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->bb.entries);
  56                         free(d);
  57                 }
  58                 free(sra->bb.entries);
  59                 free(sra);
  60                 sra = sra2;
  61         }
  62 }
  63
  64 int sysfs_open(char *devnm, char *devname, char *attr)
  65 {
  66         char fname[MAX_SYSFS_PATH_LEN];
  67         int fd;
  68
  69         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/", devnm);
  70         if (devname) {
  71                 strncat(fname, devname, MAX_SYSFS_PATH_LEN - strlen(fname));
  72                 strncat(fname, "/", MAX_SYSFS_PATH_LEN - strlen(fname));
  73         }
  74         strncat(fname, attr, MAX_SYSFS_PATH_LEN - strlen(fname));
  75         fd = open(fname, O_RDWR);
  76         if (fd < 0 && errno == EACCES)
  77                 fd = open(fname, O_RDONLY);
  78         return fd;
  79 }
  80
  81 void sysfs_init_dev(struct mdinfo *mdi, unsigned long devid)
  82 {
  83         snprintf(mdi->sys_name,
  84                  sizeof(mdi->sys_name), "dev-%s", devid2kname(devid));
  85 }
  86
  87 void sysfs_init(struct mdinfo *mdi, int fd, char *devnm)
  88 {
  89         mdi->sys_name[0] = 0;
  90         if (fd >= 0) {
  91                 mdu_version_t vers;
  92                 if (ioctl(fd, RAID_VERSION, &vers) != 0)
  93                         return;
  94                 devnm = fd2devnm(fd);
  95         }
  96         if (devnm == NULL)
  97                 return;
  98         strcpy(mdi->sys_name, devnm);
  99 }
 100
 101 struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
 102 {
 103         char fname[PATH_MAX];
 104         char buf[PATH_MAX];
 105         char *base;
 106         char *dbase;
 107         struct mdinfo *sra;
 108         struct mdinfo *dev, **devp;
 109         DIR *dir = NULL;
 110         struct dirent *de;
 111
 112         sra = xcalloc(1, sizeof(*sra));
 113         sysfs_init(sra, fd, devnm);
 114         if (sra->sys_name[0] == 0) {
 115                 free(sra);
 116                 return NULL;
 117         }
 118
 119         sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
 120         base = fname + strlen(fname);
 121
 122         sra->devs = NULL;
 123         if (options & GET_VERSION) {
 124                 strcpy(base, "metadata_version");
 125                 if (load_sys(fname, buf, sizeof(buf)))
 126                         goto abort;
 127                 if (strncmp(buf, "none", 4) == 0) {
 128                         sra->array.major_version =
 129                                 sra->array.minor_version = -1;
 130                         strcpy(sra->text_version, "");
 131                 } else if (strncmp(buf, "external:", 9) == 0) {
 132                         sra->array.major_version = -1;
 133                         sra->array.minor_version = -2;
 134                         strcpy(sra->text_version, buf+9);
 135                 } else {
 136                         sscanf(buf, "%d.%d",
 137                                &sra->array.major_version,
 138                                &sra->array.minor_version);
 139                         strcpy(sra->text_version, buf);
 140                 }
 141         }
 142         if (options & GET_LEVEL) {
 143                 strcpy(base, "level");
 144                 if (load_sys(fname, buf, sizeof(buf)))
 145                         goto abort;
 146                 sra->array.level = map_name(pers, buf);
 147         }
 148         if (options & GET_LAYOUT) {
 149                 strcpy(base, "layout");
 150                 if (load_sys(fname, buf, sizeof(buf)))
 151                         goto abort;
 152                 sra->array.layout = strtoul(buf, NULL, 0);
 153         }
 154         if (options & GET_DISKS) {
 155                 strcpy(base, "raid_disks");
 156                 if (load_sys(fname, buf, sizeof(buf)))
 157                         goto abort;
 158                 sra->array.raid_disks = strtoul(buf, NULL, 0);
 159         }
 160         if (options & GET_DEGRADED) {
 161                 strcpy(base, "degraded");
 162                 if (load_sys(fname, buf, sizeof(buf)))
 163                         goto abort;
 164                 sra->array.failed_disks = strtoul(buf, NULL, 0);
 165         }
 166         if (options & GET_COMPONENT) {
 167                 strcpy(base, "component_size");
 168                 if (load_sys(fname, buf, sizeof(buf)))
