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 #define MAX_SYSFS_PATH_LEN      120
  31
  32 int load_sys(char *path, char *buf, int len)
  33 {
  34         int fd = open(path, O_RDONLY);
  35         int n;
  36         if (fd < 0)
  37                 return -1;
  38         n = read(fd, buf, len);
  39         close(fd);
  40         if (n <0 || n >= len)
  41                 return -1;
  42         buf[n] = 0;
  43         if (n && buf[n-1] == '\n')
  44                 buf[n-1] = 0;
  45         return 0;
  46 }
  47
  48 void sysfs_free(struct mdinfo *sra)
  49 {
  50         while (sra) {
  51                 struct mdinfo *sra2 = sra->next;
  52                 while (sra->devs) {
  53                         struct mdinfo *d = sra->devs;
  54                         sra->devs = d->next;
  55                         free(d);
  56                 }
  57                 free(sra);
  58                 sra = sra2;
  59         }
  60 }
  61
  62 int sysfs_open(char *devnm, char *devname, char *attr)
  63 {
  64         char fname[MAX_SYSFS_PATH_LEN];
  65         int fd;
  66
  67         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/", devnm);
  68         if (devname) {
  69                 strncat(fname, devname, MAX_SYSFS_PATH_LEN - strlen(fname));
  70                 strncat(fname, "/", MAX_SYSFS_PATH_LEN - strlen(fname));
  71         }
  72         strncat(fname, attr, MAX_SYSFS_PATH_LEN - strlen(fname));
  73         fd = open(fname, O_RDWR);
  74         if (fd < 0 && errno == EACCES)
  75                 fd = open(fname, O_RDONLY);
  76         return fd;
  77 }
  78
  79 void sysfs_init_dev(struct mdinfo *mdi, unsigned long devid)
  80 {
  81         snprintf(mdi->sys_name,
  82                  sizeof(mdi->sys_name), "dev-%s", devid2kname(devid));
  83 }
  84
  85 void sysfs_init(struct mdinfo *mdi, int fd, char *devnm)
  86 {
  87         mdi->sys_name[0] = 0;
  88         if (fd >= 0) {
  89                 mdu_version_t vers;
  90                 if (ioctl(fd, RAID_VERSION, &vers) != 0)
  91                         return;
  92                 devnm = fd2devnm(fd);
  93         }
  94         if (devnm == NULL)
  95                 return;
  96         strcpy(mdi->sys_name, devnm);
  97 }
  98
  99 struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
 100 {
 101         char fname[PATH_MAX];
 102         char buf[PATH_MAX];
 103         char *base;
 104         char *dbase;
 105         struct mdinfo *sra;
 106         struct mdinfo *dev, **devp;
 107         DIR *dir = NULL;
 108         struct dirent *de;
 109
 110         sra = xcalloc(1, sizeof(*sra));
 111         sysfs_init(sra, fd, devnm);
 112         if (sra->sys_name[0] == 0) {
 113                 free(sra);
 114                 return NULL;
 115         }
 116
 117         sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
 118         base = fname + strlen(fname);
 119
 120         sra->devs = NULL;
 121         if (options & GET_VERSION) {
 122                 strcpy(base, "metadata_version");
 123                 if (load_sys(fname, buf, sizeof(buf)))
 124                         goto abort;
 125                 if (strncmp(buf, "none", 4) == 0) {
 126                         sra->array.major_version =
 127                                 sra->array.minor_version = -1;
 128                         strcpy(sra->text_version, "");
 129                 } else if (strncmp(buf, "external:", 9) == 0) {
 130                         sra->array.major_version = -1;
 131                         sra->array.minor_version = -2;
 132                         strcpy(sra->text_version, buf+9);
 133                 } else {
 134                         sscanf(buf, "%d.%d",
 135                                &sra->array.major_version,
 136                                &sra->array.minor_version);
 137                         strcpy(sra->text_version, buf);
 138                 }
 139         }
 140         if (options & GET_LEVEL) {
 141                 strcpy(base, "level");
 142                 if (load_sys(fname, buf, sizeof(buf)))
 143                         goto abort;
 144                 sra->array.level = map_name(pers, buf);
 145         }
 146         if (options & GET_LAYOUT) {
 147                 strcpy(base, "layout");
 148                 if (load_sys(fname, buf, sizeof(buf)))
 149                         goto abort;
 150                 sra->array.layout = strtoul(buf, NULL, 0);
 151         }
 152         if (options & GET_DISKS) {
 153                 strcpy(base, "raid_disks");
 154                 if (load_sys(fname, buf, sizeof(buf)))
 155                         goto abort;
 156                 sra->array.raid_disks = strtoul(buf, NULL, 0);
 157         }
 158         if (options & GET_DEGRADED) {
