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 int sysfs_init(struct mdinfo *mdi, int fd, char *devnm)
  88 {
  89         struct stat stb;
  90         char fname[MAX_SYSFS_PATH_LEN];
  91         int retval = -ENODEV;
  92
  93         mdi->sys_name[0] = 0;
  94         if (fd >= 0)
  95                 devnm = fd2devnm(fd);
  96
  97         if (devnm == NULL)
  98                 goto out;
  99
 100         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md", devnm);
 101
 102         if (stat(fname, &stb))
 103                 goto out;
 104         if (!S_ISDIR(stb.st_mode))
 105                 goto out;
 106         strcpy(mdi->sys_name, devnm);
 107
 108         retval = 0;
 109 out:
 110         return retval;
 111 }
 112
 113 struct mdinfo *sysfs_read(int fd, char *devnm, unsigned long options)
 114 {
 115         char fname[PATH_MAX];
 116         char buf[PATH_MAX];
 117         char *base;
 118         char *dbase;
 119         struct mdinfo *sra;
 120         struct mdinfo *dev, **devp;
 121         DIR *dir = NULL;
 122         struct dirent *de;
 123
 124         sra = xcalloc(1, sizeof(*sra));
 125         if (sysfs_init(sra, fd, devnm)) {
 126                 free(sra);
 127                 return NULL;
 128         }
 129
 130         sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
 131         base = fname + strlen(fname);
 132
 133         sra->devs = NULL;
 134         if (options & GET_VERSION) {
 135                 strcpy(base, "metadata_version");
 136                 if (load_sys(fname, buf, sizeof(buf)))
 137                         goto abort;
 138                 if (strncmp(buf, "none", 4) == 0) {
 139                         sra->array.major_version =
 140                                 sra->array.minor_version = -1;
 141                         strcpy(sra->text_version, "");
 142                 } else if (strncmp(buf, "external:", 9) == 0) {
 143                         sra->array.major_version = -1;
 144                         sra->array.minor_version = -2;
 145                         strcpy(sra->text_version, buf+9);
 146                 } else {
 147                         sscanf(buf, "%d.%d",
 148                                &sra->array.major_version,
 149                                &sra->array.minor_version);
 150                         strcpy(sra->text_version, buf);
 151                 }
 152         }
 153         if (options & GET_LEVEL) {
 154                 strcpy(base, "level");
 155                 if (load_sys(fname, buf, sizeof(buf)))
 156                         goto abort;
 157                 sra->array.level = map_name(pers, buf);
 158         }
 159         if (options & GET_LAYOUT) {
 160                 strcpy(base, "layout");
 161                 if (load_sys(fname, buf, sizeof(buf)))
 162                         goto abort;
 163                 sra->array.layout = strtoul(buf, NULL, 0);
 164         }
 165         if (options & GET_DISKS) {
 166                 strcpy(base, "raid_disks");
 167                 if (load_sys(fname, buf, sizeof(buf)))
 168                         goto abort;
 169                 sra->array.raid_disks = strtoul(buf, NULL, 0);
 170         }
 171         if (options & GET_DEGRADED) {
 172                 strcpy(base, "degraded");
 173                 if (load_sys(fname, buf, sizeof(buf)))
 174                         goto abort;
 175                 sra->array.failed_disks = strtoul(buf, NULL, 0);
 176         }
 177         if (options & GET_COMPONENT) {
 178                 strcpy(base, "component_size");
 179                 if (load_sys(fname, buf, sizeof(buf)))
 180                         goto abort;
 181                 sra->component_size = strtoull(buf, NULL, 0);
 182                 /* sysfs reports "K", but we want sectors */
 183                 sra->component_size *= 2;
 184         }
 185         if (options & GET_CHUNK) {
 186                 strcpy(base, "chunk_size");
 187                 if (load_sys(fname, buf, sizeof(buf)))
 188                         goto abort;
 189                 sra->array.chunk_size = strtoul(buf, NULL, 0);
 190         }
 191         if (options & GET_CACHE) {
 192                 strcpy(base, "stripe_cache_size");
