofpart = curr_part_end;
1198                 }
1199
1200         }
1201         return 1;
1202 }
1203
1204 /* Sets endofpart parameter to the last block used by the last partition on the device.
1205  * Returns: 1 if successful
1206  *         -1 for unknown partition type
1207  *          0 for other errors
1208  */
1209 static int get_last_partition_end(int fd, unsigned long long *endofpart)
1210 {
1211         struct MBR boot_sect;
1212         struct MBR_part_record *part;
1213         unsigned long long curr_part_end;
1214         unsigned part_nr;
1215         int retval = 0;
1216
1217         *endofpart = 0;
1218
1219         BUILD_BUG_ON(sizeof(boot_sect) != 512);
1220         /* read MBR */
1221         lseek(fd, 0, 0);
1222         if (read(fd, &boot_sect, 512) != 512)
1223                 goto abort;
1224
1225         /* check MBP signature */
1226         if (boot_sect.magic == MBR_SIGNATURE_MAGIC) {
1227                 retval = 1;
1228                 /* found the correct signature */
1229                 part = boot_sect.parts;
1230
1231                 for (part_nr = 0; part_nr < MBR_PARTITIONS; part_nr++) {
1232                         /* check for GPT type */
1233                         if (part->part_type == MBR_GPT_PARTITION_TYPE) {
1234                                 retval = get_gpt_last_partition_end(fd, endofpart);
1235                                 break;
1236                         }
1237                         /* check the last used lba for the current partition  */
1238                         curr_part_end = __le32_to_cpu(part->first_sect_lba) +
1239                                 __le32_to_cpu(part->blocks_num);
1240                         if (curr_part_end > *endofpart)
1241                                 *endofpart = curr_part_end;
1242
1243                         part++;
1244                 }
1245         } else {
1246                 /* Unknown partition table */
1247                 retval = -1;
1248         }
1249  abort:
1250         return retval;
1251 }
1252
1253 int check_partitions(int fd, char *dname, unsigned long long freesize,
1254                         unsigned long long size)
1255 {
1256         /*
1257          * Check where the last partition ends
1258          */
1259         unsigned long long endofpart;
1260         int ret;
1261
1262         if ((ret = get_last_partition_end(fd, &endofpart)) > 0) {
1263                 /* There appears to be a partition table here */
1264                 if (freesize == 0) {
1265                         /* partitions will not be visible in new device */
1266                         pr_err("partition table exists on %s but will be lost or\n"
1267                                "       meaningless after creating array\n",
1268                                dname);
1269                         return 1;
1270                 } else if (endofpart > freesize) {
1271                         /* last partition overlaps metadata */
1272                         pr_err("metadata will over-write last partition on %s.\n",
1273                                dname);
1274                         return 1;
1275                 } else if (size && endofpart > size) {
1276                         /* partitions will be truncated in new device */
1277                         pr_err("array size is too small to cover all partitions on %s.\n",
1278                                dname);
1279                         return 1;
1280                 }
1281         }
1282         return 0;
1283 }
1284
1285 void get_one_disk(int mdfd, mdu_array_info_t *ainf, mdu_disk_info_t *disk)
1286 {
1287         int d;
1288
1289         ioctl(mdfd, GET_ARRAY_INFO, ainf);
1290         for (d = 0 ; d < MAX_DISKS ; d++) {
1291                 if (ioctl(mdfd, GET_DISK_INFO, disk) == 0 &&
1292                     (disk->major || disk->minor))
1293                         return;
1294         }
1295 }
1296
1297 int open_container(int fd)
1298 {
1299         /* 'fd' is a block device.  Find out if it is in use
1300          * by a container, and return an open fd on that container.
1301          */
1302         char path[256];
1303         char *e;
1304         DIR *dir;
1305         struct dirent *de;
1306         int dfd, n;
1307         char buf[200];
1308         int major, minor;
1309         struct stat st;
1310
1311         if (fstat(fd, &st) != 0)
1312                 return -1;
1313         sprintf(path, "/sys/dev/block/%d:%d/holders",
1314                 (int)major(st.st_rdev), (int)minor(st.st_rdev));
1315         e = path + strlen(path);
1316
1317         dir = opendir(path);
1318         if (!dir)
1319                 return -1;
1320         while ((de = readdir(dir))) {
1321                 if (de->d_ino == 0)
1322                         continue;
1323                 if (de->d_name[0] == '.')
