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/socket.h>
  33 #include        <sys/utsname.h>
  34 #include        <sys/wait.h>
  35 #include        <sys/un.h>
  36 #include        <ctype.h>
  37 #include        <dirent.h>
  38 #include        <signal.h>
  39
  40 /*
  41  * following taken from linux/blkpg.h because they aren't
  42  * anywhere else and it isn't safe to #include linux/ * stuff.
  43  */
  44
  45 #define BLKPG      _IO(0x12,105)
  46
  47 /* The argument structure */
  48 struct blkpg_ioctl_arg {
  49         int op;
  50         int flags;
  51         int datalen;
  52         void *data;
  53 };
  54
  55 /* The subfunctions (for the op field) */
  56 #define BLKPG_ADD_PARTITION     1
  57 #define BLKPG_DEL_PARTITION     2
  58
  59 /* Sizes of name fields. Unused at present. */
  60 #define BLKPG_DEVNAMELTH        64
  61 #define BLKPG_VOLNAMELTH        64
  62
  63 /* The data structure for ADD_PARTITION and DEL_PARTITION */
  64 struct blkpg_partition {
  65         long long start;                /* starting offset in bytes */
  66         long long length;               /* length in bytes */
  67         int pno;                        /* partition number */
  68         char devname[BLKPG_DEVNAMELTH]; /* partition name, like sda5 or c0d1p2,
  69                                            to be used in kernel messages */
  70         char volname[BLKPG_VOLNAMELTH]; /* volume label */
  71 };
  72
  73 /*
  74  * Parse a 128 bit uuid in 4 integers
  75  * format is 32 hexx nibbles with options :.<space> separator
  76  * If not exactly 32 hex digits are found, return 0
  77  * else return 1
  78  */
  79 int parse_uuid(char *str, int uuid[4])
  80 {
  81         int hit = 0; /* number of Hex digIT */
  82         int i;
  83         char c;
  84         for (i=0; i<4; i++) uuid[i]=0;
  85
  86         while ((c= *str++)) {
  87                 int n;
  88                 if (c>='0' && c<='9')
  89                         n = c-'0';
  90                 else if (c>='a' && c <= 'f')
  91                         n = 10 + c - 'a';
  92                 else if (c>='A' && c <= 'F')
  93                         n = 10 + c - 'A';
  94                 else if (strchr(":. -", c))
  95                         continue;
  96                 else return 0;
  97
  98                 if (hit<32) {
  99                         uuid[hit/8] <<= 4;
 100                         uuid[hit/8] += n;
 101                 }
 102                 hit++;
 103         }
 104         if (hit == 32)
 105                 return 1;
 106         return 0;
 107 }
 108
 109
 110 /*
 111  * Get the md version number.
 112  * We use the RAID_VERSION ioctl if it is supported
 113  * If not, but we have a block device with major '9', we assume
 114  * 0.36.0
 115  *
 116  * Return version number as 24 but number - assume version parts
 117  * always < 255
 118  */
 119
 120 int md_get_version(int fd)
 121 {
 122     struct stat stb;
 123     mdu_version_t vers;
 124
 125     if (fstat(fd, &stb)<0)
 126         return -1;
 127     if ((S_IFMT&stb.st_mode) != S_IFBLK)
 128         return -1;
 129
 130     if (ioctl(fd, RAID_VERSION, &vers) == 0)
 131         return  (vers.major*10000) + (vers.minor*100) + vers.patchlevel;
 132     if (errno == EACCES)
 133             return -1;
 134     if (major(stb.st_rdev) == MD_MAJOR)
 135         return (3600);
 136     return -1;
 137 }
 138
 139 int get_linux_version()
 140 {
 141         struct utsname name;
 142         char *cp;
 143         int a,b,c;
 144         if (uname(&name) <0)
 145                 return -1;
 146
 147         cp = name.release;
 148         a = strtoul(cp, &cp, 10);
 149         if (*cp != '.') return -1;
 150         b = strtoul(cp+1, &cp, 10);
 151         if (*cp != '.') return -1;
 152         c = strtoul(cp+1, NULL, 10);
 153
 154         return (a*1000000)+(b*1000)+c;
 155 }
 156
 157 void remove_partitions(int fd)
 158 {
 159         /* remove partitions from this block devices.
 160          * This is used for components added to an array
 161          */
 162 #ifdef BLKPG_DEL_PARTITION
 163         struct blkpg_ioctl_arg a;
 164         struct blkpg_partition p;
 165
 166         a.op = BLKPG_DEL_PARTITION;
 167         a.data = (void*)&p;
 168         a.datalen = sizeof(p);
 169         a.flags = 0;
 170         memset(a.data, 0, a.datalen);
 171         for (p.pno=0; p.pno < 16; p.pno++)
 172                 ioctl(fd, BLKPG, &a);
 173 #endif
 174 }
 175
 176 int enough(int level, int raid_disks, int layout, int clean,
 177            char *avail, int avail_disks)
 178 {
 179         int copies, first;
 180         switch (level) {
 181         case 10:
 182                 /* This is the tricky one - we need to check
 183                  * which actual disks are present.
