sysfs.c

   1 /*
   2  * sysfs - extract md related information from sysfs.  Part of:
   3  * mdadm - manage Linux "md" devices aka RAID arrays.
   4  *
   5  * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de>
   6  *
   7  *
   8  *    This program is free software; you can redistribute it and/or modify
   9  *    it under the terms of the GNU General Public License as published by
  10  *    the Free Software Foundation; either version 2 of the License, or
  11  *    (at your option) any later version.
  12  *
  13  *    This program is distributed in the hope that it will be useful,
  14  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  *    GNU General Public License for more details.
  17  *
  18  *    You should have received a copy of the GNU General Public License
  19  *    along with this program; if not, write to the Free Software
  20  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  21  *
  22  *    Author: Neil Brown
  23  *    Email: <neilb@suse.de>
  24  */
  25
  26 #include        "mdadm.h"
  27 #include        <dirent.h>
  28 #include        <ctype.h>
  29
  30 int load_sys(char *path, char *buf, int len)
  31 {
  32         int fd = open(path, O_RDONLY);
  33         int n;
  34         if (fd < 0)
  35                 return -1;
  36         n = read(fd, buf, len);
  37         close(fd);
  38         if (n <0 || n >= len)
  39                 return -1;
  40         buf[n] = 0;
  41         if (n && buf[n-1] == '\n')
  42                 buf[n-1] = 0;
  43         return 0;
  44 }
  45
  46 void sysfs_free(struct mdinfo *sra)
  47 {
  48         while (sra) {
  49                 struct mdinfo *sra2 = sra->next;
  50                 while (sra->devs) {
  51                         struct mdinfo *d = sra->devs;
  52                         sra->devs = d->next;
  53                         free(d);
  54                 }
  55                 free(sra);
  56                 sra = sra2;
  57         }
  58 }
  59
  60 int sysfs_open(char *devnm, char *devname, char *attr)
  61 {
  62         char fname[50];
  63         int fd;
  64
  65         sprintf(fname, "/sys/block/%s/md/", devnm);
  66         if (devname) {
  67                 strcat(fname, devname);
  68                 strcat(fname, "/");
  69         }
  70         strcat(fname, attr);
  71         fd = open(fname, O_RDWR);
  72         if (fd < 0 && errno == EACCES)
  73                 fd = open(fname, O_RDONLY);
  74         return fd;
  75 }
  76
  77 void sysfs_init_dev(struct mdinfo *mdi, unsigned long devid)
  78 {
  79         snprintf(mdi->sys_name,
  80                  sizeof(mdi->sys_name), "dev-%s", devid2kname(devid));
  81 }
  82
  83 void sysfs_init(struct mdinfo *mdi, int fd, char *devnm)
  84 {
  85         mdi->sys_name[0] = 0;
  86         if (fd >= 0) {
  87                 mdu_version_t vers;
  88                 if (ioctl(fd, RAID_VERSION, &vers) != 0)
  89                         return;
  90                 devnm = fd2devnm(fd);
  91         }
  92         if (devnm == NULL)
  93                 return;
  94         strcpy(mdi->sys_name, devnm);
  95 }
  96
  97 struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
  98 {
  99         char fname[PATH_MAX];
 100         char buf[PATH_MAX];
 101         char *base;
 102         char *dbase;
 103         struct mdinfo *sra;
 104         struct mdinfo *dev, **devp;
 105         DIR *dir = NULL;
 106         struct dirent *de;
 107
 108         sra = xcalloc(1, sizeof(*sra));
 109         sysfs_init(sra, fd, devnm);
 110         if (sra->sys_name[0] == 0) {
 111                 free(sra);
 112                 return NULL;
 113         }
 114
 115         sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
 116         base = fname + strlen(fname);
 117
 118         sra->devs = NULL;
 119         if (options & GET_VERSION) {
 120                 strcpy(base, "metadata_version");
 121                 if (load_sys(fname, buf, sizeof(buf)))
 122                         goto abort;
 123                 if (strncmp(buf, "none", 4) == 0) {
 124                         sra->array.major_version =
 125                                 sra->array.minor_version = -1;
 126                         strcpy(sra->text_version, "");
 127                 } else if (strncmp(buf, "external:", 9) == 0) {
 128                         sra->array.major_version = -1;
 129                         sra->array.minor_version = -2;
 130                         strcpy(sra->text_version, buf+9);
 131                 } else {
 132                         sscanf(buf, "%d.%d",
 133                                &sra->array.major_version,
 134                                &sra->array.minor_version);
 135                         strcpy(sra->text_version, buf);
 136                 }
 137         }
 138         if (options & GET_LEVEL) {
 139                 strcpy(base, "level");
 140                 if (load_sys(fname, buf, sizeof(buf)))
 141                         goto abort;
 142                 sra->array.level = map_name(pers, buf);
 143         }
 144         if (options & GET_LAYOUT) {
 145                 strcpy(base, "layout");
