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         if (major(stb.st_rdev) != (unsigned)get_mdp_major())
 404                 snprintf(fname, MAX_SYSFS_PATH_LEN,
 405                         "/sys/block/md%d/md/component_size",
 406                         (int)minor(stb.st_rdev));
 407         else
 408                 snprintf(fname, MAX_SYSFS_PATH_LEN,
 409                         "/sys/block/md_d%d/md/component_size",
 410                         (int)minor(stb.st_rdev)>>MdpMinorShift);
 411         fd = open(fname, O_RDONLY);
 412         if (fd < 0)
 413                 return 0;
 414         n = read(fd, fname, sizeof(fname));
 415         close(fd);
 416         if (n < 0 || n == sizeof(fname))
 417                 return 0;
 418         fname[n] = 0;
 419         return strtoull(fname, NULL, 10) * 2;
 420 }
 421
 422 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 423                   char *name, char *val)
 424 {
 425         char fname[MAX_SYSFS_PATH_LEN];
 426         unsigned int n;
 427         int fd;
 428
 429         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 430                 sra->sys_name, dev?dev->sys_name:"", name);
 431         fd = open(fname, O_WRONLY);
 432         if (fd < 0)
 433                 return -1;
 434         n = write(fd, val, strlen(val));
 435         close(fd);
 436         if (n != strlen(val)) {
 437                 dprintf("failed to write '%s' to '%s' (%s)\n",
 438                         val, fname, strerror(errno));
 439                 return -1;
 440         }
 441         return 0;
 442 }
 443
 444 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 445                   char *name, unsigned long long val)
 446 {
 447         char valstr[50];
 448         sprintf(valstr, "%llu", val);
 449         return sysfs_set_str(sra, dev, name, valstr);
 450 }
 451
 452 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
 453                          char *name, long long val)
 454 {
 455         char valstr[50];
 456         sprintf(valstr, "%lli", val);
 457         return sysfs_set_str(sra, dev, name, valstr);
 458 }
 459
 460 int sysfs_uevent(struct mdinfo *sra, char *event)
 461 {
 462         char fname[MAX_SYSFS_PATH_LEN];
 463         int n;
 464         int fd;
 465
 466         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/uevent",
 467                 sra->sys_name);
 468         fd = open(fname, O_WRONLY);
 469         if (fd < 0)
 470                 return -1;
 471         n = write(fd, event, strlen(event));
 472         close(fd);
 473         if (n != (int)strlen(event)) {
 474                 dprintf("failed to write '%s' to '%s' (%s)\n",
 475                         event, fname, strerror(errno));
 476                 return -1;
 477         }
 478         return 0;
 479 }
 480
 481 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
 482 {
 483         char fname[MAX_SYSFS_PATH_LEN];
 484         struct stat st;
 485
 486         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 487                 sra->sys_name, dev?dev->sys_name:"", name);
 488
 489         return stat(fname, &st) == 0;
 490 }
 491
 492 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 493                        char *name)
 494 {
 495         char fname[MAX_SYSFS_PATH_LEN];
 496         int fd;
 497
 498         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 499                 sra->sys_name, dev?dev->sys_name:"", name);
 500         fd = open(fname, O_RDWR);
 501         if (fd < 0)
 502                 fd = open(fname, O_RDONLY);
 503         return fd;
 504 }
 505
 506 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 507 {
 508         char buf[50];
 509         int n;
 510         char *ep;
 511
 512         lseek(fd, 0, 0);
 513         n = read(fd, buf, sizeof(buf));
 514         if (n <= 0 || n == sizeof(buf))
 515                 return -2;
 516         buf[n] = 0;
 517         *val = strtoull(buf, &ep, 0);
 518         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 519                 return -1;
 520         return 0;
 521 }
 522
 523 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 524                        char *name, unsigned long long *val)
 525 {
 526         int n;
 527         int fd;
 528
 529         fd = sysfs_get_fd(sra, dev, name);
 530         if (fd < 0)
 531                 return -1;
 532         n = sysfs_fd_get_ll(fd, val);
 533         close(fd);
 534         return n;
 535 }
 536
 537 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
 538 {
 539         /* two numbers in this sysfs file, either
 540          *  NNN (NNN)
 541          * or
 542          *  NNN / NNN
 543          */
 544         char buf[80];
 545         int n;
 546         char *ep, *ep2;
 547
 548         lseek(fd, 0, 0);
 549         n = read(fd, buf, sizeof(buf));
 550         if (n <= 0 || n == sizeof(buf))
 551                 return -2;
 552         buf[n] = 0;
 553         *v1 = strtoull(buf, &ep, 0);