 184                  */
 185                 copies = (layout&255)* ((layout>>8) & 255);
 186                 first=0;
 187                 do {
 188                         /* there must be one of the 'copies' form 'first' */
 189                         int n = copies;
 190                         int cnt=0;
 191                         while (n--) {
 192                                 if (avail[first])
 193                                         cnt++;
 194                                 first = (first+1) % raid_disks;
 195                         }
 196                         if (cnt == 0)
 197                                 return 0;
 198
 199                 } while (first != 0);
 200                 return 1;
 201
 202         case -4:
 203                 return avail_disks>= 1;
 204         case -1:
 205         case 0:
 206                 return avail_disks == raid_disks;
 207         case 1:
 208                 return avail_disks >= 1;
 209         case 4:
 210         case 5:
 211                 if (clean)
 212                         return avail_disks >= raid_disks-1;
 213                 else
 214                         return avail_disks >= raid_disks;
 215         case 6:
 216                 if (clean)
 217                         return avail_disks >= raid_disks-2;
 218                 else
 219                         return avail_disks >= raid_disks;
 220         default:
 221                 return 0;
 222         }
 223 }
 224
 225 const int uuid_match_any[4] = { ~0, ~0, ~0, ~0 };
 226 int same_uuid(int a[4], int b[4], int swapuuid)
 227 {
 228         if (memcmp(a, uuid_match_any, sizeof(int[4])) == 0 ||
 229             memcmp(b, uuid_match_any, sizeof(int[4])) == 0)
 230                 return 1;
 231
 232         if (swapuuid) {
 233                 /* parse uuids are hostendian.
 234                  * uuid's from some superblocks are big-ending
 235                  * if there is a difference, we need to swap..
 236                  */
 237                 unsigned char *ac = (unsigned char *)a;
 238                 unsigned char *bc = (unsigned char *)b;
 239                 int i;
 240                 for (i=0; i<16; i+= 4) {
 241                         if (ac[i+0] != bc[i+3] ||
 242                             ac[i+1] != bc[i+2] ||
 243                             ac[i+2] != bc[i+1] ||
 244                             ac[i+3] != bc[i+0])
 245                                 return 0;
 246                 }
 247                 return 1;
 248         } else {
 249                 if (a[0]==b[0] &&
 250                     a[1]==b[1] &&
 251                     a[2]==b[2] &&
 252                     a[3]==b[3])
 253                         return 1;
 254                 return 0;
 255         }
 256 }
 257 void copy_uuid(void *a, int b[4], int swapuuid)
 258 {
 259         if (swapuuid) {
 260                 /* parse uuids are hostendian.
 261                  * uuid's from some superblocks are big-ending
 262                  * if there is a difference, we need to swap..
 263                  */
 264                 unsigned char *ac = (unsigned char *)a;
 265                 unsigned char *bc = (unsigned char *)b;
 266                 int i;
 267                 for (i=0; i<16; i+= 4) {
 268                         ac[i+0] = bc[i+3];
 269                         ac[i+1] = bc[i+2];
 270                         ac[i+2] = bc[i+1];
 271                         ac[i+3] = bc[i+0];
 272                 }
 273         } else
 274                 memcpy(a, b, 16);
 275 }
 276
 277 char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep)
 278 {
 279         int i, j;
 280         int id;
 281         char uuid[16];
 282         char *c = buf;
 283         strcpy(c, "UUID-");
 284         c += strlen(c);
 285         copy_uuid(uuid, info->uuid, st->ss->swapuuid);
 286         for (i = 0; i < 4; i++) {
 287                 id = uuid[i];
 288                 if (i)
 289                         *c++ = sep;
 290                 for (j = 3; j >= 0; j--) {
 291                         sprintf(c,"%02x", (unsigned char) uuid[j+4*i]);
 292                         c+= 2;
 293                 }
 294         }
 295         return buf;
 296 }
 297
 298 #ifndef MDASSEMBLE
 299 int check_ext2(int fd, char *name)
 300 {
 301         /*
 302          * Check for an ext2fs file system.
 303          * Superblock is always 1K at 1K offset
 304          *
 305          * s_magic is le16 at 56 == 0xEF53
 306          * report mtime - le32 at 44
 307          * blocks - le32 at 4
 308          * logblksize - le32 at 24
 309          */
 310         unsigned char sb[1024];
 311         time_t mtime;
 312         int size, bsize;
 313         if (lseek(fd, 1024,0)!= 1024)
 314                 return 0;
 315         if (read(fd, sb, 1024)!= 1024)
 316                 return 0;
 317         if (sb[56] != 0x53 || sb[57] != 0xef)
 318                 return 0;
 319
 320         mtime = sb[44]|(sb[45]|(sb[46]|sb[47]<<8)<<8)<<8;
 321         bsize = sb[24]|(sb[25]|(sb[26]|sb[27]<<8)<<8)<<8;
 322         size = sb[4]|(sb[5]|(sb[6]|sb[7]<<8)<<8)<<8;
 323         fprintf(stderr, Name ": %s appears to contain an ext2fs file system\n",
 324                 name);
 325         fprintf(stderr,"    size=%dK  mtime=%s",
 326                 size*(1<<bsize), ctime(&mtime));
 327         return 1;
 328 }
 329
 330 int check_reiser(int fd, char *name)
 331 {
 332         /*
 333          * superblock is at 64K
 334          * size is 1024;
 335          * Magic string "ReIsErFs" or "ReIsEr2Fs" at 52
 336          *
 337          */
 338         unsigned char sb[1024];
 339         unsigned long size;
 340         if (lseek(fd, 64*1024, 0) != 64*1024)
 341                 return 0;
 342         if (read(fd, sb, 1024) != 1024)
 343                 return 0;
