util.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2006 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@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        "mdadm.h"
  31 #include        "md_p.h"
  32 #include        <sys/utsname.h>
  33 #include        <ctype.h>
  34 #include        <dirent.h>
  35
  36 /*
  37  * following taken from linux/blkpg.h because they aren't
  38  * anywhere else and it isn't safe to #include linux/ * stuff.
  39  */
  40
  41 #define BLKPG      _IO(0x12,105)
  42
  43 /* The argument structure */
  44 struct blkpg_ioctl_arg {
  45         int op;
  46         int flags;
  47         int datalen;
  48         void *data;
  49 };
  50
  51 /* The subfunctions (for the op field) */
  52 #define BLKPG_ADD_PARTITION     1
  53 #define BLKPG_DEL_PARTITION     2
  54
  55 /* Sizes of name fields. Unused at present. */
  56 #define BLKPG_DEVNAMELTH        64
  57 #define BLKPG_VOLNAMELTH        64
  58
  59 /* The data structure for ADD_PARTITION and DEL_PARTITION */
  60 struct blkpg_partition {
  61         long long start;                /* starting offset in bytes */
  62         long long length;               /* length in bytes */
  63         int pno;                        /* partition number */
  64         char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2,
  65                                            to be used in kernel messages */
  66         char volname[BLKPG_VOLNAMELTH]; /* volume label */
  67 };
  68
  69 /*
  70  * Parse a 128 bit uuid in 4 integers
  71  * format is 32 hexx nibbles with options :.<space> separator
  72  * If not exactly 32 hex digits are found, return 0
  73  * else return 1
  74  */
  75 int parse_uuid(char *str, int uuid[4])
  76 {
  77         int hit = 0; /* number of Hex digIT */
  78         int i;
  79         char c;
  80         for (i=0; i<4; i++) uuid[i]=0;
  81
  82         while ((c= *str++)) {
  83                 int n;
  84                 if (c>='0' && c<='9')
  85                         n = c-'0';
  86                 else if (c>='a' && c <= 'f')
  87                         n = 10 + c - 'a';
  88                 else if (c>='A' && c <= 'F')
  89                         n = 10 + c - 'A';
  90                 else if (strchr(":. -", c))
  91                         continue;
  92                 else return 0;
  93
  94                 if (hit<32) {
  95                         uuid[hit/8] <<= 4;
  96                         uuid[hit/8] += n;
  97                 }
  98                 hit++;
  99         }
 100         if (hit == 32)
 101                 return 1;
 102         return 0;
 103 }
 104
 105
 106 /*
 107  * Get the md version number.
 108  * We use the RAID_VERSION ioctl if it is supported
 109  * If not, but we have a block device with major '9', we assume
 110  * 0.36.0
 111  *
 112  * Return version number as 24 but number - assume version parts
 113  * always < 255
 114  */
 115
 116 int md_get_version(int fd)
 117 {
 118     struct stat stb;
 119     mdu_version_t vers;
 120
 121     if (fstat(fd, &stb)<0)
 122         return -1;
 123     if ((S_IFMT&stb.st_mode) != S_IFBLK)
 124         return -1;
 125
 126     if (ioctl(fd, RAID_VERSION, &vers) == 0)
 127         return  (vers.major*10000) + (vers.minor*100) + vers.patchlevel;
 128     if (errno == EACCES)
 129             return -1;
 130     if (major(stb.st_rdev) == MD_MAJOR)
 131         return (3600);
 132     return -1;
 133 }
 134
 135 int get_linux_version()
 136 {
 137         struct utsname name;
 138         char *cp;
 139         int a,b,c;
 140         if (uname(&name) <0)
 141                 return -1;
 142
 143         cp = name.release;
 144         a = strtoul(cp, &cp, 10);
 145         if (*cp != '.') return -1;
 146         b = strtoul(cp+1, &cp, 10);
 147         if (*cp != '.') return -1;
 148         c = strtoul(cp+1, NULL, 10);
 149
 150         return (a*1000000)+(b*1000)+c;
 151 }
 152
 153 void remove_partitions(int fd)
 154 {
 155         /* remove partitions from this block devices.