 146                 if (load_sys(fname, buf, sizeof(buf)))
 147                         goto abort;
 148                 sra->array.layout = strtoul(buf, NULL, 0);
 149         }
 150         if (options & GET_DISKS) {
 151                 strcpy(base, "raid_disks");
 152                 if (load_sys(fname, buf, sizeof(buf)))
 153                         goto abort;
 154                 sra->array.raid_disks = strtoul(buf, NULL, 0);
 155         }
 156         if (options & GET_DEGRADED) {
 157                 strcpy(base, "degraded");
 158                 if (load_sys(fname, buf, sizeof(buf)))
 159                         goto abort;
 160                 sra->array.failed_disks = strtoul(buf, NULL, 0);
 161         }
 162         if (options & GET_COMPONENT) {
 163                 strcpy(base, "component_size");
 164                 if (load_sys(fname, buf, sizeof(buf)))
 165                         goto abort;
 166                 sra->component_size = strtoull(buf, NULL, 0);
 167                 /* sysfs reports "K", but we want sectors */
 168                 sra->component_size *= 2;
 169         }
 170         if (options & GET_CHUNK) {
 171                 strcpy(base, "chunk_size");
 172                 if (load_sys(fname, buf, sizeof(buf)))
 173                         goto abort;
 174                 sra->array.chunk_size = strtoul(buf, NULL, 0);
 175         }
 176         if (options & GET_CACHE) {
 177                 strcpy(base, "stripe_cache_size");
 178                 if (load_sys(fname, buf, sizeof(buf)))
 179                         /* Probably level doesn't support it */
 180                         sra->cache_size = 0;
 181                 else
 182                         sra->cache_size = strtoul(buf, NULL, 0);
 183         }
 184         if (options & GET_MISMATCH) {
 185                 strcpy(base, "mismatch_cnt");
 186                 if (load_sys(fname, buf, sizeof(buf)))
 187                         goto abort;
 188                 sra->mismatch_cnt = strtoul(buf, NULL, 0);
 189         }
 190         if (options & GET_SAFEMODE) {
 191                 int scale = 1;
 192                 int dot = 0;
 193                 unsigned i;
 194                 unsigned long msec;
 195                 size_t len;
 196
 197                 strcpy(base, "safe_mode_delay");
 198                 if (load_sys(fname, buf, sizeof(buf)))
 199                         goto abort;
 200
 201                 /* remove a period, and count digits after it */
 202                 len = strlen(buf);
 203                 for (i = 0; i < len; i++) {
 204                         if (dot) {
 205                                 if (isdigit(buf[i])) {
 206                                         buf[i-1] = buf[i];
 207                                         scale *= 10;
 208                                 }
 209                                 buf[i] = 0;
 210                         } else if (buf[i] == '.') {
 211                                 dot=1;
 212                                 buf[i] = 0;
 213                         }
 214                 }
 215                 msec = strtoul(buf, NULL, 10);
 216                 msec = (msec * 1000) / scale;
 217                 sra->safe_mode_delay = msec;
 218         }
 219         if (options & GET_BITMAP_LOCATION) {
 220                 strcpy(base, "bitmap/location");
 221                 if (load_sys(fname, buf, sizeof(buf)))
 222                         goto abort;
 223                 if (strncmp(buf, "file", 4) == 0)
 224                         sra->bitmap_offset = 1;
 225                 else if (strncmp(buf, "none", 4) == 0)
 226                         sra->bitmap_offset = 0;
 227                 else if (buf[0] == '+')
 228                         sra->bitmap_offset = strtol(buf+1, NULL, 10);
 229                 else
 230                         goto abort;
 231         }
 232
 233         if (options & GET_ARRAY_STATE) {
 234                 strcpy(base, "array_state");
 235                 if (load_sys(fname, sra->sysfs_array_state,
 236                              sizeof(sra->sysfs_array_state)))
 237                         goto abort;
 238         } else
 239                 sra->sysfs_array_state[0] = 0;
 240
 241         if (! (options & GET_DEVS))
 242                 return sra;
 243
 244         /* Get all the devices as well */
 245         *base = 0;
 246         dir = opendir(fname);
 247         if (!dir)
 248                 goto abort;
 249         sra->array.spare_disks = 0;
 250
 251         devp = &sra->devs;
 252         sra->devs = NULL;
 253         while ((de = readdir(dir)) != NULL) {
 254                 char *ep;
 255                 if (de->d_ino == 0 ||
 256                     strncmp(de->d_name, "dev-", 4) != 0)
 257                         continue;
 258                 strcpy(base, de->d_name);
 259                 dbase = base + strlen(base);
 260                 *dbase++ = '/';
 261
 262                 dev = xmalloc(sizeof(*dev));
 263
 264                 /* Always get slot, major, minor */
 265                 strcpy(dbase, "slot");
 266                 if (load_sys(fname, buf, sizeof(buf))) {