 159                 strcpy(base, "degraded");
 160                 if (load_sys(fname, buf, sizeof(buf)))
 161                         goto abort;
 162                 sra->array.failed_disks = strtoul(buf, NULL, 0);
 163         }
 164         if (options & GET_COMPONENT) {
 165                 strcpy(base, "component_size");
 166                 if (load_sys(fname, buf, sizeof(buf)))
 167                         goto abort;
 168                 sra->component_size = strtoull(buf, NULL, 0);
 169                 /* sysfs reports "K", but we want sectors */
 170                 sra->component_size *= 2;
 171         }
 172         if (options & GET_CHUNK) {
 173                 strcpy(base, "chunk_size");
 174                 if (load_sys(fname, buf, sizeof(buf)))
 175                         goto abort;
 176                 sra->array.chunk_size = strtoul(buf, NULL, 0);
 177         }
 178         if (options & GET_CACHE) {
 179                 strcpy(base, "stripe_cache_size");
 180                 if (load_sys(fname, buf, sizeof(buf)))
 181                         /* Probably level doesn't support it */
 182                         sra->cache_size = 0;
 183                 else
 184                         sra->cache_size = strtoul(buf, NULL, 0);
 185         }
 186         if (options & GET_MISMATCH) {
 187                 strcpy(base, "mismatch_cnt");
 188                 if (load_sys(fname, buf, sizeof(buf)))
 189                         goto abort;
 190                 sra->mismatch_cnt = strtoul(buf, NULL, 0);
 191         }
 192         if (options & GET_SAFEMODE) {
 193                 int scale = 1;
 194                 int dot = 0;
 195                 unsigned i;
 196                 unsigned long msec;
 197                 size_t len;
 198
 199                 strcpy(base, "safe_mode_delay");
 200                 if (load_sys(fname, buf, sizeof(buf)))
 201                         goto abort;
 202
 203                 /* remove a period, and count digits after it */
 204                 len = strlen(buf);
 205                 for (i = 0; i < len; i++) {
 206                         if (dot) {
 207                                 if (isdigit(buf[i])) {
 208                                         buf[i-1] = buf[i];
 209                                         scale *= 10;
 210                                 }
 211                                 buf[i] = 0;
 212                         } else if (buf[i] == '.') {
 213                                 dot=1;
 214                                 buf[i] = 0;
 215                         }
 216                 }
 217                 msec = strtoul(buf, NULL, 10);
 218                 msec = (msec * 1000) / scale;
 219                 sra->safe_mode_delay = msec;
 220         }
 221         if (options & GET_BITMAP_LOCATION) {
 222                 strcpy(base, "bitmap/location");
 223                 if (load_sys(fname, buf, sizeof(buf)))
 224                         goto abort;
 225                 if (strncmp(buf, "file", 4) == 0)
 226                         sra->bitmap_offset = 1;
 227                 else if (strncmp(buf, "none", 4) == 0)
 228                         sra->bitmap_offset = 0;
 229                 else if (buf[0] == '+')
 230                         sra->bitmap_offset = strtol(buf+1, NULL, 10);
 231                 else
 232                         goto abort;
 233         }
 234
 235         if (options & GET_ARRAY_STATE) {
 236                 strcpy(base, "array_state");
 237                 if (load_sys(fname, sra->sysfs_array_state,
 238                              sizeof(sra->sysfs_array_state)))
 239                         goto abort;
 240         } else
 241                 sra->sysfs_array_state[0] = 0;
 242
 243         if (! (options & GET_DEVS))
 244                 return sra;
 245
 246         /* Get all the devices as well */
 247         *base = 0;
 248         dir = opendir(fname);
 249         if (!dir)
 250                 goto abort;
 251         sra->array.spare_disks = 0;
 252
 253         devp = &sra->devs;
 254         sra->devs = NULL;
 255         while ((de = readdir(dir)) != NULL) {
 256                 char *ep;
 257                 if (de->d_ino == 0 ||
 258                     strncmp(de->d_name, "dev-", 4) != 0)
 259                         continue;
 260                 strcpy(base, de->d_name);
 261                 dbase = base + strlen(base);
 262                 *dbase++ = '/';
 263
 264                 dev = xmalloc(sizeof(*dev));
 265
 266                 /* Always get slot, major, minor */
 267                 strcpy(dbase, "slot");
 268                 if (load_sys(fname, buf, sizeof(buf))) {
 269                         /* hmm... unable to read 'slot' maybe the device
 270                          * is going away?