 156          * This is used for components added to an array
 157          */
 158 #ifdef BLKPG_DEL_PARTITION
 159         struct blkpg_ioctl_arg a;
 160         struct blkpg_partition p;
 161
 162         a.op = BLKPG_DEL_PARTITION;
 163         a.data = (void*)&p;
 164         a.datalen = sizeof(p);
 165         a.flags = 0;
 166         memset(a.data, 0, a.datalen);
 167         for (p.pno=0; p.pno < 16; p.pno++)
 168                 ioctl(fd, BLKPG, &a);
 169 #endif
 170 }
 171
 172 int enough(int level, int raid_disks, int layout, int clean,
 173            char *avail, int avail_disks)
 174 {
 175         int copies, first;
 176         switch (level) {
 177         case 10:
 178                 /* This is the tricky one - we need to check
 179                  * which actual disks are present.
 180                  */
 181                 copies = (layout&255)* ((layout>>8) & 255);
 182                 first=0;
 183                 do {
 184                         /* there must be one of the 'copies' form 'first' */
 185                         int n = copies;
 186                         int cnt=0;
 187                         while (n--) {
 188                                 if (avail[first])
 189                                         cnt++;
 190                                 first = (first+1) % raid_disks;
 191                         }
 192                         if (cnt == 0)
 193                                 return 0;
 194
 195                 } while (first != 0);
 196                 return 1;
 197
 198         case -4:
 199                 return avail_disks>= 1;
 200         case -1:
 201         case 0:
 202                 return avail_disks == raid_disks;
 203         case 1:
 204                 return avail_disks >= 1;
 205         case 4:
 206         case 5:
 207                 if (clean)
 208                         return avail_disks >= raid_disks-1;
 209                 else
 210                         return avail_disks >= raid_disks;
 211         case 6:
 212                 if (clean)
 213                         return avail_disks >= raid_disks-2;
 214                 else
 215                         return avail_disks >= raid_disks;
 216         default:
 217                 return 0;
 218         }
 219 }
 220
 221 int same_uuid(int a[4], int b[4], int swapuuid)
 222 {
 223         if (swapuuid) {
 224                 /* parse uuids are hostendian.
 225                  * uuid's from some superblocks are big-ending
 226                  * if there is a difference, we need to swap..
 227                  */
 228                 unsigned char *ac = (unsigned char *)a;
 229                 unsigned char *bc = (unsigned char *)b;
 230                 int i;
 231                 for (i=0; i<16; i+= 4) {
 232                         if (ac[i+0] != bc[i+3] ||
 233                             ac[i+1] != bc[i+2] ||
 234                             ac[i+2] != bc[i+1] ||
 235                             ac[i+3] != bc[i+0])
 236                                 return 0;
 237                 }
 238                 return 1;
 239         } else {
 240                 if (a[0]==b[0] &&
 241                     a[1]==b[1] &&
 242                     a[2]==b[2] &&
 243                     a[3]==b[3])
 244                         return 1;
 245                 return 0;
 246         }
 247 }
 248 void copy_uuid(void *a, int b[4], int swapuuid)
 249 {
 250         if (swapuuid) {
 251                 /* parse uuids are hostendian.
 252                  * uuid's from some superblocks are big-ending
 253                  * if there is a difference, we need to swap..
 254                  */
 255                 unsigned char *ac = (unsigned char *)a;
 256                 unsigned char *bc = (unsigned char *)b;
 257                 int i;
 258                 for (i=0; i<16; i+= 4) {
 259                         ac[i+0] = bc[i+3];
 260                         ac[i+1] = bc[i+2];
 261                         ac[i+2] = bc[i+1];
 262                         ac[i+3] = bc[i+0];
 263                 }
 264         } else
 265                 memcpy(a, b, 16);
 266 }
 267
 268 #ifndef MDASSEMBLE
 269 int check_ext2(int fd, char *name)
 270 {
 271         /*
 272          * Check for an ext2fs file system.