 193                 if (load_sys(fname, buf, sizeof(buf)))
 194                         /* Probably level doesn't support it */
 195                         sra->cache_size = 0;
 196                 else
 197                         sra->cache_size = strtoul(buf, NULL, 0);
 198         }
 199         if (options & GET_MISMATCH) {
 200                 strcpy(base, "mismatch_cnt");
 201                 if (load_sys(fname, buf, sizeof(buf)))
 202                         goto abort;
 203                 sra->mismatch_cnt = strtoul(buf, NULL, 0);
 204         }
 205         if (options & GET_SAFEMODE) {
 206                 int scale = 1;
 207                 int dot = 0;
 208                 unsigned i;
 209                 unsigned long msec;
 210                 size_t len;
 211
 212                 strcpy(base, "safe_mode_delay");
 213                 if (load_sys(fname, buf, sizeof(buf)))
 214                         goto abort;
 215
 216                 /* remove a period, and count digits after it */
 217                 len = strlen(buf);
 218                 for (i = 0; i < len; i++) {
 219                         if (dot) {
 220                                 if (isdigit(buf[i])) {
 221                                         buf[i-1] = buf[i];
 222                                         scale *= 10;
 223                                 }
 224                                 buf[i] = 0;
 225                         } else if (buf[i] == '.') {
 226                                 dot=1;
 227                                 buf[i] = 0;
 228                         }
 229                 }
 230                 msec = strtoul(buf, NULL, 10);
 231                 msec = (msec * 1000) / scale;
 232                 sra->safe_mode_delay = msec;
 233         }
 234         if (options & GET_BITMAP_LOCATION) {
 235                 strcpy(base, "bitmap/location");
 236                 if (load_sys(fname, buf, sizeof(buf)))
 237                         goto abort;
 238                 if (strncmp(buf, "file", 4) == 0)
 239                         sra->bitmap_offset = 1;
 240                 else if (strncmp(buf, "none", 4) == 0)
 241                         sra->bitmap_offset = 0;
 242                 else if (buf[0] == '+')
 243                         sra->bitmap_offset = strtol(buf+1, NULL, 10);
 244                 else
 245                         goto abort;
 246         }
 247
 248         if (options & GET_ARRAY_STATE) {
 249                 strcpy(base, "array_state");
 250                 if (load_sys(fname, buf, sizeof(buf)))
 251                         goto abort;
 252                 sra->array_state = map_name(sysfs_array_states, buf);
 253         }
 254
 255         if (options & GET_CONSISTENCY_POLICY) {
 256                 strcpy(base, "consistency_policy");
 257                 if (load_sys(fname, buf, sizeof(buf)))
 258                         sra->consistency_policy = CONSISTENCY_POLICY_UNKNOWN;
 259                 else
 260                         sra->consistency_policy = map_name(consistency_policies,
 261                                                            buf);
 262         }
 263
 264         if (! (options & GET_DEVS))
 265                 return sra;
 266
 267         /* Get all the devices as well */
 268         *base = 0;
 269         dir = opendir(fname);
 270         if (!dir)
 271                 goto abort;
 272         sra->array.spare_disks = 0;
 273
 274         devp = &sra->devs;
 275         sra->devs = NULL;
 276         while ((de = readdir(dir)) != NULL) {
 277                 char *ep;
 278                 if (de->d_ino == 0 ||
 279                     strncmp(de->d_name, "dev-", 4) != 0)
 280                         continue;
 281                 strcpy(base, de->d_name);
 282                 dbase = base + strlen(base);
 283                 *dbase++ = '/';
 284
 285                 dev = xcalloc(1, sizeof(*dev));
 286
 287                 /* Always get slot, major, minor */
 288                 strcpy(dbase, "slot");
 289                 if (load_sys(fname, buf, sizeof(buf))) {
 290                         /* hmm... unable to read 'slot' maybe the device
 291                          * is going away?