 169                         goto abort;
 170                 sra->component_size = strtoull(buf, NULL, 0);
 171                 /* sysfs reports "K", but we want sectors */
 172                 sra->component_size *= 2;
 173         }
 174         if (options & GET_CHUNK) {
 175                 strcpy(base, "chunk_size");
 176                 if (load_sys(fname, buf, sizeof(buf)))
 177                         goto abort;
 178                 sra->array.chunk_size = strtoul(buf, NULL, 0);
 179         }
 180         if (options & GET_CACHE) {
 181                 strcpy(base, "stripe_cache_size");
 182                 if (load_sys(fname, buf, sizeof(buf)))
 183                         /* Probably level doesn't support it */
 184                         sra->cache_size = 0;
 185                 else
 186                         sra->cache_size = strtoul(buf, NULL, 0);
 187         }
 188         if (options & GET_MISMATCH) {
 189                 strcpy(base, "mismatch_cnt");
 190                 if (load_sys(fname, buf, sizeof(buf)))
 191                         goto abort;
 192                 sra->mismatch_cnt = strtoul(buf, NULL, 0);
 193         }
 194         if (options & GET_SAFEMODE) {
 195                 int scale = 1;
 196                 int dot = 0;
 197                 unsigned i;
 198                 unsigned long msec;
 199                 size_t len;
 200
 201                 strcpy(base, "safe_mode_delay");
 202                 if (load_sys(fname, buf, sizeof(buf)))
 203                         goto abort;
 204
 205                 /* remove a period, and count digits after it */
 206                 len = strlen(buf);
 207                 for (i = 0; i < len; i++) {
 208                         if (dot) {
 209                                 if (isdigit(buf[i])) {
 210                                         buf[i-1] = buf[i];
 211                                         scale *= 10;
 212                                 }
 213                                 buf[i] = 0;
 214                         } else if (buf[i] == '.') {
 215                                 dot=1;
 216                                 buf[i] = 0;
 217                         }
 218                 }
 219                 msec = strtoul(buf, NULL, 10);
 220                 msec = (msec * 1000) / scale;
 221                 sra->safe_mode_delay = msec;
 222         }
 223         if (options & GET_BITMAP_LOCATION) {
 224                 strcpy(base, "bitmap/location");
 225                 if (load_sys(fname, buf, sizeof(buf)))
 226                         goto abort;
 227                 if (strncmp(buf, "file", 4) == 0)
 228                         sra->bitmap_offset = 1;
 229                 else if (strncmp(buf, "none", 4) == 0)
 230                         sra->bitmap_offset = 0;
 231                 else if (buf[0] == '+')
 232                         sra->bitmap_offset = strtol(buf+1, NULL, 10);
 233                 else
 234                         goto abort;
 235         }
 236
 237         if (options & GET_ARRAY_STATE) {
 238                 strcpy(base, "array_state");
 239                 if (load_sys(fname, sra->sysfs_array_state,
 240                              sizeof(sra->sysfs_array_state)))
 241                         goto abort;
 242         } else
 243                 sra->sysfs_array_state[0] = 0;
 244
 245         if (! (options & GET_DEVS))
 246                 return sra;
 247
 248         /* Get all the devices as well */
 249         *base = 0;
 250         dir = opendir(fname);
 251         if (!dir)
 252                 goto abort;
 253         sra->array.spare_disks = 0;
 254
 255         devp = &sra->devs;
 256         sra->devs = NULL;
 257         while ((de = readdir(dir)) != NULL) {
 258                 char *ep;
 259                 if (de->d_ino == 0 ||
 260                     strncmp(de->d_name, "dev-", 4) != 0)
 261                         continue;
 262                 strcpy(base, de->d_name);
 263                 dbase = base + strlen(base);
 264                 *dbase++ = '/';
 265
 266                 dev = xcalloc(1, sizeof(*dev));
 267
 268                 /* Always get slot, major, minor */
 269                 strcpy(dbase, "slot");
 270                 if (load_sys(fname, buf, sizeof(buf))) {
 271                         /* hmm... unable to read 'slot' maybe the device
 272                          * is going away?