 273          * Superblock is always 1K at 1K offset
 274          *
 275          * s_magic is le16 at 56 == 0xEF53
 276          * report mtime - le32 at 44
 277          * blocks - le32 at 4
 278          * logblksize - le32 at 24
 279          */
 280         unsigned char sb[1024];
 281         time_t mtime;
 282         int size, bsize;
 283         if (lseek(fd, 1024,0)!= 1024)
 284                 return 0;
 285         if (read(fd, sb, 1024)!= 1024)
 286                 return 0;
 287         if (sb[56] != 0x53 || sb[57] != 0xef)
 288                 return 0;
 289
 290         mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8;
 291         bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8;
 292         size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8;
 293         fprintf(stderr, Name ": %s appears to contain an ext2fs file system\n",
 294                 name);
 295         fprintf(stderr,"    size=%dK  mtime=%s",
 296                 size*(1<<bsize), ctime(&mtime));
 297         return 1;
 298 }
 299
 300 int check_reiser(int fd, char *name)
 301 {
 302         /*
 303          * superblock is at 64K
 304          * size is 1024;
 305          * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
 306          *
 307          */
 308         unsigned char sb[1024];
 309         unsigned long size;
 310         if (lseek(fd, 64*1024, 0) != 64*1024)
 311                 return 0;
 312         if (read(fd, sb, 1024) != 1024)
 313                 return 0;
 314         if (strncmp((char*)sb+52, "ReIsErFs",8)!=0 &&
 315             strncmp((char*)sb+52, "ReIsEr2Fs",9)!=0)
 316                 return 0;
 317         fprintf(stderr, Name ": %s appears to contain a reiserfs file system\n",name);
 318         size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8;
 319         fprintf(stderr, "    size = %luK\n", size*4);
 320
 321         return 1;
 322 }
 323
 324 int check_raid(int fd, char *name)
 325 {
 326         struct mdinfo info;
 327         time_t crtime;
 328         char *level;
 329         struct supertype *st = guess_super(fd);
 330
 331         if (!st) return 0;
 332         st->ss->load_super(st, fd, name);
 333         /* Looks like a raid array .. */
 334         fprintf(stderr, Name ": %s appears to be part of a raid array:\n",
 335                 name);
 336         st->ss->getinfo_super(st, &info);
 337         st->ss->free_super(st);
 338         crtime = info.array.ctime;
 339         level = map_num(pers, info.array.level);
 340         if (!level) level = "-unknown-";
 341         fprintf(stderr, "    level=%s devices=%d ctime=%s",
 342                 level, info.array.raid_disks, ctime(&crtime));
 343         return 1;
 344 }
 345
 346 int ask(char *mesg)
 347 {
 348         char *add = "";
 349         int i;
 350         for (i=0; i<5; i++) {
 351                 char buf[100];
 352                 fprintf(stderr, "%s%s", mesg, add);
 353                 fflush(stderr);
 354                 if (fgets(buf, 100, stdin)==NULL)
 355                         return 0;
 356                 if (buf[0]=='y' || buf[0]=='Y')
 357                         return 1;
 358                 if (buf[0]=='n' || buf[0]=='N')
 359                         return 0;
 360                 add = "(y/n) ";
 361         }
 362         fprintf(stderr, Name ": assuming 'no'\n");
 363         return 0;
 364 }
 365 #endif /* MDASSEMBLE */
 366
 367 char *map_num(mapping_t *map, int num)
 368 {
 369         while (map->name) {
 370                 if (map->num == num)
 371                         return map->name;
 372                 map++;
 373         }
 374         return NULL;
 375 }
 376
 377 int map_name(mapping_t *map, char *name)
 378 {
 379         while (map->name) {
 380                 if (strcmp(map->name, name)==0)
 381                         return map->num;
 382                 map++;
 383         }
 384         return UnSet;
 385 }
 386
 387
 388 int is_standard(char *dev, int *nump)
 389 {
 390         /* tests if dev is a "standard" md dev name.