 554         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 555                 return -1;
 556         while (*ep == ' ' || *ep == '/' || *ep == '(')
 557                 ep++;
 558         *v2 = strtoull(ep, &ep2, 0);
 559         if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
 560                 *v2 = *v1;
 561                 return 1;
 562         }
 563         return 2;
 564 }
 565
 566 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
 567                   char *name, unsigned long long *v1, unsigned long long *v2)
 568 {
 569         int n;
 570         int fd;
 571
 572         fd = sysfs_get_fd(sra, dev, name);
 573         if (fd < 0)
 574                 return -1;
 575         n = sysfs_fd_get_two(fd, v1, v2);
 576         close(fd);
 577         return n;
 578 }
 579
 580 int sysfs_fd_get_str(int fd, char *val, int size)
 581 {
 582         int n;
 583
 584         lseek(fd, 0, 0);
 585         n = read(fd, val, size);
 586         if (n <= 0 || n == size)
 587                 return -1;
 588         val[n] = 0;
 589         return n;
 590 }
 591
 592 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 593                        char *name, char *val, int size)
 594 {
 595         int n;
 596         int fd;
 597
 598         fd = sysfs_get_fd(sra, dev, name);
 599         if (fd < 0)
 600                 return -1;
 601         n = sysfs_fd_get_str(fd, val, size);
 602         close(fd);
 603         return n;
 604 }
 605
 606 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 607 {
 608         unsigned long sec;
 609         unsigned long msec;
 610         char delay[30];
 611
 612         sec = ms / 1000;
 613         msec = ms % 1000;
 614
 615         sprintf(delay, "%ld.%03ld\n", sec, msec);
 616         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 617         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 618 }
 619
 620 int sysfs_set_array(struct mdinfo *info, int vers)
 621 {
 622         int rv = 0;
 623         char ver[100];
 624         int raid_disks = info->array.raid_disks;
 625
 626         ver[0] = 0;
 627         if (info->array.major_version == -1 &&
 628             info->array.minor_version == -2) {
 629                 char buf[1024];
 630
 631                 strcat(strcpy(ver, "external:"), info->text_version);
 632
 633                 /* meta version might already be set if we are setting
 634                  * new geometry for a reshape.  In that case we don't
 635                  * want to over-write the 'readonly' flag that is
 636                  * stored in the metadata version.  So read the current
 637                  * version first, and preserve the flag
 638                  */
 639                 if (sysfs_get_str(info, NULL, "metadata_version",
 640                                   buf, 1024) > 0)
 641                         if (strlen(buf) >= 9 && buf[9] == '-')
 642                                 ver[9] = '-';
 643
 644                 if ((vers % 100) < 2 ||
 645                     sysfs_set_str(info, NULL, "metadata_version",
 646                                   ver) < 0) {
 647                         pr_err("This kernel does not support external metadata.\n");
 648                         return 1;
 649                 }
 650         }
 651         if (info->array.level < 0)
 652                 return 0; /* FIXME */
 653         rv |= sysfs_set_str(info, NULL, "level",
 654                             map_num(pers, info->array.level));
 655         if (info->reshape_active && info->delta_disks != UnSet)
 656                 raid_disks -= info->delta_disks;
 657         rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
 658         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 659         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 660         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 661         if (info->custom_array_size) {
 662                 int rc;
 663
 664                 rc = sysfs_set_num(info, NULL, "array_size",
 665                                    info->custom_array_size/2);
 666                 if (rc && errno == ENOENT) {
 667                         pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
 668                         rc = 0;
 669                 }
 670                 rv |= rc;
 671         }
 672
 673         if (info->array.level > 0)
 674                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 675
 676         if (info->reshape_active) {
 677                 rv |= sysfs_set_num(info, NULL, "reshape_position",
 678                                     info->reshape_progress);
 679                 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
 680                 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
 681                 rv |= sysfs_set_num(info, NULL, "raid_disks",
 682                                     info->array.raid_disks);
 683                 /* We don't set 'new_level' here.  That can only happen
 684                  * once the reshape completes.