1324                         continue;
1325                 sprintf(e, "/%s/dev", de->d_name);
1326                 dfd = open(path, O_RDONLY);
1327                 if (dfd < 0)
1328                         continue;
1329                 n = read(dfd, buf, sizeof(buf));
1330                 close(dfd);
1331                 if (n <= 0 || (unsigned)n >= sizeof(buf))
1332                         continue;
1333                 buf[n] = 0;
1334                 if (sscanf(buf, "%d:%d", &major, &minor) != 2)
1335                         continue;
1336                 sprintf(buf, "%d:%d", major, minor);
1337                 dfd = dev_open(buf, O_RDONLY);
1338                 if (dfd >= 0) {
1339                         closedir(dir);
1340                         return dfd;
1341                 }
1342         }
1343         closedir(dir);
1344         return -1;
1345 }
1346
1347 struct superswitch *version_to_superswitch(char *vers)
1348 {
1349         int i;
1350
1351         for (i = 0; superlist[i]; i++) {
1352                 struct superswitch *ss = superlist[i];
1353
1354                 if (strcmp(vers, ss->name) == 0)
1355                         return ss;
1356         }
1357
1358         return NULL;
1359 }
1360
1361 int is_container_member(struct mdstat_ent *mdstat, char *container)
1362 {
1363         if (mdstat->metadata_version == NULL ||
1364             strncmp(mdstat->metadata_version, "external:", 9) != 0 ||
1365             !is_subarray(mdstat->metadata_version+9) ||
1366             strncmp(mdstat->metadata_version+10, container, strlen(container)) != 0 ||
1367             mdstat->metadata_version[10+strlen(container)] != '/')
1368                 return 0;
1369
1370         return 1;
1371 }
1372
1373 int is_subarray_active(char *subarray, char *container)
1374 {
1375         struct mdstat_ent *mdstat = mdstat_read(0, 0);
1376         struct mdstat_ent *ent;
1377
1378         for (ent = mdstat; ent; ent = ent->next)
1379                 if (is_container_member(ent, container))
1380                         if (strcmp(to_subarray(ent, container), subarray) == 0)
1381                                 break;
1382
1383         free_mdstat(mdstat);
1384
1385         return ent != NULL;
1386 }
1387
1388 /* open_subarray - opens a subarray in a container
1389  * @dev: container device name
1390  * @st: empty supertype
1391  * @quiet: block reporting errors flag
1392  *
1393  * On success returns an fd to a container and fills in *st
1394  */
1395 int open_subarray(char *dev, char *subarray, struct supertype *st, int quiet)
1396 {
1397         struct mdinfo *mdi;
1398         struct mdinfo *info;
1399         int fd, err = 1;
1400
1401         fd = open(dev, O_RDWR|O_EXCL);
1402         if (fd < 0) {
1403                 if (!quiet)
1404                         pr_err("Couldn't open %s, aborting\n",
1405                                 dev);
1406                 return -1;
1407         }
1408
1409         st->devnum = fd2devnum(fd);
1410         if (st->devnum == NoMdDev) {
1411                 if (!quiet)
1412                         pr_err("Failed to determine device number for %s\n",
1413                                dev);
1414                 goto close_fd;
1415         }
1416
1417         mdi = sysfs_read(fd, st->devnum, GET_VERSION|GET_LEVEL);
1418         if (!mdi) {
1419                 if (!quiet)
1420                         pr_err("Failed to read sysfs for %s\n",
1421                                 dev);
1422                 goto close_fd;
1423         }
1424
1425         if (mdi->array.level != UnSet) {
1426                 if (!quiet)
1427                         pr_err("%s is not a container\n", dev);
1428                 goto free_sysfs;
1429         }
1430
1431         st->ss = version_to_superswitch(mdi->text_version);
1432         if (!st->ss) {
1433                 if (!quiet)
1434                         pr_err("Operation not supported for %s metadata\n",
1435                                mdi->text_version);
1436                 goto free_sysfs;
1437         }
1438
1439         st->devname = devnum2devname(st->devnum);
1440         if (!st->devname) {
1441                 if (!quiet)
1442                         pr_err("Failed to allocate device name\n");
1443                 goto free_sysfs;
1444         }
1445
1446         if (!st->ss->load_container) {
1447                 if (!quiet)
1448                         pr_err("%s is not a container\n", dev);
1449                 goto free_name;
1450         }
1451
1452         if (st->ss->load_container(st, fd, NULL)) {