 391          * i.e if the last component is "/dNN" or "/mdNN",
 392          * where NN is a string of digits
 393          */
 394         char *d = strrchr(dev, '/');
 395         int type=0;
 396         int num;
 397         if (!d)
 398                 return 0;
 399         if (strncmp(d, "/d",2)==0)
 400                 d += 2, type=1; /* /dev/md/dN{pM} */
 401         else if (strncmp(d, "/md_d", 5)==0)
 402                 d += 5, type=1; /* /dev/md_dNpM */
 403         else if (strncmp(d, "/md", 3)==0)
 404                 d += 3, type=-1; /* /dev/mdN */
 405         else if (d-dev > 3 && strncmp(d-2, "md/", 3)==0)
 406                 d += 1, type=-1; /* /dev/md/N */
 407         else
 408                 return 0;
 409         if (!*d)
 410                 return 0;
 411         num = atoi(d);
 412         while (isdigit(*d))
 413                 d++;
 414         if (*d)
 415                 return 0;
 416         if (nump) *nump = num;
 417
 418         return type;
 419 }
 420
 421
 422 /*
 423  * convert a major/minor pair for a block device into a name in /dev, if possible.
 424  * On the first call, walk /dev collecting name.
 425  * Put them in a simple linked listfor now.
 426  */
 427 struct devmap {
 428     int major, minor;
 429     char *name;
 430     struct devmap *next;
 431 } *devlist = NULL;
 432 int devlist_ready = 0;
 433
 434 int add_dev(const char *name, const struct stat *stb, int flag, struct FTW *s)
 435 {
 436         struct stat st;
 437         if (S_ISLNK(stb->st_mode)) {
 438                 stat(name, &st);
 439                 stb = &st;
 440         }
 441
 442         if ((stb->st_mode&S_IFMT)== S_IFBLK) {
 443                 char *n = strdup(name);
 444                 struct devmap *dm = malloc(sizeof(*dm));
 445                 if (strncmp(n, "/dev/./", 7)==0)
 446                         strcpy(n+4, name+6);
 447                 if (dm) {
 448                         dm->major = major(stb->st_rdev);
 449                         dm->minor = minor(stb->st_rdev);
 450                         dm->name = n;
 451                         dm->next = devlist;
 452                         devlist = dm;
 453                 }
 454         }
 455         return 0;
 456 }
 457
 458 #ifndef HAVE_NFTW
 459 #ifdef HAVE_FTW
 460 int add_dev_1(const char *name, const struct stat *stb, int flag)
 461 {
 462         return add_dev(name, stb, flag, NULL);
 463 }
 464 int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags)
 465 {
 466         return ftw(path, add_dev_1, nopenfd);
 467 }
 468 #else
 469 int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags)
 470 {
 471         return 0;
 472 }
 473 #endif /* HAVE_FTW */
 474 #endif /* HAVE_NFTW */
 475
 476 /*
 477  * Find a block device with the right major/minor number.
 478  * If we find multiple names, choose the shortest.
 479  * If we find a non-standard name, it is probably there
 480  * deliberately so prefer it over a standard name.
 481  * This applies only to names for MD devices.
 482  */
 483 char *map_dev(int major, int minor, int create)
 484 {
 485         struct devmap *p;
 486         char *std = NULL, *nonstd=NULL;
 487         int did_check = 0;
 488
 489         if (major == 0 && minor == 0)
 490                         return NULL;
 491
 492  retry:
 493         if (!devlist_ready) {
 494                 char *dev = "/dev";
 495                 struct stat stb;
 496                 while(devlist) {
 497                         struct devmap *d = devlist;
 498                         devlist = d->next;
 499                         free(d->name);
 500                         free(d);
 501                 }
 502                 if (lstat(dev, &stb)==0 &&
 503                     S_ISLNK(stb.st_mode))
 504                         dev = "/dev/.";
 505                 nftw(dev, add_dev, 10, FTW_PHYS);
 506                 devlist_ready=1;
 507                 did_check = 1;
 508         }
 509
 510         for (p=devlist; p; p=p->next)
 511                 if (p->major == major &&
 512                     p->minor == minor) {
 513                         if (is_standard(p->name, NULL)) {
 514                                 if (std == NULL ||
 515                                     strlen(p->name) < strlen(std))
 516                                         std = p->name;
 517                         } else {
 518                                 if (nonstd == NULL ||
 519                                     strlen(p->name) < strlen(nonstd))
 520                                         nonstd = p->name;
 521                         }
 522                 }
 523         if (!std && !nonstd && !did_check) {
 524                 devlist_ready = 0;
 525                 goto retry;
 526         }
 527         if (create && !std && !nonstd) {
 528                 static char buf[30];
 529                 snprintf(buf, sizeof(buf), "%d:%d", major, minor);
 530                 nonstd = buf;
 531         }
 532
 533         return nonstd ? nonstd : std;
 534 }
 535
 536 unsigned long calc_csum(void *super, int bytes)
 537 {
 538         unsigned long long newcsum = 0;
 539         int i;
 540         unsigned int csum;
 541         unsigned int *superc = (unsigned int*) super;
 542
 543         for(i=0; i<bytes/4; i++)
 544                 newcsum+= superc[i];
 545         csum = (newcsum& 0xffffffff) + (newcsum>>32);
 546 #ifdef __alpha__
 547 /* The in-kernel checksum calculation is always 16bit on
 548  * the alpha, though it is 32 bit on i386...