 273                          */
 274                         strcpy(dbase, "block");
 275                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 276                             errno != ENAMETOOLONG) {
 277                                 /* ...yup device is gone */
 278                                 free(dev);
 279                                 continue;
 280                         } else {
 281                                 /* slot is unreadable but 'block' link
 282                                  * still intact... something bad is happening
 283                                  * so abort
 284                                  */
 285                                 free(dev);
 286                                 goto abort;
 287                         }
 288
 289                 }
 290                 strcpy(dev->sys_name, de->d_name);
 291                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 292                 if (*ep) dev->disk.raid_disk = -1;
 293
 294                 strcpy(dbase, "block/dev");
 295                 if (load_sys(fname, buf, sizeof(buf))) {
 296                         /* assume this is a stale reference to a hot
 297                          * removed device
 298                          */
 299                         free(dev);
 300                         continue;
 301                 }
 302                 sra->array.nr_disks++;
 303                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 304
 305                 /* special case check for block devices that can go 'offline' */
 306                 strcpy(dbase, "block/device/state");
 307                 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
 308                     strncmp(buf, "offline", 7) == 0) {
 309                         free(dev);
 310                         continue;
 311                 }
 312
 313                 /* finally add this disk to the array */
 314                 *devp = dev;
 315                 devp = & dev->next;
 316                 dev->next = NULL;
 317
 318                 if (options & GET_OFFSET) {
 319                         strcpy(dbase, "offset");
 320                         if (load_sys(fname, buf, sizeof(buf)))
 321                                 goto abort;
 322                         dev->data_offset = strtoull(buf, NULL, 0);
 323                         strcpy(dbase, "new_offset");
 324                         if (load_sys(fname, buf, sizeof(buf)) == 0)
 325                                 dev->new_data_offset = strtoull(buf, NULL, 0);
 326                         else
 327                                 dev->new_data_offset = dev->data_offset;
 328                 }
 329                 if (options & GET_SIZE) {
 330                         strcpy(dbase, "size");
 331                         if (load_sys(fname, buf, sizeof(buf)))
 332                                 goto abort;
 333                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 334                 }
 335                 if (options & GET_STATE) {
 336                         dev->disk.state = 0;
 337                         strcpy(dbase, "state");
 338                         if (load_sys(fname, buf, sizeof(buf)))
 339                                 goto abort;
 340                         if (strstr(buf, "in_sync"))
 341                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 342                         if (strstr(buf, "faulty"))
 343                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 344                         if (dev->disk.state == 0)
 345                                 sra->array.spare_disks++;
 346                 }
 347                 if (options & GET_ERROR) {
 348                         strcpy(buf, "errors");
 349                         if (load_sys(fname, buf, sizeof(buf)))
 350                                 goto abort;
 351                         dev->errors = strtoul(buf, NULL, 0);
 352                 }
 353         }
 354         closedir(dir);
 355         return sra;
 356
 357  abort:
 358         if (dir)
 359                 closedir(dir);
 360         sysfs_free(sra);
 361         return NULL;
 362 }
 363
 364 int sysfs_attr_match(const char *attr, const char *str)
 365 {
 366         /* See if attr, read from a sysfs file, matches
 367          * str.  They must either be the same, or attr can
 368          * have a trailing newline or comma
 369          */
 370         while (*attr && *str && *attr == *str) {
 371                 attr++;
 372                 str++;
 373         }
 374
 375         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 376                 return 0;
 377         return 1;
 378 }
 379
 380 int sysfs_match_word(const char *word, char **list)
 381 {
 382         int n;
 383         for (n=0; list[n]; n++)
 384                 if (sysfs_attr_match(word, list[n]))
 385                         break;
 386         return n;
 387 }
 388
 389 unsigned long long get_component_size(int fd)
 390 {
 391         /* Find out the component size of the array.
 392          * We cannot trust GET_ARRAY_INFO ioctl as it's
 393          * size field is only 32bits.
 394          * So look in /sys/block/mdXXX/md/component_size
 395          *
 396          * This returns in units of sectors.