1453                 if (!quiet)
1454                         pr_err("Failed to load metadata for %s\n",
1455                                 dev);
1456                 goto free_name;
1457         }
1458
1459         info = st->ss->container_content(st, subarray);
1460         if (!info) {
1461                 if (!quiet)
1462                         pr_err("Failed to find subarray-%s in %s\n",
1463                                 subarray, dev);
1464                 goto free_super;
1465         }
1466         free(info);
1467
1468         err = 0;
1469
1470  free_super:
1471         if (err)
1472                 st->ss->free_super(st);
1473  free_name:
1474         if (err)
1475                 free(st->devname);
1476  free_sysfs:
1477         sysfs_free(mdi);
1478  close_fd:
1479         if (err)
1480                 close(fd);
1481
1482         if (err)
1483                 return -1;
1484         else
1485                 return fd;
1486 }
1487
1488 int add_disk(int mdfd, struct supertype *st,
1489              struct mdinfo *sra, struct mdinfo *info)
1490 {
1491         /* Add a device to an array, in one of 2 ways. */
1492         int rv;
1493 #ifndef MDASSEMBLE
1494         if (st->ss->external) {
1495                 if (info->disk.state & (1<<MD_DISK_SYNC))
1496                         info->recovery_start = MaxSector;
1497                 else
1498                         info->recovery_start = 0;
1499                 rv = sysfs_add_disk(sra, info, 0);
1500                 if (! rv) {
1501                         struct mdinfo *sd2;
1502                         for (sd2 = sra->devs; sd2; sd2=sd2->next)
1503                                 if (sd2 == info)
1504                                         break;
1505                         if (sd2 == NULL) {
1506                                 sd2 = xmalloc(sizeof(*sd2));
1507                                 *sd2 = *info;
1508                                 sd2->next = sra->devs;
1509                                 sra->devs = sd2;
1510                         }
1511                 }
1512         } else
1513 #endif
1514                 rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk);
1515         return rv;
1516 }
1517
1518 int remove_disk(int mdfd, struct supertype *st,
1519                 struct mdinfo *sra, struct mdinfo *info)
1520 {
1521         int rv;
1522         /* Remove the disk given by 'info' from the array */
1523 #ifndef MDASSEMBLE
1524         if (st->ss->external)
1525                 rv = sysfs_set_str(sra, info, "slot", "none");
1526         else
1527 #endif
1528                 rv = ioctl(mdfd, HOT_REMOVE_DISK, makedev(info->disk.major,
1529                                                           info->disk.minor));
1530         return rv;
1531 }
1532
1533 int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info)
1534 {
1535         /* Initialise kernel's knowledge of array.
1536          * This varies between externally managed arrays
1537          * and older kernels
1538          */
1539         int vers = md_get_version(mdfd);
1540         int rv;
1541
1542 #ifndef MDASSEMBLE
1543         if (st->ss->external)
1544                 rv = sysfs_set_array(info, vers);
1545         else
1546 #endif
1547                 if ((vers % 100) >= 1) { /* can use different versions */
1548                 mdu_array_info_t inf;
1549                 memset(&inf, 0, sizeof(inf));
1550                 inf.major_version = info->array.major_version;
1551                 inf.minor_version = info->array.minor_version;
1552                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
1553         } else
1554                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
1555         return rv;
1556 }
1557
1558 unsigned long long min_recovery_start(struct mdinfo *array)
1559 {
1560         /* find the minimum recovery_start in an array for metadata
1561          * formats that only record per-array recovery progress instead
1562          * of per-device
1563          */
1564         unsigned long long recovery_start = MaxSector;
1565         struct mdinfo *d;
1566
1567         for (d = array->devs; d; d = d->next)
1568                 recovery_start = min(recovery_start, d->recovery_start);
1569
1570         return recovery_start;
1571 }
1572
1573 int mdmon_pid(int devnum)
1574 {
1575         char path[100];
1576         char pid[10];
1577         int fd;
1578         int n;
1579         char *devname = devnum2devname(devnum);
1580
1581         sprintf(path, "%s/%s.pid", MDMON_DIR, devname);