 344         if (strncmp((char*)sb+52, "ReIsErFs",8)!=0 &&
 345             strncmp((char*)sb+52, "ReIsEr2Fs",9)!=0)
 346                 return 0;
 347         fprintf(stderr, Name ": %s appears to contain a reiserfs file system\n",name);
 348         size = sb[0]|(sb[1]|(sb[2]|sb[3]<<8)<<8)<<8;
 349         fprintf(stderr, "    size = %luK\n", size*4);
 350
 351         return 1;
 352 }
 353
 354 int check_raid(int fd, char *name)
 355 {
 356         struct mdinfo info;
 357         time_t crtime;
 358         char *level;
 359         struct supertype *st = guess_super(fd);
 360
 361         if (!st) return 0;
 362         st->ss->load_super(st, fd, name);
 363         /* Looks like a raid array .. */
 364         fprintf(stderr, Name ": %s appears to be part of a raid array:\n",
 365                 name);
 366         st->ss->getinfo_super(st, &info);
 367         st->ss->free_super(st);
 368         crtime = info.array.ctime;
 369         level = map_num(pers, info.array.level);
 370         if (!level) level = "-unknown-";
 371         fprintf(stderr, "    level=%s devices=%d ctime=%s",
 372                 level, info.array.raid_disks, ctime(&crtime));
 373         return 1;
 374 }
 375
 376 int ask(char *mesg)
 377 {
 378         char *add = "";
 379         int i;
 380         for (i=0; i<5; i++) {
 381                 char buf[100];
 382                 fprintf(stderr, "%s%s", mesg, add);
 383                 fflush(stderr);
 384                 if (fgets(buf, 100, stdin)==NULL)
 385                         return 0;
 386                 if (buf[0]=='y' || buf[0]=='Y')
 387                         return 1;
 388                 if (buf[0]=='n' || buf[0]=='N')
 389                         return 0;
 390                 add = "(y/n) ";
 391         }
 392         fprintf(stderr, Name ": assuming 'no'\n");
 393         return 0;
 394 }
 395 #endif /* MDASSEMBLE */
 396
 397 char *map_num(mapping_t *map, int num)
 398 {
 399         while (map->name) {
 400                 if (map->num == num)
 401                         return map->name;
 402                 map++;
 403         }
 404         return NULL;
 405 }
 406
 407 int map_name(mapping_t *map, char *name)
 408 {
 409         while (map->name) {
 410                 if (strcmp(map->name, name)==0)
 411                         return map->num;
 412                 map++;
 413         }
 414         return UnSet;
 415 }
 416
 417
 418 int is_standard(char *dev, int *nump)
 419 {
 420         /* tests if dev is a "standard" md dev name.
 421          * i.e if the last component is "/dNN" or "/mdNN",
 422          * where NN is a string of digits
 423          * Returns 1 if a partitionable standard,
 424          *   -1 if non-partitonable,
 425          *   0 if not a standard name.
 426          */
 427         char *d = strrchr(dev, '/');
 428         int type=0;
 429         int num;
 430         if (!d)
 431                 return 0;
 432         if (strncmp(d, "/d",2)==0)
 433                 d += 2, type=1; /* /dev/md/dN{pM} */
 434         else if (strncmp(d, "/md_d", 5)==0)
 435                 d += 5, type=1; /* /dev/md_dN{pM} */
 436         else if (strncmp(d, "/md", 3)==0)
 437                 d += 3, type=-1; /* /dev/mdN */
 438         else if (d-dev > 3 && strncmp(d-2, "md/", 3)==0)
 439                 d += 1, type=-1; /* /dev/md/N */
 440         else
 441                 return 0;
 442         if (!*d)
 443                 return 0;
 444         num = atoi(d);
 445         while (isdigit(*d))
 446                 d++;
 447         if (*d)
 448                 return 0;
 449         if (nump) *nump = num;
 450
 451         return type;
 452 }
 453
 454
 455 /*
 456  * convert a major/minor pair for a block device into a name in /dev, if possible.
 457  * On the first call, walk /dev collecting name.
 458  * Put them in a simple linked listfor now.
 459  */
 460 struct devmap {
 461     int major, minor;
 462     char *name;
 463     struct devmap *next;
 464 } *devlist = NULL;
 465 int devlist_ready = 0;
 466
 467 int add_dev(const char *name, const struct stat *stb, int flag, struct FTW *s)
 468 {
 469         struct stat st;
 470
 471         if (S_ISLNK(stb->st_mode)) {
 472                 if (stat(name, &st) != 0)
 473                         return 0;
 474                 stb = &st;
 475         }
 476
 477         if ((stb->st_mode&S_IFMT)== S_IFBLK) {
 478                 char *n = strdup(name);
 479                 struct devmap *dm = malloc(sizeof(*dm));
 480                 if (strncmp(n, "/dev/./", 7)==0)
 481                         strcpy(n+4, name+6);
 482                 if (dm) {
 483                         dm->major = major(stb->st_rdev);
 484                         dm->minor = minor(stb->st_rdev);
 485                         dm->name = n;
 486                         dm->next = devlist;
 487                         devlist = dm;
 488                 }
 489         }
 490         return 0;
 491 }
 492
 493 #ifndef HAVE_NFTW
 494 #ifdef HAVE_FTW
 495 int add_dev_1(const char *name, const struct stat *stb, int flag)
 496 {
 497         return add_dev(name, stb, flag, NULL);
 498 }
 499 int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags)
 500 {
 501         return ftw(path, add_dev_1, nopenfd);
 502 }
 503 #else
 504 int nftw(const char *path, int (*han)(const char *name, const struct stat *stb, int flag, struct FTW *s), int nopenfd, int flags)
 505 {
 506         return 0;
 507 }
 508 #endif /* HAVE_FTW */
 509 #endif /* HAVE_NFTW */
 510
 511 /*
 512  * Find a block device with the right major/minor number.
 513  * If we find multiple names, choose the shortest.
 514  * If we find a non-standard name, it is probably there
 515  * deliberately so prefer it over a standard name.
 516  * This applies only to names for MD devices.