 397          */
 398         struct stat stb;
 399         char fname[MAX_SYSFS_PATH_LEN];
 400         int n;
 401         if (fstat(fd, &stb))
 402                 return 0;
 403         snprintf(fname, MAX_SYSFS_PATH_LEN,
 404                  "/sys/block/%s/md/component_size", stat2devnm(&stb));
 405         fd = open(fname, O_RDONLY);
 406         if (fd < 0)
 407                 return 0;
 408         n = read(fd, fname, sizeof(fname));
 409         close(fd);
 410         if (n < 0 || n == sizeof(fname))
 411                 return 0;
 412         fname[n] = 0;
 413         return strtoull(fname, NULL, 10) * 2;
 414 }
 415
 416 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 417                   char *name, char *val)
 418 {
 419         char fname[MAX_SYSFS_PATH_LEN];
 420         unsigned int n;
 421         int fd;
 422
 423         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 424                 sra->sys_name, dev?dev->sys_name:"", name);
 425         fd = open(fname, O_WRONLY);
 426         if (fd < 0)
 427                 return -1;
 428         n = write(fd, val, strlen(val));
 429         close(fd);
 430         if (n != strlen(val)) {
 431                 dprintf("failed to write '%s' to '%s' (%s)\n",
 432                         val, fname, strerror(errno));
 433                 return -1;
 434         }
 435         return 0;
 436 }
 437
 438 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 439                   char *name, unsigned long long val)
 440 {
 441         char valstr[50];
 442         sprintf(valstr, "%llu", val);
 443         return sysfs_set_str(sra, dev, name, valstr);
 444 }
 445
 446 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
 447                          char *name, long long val)
 448 {
 449         char valstr[50];
 450         sprintf(valstr, "%lli", val);
 451         return sysfs_set_str(sra, dev, name, valstr);
 452 }
 453
 454 int sysfs_uevent(struct mdinfo *sra, char *event)
 455 {
 456         char fname[MAX_SYSFS_PATH_LEN];
 457         int n;
 458         int fd;
 459
 460         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/uevent",
 461                 sra->sys_name);
 462         fd = open(fname, O_WRONLY);
 463         if (fd < 0)
 464                 return -1;
 465         n = write(fd, event, strlen(event));
 466         close(fd);
 467         if (n != (int)strlen(event)) {
 468                 dprintf("failed to write '%s' to '%s' (%s)\n",
 469                         event, fname, strerror(errno));
 470                 return -1;
 471         }
 472         return 0;
 473 }
 474
 475 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
 476 {
 477         char fname[MAX_SYSFS_PATH_LEN];
 478         struct stat st;
 479
 480         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 481                 sra->sys_name, dev?dev->sys_name:"", name);
 482
 483         return stat(fname, &st) == 0;
 484 }
 485
 486 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 487                        char *name)
 488 {
 489         char fname[MAX_SYSFS_PATH_LEN];
 490         int fd;
 491
 492         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 493                 sra->sys_name, dev?dev->sys_name:"", name);
 494         fd = open(fname, O_RDWR);
 495         if (fd < 0)
 496                 fd = open(fname, O_RDONLY);
 497         return fd;
 498 }
 499
 500 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 501 {
 502         char buf[50];
 503         int n;
 504         char *ep;
 505
 506         lseek(fd, 0, 0);
 507         n = read(fd, buf, sizeof(buf));
 508         if (n <= 0 || n == sizeof(buf))
 509                 return -2;
 510         buf[n] = 0;
 511         *val = strtoull(buf, &ep, 0);
 512         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 513                 return -1;
 514         return 0;
 515 }
 516
 517 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 518                        char *name, unsigned long long *val)
 519 {
 520         int n;
 521         int fd;
 522
 523         fd = sysfs_get_fd(sra, dev, name);
 524         if (fd < 0)
 525                 return -1;
 526         n = sysfs_fd_get_ll(fd, val);
 527         close(fd);
 528         return n;
 529 }
 530
 531 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
 532 {
 533         /* two numbers in this sysfs file, either
 534          *  NNN (NNN)
 535          * or
 536          *  NNN / NNN
 537          */
 538         char buf[80];
 539         int n;
 540         char *ep, *ep2;
 541
 542         lseek(fd, 0, 0);
 543         n = read(fd, buf, sizeof(buf));
 544         if (n <= 0 || n == sizeof(buf))
 545                 return -2;
 546         buf[n] = 0;
 547         *v1 = strtoull(buf, &ep, 0);
 548         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 549                 return -1;
 550         while (*ep == ' ' || *ep == '/' || *ep == '(')