1582         free(devname);
1583
1584         fd = open(path, O_RDONLY | O_NOATIME, 0);
1585
1586         if (fd < 0)
1587                 return -1;
1588         n = read(fd, pid, 9);
1589         close(fd);
1590         if (n <= 0)
1591                 return -1;
1592         return atoi(pid);
1593 }
1594
1595 int mdmon_running(int devnum)
1596 {
1597         int pid = mdmon_pid(devnum);
1598         if (pid <= 0)
1599                 return 0;
1600         if (kill(pid, 0) == 0)
1601                 return 1;
1602         return 0;
1603 }
1604
1605 int start_mdmon(int devnum)
1606 {
1607         int i, skipped;
1608         int len;
1609         pid_t pid;
1610         int status;
1611         char pathbuf[1024];
1612         char *paths[4] = {
1613                 pathbuf,
1614                 "/sbin/mdmon",
1615                 "mdmon",
1616                 NULL
1617         };
1618
1619         if (check_env("MDADM_NO_MDMON"))
1620                 return 0;
1621
1622         len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf)-1);
1623         if (len > 0) {
1624                 char *sl;
1625                 pathbuf[len] = 0;
1626                 sl = strrchr(pathbuf, '/');
1627                 if (sl)
1628                         sl++;
1629                 else
1630                         sl = pathbuf;
1631                 strcpy(sl, "mdmon");
1632         } else
1633                 pathbuf[0] = '\0';
1634
1635         switch(fork()) {
1636         case 0:
1637                 /* FIXME yuk. CLOSE_EXEC?? */
1638                 skipped = 0;
1639                 for (i = 3; skipped < 20; i++)
1640                         if (close(i) < 0)
1641                                 skipped++;
1642                         else
1643                                 skipped = 0;
1644
1645                 for (i = 0; paths[i]; i++)
1646                         if (paths[i][0]) {
1647                                 if (__offroot) {
1648                                         execl(paths[i], "mdmon", "--offroot",
1649                                               devnum2devname(devnum),
1650                                               NULL);
1651                                 } else {
1652                                         execl(paths[i], "mdmon",
1653                                               devnum2devname(devnum),
1654                                               NULL);
1655                                 }
1656                         }
1657                 exit(1);
1658         case -1: pr_err("cannot run mdmon. "
1659                          "Array remains readonly\n");
1660                 return -1;
1661         default: /* parent - good */
1662                 pid = wait(&status);
1663                 if (pid < 0 || status != 0)
1664                         return -1;
1665         }
1666         return 0;
1667 }
1668
1669 int check_env(char *name)
1670 {
1671         char *val = getenv(name);
1672
1673         if (val && atoi(val) == 1)
1674                 return 1;
1675
1676         return 0;
1677 }
1678
1679 __u32 random32(void)
1680 {
1681         __u32 rv;
1682         int rfd = open("/dev/urandom", O_RDONLY);
1683         if (rfd < 0 || read(rfd, &rv, 4) != 4)
1684                 rv = random();
1685         if (rfd >= 0)
1686                 close(rfd);
1687         return rv;
1688 }
1689
1690 #ifndef MDASSEMBLE
1691 int flush_metadata_updates(struct supertype *st)
1692 {
1693         int sfd;
1694         if (!st->updates) {
1695                 st->update_tail = NULL;
1696                 return -1;
1697         }
1698
1699         sfd = connect_monitor(devnum2devname(st->container_dev));
1700         if (sfd < 0)
1701                 return -1;
1702
1703         while (st->updates) {
1704                 struct metadata_update *mu = st->updates;
1705                 st->updates = mu->next;
1706
1707                 send_message(sfd, mu, 0);
1708                 wait_reply(sfd, 0);
1709                 free(mu->buf);
1710                 free(mu);
1711         }
1712         ack(sfd, 0);
1713         wait_reply(sfd, 0);
1714         close(sfd);
1715         st->update_tail = NULL;
1716         return 0;
1717 }
1718
1719 void append_metadata_update(struct supertype *st, void *buf, int len)
1720 {
1721
1722         struct metadata_update *mu = xmalloc(sizeof(*mu));
1723
1724         mu->buf = buf;
1725         mu->len = len;
1726         mu->space = NULL;
1727         mu->space_list = NULL;
1728         mu->next = NULL;
1729         *st->update_tail = mu;