 517  */
 518 char *map_dev(int major, int minor, int create)
 519 {
 520         struct devmap *p;
 521         char *std = NULL, *nonstd=NULL;
 522         int did_check = 0;
 523
 524         if (major == 0 && minor == 0)
 525                         return NULL;
 526
 527  retry:
 528         if (!devlist_ready) {
 529                 char *dev = "/dev";
 530                 struct stat stb;
 531                 while(devlist) {
 532                         struct devmap *d = devlist;
 533                         devlist = d->next;
 534                         free(d->name);
 535                         free(d);
 536                 }
 537                 if (lstat(dev, &stb)==0 &&
 538                     S_ISLNK(stb.st_mode))
 539                         dev = "/dev/.";
 540                 nftw(dev, add_dev, 10, FTW_PHYS);
 541                 devlist_ready=1;
 542                 did_check = 1;
 543         }
 544
 545         for (p=devlist; p; p=p->next)
 546                 if (p->major == major &&
 547                     p->minor == minor) {
 548                         if (is_standard(p->name, NULL)) {
 549                                 if (std == NULL ||
 550                                     strlen(p->name) < strlen(std))
 551                                         std = p->name;
 552                         } else {
 553                                 if (nonstd == NULL ||
 554                                     strlen(p->name) < strlen(nonstd))
 555                                         nonstd = p->name;
 556                         }
 557                 }
 558         if (!std && !nonstd && !did_check) {
 559                 devlist_ready = 0;
 560                 goto retry;
 561         }
 562         if (create && !std && !nonstd) {
 563                 static char buf[30];
 564                 snprintf(buf, sizeof(buf), "%d:%d", major, minor);
 565                 nonstd = buf;
 566         }
 567
 568         return nonstd ? nonstd : std;
 569 }
 570
 571 unsigned long calc_csum(void *super, int bytes)
 572 {
 573         unsigned long long newcsum = 0;
 574         int i;
 575         unsigned int csum;
 576         unsigned int *superc = (unsigned int*) super;
 577
 578         for(i=0; i<bytes/4; i++)
 579                 newcsum+= superc[i];
 580         csum = (newcsum& 0xffffffff) + (newcsum>>32);
 581 #ifdef __alpha__
 582 /* The in-kernel checksum calculation is always 16bit on
 583  * the alpha, though it is 32 bit on i386...
 584  * I wonder what it is elsewhere... (it uses and API in
 585  * a way that it shouldn't).
 586  */
 587         csum = (csum & 0xffff) + (csum >> 16);
 588         csum = (csum & 0xffff) + (csum >> 16);
 589 #endif
 590         return csum;
 591 }
 592
 593 #ifndef MDASSEMBLE
 594 char *human_size(long long bytes)
 595 {
 596         static char buf[30];
 597
 598         /* We convert bytes to either centi-M{ega,ibi}bytes or
 599          * centi-G{igi,ibi}bytes, with appropriate rounding,
 600          * and then print 1/100th of those as a decimal.
 601          * We allow upto 2048Megabytes before converting to
 602          * gigabytes, as that shows more precision and isn't
 603          * too large a number.
 604          * Terrabytes are not yet handled.
 605          */
 606
 607         if (bytes < 5000*1024)
 608                 buf[0]=0;
 609         else if (bytes < 2*1024LL*1024LL*1024LL) {
 610                 long cMiB = (bytes / ( (1LL<<20) / 200LL ) +1) /2;
 611                 long cMB  = (bytes / ( 1000000LL / 200LL ) +1) /2;
 612                 snprintf(buf, sizeof(buf), " (%ld.%02ld MiB %ld.%02ld MB)",
 613                         cMiB/100 , cMiB % 100,
 614                         cMB/100, cMB % 100);
 615         } else {
 616                 long cGiB = (bytes / ( (1LL<<30) / 200LL ) +1) /2;
 617                 long cGB  = (bytes / (1000000000LL/200LL ) +1) /2;
 618                 snprintf(buf, sizeof(buf), " (%ld.%02ld GiB %ld.%02ld GB)",
 619                         cGiB/100 , cGiB % 100,
 620                         cGB/100, cGB % 100);
 621         }
 622         return buf;
 623 }
 624
 625 char *human_size_brief(long long bytes)
 626 {
 627         static char buf[30];
 628
 629         if (bytes < 5000*1024)
 630                 snprintf(buf, sizeof(buf), "%ld.%02ldKiB",
 631                         (long)(bytes>>10), (long)(((bytes&1023)*100+512)/1024)
 632                         );
 633         else if (bytes < 2*1024LL*1024LL*1024LL)
 634                 snprintf(buf, sizeof(buf), "%ld.%02ldMiB",
 635                         (long)(bytes>>20),
 636                         (long)((bytes&0xfffff)+0x100000/200)/(0x100000/100)
 637                         );
 638         else
 639                 snprintf(buf, sizeof(buf), "%ld.%02ldGiB",
 640                         (long)(bytes>>30),
 641                         (long)(((bytes>>10)&0xfffff)+0x100000/200)/(0x100000/100)
 642                         );
 643         return buf;
 644 }
 645
 646 void print_r10_layout(int layout)
 647 {
 648         int near = layout & 255;
 649         int far = (layout >> 8) & 255;
 650         int offset = (layout&0x10000);
 651         char *sep = "";
 652
 653         if (near != 1) {
 654                 printf("%s near=%d", sep, near);
 655                 sep = ",";
 656         }
 657         if (far != 1)
 658                 printf("%s %s=%d", sep, offset?"offset":"far", far);
 659         if (near*far == 1)
 660                 printf("NO REDUNDANCY");
 661 }
 662 #endif
 663
 664 unsigned long long calc_array_size(int level, int raid_disks, int layout,
 665                                    int chunksize, unsigned long long devsize)
 666 {
 667         int data_disks = 0;
 668         switch (level) {
 669         case 0: data_disks = raid_disks; break;
 670         case 1: data_disks = 1; break;
 671         case 4:
 672         case 5: data_disks = raid_disks - 1; break;
 673         case 6: data_disks = raid_disks - 2; break;
 674         case 10: data_disks = raid_disks / (layout & 255) / ((layout>>8)&255);
 675                 break;
 676         }
 677         devsize &= ~(unsigned long long)((chunksize>>9)-1);
 678         return data_disks * devsize;
 679 }
 680
 681 int get_mdp_major(void)
 682 {
 683 static int mdp_major = -1;
 684         FILE *fl;
 685         char *w;