 292                          */
 293                         strcpy(dbase, "block");
 294                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 295                             errno != ENAMETOOLONG) {
 296                                 /* ...yup device is gone */
 297                                 free(dev);
 298                                 continue;
 299                         } else {
 300                                 /* slot is unreadable but 'block' link
 301                                  * still intact... something bad is happening
 302                                  * so abort
 303                                  */
 304                                 free(dev);
 305                                 goto abort;
 306                         }
 307
 308                 }
 309                 strcpy(dev->sys_name, de->d_name);
 310                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 311                 if (*ep) dev->disk.raid_disk = -1;
 312
 313                 strcpy(dbase, "block/dev");
 314                 if (load_sys(fname, buf, sizeof(buf))) {
 315                         /* assume this is a stale reference to a hot
 316                          * removed device
 317                          */
 318                         free(dev);
 319                         continue;
 320                 }
 321                 sra->array.nr_disks++;
 322                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 323
 324                 /* special case check for block devices that can go 'offline' */
 325                 strcpy(dbase, "block/device/state");
 326                 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
 327                     strncmp(buf, "offline", 7) == 0) {
 328                         free(dev);
 329                         continue;
 330                 }
 331
 332                 /* finally add this disk to the array */
 333                 *devp = dev;
 334                 devp = & dev->next;
 335                 dev->next = NULL;
 336
 337                 if (options & GET_OFFSET) {
 338                         strcpy(dbase, "offset");
 339                         if (load_sys(fname, buf, sizeof(buf)))
 340                                 goto abort;
 341                         dev->data_offset = strtoull(buf, NULL, 0);
 342                         strcpy(dbase, "new_offset");
 343                         if (load_sys(fname, buf, sizeof(buf)) == 0)
 344                                 dev->new_data_offset = strtoull(buf, NULL, 0);
 345                         else
 346                                 dev->new_data_offset = dev->data_offset;
 347                 }
 348                 if (options & GET_SIZE) {
 349                         strcpy(dbase, "size");
 350                         if (load_sys(fname, buf, sizeof(buf)))
 351                                 goto abort;
 352                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 353                 }
 354                 if (options & GET_STATE) {
 355                         dev->disk.state = 0;
 356                         strcpy(dbase, "state");
 357                         if (load_sys(fname, buf, sizeof(buf)))
 358                                 goto abort;
 359                         if (strstr(buf, "in_sync"))
 360                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 361                         if (strstr(buf, "faulty"))
 362                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 363                         if (dev->disk.state == 0)
 364                                 sra->array.spare_disks++;
 365                 }
 366                 if (options & GET_ERROR) {
 367                         strcpy(buf, "errors");
 368                         if (load_sys(fname, buf, sizeof(buf)))
 369                                 goto abort;
 370                         dev->errors = strtoul(buf, NULL, 0);
 371                 }
 372         }
 373         closedir(dir);
 374         return sra;
 375
 376  abort:
 377         if (dir)
 378                 closedir(dir);
 379         sysfs_free(sra);
 380         return NULL;
 381 }
 382
 383 int sysfs_attr_match(const char *attr, const char *str)
 384 {
 385         /* See if attr, read from a sysfs file, matches
 386          * str.  They must either be the same, or attr can
 387          * have a trailing newline or comma
 388          */
 389         while (*attr && *str && *attr == *str) {
 390                 attr++;
 391                 str++;
 392         }
 393
 394         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 395                 return 0;
 396         return 1;
 397 }
 398
 399 int sysfs_match_word(const char *word, char **list)
 400 {
 401         int n;
 402         for (n=0; list[n]; n++)
 403                 if (sysfs_attr_match(word, list[n]))
 404                         break;
 405         return n;
 406 }
 407
 408 unsigned long long get_component_size(int fd)
 409 {
 410         /* Find out the component size of the array.
 411          * We cannot trust GET_ARRAY_INFO ioctl as it's
 412          * size field is only 32bits.
 413          * So look in /sys/block/mdXXX/md/component_size
 414          *
 415          * This returns in units of sectors.