 271                          */
 272                         strcpy(dbase, "block");
 273                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 274                             errno != ENAMETOOLONG) {
 275                                 /* ...yup device is gone */
 276                                 free(dev);
 277                                 continue;
 278                         } else {
 279                                 /* slot is unreadable but 'block' link
 280                                  * still intact... something bad is happening
 281                                  * so abort
 282                                  */
 283                                 free(dev);
 284                                 goto abort;
 285                         }
 286
 287                 }
 288                 strcpy(dev->sys_name, de->d_name);
 289                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 290                 if (*ep) dev->disk.raid_disk = -1;
 291
 292                 strcpy(dbase, "block/dev");
 293                 if (load_sys(fname, buf, sizeof(buf))) {
 294                         /* assume this is a stale reference to a hot
 295                          * removed device
 296                          */
 297                         free(dev);
 298                         continue;
 299                 }
 300                 sra->array.nr_disks++;
 301                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 302
 303                 /* special case check for block devices that can go 'offline' */
 304                 strcpy(dbase, "block/device/state");
 305                 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
 306                     strncmp(buf, "offline", 7) == 0) {
 307                         free(dev);
 308                         continue;
 309                 }
 310
 311                 /* finally add this disk to the array */
 312                 *devp = dev;
 313                 devp = & dev->next;
 314                 dev->next = NULL;
 315
 316                 if (options & GET_OFFSET) {
 317                         strcpy(dbase, "offset");
 318                         if (load_sys(fname, buf, sizeof(buf)))
 319                                 goto abort;
 320                         dev->data_offset = strtoull(buf, NULL, 0);
 321                         strcpy(dbase, "new_offset");
 322                         if (load_sys(fname, buf, sizeof(buf)) == 0)
 323                                 dev->new_data_offset = strtoull(buf, NULL, 0);
 324                         else
 325                                 dev->new_data_offset = dev->data_offset;
 326                 }
 327                 if (options & GET_SIZE) {
 328                         strcpy(dbase, "size");
 329                         if (load_sys(fname, buf, sizeof(buf)))
 330                                 goto abort;
 331                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 332                 }
 333                 if (options & GET_STATE) {
 334                         dev->disk.state = 0;
 335                         strcpy(dbase, "state");
 336                         if (load_sys(fname, buf, sizeof(buf)))
 337                                 goto abort;
 338                         if (strstr(buf, "in_sync"))
 339                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 340                         if (strstr(buf, "faulty"))
 341                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 342                         if (dev->disk.state == 0)
 343                                 sra->array.spare_disks++;
 344                 }
 345                 if (options & GET_ERROR) {
 346                         strcpy(buf, "errors");
 347                         if (load_sys(fname, buf, sizeof(buf)))
 348                                 goto abort;
 349                         dev->errors = strtoul(buf, NULL, 0);
 350                 }
 351         }
 352         closedir(dir);
 353         return sra;
 354
 355  abort:
 356         if (dir)
 357                 closedir(dir);
 358         sysfs_free(sra);
 359         return NULL;
 360 }
 361
 362 int sysfs_attr_match(const char *attr, const char *str)
 363 {
 364         /* See if attr, read from a sysfs file, matches
 365          * str.  They must either be the same, or attr can
 366          * have a trailing newline or comma
 367          */
 368         while (*attr && *str && *attr == *str) {
 369                 attr++;
 370                 str++;
 371         }
 372
 373         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 374                 return 0;
 375         return 1;
 376 }
 377
 378 int sysfs_match_word(const char *word, char **list)
 379 {
 380         int n;
 381         for (n=0; list[n]; n++)
 382                 if (sysfs_attr_match(word, list[n]))
 383                         break;
 384         return n;
 385 }
 386
 387 unsigned long long get_component_size(int fd)
 388 {
 389         /* Find out the component size of the array.