 267                         /* hmm... unable to read 'slot' maybe the device
 268                          * is going away?
 269                          */
 270                         strcpy(dbase, "block");
 271                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 272                             errno != ENAMETOOLONG) {
 273                                 /* ...yup device is gone */
 274                                 free(dev);
 275                                 continue;
 276                         } else {
 277                                 /* slot is unreadable but 'block' link
 278                                  * still intact... something bad is happening
 279                                  * so abort
 280                                  */
 281                                 free(dev);
 282                                 goto abort;
 283                         }
 284
 285                 }
 286                 strcpy(dev->sys_name, de->d_name);
 287                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 288                 if (*ep) dev->disk.raid_disk = -1;
 289
 290                 strcpy(dbase, "block/dev");
 291                 if (load_sys(fname, buf, sizeof(buf))) {
 292                         /* assume this is a stale reference to a hot
 293                          * removed device
 294                          */
 295                         free(dev);
 296                         continue;
 297                 }
 298                 sra->array.nr_disks++;
 299                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 300
 301                 /* special case check for block devices that can go 'offline' */
 302                 strcpy(dbase, "block/device/state");
 303                 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
 304                     strncmp(buf, "offline", 7) == 0) {
 305                         free(dev);
 306                         continue;
 307                 }
 308
 309                 /* finally add this disk to the array */
 310                 *devp = dev;
 311                 devp = & dev->next;
 312                 dev->next = NULL;
 313
 314                 if (options & GET_OFFSET) {
 315                         strcpy(dbase, "offset");
 316                         if (load_sys(fname, buf, sizeof(buf)))
 317                                 goto abort;
 318                         dev->data_offset = strtoull(buf, NULL, 0);
 319                         strcpy(dbase, "new_offset");
 320                         if (load_sys(fname, buf, sizeof(buf)) == 0)
 321                                 dev->new_data_offset = strtoull(buf, NULL, 0);
 322                         else
 323                                 dev->new_data_offset = dev->data_offset;
 324                 }
 325                 if (options & GET_SIZE) {
 326                         strcpy(dbase, "size");
 327                         if (load_sys(fname, buf, sizeof(buf)))
 328                                 goto abort;
 329                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 330                 }
 331                 if (options & GET_STATE) {
 332                         dev->disk.state = 0;
 333                         strcpy(dbase, "state");
 334                         if (load_sys(fname, buf, sizeof(buf)))
 335                                 goto abort;
 336                         if (strstr(buf, "in_sync"))
 337                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 338                         if (strstr(buf, "faulty"))
 339                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 340                         if (dev->disk.state == 0)
 341                                 sra->array.spare_disks++;
 342                 }
 343                 if (options & GET_ERROR) {
 344                         strcpy(buf, "errors");
 345                         if (load_sys(fname, buf, sizeof(buf)))
 346                                 goto abort;
 347                         dev->errors = strtoul(buf, NULL, 0);
 348                 }
 349         }
 350         closedir(dir);
 351         return sra;
 352
 353  abort:
 354         if (dir)
 355                 closedir(dir);
 356         sysfs_free(sra);
 357         return NULL;
 358 }
 359
 360 int sysfs_attr_match(const char *attr, const char *str)
 361 {
 362         /* See if attr, read from a sysfs file, matches
 363          * str.  They must either be the same, or attr can
 364          * have a trailing newline or comma
 365          */
 366         while (*attr && *str && *attr == *str) {
 367                 attr++;
 368                 str++;
 369         }
 370
 371         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 372                 return 0;
 373         return 1;
 374 }
 375
 376 int sysfs_match_word(const char *word, char **list)
 377 {
 378         int n;
 379         for (n=0; list[n]; n++)
 380                 if (sysfs_attr_match(word, list[n]))
 381                         break;
 382         return n;
 383 }
 384
 385 unsigned long long get_component_size(int fd)
 386 {
 387         /* Find out the component size of the array.