 549  * I wonder what it is elsewhere... (it uses and API in
 550  * a way that it shouldn't).
 551  */
 552         csum = (csum & 0xffff) + (csum >> 16);
 553         csum = (csum & 0xffff) + (csum >> 16);
 554 #endif
 555         return csum;
 556 }
 557
 558 #ifndef MDASSEMBLE
 559 char *human_size(long long bytes)
 560 {
 561         static char buf[30];
 562
 563         /* We convert bytes to either centi-M{ega,ibi}bytes or
 564          * centi-G{igi,ibi}bytes, with appropriate rounding,
 565          * and then print 1/100th of those as a decimal.
 566          * We allow upto 2048Megabytes before converting to
 567          * gigabytes, as that shows more precision and isn't
 568          * too large a number.
 569          * Terrabytes are not yet handled.
 570          */
 571
 572         if (bytes < 5000*1024)
 573                 buf[0]=0;
 574         else if (bytes < 2*1024LL*1024LL*1024LL) {
 575                 long cMiB = (bytes / ( (1LL<<20) / 200LL ) +1) /2;
 576                 long cMB  = (bytes / ( 1000000LL / 200LL ) +1) /2;
 577                 snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)",
 578                         cMiB/100 , cMiB % 100,
 579                         cMB/100, cMB % 100);
 580         } else {
 581                 long cGiB = (bytes / ( (1LL<<30) / 200LL ) +1) /2;
 582                 long cGB  = (bytes / (1000000000LL/200LL ) +1) /2;
 583                 snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)",
 584                         cGiB/100 , cGiB % 100,
 585                         cGB/100, cGB % 100);
 586         }
 587         return buf;
 588 }
 589
 590 char *human_size_brief(long long bytes)
 591 {
 592         static char buf[30];
 593
 594         if (bytes < 5000*1024)
 595                 snprintf(buf, sizeof(buf), "%ld.%02ldKiB",
 596                         (long)(bytes>>10), (long)(((bytes&1023)*100+512)/1024)
 597                         );
 598         else if (bytes < 2*1024LL*1024LL*1024LL)
 599                 snprintf(buf, sizeof(buf), "%ld.%02ldMiB",
 600                         (long)(bytes>>20),
 601                         (long)((bytes&0xfffff)+0x100000/200)/(0x100000/100)
 602                         );
 603         else
 604                 snprintf(buf, sizeof(buf), "%ld.%02ldGiB",
 605                         (long)(bytes>>30),
 606                         (long)(((bytes>>10)&0xfffff)+0x100000/200)/(0x100000/100)
 607                         );
 608         return buf;
 609 }
 610 #endif
 611
 612 unsigned long long calc_array_size(int level, int raid_disks, int layout,
 613                                    int chunksize, unsigned long long devsize)
 614 {
 615         int data_disks = 0;
 616         switch (level) {
 617         case 0: data_disks = raid_disks; break;
 618         case 1: data_disks = 1; break;
 619         case 4:
 620         case 5: data_disks = raid_disks - 1; break;
 621         case 6: data_disks = raid_disks - 2; break;
 622         case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255);
 623                 break;
 624         }
 625         devsize &= ~(unsigned long long)((chunksize>>9)-1);
 626         return data_disks * devsize;
 627 }
 628
 629 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
 630 int get_mdp_major(void)
 631 {
 632 static int mdp_major = -1;
 633         FILE *fl;
 634         char *w;
 635         int have_block = 0;
 636         int have_devices = 0;
 637         int last_num = -1;
 638
 639         if (mdp_major != -1)
 640                 return mdp_major;
 641         fl = fopen("/proc/devices", "r");
 642         if (!fl)
 643                 return -1;
 644         while ((w = conf_word(fl, 1))) {
 645                 if (have_block && strcmp(w, "devices:")==0)
 646                         have_devices = 1;
 647                 have_block =  (strcmp(w, "Block")==0);
 648                 if (isdigit(w[0]))