 390          * We cannot trust GET_ARRAY_INFO ioctl as it's
 391          * size field is only 32bits.
 392          * So look in /sys/block/mdXXX/md/component_size
 393          *
 394          * This returns in units of sectors.
 395          */
 396         struct stat stb;
 397         char fname[MAX_SYSFS_PATH_LEN];
 398         int n;
 399         if (fstat(fd, &stb))
 400                 return 0;
 401         if (major(stb.st_rdev) != (unsigned)get_mdp_major())
 402                 snprintf(fname, MAX_SYSFS_PATH_LEN,
 403                         "/sys/block/md%d/md/component_size",
 404                         (int)minor(stb.st_rdev));
 405         else
 406                 snprintf(fname, MAX_SYSFS_PATH_LEN,
 407                         "/sys/block/md_d%d/md/component_size",
 408                         (int)minor(stb.st_rdev)>>MdpMinorShift);
 409         fd = open(fname, O_RDONLY);
 410         if (fd < 0)
 411                 return 0;
 412         n = read(fd, fname, sizeof(fname));
 413         close(fd);
 414         if (n < 0 || n == sizeof(fname))
 415                 return 0;
 416         fname[n] = 0;
 417         return strtoull(fname, NULL, 10) * 2;
 418 }
 419
 420 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 421                   char *name, char *val)
 422 {
 423         char fname[MAX_SYSFS_PATH_LEN];
 424         unsigned int n;
 425         int fd;
 426
 427         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 428                 sra->sys_name, dev?dev->sys_name:"", name);
 429         fd = open(fname, O_WRONLY);
 430         if (fd < 0)
 431                 return -1;
 432         n = write(fd, val, strlen(val));
 433         close(fd);
 434         if (n != strlen(val)) {
 435                 dprintf("failed to write '%s' to '%s' (%s)\n",
 436                         val, fname, strerror(errno));
 437                 return -1;
 438         }
 439         return 0;
 440 }
 441
 442 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 443                   char *name, unsigned long long val)
 444 {
 445         char valstr[50];
 446         sprintf(valstr, "%llu", val);
 447         return sysfs_set_str(sra, dev, name, valstr);
 448 }
 449
 450 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
 451                          char *name, long long val)
 452 {
 453         char valstr[50];
 454         sprintf(valstr, "%lli", val);
 455         return sysfs_set_str(sra, dev, name, valstr);
 456 }
 457
 458 int sysfs_uevent(struct mdinfo *sra, char *event)
 459 {
 460         char fname[MAX_SYSFS_PATH_LEN];
 461         int n;
 462         int fd;
 463
 464         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/uevent",
 465                 sra->sys_name);
 466         fd = open(fname, O_WRONLY);
 467         if (fd < 0)
 468                 return -1;
 469         n = write(fd, event, strlen(event));
 470         close(fd);
 471         if (n != (int)strlen(event)) {
 472                 dprintf("failed to write '%s' to '%s' (%s)\n",
 473                         event, fname, strerror(errno));
 474                 return -1;
 475         }
 476         return 0;
 477 }
 478
 479 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
 480 {
 481         char fname[MAX_SYSFS_PATH_LEN];
 482         struct stat st;
 483
 484         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 485                 sra->sys_name, dev?dev->sys_name:"", name);
 486
 487         return stat(fname, &st) == 0;
 488 }
 489
 490 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 491                        char *name)
 492 {
 493         char fname[MAX_SYSFS_PATH_LEN];
 494         int fd;
 495
 496         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 497                 sra->sys_name, dev?dev->sys_name:"", name);
 498         fd = open(fname, O_RDWR);
 499         if (fd < 0)
 500                 fd = open(fname, O_RDONLY);
 501         return fd;
 502 }
 503
 504 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 505 {
 506         char buf[50];
 507         int n;
 508         char *ep;
 509
 510         lseek(fd, 0, 0);
 511         n = read(fd, buf, sizeof(buf));
 512         if (n <= 0 || n == sizeof(buf))
 513                 return -2;
 514         buf[n] = 0;
 515         *val = strtoull(buf, &ep, 0);
 516         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 517                 return -1;
 518         return 0;
 519 }
 520
 521 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 522                        char *name, unsigned long long *val)
 523 {
 524         int n;
 525         int fd;
 526
 527         fd = sysfs_get_fd(sra, dev, name);
 528         if (fd < 0)
 529                 return -1;
 530         n = sysfs_fd_get_ll(fd, val);
 531         close(fd);
 532         return n;
 533 }
 534
 535 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
 536 {
 537         /* two numbers in this sysfs file, either
 538          *  NNN (NNN)
 539          * or
 540          *  NNN / NNN
 541          */
 542         char buf[80];
 543         int n;
 544         char *ep, *ep2;
 545
 546         lseek(fd, 0, 0);
 547         n = read(fd, buf, sizeof(buf));