 686         int have_block = 0;
 687         int have_devices = 0;
 688         int last_num = -1;
 689
 690         if (mdp_major != -1)
 691                 return mdp_major;
 692         fl = fopen("/proc/devices", "r");
 693         if (!fl)
 694                 return -1;
 695         while ((w = conf_word(fl, 1))) {
 696                 if (have_block && strcmp(w, "devices:")==0)
 697                         have_devices = 1;
 698                 have_block =  (strcmp(w, "Block")==0);
 699                 if (isdigit(w[0]))
 700                         last_num = atoi(w);
 701                 if (have_devices && strcmp(w, "mdp")==0)
 702                         mdp_major = last_num;
 703                 free(w);
 704         }
 705         fclose(fl);
 706         return mdp_major;
 707 }
 708
 709 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
 710 char *get_md_name(int dev)
 711 {
 712         /* find /dev/md%d or /dev/md/%d or make a device /dev/.tmp.md%d */
 713         /* if dev < 0, want /dev/md/d%d or find mdp in /proc/devices ... */
 714         static char devname[50];
 715         struct stat stb;
 716         dev_t rdev;
 717         char *dn;
 718
 719         if (dev < 0) {
 720                 int mdp =  get_mdp_major();
 721                 if (mdp < 0) return NULL;
 722                 rdev = makedev(mdp, (-1-dev)<<6);
 723                 snprintf(devname, sizeof(devname), "/dev/md/d%d", -1-dev);
 724                 if (stat(devname, &stb) == 0
 725                     && (S_IFMT&stb.st_mode) == S_IFBLK
 726                     && (stb.st_rdev == rdev))
 727                         return devname;
 728         } else {
 729                 rdev = makedev(MD_MAJOR, dev);
 730                 snprintf(devname, sizeof(devname), "/dev/md%d", dev);
 731                 if (stat(devname, &stb) == 0
 732                     && (S_IFMT&stb.st_mode) == S_IFBLK
 733                     && (stb.st_rdev == rdev))
 734                         return devname;
 735
 736                 snprintf(devname, sizeof(devname), "/dev/md/%d", dev);
 737                 if (stat(devname, &stb) == 0
 738                     && (S_IFMT&stb.st_mode) == S_IFBLK
 739                     && (stb.st_rdev == rdev))
 740                         return devname;
 741         }
 742         dn = map_dev(major(rdev), minor(rdev), 0);
 743         if (dn)
 744                 return dn;
 745         snprintf(devname, sizeof(devname), "/dev/.tmp.md%d", dev);
 746         if (mknod(devname, S_IFBLK | 0600, rdev) == -1)
 747                 if (errno != EEXIST)
 748                         return NULL;
 749
 750         if (stat(devname, &stb) == 0
 751             && (S_IFMT&stb.st_mode) == S_IFBLK
 752             && (stb.st_rdev == rdev))
 753                 return devname;
 754         unlink(devname);
 755         return NULL;
 756 }
 757
 758 void put_md_name(char *name)
 759 {
 760         if (strncmp(name, "/dev/.tmp.md", 12)==0)
 761                 unlink(name);
 762 }
 763
 764 int find_free_devnum(int use_partitions)
 765 {
 766         int devnum;
 767         for (devnum = 127; devnum != 128;
 768              devnum = devnum ? devnum-1 : (1<<20)-1) {
 769                 char *dn;
 770                 int _devnum;
 771
 772                 _devnum = use_partitions ? (-1-devnum) : devnum;
 773                 if (mddev_busy(_devnum))
 774                         continue;
 775                 /* make sure it is new to /dev too, at least as a
 776                  * non-standard */
 777                 dn = map_dev(dev2major(_devnum), dev2minor(_devnum), 0);
 778                 if (dn && ! is_standard(dn, NULL))
 779                         continue;
 780                 break;
 781         }
 782         if (devnum == 128)
 783                 return NoMdDev;
 784         return use_partitions ? (-1-devnum) : devnum;
 785 }
 786 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
 787
 788 int dev_open(char *dev, int flags)
 789 {
 790         /* like 'open', but if 'dev' matches %d:%d, create a temp
 791          * block device and open that
 792          */
 793         char *e;
 794         int fd = -1;
 795         char devname[32];
 796         int major;
 797         int minor;
 798
 799         if (!dev) return -1;
 800
 801         major = strtoul(dev, &e, 0);
 802         if (e > dev && *e == ':' && e[1] &&
 803             (minor = strtoul(e+1, &e, 0)) >= 0 &&
 804             *e == 0) {
 805                 snprintf(devname, sizeof(devname), "/dev/.tmp.md.%d:%d:%d",
 806                          (int)getpid(), major, minor);
 807                 if (mknod(devname, S_IFBLK|0600, makedev(major, minor))==0) {
 808                         fd = open(devname, flags|O_DIRECT);
 809                         unlink(devname);
 810                 }
 811         } else
 812                 fd = open(dev, flags|O_DIRECT);
 813         return fd;
 814 }
 815
 816 int open_dev(int devnum)
 817 {
 818         char buf[20];
 819
 820         sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
 821         return dev_open(buf, O_RDWR);
 822 }
 823
 824 int open_dev_excl(int devnum)
 825 {
 826         char buf[20];
 827         int i;
 828
 829         sprintf(buf, "%d:%d", dev2major(devnum), dev2minor(devnum));
 830         for (i=0 ; i<25 ; i++) {
 831                 int fd = dev_open(buf, O_RDWR|O_EXCL);
 832                 if (fd >= 0)
 833                         return fd;
 834                 if (errno != EBUSY)
 835                         return fd;
 836                 usleep(200000);
 837         }
 838         return -1;
 839 }
 840
 841 int same_dev(char *one, char *two)
 842 {
 843         struct stat st1, st2;
 844         if (stat(one, &st1) != 0)
 845                 return 0;
 846         if (stat(two, &st2) != 0)
 847                 return 0;
 848         if ((st1.st_mode & S_IFMT) != S_IFBLK)
 849                 return 0;
 850         if ((st2.st_mode & S_IFMT) != S_IFBLK)
 851                 return 0;
 852         return st1.st_rdev == st2.st_rdev;
 853 }
 854
 855 void wait_for(char *dev, int fd)
 856 {
 857         int i;
 858         struct stat stb_want;
 859
 860         if (fstat(fd, &stb_want) != 0 ||
 861             (stb_want.st_mode & S_IFMT) != S_IFBLK)
 862                 return;
 863
 864         for (i=0 ; i<25 ; i++) {