 685                  */
 686         }
 687         return rv;
 688 }
 689
 690 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 691 {
 692         char dv[PATH_MAX];
 693         char nm[PATH_MAX];
 694         char *dname;
 695         int rv;
 696         int i;
 697
 698         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 699         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 700         if (rv)
 701                 return rv;
 702
 703         memset(nm, 0, sizeof(nm));
 704         dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
 705         strcpy(sd->sys_name, "dev-");
 706         strcpy(sd->sys_name+4, dname);
 707
 708         /* test write to see if 'recovery_start' is available */
 709         if (resume && sd->recovery_start < MaxSector &&
 710             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 711                 sysfs_set_str(sra, sd, "state", "remove");
 712                 return -1;
 713         }
 714
 715         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 716         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 717         if (sra->array.level != LEVEL_CONTAINER) {
 718                 if (sd->recovery_start == MaxSector)
 719                         /* This can correctly fail if array isn't started,
 720                          * yet, so just ignore status for now.
 721                          */
 722                         sysfs_set_str(sra, sd, "state", "insync");
 723                 if (sd->disk.raid_disk >= 0)
 724                         rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 725                 if (resume)
 726                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 727         }
 728         if (sd->bb.supported) {
 729                 if (sysfs_set_str(sra, sd, "state", "external_bbl")) {
 730                         /*
 731                          * backward compatibility - if kernel doesn't support
 732                          * bad blocks for external metadata, let it continue
 733                          * as long as there are none known so far
 734                          */
 735                         if (sd->bb.count) {
 736                                 pr_err("The kernel has no support for bad blocks in external metadata\n");
 737                                 return -1;
 738                         }
 739                 }
 740
 741                 for (i = 0; i < sd->bb.count; i++) {
 742                         char s[30];
 743                         const struct md_bb_entry *entry = &sd->bb.entries[i];
 744
 745                         snprintf(s, sizeof(s) - 1, "%llu %d\n", entry->sector,
 746                                  entry->length);
 747                         rv |= sysfs_set_str(sra, sd, "bad_blocks", s);
 748                 }
 749         }
 750         return rv;
 751 }
 752
 753 #if 0
 754 int sysfs_disk_to_sg(int fd)
 755 {
 756         /* from an open block device, try find and open its corresponding
 757          * scsi_generic interface
 758          */
 759         struct stat st;
 760         char path[256];
 761         char sg_path[256];
 762         char sg_major_minor[10];
 763         char *c;
 764         DIR *dir;
 765         struct dirent *de;
 766         int major, minor, rv;
 767
 768         if (fstat(fd, &st))
 769                 return -1;
 770
 771         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 772                  major(st.st_rdev), minor(st.st_rdev));
 773
 774         dir = opendir(path);
 775         if (!dir)
 776                 return -1;
 777
 778         de = readdir(dir);
 779         while (de) {
 780                 if (strncmp("scsi_generic:", de->d_name,
 781                             strlen("scsi_generic:")) == 0)
 782                         break;
 783                 de = readdir(dir);
 784         }
 785         closedir(dir);
 786
 787         if (!de)
 788                 return -1;
 789
 790         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 791         fd = open(sg_path, O_RDONLY);
 792         if (fd < 0)
 793                 return fd;
 794
 795         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 796         close(fd);
 797         if (rv < 0 || rv == sizeof(sg_major_minor))
 798                 return -1;
 799         else
 800                 sg_major_minor[rv - 1] = '\0';
 801
 802         c = strchr(sg_major_minor, ':');
 803         *c = '\0';
 804         c++;
 805         major = strtol(sg_major_minor, NULL, 10);
 806         minor = strtol(c, NULL, 10);
 807         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 808                  (int) getpid(), major, minor);
 809         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 810                         fd = open(path, O_RDONLY);
 811                         unlink(path);
 812                         return fd;
 813         }
 814
 815         return -1;
 816 }
 817 #endif
 818
 819 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 820 {
 821         /* from an open block device, try to retrieve it scsi_id */
 822         struct stat st;
 823         char path[256];
 824         DIR *dir;
 825         struct dirent *de;
 826         int host, bus, target, lun;
 827
 828         if (fstat(fd, &st))
 829                 return 1;
 830
 831         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
 832                  major(st.st_rdev), minor(st.st_rdev));
 833
 834         dir = opendir(path);
 835         if (!dir)
 836                 return 1;
 837
 838         for (de = readdir(dir); de; de = readdir(dir)) {
 839                 int count;
 840
 841                 if (de->d_type != DT_DIR)
 842                         continue;
 843
 844                 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
 845                 if (count == 4)
 846                         break;
 847         }
 848         closedir(dir);
 849
 850         if (!de)
 851                 return 1;
 852
 853         *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
 854         return 0;
 855 }
 856
 857 int sysfs_unique_holder(char *devnm, long rdev)
 858 {
 859         /* Check that devnm is a holder of rdev,
 860          * and is the only holder.