 388          * We cannot trust GET_ARRAY_INFO ioctl as it's
 389          * size field is only 32bits.
 390          * So look in /sys/block/mdXXX/md/component_size
 391          *
 392          * This returns in units of sectors.
 393          */
 394         struct stat stb;
 395         char fname[50];
 396         int n;
 397         if (fstat(fd, &stb)) return 0;
 398         if (major(stb.st_rdev) != (unsigned)get_mdp_major())
 399                 sprintf(fname, "/sys/block/md%d/md/component_size",
 400                         (int)minor(stb.st_rdev));
 401         else
 402                 sprintf(fname, "/sys/block/md_d%d/md/component_size",
 403                         (int)minor(stb.st_rdev)>>MdpMinorShift);
 404         fd = open(fname, O_RDONLY);
 405         if (fd < 0)
 406                 return 0;
 407         n = read(fd, fname, sizeof(fname));
 408         close(fd);
 409         if (n < 0 || n == sizeof(fname))
 410                 return 0;
 411         fname[n] = 0;
 412         return strtoull(fname, NULL, 10) * 2;
 413 }
 414
 415 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 416                   char *name, char *val)
 417 {
 418         char fname[50];
 419         unsigned int n;
 420         int fd;
 421
 422         sprintf(fname, "/sys/block/%s/md/%s/%s",
 423                 sra->sys_name, dev?dev->sys_name:"", name);
 424         fd = open(fname, O_WRONLY);
 425         if (fd < 0)
 426                 return -1;
 427         n = write(fd, val, strlen(val));
 428         close(fd);
 429         if (n != strlen(val)) {
 430                 dprintf("failed to write '%s' to '%s' (%s)\n",
 431                         val, fname, strerror(errno));
 432                 return -1;
 433         }
 434         return 0;
 435 }
 436
 437 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 438                   char *name, unsigned long long val)
 439 {
 440         char valstr[50];
 441         sprintf(valstr, "%llu", val);
 442         return sysfs_set_str(sra, dev, name, valstr);
 443 }
 444
 445 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
 446                          char *name, long long val)
 447 {
 448         char valstr[50];
 449         sprintf(valstr, "%lli", val);
 450         return sysfs_set_str(sra, dev, name, valstr);
 451 }
 452
 453 int sysfs_uevent(struct mdinfo *sra, char *event)
 454 {
 455         char fname[50];
 456         int n;
 457         int fd;
 458
 459         sprintf(fname, "/sys/block/%s/uevent",
 460                 sra->sys_name);
 461         fd = open(fname, O_WRONLY);
 462         if (fd < 0)
 463                 return -1;
 464         n = write(fd, event, strlen(event));
 465         close(fd);
 466         if (n != (int)strlen(event)) {
 467                 dprintf("failed to write '%s' to '%s' (%s)\n",
 468                         event, fname, strerror(errno));
 469                 return -1;
 470         }
 471         return 0;
 472 }
 473
 474 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
 475 {
 476         char fname[50];
 477         struct stat st;
 478
 479         sprintf(fname, "/sys/block/%s/md/%s/%s",
 480                 sra->sys_name, dev?dev->sys_name:"", name);
 481
 482         return stat(fname, &st) == 0;
 483 }
 484
 485 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 486                        char *name)
 487 {
 488         char fname[50];
 489         int fd;
 490
 491         sprintf(fname, "/sys/block/%s/md/%s/%s",
 492                 sra->sys_name, dev?dev->sys_name:"", name);
 493         fd = open(fname, O_RDWR);
 494         if (fd < 0)
 495                 fd = open(fname, O_RDONLY);
 496         return fd;
 497 }
 498
 499 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 500 {
 501         char buf[50];
 502         int n;
 503         char *ep;
 504
 505         lseek(fd, 0, 0);
 506         n = read(fd, buf, sizeof(buf));
 507         if (n <= 0 || n == sizeof(buf))
 508                 return -2;
 509         buf[n] = 0;
 510         *val = strtoull(buf, &ep, 0);
 511         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 512                 return -1;
 513         return 0;
 514 }
 515
 516 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 517                        char *name, unsigned long long *val)
 518 {
 519         int n;
 520         int fd;
 521
 522         fd = sysfs_get_fd(sra, dev, name);
 523         if (fd < 0)
 524                 return -1;
 525         n = sysfs_fd_get_ll(fd, val);
 526         close(fd);
 527         return n;
 528 }
 529
 530 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
 531 {
 532         /* two numbers in this sysfs file, either
 533          *  NNN (NNN)
 534          * or
 535          *  NNN / NNN
 536          */
 537         char buf[80];
 538         int n;
 539         char *ep, *ep2;
 540
 541         lseek(fd, 0, 0);
 542         n = read(fd, buf, sizeof(buf));
 543         if (n <= 0 || n == sizeof(buf))
 544                 return -2;
 545         buf[n] = 0;
 546         *v1 = strtoull(buf, &ep, 0);