 416          */
 417         struct stat stb;
 418         char fname[MAX_SYSFS_PATH_LEN];
 419         int n;
 420         if (fstat(fd, &stb))
 421                 return 0;
 422         snprintf(fname, MAX_SYSFS_PATH_LEN,
 423                  "/sys/block/%s/md/component_size", stat2devnm(&stb));
 424         fd = open(fname, O_RDONLY);
 425         if (fd < 0)
 426                 return 0;
 427         n = read(fd, fname, sizeof(fname));
 428         close(fd);
 429         if (n < 0 || n == sizeof(fname))
 430                 return 0;
 431         fname[n] = 0;
 432         return strtoull(fname, NULL, 10) * 2;
 433 }
 434
 435 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 436                   char *name, char *val)
 437 {
 438         char fname[MAX_SYSFS_PATH_LEN];
 439         unsigned int n;
 440         int fd;
 441
 442         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 443                 sra->sys_name, dev?dev->sys_name:"", name);
 444         fd = open(fname, O_WRONLY);
 445         if (fd < 0)
 446                 return -1;
 447         n = write(fd, val, strlen(val));
 448         close(fd);
 449         if (n != strlen(val)) {
 450                 dprintf("failed to write '%s' to '%s' (%s)\n",
 451                         val, fname, strerror(errno));
 452                 return -1;
 453         }
 454         return 0;
 455 }
 456
 457 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 458                   char *name, unsigned long long val)
 459 {
 460         char valstr[50];
 461         sprintf(valstr, "%llu", val);
 462         return sysfs_set_str(sra, dev, name, valstr);
 463 }
 464
 465 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
 466                          char *name, long long val)
 467 {
 468         char valstr[50];
 469         sprintf(valstr, "%lli", val);
 470         return sysfs_set_str(sra, dev, name, valstr);
 471 }
 472
 473 int sysfs_uevent(struct mdinfo *sra, char *event)
 474 {
 475         char fname[MAX_SYSFS_PATH_LEN];
 476         int n;
 477         int fd;
 478
 479         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/uevent",
 480                 sra->sys_name);
 481         fd = open(fname, O_WRONLY);
 482         if (fd < 0)
 483                 return -1;
 484         n = write(fd, event, strlen(event));
 485         close(fd);
 486         if (n != (int)strlen(event)) {
 487                 dprintf("failed to write '%s' to '%s' (%s)\n",
 488                         event, fname, strerror(errno));
 489                 return -1;
 490         }
 491         return 0;
 492 }
 493
 494 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
 495 {
 496         char fname[MAX_SYSFS_PATH_LEN];
 497         struct stat st;
 498
 499         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 500                 sra->sys_name, dev?dev->sys_name:"", name);
 501
 502         return stat(fname, &st) == 0;
 503 }
 504
 505 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 506                        char *name)
 507 {
 508         char fname[MAX_SYSFS_PATH_LEN];
 509         int fd;
 510
 511         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 512                 sra->sys_name, dev?dev->sys_name:"", name);
 513         fd = open(fname, O_RDWR);
 514         if (fd < 0)
 515                 fd = open(fname, O_RDONLY);
 516         return fd;
 517 }
 518
 519 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 520 {
 521         char buf[50];
 522         int n;
 523         char *ep;
 524
 525         lseek(fd, 0, 0);
 526         n = read(fd, buf, sizeof(buf));
 527         if (n <= 0 || n == sizeof(buf))
 528                 return -2;
 529         buf[n] = 0;
 530         *val = strtoull(buf, &ep, 0);
 531         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 532                 return -1;
 533         return 0;
 534 }
 535
 536 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 537                        char *name, unsigned long long *val)
 538 {
 539         int n;
 540         int fd;
 541
 542         fd = sysfs_get_fd(sra, dev, name);
 543         if (fd < 0)
 544                 return -1;
 545         n = sysfs_fd_get_ll(fd, val);
 546         close(fd);
 547         return n;
 548 }
 549
 550 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
 551 {
 552         /* two numbers in this sysfs file, either
 553          *  NNN (NNN)
 554          * or
 555          *  NNN / NNN
 556          */
 557         char buf[80];
 558         int n;
 559         char *ep, *ep2;
 560
 561         lseek(fd, 0, 0);
 562         n = read(fd, buf, sizeof(buf));
 563         if (n <= 0 || n == sizeof(buf))
 564                 return -2;
 565         buf[n] = 0;
 566         *v1 = strtoull(buf, &ep, 0);
 567         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 568                 return -1;
 569         while (*ep == ' ' || *ep == '/' || *ep == '(')