1730         st->update_tail = &mu->next;
1731 }
1732 #endif /* MDASSEMBLE */
1733
1734 #ifdef __TINYC__
1735 /* tinyc doesn't optimize this check in ioctl.h out ... */
1736 unsigned int __invalid_size_argument_for_IOC = 0;
1737 #endif
1738
1739 int experimental(void)
1740 {
1741         if (check_env("MDADM_EXPERIMENTAL"))
1742                 return 1;
1743         else {
1744                 pr_err("To use this feature MDADM_EXPERIMENTAL"
1745                                 " environment variable has to be defined.\n");
1746                 return 0;
1747         }
1748 }
1749
1750 /* Pick all spares matching given criteria from a container
1751  * if min_size == 0 do not check size
1752  * if domlist == NULL do not check domains
1753  * if spare_group given add it to domains of each spare
1754  * metadata allows to test domains using metadata of destination array */
1755 struct mdinfo *container_choose_spares(struct supertype *st,
1756                                        unsigned long long min_size,
1757                                        struct domainlist *domlist,
1758                                        char *spare_group,
1759                                        const char *metadata, int get_one)
1760 {
1761         struct mdinfo *d, **dp, *disks = NULL;
1762
1763         /* get list of all disks in container */
1764         if (st->ss->getinfo_super_disks)
1765                 disks = st->ss->getinfo_super_disks(st);
1766
1767         if (!disks)
1768                 return disks;
1769         /* find spare devices on the list */
1770         dp = &disks->devs;
1771         disks->array.spare_disks = 0;
1772         while (*dp) {
1773                 int found = 0;
1774                 d = *dp;
1775                 if (d->disk.state == 0) {
1776                         /* check if size is acceptable */
1777                         unsigned long long dev_size;
1778                         dev_t dev = makedev(d->disk.major,d->disk.minor);
1779
1780                         if (!min_size ||
1781                            (dev_size_from_id(dev,  &dev_size) &&
1782                             dev_size >= min_size))
1783                                 found = 1;
1784                         /* check if domain matches */
1785                         if (found && domlist) {
1786                                 struct dev_policy *pol = devnum_policy(dev);
1787                                 if (spare_group)
1788                                         pol_add(&pol, pol_domain,
1789                                                 spare_group, NULL);
1790                                 if (domain_test(domlist, pol, metadata) != 1)
1791                                         found = 0;
1792                                 dev_policy_free(pol);
1793                         }
1794                 }
1795                 if (found) {
1796                         dp = &d->next;
1797                         disks->array.spare_disks++;
1798                         if (get_one) {
1799                                 sysfs_free(*dp);
1800                                 d->next = NULL;
1801                         }
1802                 } else {
1803                         *dp = d->next;
1804                         d->next = NULL;
1805                         sysfs_free(d);
1806                 }
1807         }
1808         return disks;
1809 }
1810
1811 void *xmalloc(size_t len)
1812 {
1813         void *rv = malloc(len);
1814         char *msg;
1815         if (rv)
1816                 return rv;
1817         msg = Name ": memory allocation failure - aborting\n";
1818         exit(4+!!write(2, msg, strlen(msg)));
1819 }
1820
1821 void *xrealloc(void *ptr, size_t len)
1822 {
1823         void *rv = realloc(ptr, len);
1824         char *msg;
1825         if (rv)
1826                 return rv;
1827         msg = Name ": memory allocation failure - aborting\n";
1828         exit(4+!!write(2, msg, strlen(msg)));
1829 }
1830
1831 void *xcalloc(size_t num, size_t size)
1832 {
1833         void *rv = calloc(num, size);
1834         char *msg;
1835         if (rv)
1836                 return rv;
1837         msg = Name ": memory allocation failure - aborting\n";
1838         exit(4+!!write(2, msg, strlen(msg)));
1839 }
1840
1841 char *xstrdup(const char *str)
1842 {
1843         char *rv = strdup(str);
1844         char *msg;
1845         if (rv)
1846                 return rv;
1847         msg = Name ": memory allocation failure - aborting\n";
1848         exit(4+!!write(2, msg, strlen(msg)));
1849 }