 547         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 548                 return -1;
 549         while (*ep == ' ' || *ep == '/' || *ep == '(')
 550                 ep++;
 551         *v2 = strtoull(ep, &ep2, 0);
 552         if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
 553                 *v2 = *v1;
 554                 return 1;
 555         }
 556         return 2;
 557 }
 558
 559 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
 560                   char *name, unsigned long long *v1, unsigned long long *v2)
 561 {
 562         int n;
 563         int fd;
 564
 565         fd = sysfs_get_fd(sra, dev, name);
 566         if (fd < 0)
 567                 return -1;
 568         n = sysfs_fd_get_two(fd, v1, v2);
 569         close(fd);
 570         return n;
 571 }
 572
 573 int sysfs_fd_get_str(int fd, char *val, int size)
 574 {
 575         int n;
 576
 577         lseek(fd, 0, 0);
 578         n = read(fd, val, size);
 579         if (n <= 0 || n == size)
 580                 return -1;
 581         val[n] = 0;
 582         return n;
 583 }
 584
 585 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 586                        char *name, char *val, int size)
 587 {
 588         int n;
 589         int fd;
 590
 591         fd = sysfs_get_fd(sra, dev, name);
 592         if (fd < 0)
 593                 return -1;
 594         n = sysfs_fd_get_str(fd, val, size);
 595         close(fd);
 596         return n;
 597 }
 598
 599 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 600 {
 601         unsigned long sec;
 602         unsigned long msec;
 603         char delay[30];
 604
 605         sec = ms / 1000;
 606         msec = ms % 1000;
 607
 608         sprintf(delay, "%ld.%03ld\n", sec, msec);
 609         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 610         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 611 }
 612
 613 int sysfs_set_array(struct mdinfo *info, int vers)
 614 {
 615         int rv = 0;
 616         char ver[100];
 617         int raid_disks = info->array.raid_disks;
 618
 619         ver[0] = 0;
 620         if (info->array.major_version == -1 &&
 621             info->array.minor_version == -2) {
 622                 char buf[1024];
 623
 624                 strcat(strcpy(ver, "external:"), info->text_version);
 625
 626                 /* meta version might already be set if we are setting
 627                  * new geometry for a reshape.  In that case we don't
 628                  * want to over-write the 'readonly' flag that is
 629                  * stored in the metadata version.  So read the current
 630                  * version first, and preserve the flag
 631                  */
 632                 if (sysfs_get_str(info, NULL, "metadata_version",
 633                                   buf, 1024) > 0)
 634                         if (strlen(buf) >= 9 && buf[9] == '-')
 635                                 ver[9] = '-';
 636
 637                 if ((vers % 100) < 2 ||
 638                     sysfs_set_str(info, NULL, "metadata_version",
 639                                   ver) < 0) {
 640                         pr_err("This kernel does not support external metadata.\n");
 641                         return 1;
 642                 }
 643         }
 644         if (info->array.level < 0)
 645                 return 0; /* FIXME */
 646         rv |= sysfs_set_str(info, NULL, "level",
 647                             map_num(pers, info->array.level));
 648         if (info->reshape_active && info->delta_disks != UnSet)
 649                 raid_disks -= info->delta_disks;
 650         rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
 651         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 652         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 653         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 654         if (info->custom_array_size) {
 655                 int rc;
 656
 657                 rc = sysfs_set_num(info, NULL, "array_size",
 658                                    info->custom_array_size/2);
 659                 if (rc && errno == ENOENT) {
 660                         pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
 661                         rc = 0;
 662                 }
 663                 rv |= rc;
 664         }
 665
 666         if (info->array.level > 0)
 667                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 668
 669         if (info->reshape_active) {
 670                 rv |= sysfs_set_num(info, NULL, "reshape_position",
 671                                     info->reshape_progress);
 672                 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
 673                 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
 674                 rv |= sysfs_set_num(info, NULL, "raid_disks",
 675                                     info->array.raid_disks);
 676                 /* We don't set 'new_level' here.  That can only happen
 677                  * once the reshape completes.