 570                 ep++;
 571         *v2 = strtoull(ep, &ep2, 0);
 572         if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
 573                 *v2 = *v1;
 574                 return 1;
 575         }
 576         return 2;
 577 }
 578
 579 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
 580                   char *name, unsigned long long *v1, unsigned long long *v2)
 581 {
 582         int n;
 583         int fd;
 584
 585         fd = sysfs_get_fd(sra, dev, name);
 586         if (fd < 0)
 587                 return -1;
 588         n = sysfs_fd_get_two(fd, v1, v2);
 589         close(fd);
 590         return n;
 591 }
 592
 593 int sysfs_fd_get_str(int fd, char *val, int size)
 594 {
 595         int n;
 596
 597         lseek(fd, 0, 0);
 598         n = read(fd, val, size);
 599         if (n <= 0 || n == size)
 600                 return -1;
 601         val[n] = 0;
 602         return n;
 603 }
 604
 605 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 606                        char *name, char *val, int size)
 607 {
 608         int n;
 609         int fd;
 610
 611         fd = sysfs_get_fd(sra, dev, name);
 612         if (fd < 0)
 613                 return -1;
 614         n = sysfs_fd_get_str(fd, val, size);
 615         close(fd);
 616         return n;
 617 }
 618
 619 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 620 {
 621         unsigned long sec;
 622         unsigned long msec;
 623         char delay[30];
 624
 625         sec = ms / 1000;
 626         msec = ms % 1000;
 627
 628         sprintf(delay, "%ld.%03ld\n", sec, msec);
 629         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 630         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 631 }
 632
 633 int sysfs_set_array(struct mdinfo *info, int vers)
 634 {
 635         int rv = 0;
 636         char ver[100];
 637         int raid_disks = info->array.raid_disks;
 638
 639         ver[0] = 0;
 640         if (info->array.major_version == -1 &&
 641             info->array.minor_version == -2) {
 642                 char buf[1024];
 643
 644                 strcat(strcpy(ver, "external:"), info->text_version);
 645
 646                 /* meta version might already be set if we are setting
 647                  * new geometry for a reshape.  In that case we don't
 648                  * want to over-write the 'readonly' flag that is
 649                  * stored in the metadata version.  So read the current
 650                  * version first, and preserve the flag
 651                  */
 652                 if (sysfs_get_str(info, NULL, "metadata_version",
 653                                   buf, 1024) > 0)
 654                         if (strlen(buf) >= 9 && buf[9] == '-')
 655                                 ver[9] = '-';
 656
 657                 if ((vers % 100) < 2 ||
 658                     sysfs_set_str(info, NULL, "metadata_version",
 659                                   ver) < 0) {
 660                         pr_err("This kernel does not support external metadata.\n");
 661                         return 1;
 662                 }
 663         }
 664         if (info->array.level < 0)
 665                 return 0; /* FIXME */
 666         rv |= sysfs_set_str(info, NULL, "level",
 667                             map_num(pers, info->array.level));
 668         if (info->reshape_active && info->delta_disks != UnSet)
 669                 raid_disks -= info->delta_disks;
 670         rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
 671         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 672         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 673         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 674         if (info->custom_array_size) {
 675                 int rc;
 676
 677                 rc = sysfs_set_num(info, NULL, "array_size",
 678                                    info->custom_array_size/2);
 679                 if (rc && errno == ENOENT) {
 680                         pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
 681                         rc = 0;
 682                 }
 683                 rv |= rc;
 684         }
 685
 686         if (info->array.level > 0)
 687                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 688
 689         if (info->reshape_active) {
 690                 rv |= sysfs_set_num(info, NULL, "reshape_position",
 691                                     info->reshape_progress);
 692                 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
 693                 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
 694                 rv |= sysfs_set_num(info, NULL, "raid_disks",
 695                                     info->array.raid_disks);
 696                 /* We don't set 'new_level' here.  That can only happen
 697                  * once the reshape completes.