 649                         last_num = atoi(w);
 650                 if (have_devices && strcmp(w, "mdp")==0)
 651                         mdp_major = last_num;
 652                 free(w);
 653         }
 654         fclose(fl);
 655         return mdp_major;
 656 }
 657
 658
 659
 660 char *get_md_name(int dev)
 661 {
 662         /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
 663         /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
 664         static char devname[50];
 665         struct stat stb;
 666         dev_t rdev;
 667         char *dn;
 668
 669         if (dev < 0) {
 670                 int mdp =  get_mdp_major();
 671                 if (mdp < 0) return NULL;
 672                 rdev = makedev(mdp, (-1-dev)<<6);
 673                 snprintf(devname, sizeof(devname), "/dev/md/d%d", -1-dev);
 674                 if (stat(devname, &stb) == 0
 675                     && (S_IFMT&stb.st_mode) == S_IFBLK
 676                     && (stb.st_rdev == rdev))
 677                         return devname;
 678         } else {
 679                 rdev = makedev(MD_MAJOR, dev);
 680                 snprintf(devname, sizeof(devname), "/dev/md%d", dev);
 681                 if (stat(devname, &stb) == 0
 682                     && (S_IFMT&stb.st_mode) == S_IFBLK
 683                     && (stb.st_rdev == rdev))
 684                         return devname;
 685
 686                 snprintf(devname, sizeof(devname), "/dev/md/%d", dev);
 687                 if (stat(devname, &stb) == 0
 688                     && (S_IFMT&stb.st_mode) == S_IFBLK
 689                     && (stb.st_rdev == rdev))
 690                         return devname;
 691         }
 692         dn = map_dev(major(rdev), minor(rdev), 0);
 693         if (dn)
 694                 return dn;
 695         snprintf(devname, sizeof(devname), "/dev/.tmp.md%d", dev);
 696         if (mknod(devname, S_IFBLK | 0600, rdev) == -1)
 697                 if (errno != EEXIST)
 698                         return NULL;
 699
 700         if (stat(devname, &stb) == 0
 701             && (S_IFMT&stb.st_mode) == S_IFBLK
 702             && (stb.st_rdev == rdev))
 703                 return devname;
 704         unlink(devname);
 705         return NULL;
 706 }
 707
 708 void put_md_name(char *name)
 709 {
 710         if (strncmp(name, "/dev/.tmp.md", 12)==0)
 711                 unlink(name);
 712 }
 713
 714 static int dev2major(int d)
 715 {
 716         if (d >= 0)
 717                 return MD_MAJOR;
 718         else
 719                 return get_mdp_major();
 720 }
 721
 722 static int dev2minor(int d)
 723 {
 724         if (d >= 0)
 725                 return d;
 726         return (-1-d) << MdpMinorShift;
 727 }
 728
 729 int find_free_devnum(int use_partitions)
 730 {
 731         int devnum;
 732         for (devnum = 127; devnum != 128;
 733              devnum = devnum ? devnum-1 : (1<<22)-1) {
 734                 char *dn;
 735                 int _devnum;
 736
 737                 _devnum = use_partitions ? (-1-devnum) : devnum;
 738                 if (mddev_busy(_devnum))
 739                         continue;
 740                 /* make sure it is new to /dev too, at least as a
 741                  * non-standard */
 742                 dn = map_dev(dev2major(_devnum), dev2minor(_devnum), 0);
 743                 if (dn && ! is_standard(dn, NULL))
 744                         continue;
 745                 break;
 746         }
 747         if (devnum == 128)
 748                 return NoMdDev;
 749         return use_partitions ? (-1-devnum) : devnum;
 750 }
 751 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
 752
 753 int dev_open(char *dev, int flags)
 754 {
 755         /* like 'open', but if 'dev' matches %d:%d, create a temp
 756          * block device and open that
 757          */
 758         char *e;
 759         int fd = -1;