 551                 ep++;
 552         *v2 = strtoull(ep, &ep2, 0);
 553         if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
 554                 *v2 = *v1;
 555                 return 1;
 556         }
 557         return 2;
 558 }
 559
 560 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
 561                   char *name, unsigned long long *v1, unsigned long long *v2)
 562 {
 563         int n;
 564         int fd;
 565
 566         fd = sysfs_get_fd(sra, dev, name);
 567         if (fd < 0)
 568                 return -1;
 569         n = sysfs_fd_get_two(fd, v1, v2);
 570         close(fd);
 571         return n;
 572 }
 573
 574 int sysfs_fd_get_str(int fd, char *val, int size)
 575 {
 576         int n;
 577
 578         lseek(fd, 0, 0);
 579         n = read(fd, val, size);
 580         if (n <= 0 || n == size)
 581                 return -1;
 582         val[n] = 0;
 583         return n;
 584 }
 585
 586 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 587                        char *name, char *val, int size)
 588 {
 589         int n;
 590         int fd;
 591
 592         fd = sysfs_get_fd(sra, dev, name);
 593         if (fd < 0)
 594                 return -1;
 595         n = sysfs_fd_get_str(fd, val, size);
 596         close(fd);
 597         return n;
 598 }
 599
 600 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 601 {
 602         unsigned long sec;
 603         unsigned long msec;
 604         char delay[30];
 605
 606         sec = ms / 1000;
 607         msec = ms % 1000;
 608
 609         sprintf(delay, "%ld.%03ld\n", sec, msec);
 610         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 611         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 612 }
 613
 614 int sysfs_set_array(struct mdinfo *info, int vers)
 615 {
 616         int rv = 0;
 617         char ver[100];
 618         int raid_disks = info->array.raid_disks;
 619
 620         ver[0] = 0;
 621         if (info->array.major_version == -1 &&
 622             info->array.minor_version == -2) {
 623                 char buf[1024];
 624
 625                 strcat(strcpy(ver, "external:"), info->text_version);
 626
 627                 /* meta version might already be set if we are setting
 628                  * new geometry for a reshape.  In that case we don't
 629                  * want to over-write the 'readonly' flag that is
 630                  * stored in the metadata version.  So read the current
 631                  * version first, and preserve the flag
 632                  */
 633                 if (sysfs_get_str(info, NULL, "metadata_version",
 634                                   buf, 1024) > 0)
 635                         if (strlen(buf) >= 9 && buf[9] == '-')
 636                                 ver[9] = '-';
 637
 638                 if ((vers % 100) < 2 ||
 639                     sysfs_set_str(info, NULL, "metadata_version",
 640                                   ver) < 0) {
 641                         pr_err("This kernel does not support external metadata.\n");
 642                         return 1;
 643                 }
 644         }
 645         if (info->array.level < 0)
 646                 return 0; /* FIXME */
 647         rv |= sysfs_set_str(info, NULL, "level",
 648                             map_num(pers, info->array.level));
 649         if (info->reshape_active && info->delta_disks != UnSet)
 650                 raid_disks -= info->delta_disks;
 651         rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
 652         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 653         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 654         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 655         if (info->custom_array_size) {
 656                 int rc;
 657
 658                 rc = sysfs_set_num(info, NULL, "array_size",
 659                                    info->custom_array_size/2);
 660                 if (rc && errno == ENOENT) {
 661                         pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
 662                         rc = 0;
 663                 }
 664                 rv |= rc;
 665         }
 666
 667         if (info->array.level > 0)
 668                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 669
 670         if (info->reshape_active) {
 671                 rv |= sysfs_set_num(info, NULL, "reshape_position",
 672                                     info->reshape_progress);
 673                 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
 674                 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
 675                 rv |= sysfs_set_num(info, NULL, "raid_disks",
 676                                     info->array.raid_disks);
 677                 /* We don't set 'new_level' here.  That can only happen
 678                  * once the reshape completes.