 548         if (n <= 0 || n == sizeof(buf))
 549                 return -2;
 550         buf[n] = 0;
 551         *v1 = strtoull(buf, &ep, 0);
 552         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 553                 return -1;
 554         while (*ep == ' ' || *ep == '/' || *ep == '(')
 555                 ep++;
 556         *v2 = strtoull(ep, &ep2, 0);
 557         if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
 558                 *v2 = *v1;
 559                 return 1;
 560         }
 561         return 2;
 562 }
 563
 564 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
 565                   char *name, unsigned long long *v1, unsigned long long *v2)
 566 {
 567         int n;
 568         int fd;
 569
 570         fd = sysfs_get_fd(sra, dev, name);
 571         if (fd < 0)
 572                 return -1;
 573         n = sysfs_fd_get_two(fd, v1, v2);
 574         close(fd);
 575         return n;
 576 }
 577
 578 int sysfs_fd_get_str(int fd, char *val, int size)
 579 {
 580         int n;
 581
 582         lseek(fd, 0, 0);
 583         n = read(fd, val, size);
 584         if (n <= 0 || n == size)
 585                 return -1;
 586         val[n] = 0;
 587         return n;
 588 }
 589
 590 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 591                        char *name, char *val, int size)
 592 {
 593         int n;
 594         int fd;
 595
 596         fd = sysfs_get_fd(sra, dev, name);
 597         if (fd < 0)
 598                 return -1;
 599         n = sysfs_fd_get_str(fd, val, size);
 600         close(fd);
 601         return n;
 602 }
 603
 604 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 605 {
 606         unsigned long sec;
 607         unsigned long msec;
 608         char delay[30];
 609
 610         sec = ms / 1000;
 611         msec = ms % 1000;
 612
 613         sprintf(delay, "%ld.%03ld\n", sec, msec);
 614         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 615         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 616 }
 617
 618 int sysfs_set_array(struct mdinfo *info, int vers)
 619 {
 620         int rv = 0;
 621         char ver[100];
 622         int raid_disks = info->array.raid_disks;
 623
 624         ver[0] = 0;
 625         if (info->array.major_version == -1 &&
 626             info->array.minor_version == -2) {
 627                 char buf[1024];
 628
 629                 strcat(strcpy(ver, "external:"), info->text_version);
 630
 631                 /* meta version might already be set if we are setting
 632                  * new geometry for a reshape.  In that case we don't
 633                  * want to over-write the 'readonly' flag that is
 634                  * stored in the metadata version.  So read the current
 635                  * version first, and preserve the flag
 636                  */
 637                 if (sysfs_get_str(info, NULL, "metadata_version",
 638                                   buf, 1024) > 0)
 639                         if (strlen(buf) >= 9 && buf[9] == '-')
 640                                 ver[9] = '-';
 641
 642                 if ((vers % 100) < 2 ||
 643                     sysfs_set_str(info, NULL, "metadata_version",
 644                                   ver) < 0) {
 645                         pr_err("This kernel does not support external metadata.\n");
 646                         return 1;
 647                 }
 648         }
 649         if (info->array.level < 0)
 650                 return 0; /* FIXME */
 651         rv |= sysfs_set_str(info, NULL, "level",
 652                             map_num(pers, info->array.level));
 653         if (info->reshape_active && info->delta_disks != UnSet)
 654                 raid_disks -= info->delta_disks;
 655         rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
 656         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 657         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 658         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 659         if (info->custom_array_size) {
 660                 int rc;
 661
 662                 rc = sysfs_set_num(info, NULL, "array_size",
 663                                    info->custom_array_size/2);
 664                 if (rc && errno == ENOENT) {
 665                         pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
 666                         rc = 0;
 667                 }
 668                 rv |= rc;
 669         }
 670
 671         if (info->array.level > 0)
 672                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 673
 674         if (info->reshape_active) {
 675                 rv |= sysfs_set_num(info, NULL, "reshape_position",
 676                                     info->reshape_progress);
 677                 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
 678                 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
 679                 rv |= sysfs_set_num(info, NULL, "raid_disks",
 680                                     info->array.raid_disks);
 681                 /* We don't set 'new_level' here.  That can only happen
 682                  * once the reshape completes.