 760         char devname[32];
 761         int major;
 762         int minor;
 763
 764         if (!dev) return -1;
 765
 766         major = strtoul(dev, &e, 0);
 767         if (e > dev && *e == ':' && e[1] &&
 768             (minor = strtoul(e+1, &e, 0)) >= 0 &&
 769             *e == 0) {
 770                 snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d", major, minor);
 771                 if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) {
 772                         fd = open(devname, flags);
 773                         unlink(devname);
 774                 }
 775         } else
 776                 fd = open(dev, flags);
 777         return fd;
 778 }
 779
 780 struct superswitch *superlist[] = { &super0, &super1, &super_ddf, NULL };
 781
 782 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
 783 struct supertype *super_by_fd(int fd)
 784 {
 785         mdu_array_info_t array;
 786         int vers;
 787         int minor;
 788         struct supertype *st = NULL;
 789         struct mdinfo *sra;
 790         char *verstr;
 791         char version[20];
 792         int i;
 793
 794         sra = sysfs_read(fd, 0, GET_VERSION);
 795
 796         if (sra) {
 797                 vers = sra->array.major_version;
 798                 minor = sra->array.minor_version;
 799                 verstr = sra->text_version;
 800         } else {
 801                 if (ioctl(fd, GET_ARRAY_INFO, &array))
 802                         array.major_version = array.minor_version = 0;
 803                 vers = array.major_version;
 804                 minor = array.minor_version;
 805                 verstr = "";
 806         }
 807
 808         if (vers != -1) {
 809                 sprintf(version, "%d.%d", vers, minor);
 810                 verstr = version;
 811         }
 812         for (i = 0; st == NULL && superlist[i] ; i++)
 813                 st = superlist[i]->match_metadata_desc(verstr);
 814
 815         if (sra)
 816                 sysfs_free(sra);
 817         if (st)
 818                 st->sb = NULL;
 819         return st;
 820 }
 821 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
 822
 823
 824 struct supertype *dup_super(struct supertype *st)
 825 {
 826         struct supertype *stnew = NULL;
 827         char *verstr = NULL;
 828         char version[20];
 829         int i;
 830
 831         if (!st)
 832                 return st;
 833
 834         if (st->ss->text_version)
 835                 strcpy(version, st->ss->text_version);
 836         else if (st->minor_version == -1)
 837                 sprintf(version, "%d", st->ss->major);
 838         else
 839                 sprintf(version, "%d.%d", st->ss->major, st->minor_version);
 840         verstr = version;
 841
 842         for (i = 0; stnew == NULL && superlist[i] ; i++)
 843                 stnew = superlist[i]->match_metadata_desc(verstr);
 844
 845         if (stnew)
 846                 stnew->sb = NULL;
 847         return stnew;
 848 }
 849
 850 struct supertype *guess_super(int fd)
 851 {
 852         /* try each load_super to find the best match,
 853          * and return the best superswitch
 854          */
 855         struct superswitch  *ss;
 856         struct supertype *st;
 857         unsigned long besttime = 0;
 858         int bestsuper = -1;
 859         int i;
 860
 861         st = malloc(sizeof(*st));
 862         memset(st, 0, sizeof(*st));
 863         for (i=0 ; superlist[i]; i++) {
 864                 int rv;
 865                 ss = superlist[i];
 866                 st->ss = NULL;
 867                 rv = ss->load_super(st, fd, NULL);
 868                 if (rv == 0) {
 869                         struct mdinfo info;
 870                         st->ss->getinfo_super(st, &info);
 871                         if (bestsuper == -1 ||
 872                             besttime < info.array.ctime) {