 865                 struct stat stb;
 866                 if (stat(dev, &stb) == 0 &&
 867                     (stb.st_mode & S_IFMT) == S_IFBLK &&
 868                     (stb.st_rdev == stb_want.st_rdev))
 869                         return;
 870                 usleep(200000);
 871         }
 872 }
 873
 874 struct superswitch *superlist[] = { &super0, &super1, &super_ddf, &super_imsm, NULL };
 875
 876 #if !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO)
 877
 878 struct supertype *super_by_fd(int fd)
 879 {
 880         mdu_array_info_t array;
 881         int vers;
 882         int minor;
 883         struct supertype *st = NULL;
 884         struct mdinfo *sra;
 885         char *verstr;
 886         char version[20];
 887         int i;
 888         char *subarray = NULL;
 889
 890         sra = sysfs_read(fd, 0, GET_VERSION);
 891
 892         if (sra) {
 893                 vers = sra->array.major_version;
 894                 minor = sra->array.minor_version;
 895                 verstr = sra->text_version;
 896         } else {
 897                 if (ioctl(fd, GET_ARRAY_INFO, &array))
 898                         array.major_version = array.minor_version = 0;
 899                 vers = array.major_version;
 900                 minor = array.minor_version;
 901                 verstr = "";
 902         }
 903
 904         if (vers != -1) {
 905                 sprintf(version, "%d.%d", vers, minor);
 906                 verstr = version;
 907         }
 908         if (minor == -2 && is_subarray(verstr)) {
 909                 char *dev = verstr+1;
 910                 subarray = strchr(dev, '/');
 911                 int devnum;
 912                 if (subarray)
 913                         *subarray++ = '\0';
 914                 devnum = devname2devnum(dev);
 915                 subarray = strdup(subarray);
 916                 if (sra)
 917                         sysfs_free(sra);
 918                 sra = sysfs_read(-1, devnum, GET_VERSION);
 919                 verstr = sra->text_version ? : "-no-metadata-";
 920         }
 921
 922         for (i = 0; st == NULL && superlist[i] ; i++)
 923                 st = superlist[i]->match_metadata_desc(verstr);
 924
 925         if (sra)
 926                 sysfs_free(sra);
 927         if (st) {
 928                 st->sb = NULL;
 929                 if (subarray) {
 930                         strncpy(st->subarray, subarray, 32);
 931                         st->subarray[31] = 0;
 932                         free(subarray);
 933                 } else
 934                         st->subarray[0] = 0;
 935         }
 936         return st;
 937 }
 938 #endif /* !defined(MDASSEMBLE) || defined(MDASSEMBLE) && defined(MDASSEMBLE_AUTO) */
 939
 940
 941 struct supertype *dup_super(struct supertype *orig)
 942 {
 943         struct supertype *st;
 944
 945         if (!orig)
 946                 return orig;
 947         st = malloc(sizeof(*st));
 948         if (!st)
 949                 return st;
 950         memset(st, 0, sizeof(*st));
 951         st->ss = orig->ss;
 952         st->max_devs = orig->max_devs;
 953         st->minor_version = orig->minor_version;
 954         strcpy(st->subarray, orig->subarray);
 955         st->sb = NULL;
 956         st->info = NULL;
 957         return st;
 958 }
 959
 960 struct supertype *guess_super(int fd)
 961 {
 962         /* try each load_super to find the best match,
 963          * and return the best superswitch
 964          */
 965         struct superswitch  *ss;
 966         struct supertype *st;
 967         unsigned long besttime = 0;
 968         int bestsuper = -1;
 969         int i;
 970
 971         st = malloc(sizeof(*st));
 972         for (i=0 ; superlist[i]; i++) {
 973                 int rv;
 974                 ss = superlist[i];
 975                 memset(st, 0, sizeof(*st));
 976                 rv = ss->load_super(st, fd, NULL);
 977                 if (rv == 0) {
 978                         struct mdinfo info;
 979                         st->ss->getinfo_super(st, &info);
 980                         if (bestsuper == -1 ||
 981                             besttime < info.array.ctime) {
 982                                 bestsuper = i;
 983                                 besttime = info.array.ctime;
 984                         }
 985                         ss->free_super(st);
 986                 }
 987         }
 988         if (bestsuper != -1) {
 989                 int rv;
 990                 memset(st, 0, sizeof(*st));
 991                 rv = superlist[bestsuper]->load_super(st, fd, NULL);
 992                 if (rv == 0) {
 993                         superlist[bestsuper]->free_super(st);
 994                         return st;
 995                 }
 996         }
 997         free(st);
 998         return NULL;
 999 }
1000
1001 /* Return size of device in bytes */
1002 int get_dev_size(int fd, char *dname, unsigned long long *sizep)
1003 {
1004         unsigned long long ldsize;
1005         struct stat st;
1006
1007         if (fstat(fd, &st) != -1 && S_ISREG(st.st_mode))
1008                 ldsize = (unsigned long long)st.st_size;
1009         else
1010 #ifdef BLKGETSIZE64
1011         if (ioctl(fd, BLKGETSIZE64, &ldsize) != 0)
1012 #endif
1013         {
1014                 unsigned long dsize;
1015                 if (ioctl(fd, BLKGETSIZE, &dsize) == 0) {
1016                         ldsize = dsize;
1017                         ldsize <<= 9;
1018                 } else {
1019                         if (dname)
1020                                 fprintf(stderr, Name ": Cannot get size of %s: %s\b",
1021                                         dname, strerror(errno));
1022                         return 0;
1023                 }
1024         }
1025         *sizep = ldsize;
1026         return 1;
1027 }
1028
1029 void get_one_disk(int mdfd, mdu_array_info_t *ainf, mdu_disk_info_t *disk)
1030 {
1031         int d;
1032         ioctl(mdfd, GET_ARRAY_INFO, ainf);
1033         for (d = 0 ; d < ainf->raid_disks + ainf->nr_disks ; d++)
1034                 if (ioctl(mdfd, GET_DISK_INFO, disk) == 0)
1035                         return;
1036 }
1037
1038 int open_container(int fd)
1039 {
1040         /* 'fd' is a block device.  Find out if it is in use
1041          * by a container, and return an open fd on that container.