 683                  */
 684         }
 685         return rv;
 686 }
 687
 688 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 689 {
 690         char dv[PATH_MAX];
 691         char nm[PATH_MAX];
 692         char *dname;
 693         int rv;
 694
 695         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 696         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 697         if (rv)
 698                 return rv;
 699
 700         memset(nm, 0, sizeof(nm));
 701         dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
 702         strcpy(sd->sys_name, "dev-");
 703         strcpy(sd->sys_name+4, dname);
 704
 705         /* test write to see if 'recovery_start' is available */
 706         if (resume && sd->recovery_start < MaxSector &&
 707             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 708                 sysfs_set_str(sra, sd, "state", "remove");
 709                 return -1;
 710         }
 711
 712         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 713         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 714         if (sra->array.level != LEVEL_CONTAINER) {
 715                 if (sd->recovery_start == MaxSector)
 716                         /* This can correctly fail if array isn't started,
 717                          * yet, so just ignore status for now.
 718                          */
 719                         sysfs_set_str(sra, sd, "state", "insync");
 720                 if (sd->disk.raid_disk >= 0)
 721                         rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 722                 if (resume)
 723                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 724         }
 725         return rv;
 726 }
 727
 728 #if 0
 729 int sysfs_disk_to_sg(int fd)
 730 {
 731         /* from an open block device, try find and open its corresponding
 732          * scsi_generic interface
 733          */
 734         struct stat st;
 735         char path[256];
 736         char sg_path[256];
 737         char sg_major_minor[10];
 738         char *c;
 739         DIR *dir;
 740         struct dirent *de;
 741         int major, minor, rv;
 742
 743         if (fstat(fd, &st))
 744                 return -1;
 745
 746         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 747                  major(st.st_rdev), minor(st.st_rdev));
 748
 749         dir = opendir(path);
 750         if (!dir)
 751                 return -1;
 752
 753         de = readdir(dir);
 754         while (de) {
 755                 if (strncmp("scsi_generic:", de->d_name,
 756                             strlen("scsi_generic:")) == 0)
 757                         break;
 758                 de = readdir(dir);
 759         }
 760         closedir(dir);
 761
 762         if (!de)
 763                 return -1;
 764
 765         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 766         fd = open(sg_path, O_RDONLY);
 767         if (fd < 0)
 768                 return fd;
 769
 770         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 771         close(fd);
 772         if (rv < 0 || rv == sizeof(sg_major_minor))
 773                 return -1;
 774         else
 775                 sg_major_minor[rv - 1] = '\0';
 776
 777         c = strchr(sg_major_minor, ':');
 778         *c = '\0';
 779         c++;
 780         major = strtol(sg_major_minor, NULL, 10);
 781         minor = strtol(c, NULL, 10);
 782         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 783                  (int) getpid(), major, minor);
 784         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 785                         fd = open(path, O_RDONLY);
 786                         unlink(path);
 787                         return fd;
 788         }
 789
 790         return -1;
 791 }
 792 #endif
 793
 794 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 795 {
 796         /* from an open block device, try to retrieve it scsi_id */
 797         struct stat st;
 798         char path[256];
 799         DIR *dir;
 800         struct dirent *de;
 801         int host, bus, target, lun;
 802
 803         if (fstat(fd, &st))
 804                 return 1;
 805
 806         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
 807                  major(st.st_rdev), minor(st.st_rdev));
 808
 809         dir = opendir(path);
 810         if (!dir)
 811                 return 1;
 812
 813         for (de = readdir(dir); de; de = readdir(dir)) {
 814                 int count;
 815
 816                 if (de->d_type != DT_DIR)
 817                         continue;
 818
 819                 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
 820                 if (count == 4)
 821                         break;
 822         }
 823         closedir(dir);
 824
 825         if (!de)
 826                 return 1;
 827
 828         *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
 829         return 0;
 830 }
 831
 832 int sysfs_unique_holder(char *devnm, long rdev)
 833 {
 834         /* Check that devnm is a holder of rdev,
 835          * and is the only holder.