 678                  */
 679         }
 680         return rv;
 681 }
 682
 683 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 684 {
 685         char dv[PATH_MAX];
 686         char nm[PATH_MAX];
 687         char *dname;
 688         int rv;
 689
 690         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 691         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 692         if (rv)
 693                 return rv;
 694
 695         memset(nm, 0, sizeof(nm));
 696         dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
 697         strcpy(sd->sys_name, "dev-");
 698         strcpy(sd->sys_name+4, dname);
 699
 700         /* test write to see if 'recovery_start' is available */
 701         if (resume && sd->recovery_start < MaxSector &&
 702             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 703                 sysfs_set_str(sra, sd, "state", "remove");
 704                 return -1;
 705         }
 706
 707         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 708         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 709         if (sra->array.level != LEVEL_CONTAINER) {
 710                 if (sd->recovery_start == MaxSector)
 711                         /* This can correctly fail if array isn't started,
 712                          * yet, so just ignore status for now.
 713                          */
 714                         sysfs_set_str(sra, sd, "state", "insync");
 715                 if (sd->disk.raid_disk >= 0)
 716                         rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 717                 if (resume)
 718                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 719         }
 720         return rv;
 721 }
 722
 723 #if 0
 724 int sysfs_disk_to_sg(int fd)
 725 {
 726         /* from an open block device, try find and open its corresponding
 727          * scsi_generic interface
 728          */
 729         struct stat st;
 730         char path[256];
 731         char sg_path[256];
 732         char sg_major_minor[10];
 733         char *c;
 734         DIR *dir;
 735         struct dirent *de;
 736         int major, minor, rv;
 737
 738         if (fstat(fd, &st))
 739                 return -1;
 740
 741         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 742                  major(st.st_rdev), minor(st.st_rdev));
 743
 744         dir = opendir(path);
 745         if (!dir)
 746                 return -1;
 747
 748         de = readdir(dir);
 749         while (de) {
 750                 if (strncmp("scsi_generic:", de->d_name,
 751                             strlen("scsi_generic:")) == 0)
 752                         break;
 753                 de = readdir(dir);
 754         }
 755         closedir(dir);
 756
 757         if (!de)
 758                 return -1;
 759
 760         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 761         fd = open(sg_path, O_RDONLY);
 762         if (fd < 0)
 763                 return fd;
 764
 765         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 766         close(fd);
 767         if (rv < 0 || rv == sizeof(sg_major_minor))
 768                 return -1;
 769         else
 770                 sg_major_minor[rv - 1] = '\0';
 771
 772         c = strchr(sg_major_minor, ':');
 773         *c = '\0';
 774         c++;
 775         major = strtol(sg_major_minor, NULL, 10);
 776         minor = strtol(c, NULL, 10);
 777         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 778                  (int) getpid(), major, minor);
 779         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 780                         fd = open(path, O_RDONLY);
 781                         unlink(path);
 782                         return fd;
 783         }
 784
 785         return -1;
 786 }
 787 #endif
 788
 789 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 790 {
 791         /* from an open block device, try to retrieve it scsi_id */
 792         struct stat st;
 793         char path[256];
 794         DIR *dir;
 795         struct dirent *de;
 796         int host, bus, target, lun;
 797
 798         if (fstat(fd, &st))
 799                 return 1;
 800
 801         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
 802                  major(st.st_rdev), minor(st.st_rdev));
 803
 804         dir = opendir(path);
 805         if (!dir)
 806                 return 1;
 807
 808         for (de = readdir(dir); de; de = readdir(dir)) {
 809                 int count;
 810
 811                 if (de->d_type != DT_DIR)
 812                         continue;
 813
 814                 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
 815                 if (count == 4)
 816                         break;
 817         }
 818         closedir(dir);
 819
 820         if (!de)
 821                 return 1;
 822
 823         *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
 824         return 0;
 825 }
 826
 827 int sysfs_unique_holder(char *devnm, long rdev)
 828 {
 829         /* Check that devnm is a holder of rdev,
 830          * and is the only holder.