 861          * we should be locked against races by
 862          * an O_EXCL on devnm
 863          * Return values:
 864          *  0 - not unique, not even a holder
 865          *  1 - unique, this is the only holder.
 866          *  2/3 - not unique, there is another holder
 867          * -1 - error, cannot find the holders
 868          */
 869         DIR *dir;
 870         struct dirent *de;
 871         char dirname[100];
 872         char l;
 873         int ret = 0;
 874         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 875                 major(rdev), minor(rdev));
 876         dir = opendir(dirname);
 877         if (!dir)
 878                 return -1;
 879         l = strlen(dirname);
 880         while ((de = readdir(dir)) != NULL) {
 881                 char buf[100];
 882                 char *sl;
 883                 int n;
 884
 885                 if (de->d_ino == 0)
 886                         continue;
 887                 if (de->d_name[0] == '.')
 888                         continue;
 889                 strcpy(dirname+l, "/");
 890                 strcat(dirname+l, de->d_name);
 891                 n = readlink(dirname, buf, sizeof(buf)-1);
 892                 if (n <= 0)
 893                         continue;
 894                 buf[n] = 0;
 895                 sl = strrchr(buf, '/');
 896                 if (!sl)
 897                         continue;
 898                 sl++;
 899
 900                 if (strcmp(devnm, sl) == 0)
 901                         ret |= 1;
 902                 else
 903                         ret |= 2;
 904         }
 905         closedir(dir);
 906         return ret;
 907 }
 908
 909 int sysfs_freeze_array(struct mdinfo *sra)
 910 {
 911         /* Try to freeze resync/rebuild on this array/container.
 912          * Return -1 if the array is busy,
 913          * return 0 if this kernel doesn't support 'frozen'
 914          * return 1 if it worked.
 915          */
 916         char buf[20];
 917
 918         if (!sysfs_attribute_available(sra, NULL, "sync_action"))
 919                 return 1; /* no sync_action == frozen */
 920         if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
 921                 return 0;
 922         if (strcmp(buf, "frozen\n") == 0)
 923                 /* Already frozen */
 924                 return 0;
 925         if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
 926                 return -1;
 927         if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
 928                 return 0;
 929         return 1;
 930 }
 931
 932 int sysfs_wait(int fd, int *msec)
 933 {
 934         /* Wait up to '*msec' for fd to have an exception condition.
 935          * if msec == NULL, wait indefinitely.
 936          */
 937         fd_set fds;
 938         int n;
 939         FD_ZERO(&fds);
 940         FD_SET(fd, &fds);
 941         if (msec == NULL)
 942                 n = select(fd+1, NULL, NULL, &fds, NULL);
 943         else if (*msec < 0)
 944                 n = 0;
 945         else {
 946                 struct timeval start, end, tv;
 947                 gettimeofday(&start, NULL);
 948                 if (*msec < 1000) {
 949                         tv.tv_sec = 0;
 950                         tv.tv_usec = (*msec)*1000;
 951                 } else {
 952                         tv.tv_sec = (*msec)/1000;
 953                         tv.tv_usec = 0;
 954                 }
 955                 n = select(fd+1, NULL, NULL, &fds, &tv);
 956                 gettimeofday(&end, NULL);
 957                 end.tv_sec -= start.tv_sec;
 958                 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
 959                           - start.tv_usec/1000) + 1;
 960         }
 961         return n;
 962 }