1042          */
1043         char path[256];
1044         char *e;
1045         DIR *dir;
1046         struct dirent *de;
1047         int dfd, n;
1048         char buf[200];
1049         int major, minor;
1050         struct stat st;
1051
1052         if (fstat(fd, &st) != 0)
1053                 return -1;
1054         sprintf(path, "/sys/dev/block/%d:%d/holders",
1055                 (int)major(st.st_rdev), (int)minor(st.st_rdev));
1056         e = path + strlen(path);
1057
1058         dir = opendir(path);
1059         if (!dir)
1060                 return -1;
1061         while ((de = readdir(dir))) {
1062                 if (de->d_ino == 0)
1063                         continue;
1064                 if (de->d_name[0] == '.')
1065                         continue;
1066                 sprintf(e, "/%s/dev", de->d_name);
1067                 dfd = open(path, O_RDONLY);
1068                 if (dfd < 0)
1069                         continue;
1070                 n = read(dfd, buf, sizeof(buf));
1071                 close(dfd);
1072                 if (n <= 0 || n >= sizeof(buf))
1073                         continue;
1074                 buf[n] = 0;
1075                 if (sscanf(buf, "%d:%d", &major, &minor) != 2)
1076                         continue;
1077                 sprintf(buf, "%d:%d", major, minor);
1078                 dfd = dev_open(buf, O_RDONLY);
1079                 if (dfd >= 0) {
1080                         closedir(dir);
1081                         return dfd;
1082                 }
1083         }
1084         closedir(dir);
1085         return -1;
1086 }
1087
1088 int add_disk(int mdfd, struct supertype *st,
1089              struct mdinfo *sra, struct mdinfo *info)
1090 {
1091         /* Add a device to an array, in one of 2 ways. */
1092         int rv;
1093 #ifndef MDASSEMBLE
1094         if (st->ss->external) {
1095                 rv = sysfs_add_disk(sra, info);
1096                 if (! rv) {
1097                         struct mdinfo *sd2;
1098                         for (sd2 = sra->devs; sd2; sd2=sd2->next)
1099                                 if (sd2 == info)
1100                                         break;
1101                         if (sd2 == NULL) {
1102                                 sd2 = malloc(sizeof(*sd2));
1103                                 *sd2 = *info;
1104                                 sd2->next = sra->devs;
1105                                 sra->devs = sd2;
1106                         }
1107                 }
1108         } else
1109 #endif
1110                 rv = ioctl(mdfd, ADD_NEW_DISK, &info->disk);
1111         return rv;
1112 }
1113
1114 int set_array_info(int mdfd, struct supertype *st, struct mdinfo *info)
1115 {
1116         /* Initialise kernel's knowledge of array.