 836          * we should be locked against races by
 837          * an O_EXCL on devnm
 838          * Return values:
 839          *  0 - not unique, not even a holder
 840          *  1 - unique, this is the only holder.
 841          *  2/3 - not unique, there is another holder
 842          * -1 - error, cannot find the holders
 843          */
 844         DIR *dir;
 845         struct dirent *de;
 846         char dirname[100];
 847         char l;
 848         int ret = 0;
 849         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 850                 major(rdev), minor(rdev));
 851         dir = opendir(dirname);
 852         if (!dir)
 853                 return -1;
 854         l = strlen(dirname);
 855         while ((de = readdir(dir)) != NULL) {
 856                 char buf[100];
 857                 char *sl;
 858                 int n;
 859
 860                 if (de->d_ino == 0)
 861                         continue;
 862                 if (de->d_name[0] == '.')
 863                         continue;
 864                 strcpy(dirname+l, "/");
 865                 strcat(dirname+l, de->d_name);
 866                 n = readlink(dirname, buf, sizeof(buf)-1);
 867                 if (n <= 0)
 868                         continue;
 869                 buf[n] = 0;
 870                 sl = strrchr(buf, '/');
 871                 if (!sl)
 872                         continue;
 873                 sl++;
 874
 875                 if (strcmp(devnm, sl) == 0)
 876                         ret |= 1;
 877                 else
 878                         ret |= 2;
 879         }
 880         closedir(dir);
 881         return ret;
 882 }
 883
 884 int sysfs_freeze_array(struct mdinfo *sra)
 885 {
 886         /* Try to freeze resync/rebuild on this array/container.
 887          * Return -1 if the array is busy,
 888          * return 0 if this kernel doesn't support 'frozen'
 889          * return 1 if it worked.
 890          */
 891         char buf[20];
 892
 893         if (!sysfs_attribute_available(sra, NULL, "sync_action"))
 894                 return 1; /* no sync_action == frozen */
 895         if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
 896                 return 0;
 897         if (strcmp(buf, "frozen\n") == 0)
 898                 /* Already frozen */
 899                 return 0;
 900         if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
 901                 return -1;
 902         if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
 903                 return 0;
 904         return 1;
 905 }
 906
 907 int sysfs_wait(int fd, int *msec)
 908 {
 909         /* Wait up to '*msec' for fd to have an exception condition.
 910          * if msec == NULL, wait indefinitely.
 911          */
 912         fd_set fds;
 913         int n;
 914         FD_ZERO(&fds);
 915         FD_SET(fd, &fds);
 916         if (msec == NULL)
 917                 n = select(fd+1, NULL, NULL, &fds, NULL);
 918         else if (*msec < 0)
 919                 n = 0;
 920         else {
 921                 struct timeval start, end, tv;
 922                 gettimeofday(&start, NULL);
 923                 if (*msec < 1000) {
 924                         tv.tv_sec = 0;
 925                         tv.tv_usec = (*msec)*1000;
 926                 } else {
 927                         tv.tv_sec = (*msec)/1000;
 928                         tv.tv_usec = 0;
 929                 }
 930                 n = select(fd+1, NULL, NULL, &fds, &tv);
 931                 gettimeofday(&end, NULL);
 932                 end.tv_sec -= start.tv_sec;
 933                 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
 934                           - start.tv_usec/1000) + 1;
 935         }
 936         return n;
 937 }