 679                  */
 680         }
 681         return rv;
 682 }
 683
 684 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 685 {
 686         char dv[PATH_MAX];
 687         char nm[PATH_MAX];
 688         char *dname;
 689         int rv;
 690         int i;
 691
 692         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 693         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 694         if (rv)
 695                 return rv;
 696
 697         memset(nm, 0, sizeof(nm));
 698         dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
 699         strcpy(sd->sys_name, "dev-");
 700         strcpy(sd->sys_name+4, dname);
 701
 702         /* test write to see if 'recovery_start' is available */
 703         if (resume && sd->recovery_start < MaxSector &&
 704             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 705                 sysfs_set_str(sra, sd, "state", "remove");
 706                 return -1;
 707         }
 708
 709         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 710         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 711         if (sra->array.level != LEVEL_CONTAINER) {
 712                 if (sd->recovery_start == MaxSector)
 713                         /* This can correctly fail if array isn't started,
 714                          * yet, so just ignore status for now.
 715                          */
 716                         sysfs_set_str(sra, sd, "state", "insync");
 717                 if (sd->disk.raid_disk >= 0)
 718                         rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 719                 if (resume)
 720                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 721         }
 722         if (sd->bb.supported) {
 723                 if (sysfs_set_str(sra, sd, "state", "external_bbl")) {
 724                         /*
 725                          * backward compatibility - if kernel doesn't support
 726                          * bad blocks for external metadata, let it continue
 727                          * as long as there are none known so far
 728                          */
 729                         if (sd->bb.count) {
 730                                 pr_err("The kernel has no support for bad blocks in external metadata\n");
 731                                 return -1;
 732                         }
 733                 }
 734
 735                 for (i = 0; i < sd->bb.count; i++) {
 736                         char s[30];
 737                         const struct md_bb_entry *entry = &sd->bb.entries[i];
 738
 739                         snprintf(s, sizeof(s) - 1, "%llu %d\n", entry->sector,
 740                                  entry->length);
 741                         rv |= sysfs_set_str(sra, sd, "bad_blocks", s);
 742                 }
 743         }
 744         return rv;
 745 }
 746
 747 #if 0
 748 int sysfs_disk_to_sg(int fd)
 749 {
 750         /* from an open block device, try find and open its corresponding
 751          * scsi_generic interface
 752          */
 753         struct stat st;
 754         char path[256];
 755         char sg_path[256];
 756         char sg_major_minor[10];
 757         char *c;
 758         DIR *dir;
 759         struct dirent *de;
 760         int major, minor, rv;
 761
 762         if (fstat(fd, &st))
 763                 return -1;
 764
 765         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 766                  major(st.st_rdev), minor(st.st_rdev));
 767
 768         dir = opendir(path);
 769         if (!dir)
 770                 return -1;
 771
 772         de = readdir(dir);
 773         while (de) {
 774                 if (strncmp("scsi_generic:", de->d_name,
 775                             strlen("scsi_generic:")) == 0)
 776                         break;
 777                 de = readdir(dir);
 778         }
 779         closedir(dir);
 780
 781         if (!de)
 782                 return -1;
 783
 784         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 785         fd = open(sg_path, O_RDONLY);
 786         if (fd < 0)
 787                 return fd;
 788
 789         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 790         close(fd);
 791         if (rv < 0 || rv == sizeof(sg_major_minor))
 792                 return -1;
 793         else
 794                 sg_major_minor[rv - 1] = '\0';
 795
 796         c = strchr(sg_major_minor, ':');
 797         *c = '\0';
 798         c++;
 799         major = strtol(sg_major_minor, NULL, 10);
 800         minor = strtol(c, NULL, 10);
 801         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 802                  (int) getpid(), major, minor);
 803         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 804                         fd = open(path, O_RDONLY);
 805                         unlink(path);
 806                         return fd;
 807         }
 808
 809         return -1;
 810 }
 811 #endif
 812
 813 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 814 {
 815         /* from an open block device, try to retrieve it scsi_id */
 816         struct stat st;
 817         char path[256];
 818         DIR *dir;
 819         struct dirent *de;
 820         int host, bus, target, lun;
 821
 822         if (fstat(fd, &st))
 823                 return 1;
 824
 825         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
 826                  major(st.st_rdev), minor(st.st_rdev));
 827
 828         dir = opendir(path);
 829         if (!dir)
 830                 return 1;
 831
 832         for (de = readdir(dir); de; de = readdir(dir)) {
 833                 int count;
 834
 835                 if (de->d_type != DT_DIR)
 836                         continue;
 837
 838                 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
 839                 if (count == 4)
 840                         break;
 841         }
 842         closedir(dir);
 843
 844         if (!de)
 845                 return 1;
 846
 847         *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
 848         return 0;
 849 }
 850
 851 int sysfs_unique_holder(char *devnm, long rdev)
 852 {
 853         /* Check that devnm is a holder of rdev,
 854          * and is the only holder.