 873                                 bestsuper = i;
 874                                 besttime = info.array.ctime;
 875                         }
 876                         ss->free_super(st);
 877                 }
 878         }
 879         if (bestsuper != -1) {
 880                 int rv;
 881                 st->ss = NULL;
 882                 rv = superlist[bestsuper]->load_super(st, fd, NULL);
 883                 if (rv == 0) {
 884                         superlist[bestsuper]->free_super(st);
 885                         return st;
 886                 }
 887         }
 888         free(st);
 889         return NULL;
 890 }
 891
 892 /* Return size of device in bytes */
 893 int get_dev_size(int fd, char *dname, unsigned long long *sizep)
 894 {
 895         unsigned long long ldsize;
 896         struct stat st;
 897
 898         if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
 899                 ldsize = (unsigned long long)st.st_size;
 900         else
 901 #ifdef BLKGETSIZE64
 902         if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0)
 903 #endif
 904         {
 905                 unsigned long dsize;
 906                 if (ioctl(fd, BLKGETSIZE, &dsize) == 0) {
 907                         ldsize = dsize;
 908                         ldsize <<= 9;
 909                 } else {
 910                         if (dname)
 911                                 fprintf(stderr, Name ": Cannot get size of %s: %s\b",
 912                                         dname, strerror(errno));
 913                         return 0;
 914                 }
 915         }
 916         *sizep = ldsize;
 917         return 1;
 918 }
 919
 920 void get_one_disk(int mdfd, mdu_array_info_t *ainf, mdu_disk_info_t *disk)
 921 {
 922         int d;
 923         ioctl(mdfd, GET_ARRAY_INFO, ainf);
 924         for (d = 0 ; d < ainf->raid_disks + ainf->nr_disks ; d++)
 925                 if (ioctl(mdfd, GET_DISK_INFO, disk) == 0)
 926                         return;
 927 }
 928
 929 int open_container(int fd)
 930 {
 931         /* 'fd' is a block device.  Find out if it is in use
 932          * by a container, and return an open fd on that container.
 933          */
 934         char path[256];
 935         char *e;
 936         DIR *dir;
 937         struct dirent *de;
 938         int dfd, n;
 939         char buf[200];
 940         int major, minor;
 941         struct stat st;
 942
 943         if (fstat(fd, &st) != 0)
 944                 return -1;
 945         sprintf(path, "/sys/dev/block/%d:%d/holders",
 946                 (int)major(st.st_rdev), (int)minor(st.st_rdev));
 947         e = path + strlen(path);
 948
 949         dir = opendir(path);
 950         if (!dir)
 951                 return -1;
 952         while ((de = readdir(dir))) {
 953                 if (de->d_ino == 0)
 954                         continue;
 955                 if (de->d_name[0] == '.')
 956                         continue;
 957                 sprintf(e, "/%s/dev", de->d_name);
 958                 dfd = open(path, O_RDONLY);
 959                 if (dfd < 0)
 960                         continue;
 961                 n = read(dfd, buf, sizeof(buf));
 962                 close(dfd);
 963                 if (n <= 0 || n >= sizeof(buf))
 964                         continue;
 965                 buf[n] = 0;
 966                 if (sscanf(buf, "%d:%d", &major, &minor) != 2)
 967                         continue;
 968                 sprintf(buf, "%d:%d", major, minor);
 969                 dfd = dev_open(buf, O_RDONLY);
 970                 if (dfd >= 0) {
 971                         closedir(dir);
 972                         return dfd;
 973                 }
 974         }
 975         return -1;
 976 }
 977
 978 #ifdef __TINYC__
 979 /* tinyc doesn't optimize this check in ioctl.h out ... */
 980 unsigned int __invalid_size_argument_for_IOC = 0;
 981 #endif
 982