 831          * we should be locked against races by
 832          * an O_EXCL on devnm
 833          * Return values:
 834          *  0 - not unique, not even a holder
 835          *  1 - unique, this is the only holder.
 836          *  2/3 - not unique, there is another holder
 837          * -1 - error, cannot find the holders
 838          */
 839         DIR *dir;
 840         struct dirent *de;
 841         char dirname[100];
 842         char l;
 843         int ret = 0;
 844         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 845                 major(rdev), minor(rdev));
 846         dir = opendir(dirname);
 847         if (!dir)
 848                 return -1;
 849         l = strlen(dirname);
 850         while ((de = readdir(dir)) != NULL) {
 851                 char buf[100];
 852                 char *sl;
 853                 int n;
 854
 855                 if (de->d_ino == 0)
 856                         continue;
 857                 if (de->d_name[0] == '.')
 858                         continue;
 859                 strcpy(dirname+l, "/");
 860                 strcat(dirname+l, de->d_name);
 861                 n = readlink(dirname, buf, sizeof(buf)-1);
 862                 if (n <= 0)
 863                         continue;
 864                 buf[n] = 0;
 865                 sl = strrchr(buf, '/');
 866                 if (!sl)
 867                         continue;
 868                 sl++;
 869
 870                 if (strcmp(devnm, sl) == 0)
 871                         ret |= 1;
 872                 else
 873                         ret |= 2;
 874         }
 875         closedir(dir);
 876         return ret;
 877 }
 878
 879 int sysfs_freeze_array(struct mdinfo *sra)
 880 {
 881         /* Try to freeze resync/rebuild on this array/container.
 882          * Return -1 if the array is busy,
 883          * return 0 if this kernel doesn't support 'frozen'
 884          * return 1 if it worked.
 885          */
 886         char buf[20];
 887
 888         if (!sysfs_attribute_available(sra, NULL, "sync_action"))
 889                 return 1; /* no sync_action == frozen */
 890         if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
 891                 return 0;
 892         if (strcmp(buf, "frozen\n") == 0)
 893                 /* Already frozen */
 894                 return 0;
 895         if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
 896                 return -1;
 897         if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
 898                 return 0;
 899         return 1;
 900 }
 901
 902 int sysfs_wait(int fd, int *msec)
 903 {
 904         /* Wait up to '*msec' for fd to have an exception condition.
 905          * if msec == NULL, wait indefinitely.
 906          */
 907         fd_set fds;
 908         int n;
 909         FD_ZERO(&fds);
 910         FD_SET(fd, &fds);
 911         if (msec == NULL)
 912                 n = select(fd+1, NULL, NULL, &fds, NULL);
 913         else if (*msec < 0)
 914                 n = 0;
 915         else {
 916                 struct timeval start, end, tv;
 917                 gettimeofday(&start, NULL);
 918                 if (*msec < 1000) {
 919                         tv.tv_sec = 0;
 920                         tv.tv_usec = (*msec)*1000;
 921                 } else {
 922                         tv.tv_sec = (*msec)/1000;
 923                         tv.tv_usec = 0;
 924                 }
 925                 n = select(fd+1, NULL, NULL, &fds, &tv);
 926                 gettimeofday(&end, NULL);
 927                 end.tv_sec -= start.tv_sec;
 928                 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
 929                           - start.tv_usec/1000) + 1;
 930         }
 931         return n;
 932 }