 855          * we should be locked against races by
 856          * an O_EXCL on devnm
 857          * Return values:
 858          *  0 - not unique, not even a holder
 859          *  1 - unique, this is the only holder.
 860          *  2/3 - not unique, there is another holder
 861          * -1 - error, cannot find the holders
 862          */
 863         DIR *dir;
 864         struct dirent *de;
 865         char dirname[100];
 866         char l;
 867         int ret = 0;
 868         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 869                 major(rdev), minor(rdev));
 870         dir = opendir(dirname);
 871         if (!dir)
 872                 return -1;
 873         l = strlen(dirname);
 874         while ((de = readdir(dir)) != NULL) {
 875                 char buf[100];
 876                 char *sl;
 877                 int n;
 878
 879                 if (de->d_ino == 0)
 880                         continue;
 881                 if (de->d_name[0] == '.')
 882                         continue;
 883                 strcpy(dirname+l, "/");
 884                 strcat(dirname+l, de->d_name);
 885                 n = readlink(dirname, buf, sizeof(buf)-1);
 886                 if (n <= 0)
 887                         continue;
 888                 buf[n] = 0;
 889                 sl = strrchr(buf, '/');
 890                 if (!sl)
 891                         continue;
 892                 sl++;
 893
 894                 if (strcmp(devnm, sl) == 0)
 895                         ret |= 1;
 896                 else
 897                         ret |= 2;
 898         }
 899         closedir(dir);
 900         return ret;
 901 }
 902
 903 int sysfs_freeze_array(struct mdinfo *sra)
 904 {
 905         /* Try to freeze resync/rebuild on this array/container.
 906          * Return -1 if the array is busy,
 907          * return 0 if this kernel doesn't support 'frozen'
 908          * return 1 if it worked.
 909          */
 910         char buf[20];
 911
 912         if (!sysfs_attribute_available(sra, NULL, "sync_action"))
 913                 return 1; /* no sync_action == frozen */
 914         if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
 915                 return 0;
 916         if (strcmp(buf, "frozen\n") == 0)
 917                 /* Already frozen */
 918                 return 0;
 919         if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
 920                 return -1;
 921         if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
 922                 return 0;
 923         return 1;
 924 }
 925
 926 int sysfs_wait(int fd, int *msec)
 927 {
 928         /* Wait up to '*msec' for fd to have an exception condition.
 929          * if msec == NULL, wait indefinitely.
 930          */
 931         fd_set fds;
 932         int n;
 933         FD_ZERO(&fds);
 934         FD_SET(fd, &fds);
 935         if (msec == NULL)
 936                 n = select(fd+1, NULL, NULL, &fds, NULL);
 937         else if (*msec < 0)
 938                 n = 0;
 939         else {
 940                 struct timeval start, end, tv;
 941                 gettimeofday(&start, NULL);
 942                 if (*msec < 1000) {
 943                         tv.tv_sec = 0;
 944                         tv.tv_usec = (*msec)*1000;
 945                 } else {
 946                         tv.tv_sec = (*msec)/1000;
 947                         tv.tv_usec = 0;
 948                 }
 949                 n = select(fd+1, NULL, NULL, &fds, &tv);
 950                 gettimeofday(&end, NULL);
 951                 end.tv_sec -= start.tv_sec;
 952                 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
 953                           - start.tv_usec/1000) + 1;
 954         }
 955         return n;
 956 }