1117          * This varies between externally managed arrays
1118          * and older kernels
1119          */
1120         int vers = md_get_version(mdfd);
1121         int rv;
1122
1123 #ifndef MDASSEMBLE
1124         if (st->ss->external)
1125                 rv = sysfs_set_array(info, vers);
1126         else
1127 #endif
1128                 if ((vers % 100) >= 1) { /* can use different versions */
1129                 mdu_array_info_t inf;
1130                 memset(&inf, 0, sizeof(inf));
1131                 inf.major_version = info->array.major_version;
1132                 inf.minor_version = info->array.minor_version;
1133                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
1134         } else
1135                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
1136         return rv;
1137 }
1138
1139 char *devnum2devname(int num)
1140 {
1141         char name[100];
1142         if (num > 0)
1143                 sprintf(name, "md%d", num);
1144         else
1145                 sprintf(name, "md_d%d", -1-num);
1146         return strdup(name);
1147 }
1148
1149 int devname2devnum(char *name)
1150 {
1151         char *ep;
1152         int num;
1153         if (strncmp(name, "md_d", 4)==0)
1154                 num = -1-strtoul(name+4, &ep, 10);
1155         else
1156                 num = strtoul(name+2, &ep, 10);
1157         return num;
1158 }
1159
1160 int stat2devnum(struct stat *st)
1161 {
1162         char path[30];
1163         char link[200];
1164         char *cp;
1165         int n;
1166
1167         if ((S_IFMT & st->st_mode) == S_IFBLK) {
1168                 if (major(st->st_rdev) == MD_MAJOR)
1169                         return minor(st->st_rdev);
1170                 else if (major(st->st_rdev) == get_mdp_major())
1171                         return -1- (minor(st->st_rdev)>>MdpMinorShift);
1172
1173                 /* must be an extended-minor partition. Look at the
1174                  * /sys/dev/block/%d:%d link which must look like
1175                  * ../../block/mdXXX/mdXXXpYY
1176                  */
1177                 sprintf(path, "/sys/dev/block/%d:%d", major(st->st_rdev),
1178                         minor(st->st_rdev));
1179                 n = readlink(path, link, sizeof(link)-1);
1180                 if (n <= 0)
1181                         return NoMdDev;
1182                 link[n] = 0;
1183                 cp = strrchr(link, '/');
1184                 if (cp) *cp = 0;
1185                 cp = strchr(link, '/');
1186                 if (cp && strncmp(cp, "/md", 3) == 0)
1187                         return devname2devnum(cp+1);
1188         }
1189         return NoMdDev;
1190
1191 }
1192
1193 int fd2devnum(int fd)
1194 {
1195         struct stat stb;
1196         if (fstat(fd, &stb) == 0)
1197                 return stat2devnum(&stb);
1198         return NoMdDev;
1199 }
1200
1201 int mdmon_running(int devnum)
1202 {
1203         char path[100];
1204         char pid[10];
1205         int fd;
1206         int n;
1207         sprintf(path, "/var/run/mdadm/%s.pid", devnum2devname(devnum));
1208         fd = open(path, O_RDONLY, 0);
1209
1210         if (fd < 0)
1211                 return 0;
1212         n = read(fd, pid, 9);
1213         close(fd);
1214         if (n <= 0)
1215                 return 0;
1216         if (kill(atoi(pid), 0) == 0)
1217                 return 1;
1218         return 0;
1219 }
1220
1221 int signal_mdmon(int devnum)
1222 {
1223         char path[100];
1224         char pid[10];
1225         int fd;
1226         int n;
1227         sprintf(path, "/var/run/mdadm/%s.pid", devnum2devname(devnum));
1228         fd = open(path, O_RDONLY, 0);
1229
1230         if (fd < 0)
1231                 return 0;
1232         n = read(fd, pid, 9);
1233         close(fd);
1234         if (n <= 0)
1235                 return 0;
1236         if (kill(atoi(pid), SIGUSR1) == 0)
1237                 return 1;
1238         return 0;
1239 }
1240
1241 int start_mdmon(int devnum)
1242 {
1243         int i;
1244         int len;
1245         pid_t pid;
1246         int status;
1247         char pathbuf[1024];
1248         char *paths[4] = {
1249                 pathbuf,
1250                 "/sbin/mdmon",
1251                 "mdmon",
1252                 NULL
1253         };
1254
1255         if (check_env("MDADM_NO_MDMON"))
1256                 return 0;
1257
1258         len = readlink("/proc/self/exe", pathbuf, sizeof(pathbuf));
1259         if (len > 0) {
1260                 char *sl;
1261                 pathbuf[len] = 0;
1262                 sl = strrchr(pathbuf, '/');
1263                 if (sl)
1264                         sl++;
1265                 else
1266                         sl = pathbuf;
1267                 strcpy(sl, "mdmon");
1268         } else
1269                 pathbuf[0] = '\0';
1270
1271         switch(fork()) {
1272         case 0:
1273                 /* FIXME yuk. CLOSE_EXEC?? */
1274                 for (i=3; i < 100; i++)
1275                         close(i);
1276                 for (i=0; paths[i]; i++)
1277                         if (paths[i][0])
1278                                 execl(paths[i], "mdmon",
1279                                       devnum2devname(devnum),
1280                                       NULL);
1281                 exit(1);
1282         case -1: fprintf(stderr, Name ": cannot run mdmon. "
1283                          "Array remains readonly\n");
1284                 return -1;
1285         default: /* parent - good */
1286                 pid = wait(&status);
1287                 if (pid < 0 || status != 0)
1288                         return -1;
1289         }
1290         return 0;
1291 }
1292
1293 int check_env(char *name)
1294 {
1295         char *val = getenv(name);
1296
1297         if (val && atoi(val) == 1)
1298                 return 1;
1299
1300         return 0;
1301 }
1302
1303 #ifndef MDASSEMBLE
1304 int flush_metadata_updates(struct supertype *st)
1305 {
1306         int sfd;
1307         if (!st->updates) {
1308                 st->update_tail = NULL;
1309                 return -1;
1310         }
1311
1312         sfd = connect_monitor(devnum2devname(st->container_dev));
1313         if (sfd < 0)
1314                 return -1;
1315
1316         while (st->updates) {
1317                 struct metadata_update *mu = st->updates;
1318                 st->updates = mu->next;
1319
1320                 send_message(sfd, mu, 0);
1321                 wait_reply(sfd, 0);
1322                 free(mu->buf);
1323                 free(mu);
1324         }
1325         ack(sfd, 0);
1326         wait_reply(sfd, 0);
1327         close(sfd);
1328         st->update_tail = NULL;
1329         return 0;
1330 }
1331
1332 void append_metadata_update(struct supertype *st, void *buf, int len)
1333 {
1334
1335         struct metadata_update *mu = malloc(sizeof(*mu));
1336
1337         mu->buf = buf;
1338         mu->len = len;
1339         mu->space = NULL;
1340         mu->next = NULL;
1341         *st->update_tail = mu;
1342         st->update_tail = &mu->next;
1343 }
1344 #endif /* MDASSEMBLE */
1345
1346 #ifdef __TINYC__
1347 /* tinyc doesn't optimize this check in ioctl.h out ... */
1348 unsigned int __invalid_size_argument_for_IOC = 0;
1349 #endif
1350