 698                  */
 699         }
 700
 701         if (info->consistency_policy == CONSISTENCY_POLICY_PPL) {
 702                 if (sysfs_set_str(info, NULL, "consistency_policy",
 703                                   map_num(consistency_policies,
 704                                           info->consistency_policy))) {
 705                         pr_err("This kernel does not support PPL\n");
 706                         return 1;
 707                 }
 708         }
 709
 710         return rv;
 711 }
 712
 713 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 714 {
 715         char dv[PATH_MAX];
 716         char nm[PATH_MAX];
 717         char *dname;
 718         int rv;
 719         int i;
 720
 721         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 722         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 723         if (rv)
 724                 return rv;
 725
 726         memset(nm, 0, sizeof(nm));
 727         dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
 728         strcpy(sd->sys_name, "dev-");
 729         strcpy(sd->sys_name+4, dname);
 730
 731         /* test write to see if 'recovery_start' is available */
 732         if (resume && sd->recovery_start < MaxSector &&
 733             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 734                 sysfs_set_str(sra, sd, "state", "remove");
 735                 return -1;
 736         }
 737
 738         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 739         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 740         if (sra->array.level != LEVEL_CONTAINER) {
 741                 if (sd->consistency_policy == CONSISTENCY_POLICY_PPL) {
 742                         rv |= sysfs_set_num(sra, sd, "ppl_sector", sd->ppl_sector);
 743                         rv |= sysfs_set_num(sra, sd, "ppl_size", sd->ppl_size);
 744                 }
 745                 if (sd->recovery_start == MaxSector)
 746                         /* This can correctly fail if array isn't started,
 747                          * yet, so just ignore status for now.
 748                          */
 749                         sysfs_set_str(sra, sd, "state", "insync");
 750                 if (sd->disk.raid_disk >= 0)
 751                         rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 752                 if (resume)
 753                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 754         }
 755         if (sd->bb.supported) {
 756                 if (sysfs_set_str(sra, sd, "state", "external_bbl")) {
 757                         /*
 758                          * backward compatibility - if kernel doesn't support
 759                          * bad blocks for external metadata, let it continue
 760                          * as long as there are none known so far
 761                          */
 762                         if (sd->bb.count) {
 763                                 pr_err("The kernel has no support for bad blocks in external metadata\n");
 764                                 return -1;
 765                         }
 766                 }
 767
 768                 for (i = 0; i < sd->bb.count; i++) {
 769                         char s[30];
 770                         const struct md_bb_entry *entry = &sd->bb.entries[i];
 771
 772                         snprintf(s, sizeof(s) - 1, "%llu %d\n", entry->sector,
 773                                  entry->length);
 774                         rv |= sysfs_set_str(sra, sd, "bad_blocks", s);
 775                 }
 776         }
 777         return rv;
 778 }
 779
 780 #if 0
 781 int sysfs_disk_to_sg(int fd)
 782 {
 783         /* from an open block device, try find and open its corresponding
 784          * scsi_generic interface
 785          */
 786         struct stat st;
 787         char path[256];
 788         char sg_path[256];
 789         char sg_major_minor[10];
 790         char *c;
 791         DIR *dir;
 792         struct dirent *de;
 793         int major, minor, rv;
 794
 795         if (fstat(fd, &st))
 796                 return -1;
 797
 798         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 799                  major(st.st_rdev), minor(st.st_rdev));
 800
 801         dir = opendir(path);
 802         if (!dir)
 803                 return -1;
 804
 805         de = readdir(dir);
 806         while (de) {
 807                 if (strncmp("scsi_generic:", de->d_name,
 808                             strlen("scsi_generic:")) == 0)
 809                         break;
 810                 de = readdir(dir);
 811         }
 812         closedir(dir);
 813
 814         if (!de)
 815                 return -1;
 816
 817         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 818         fd = open(sg_path, O_RDONLY);
 819         if (fd < 0)
 820                 return fd;
 821
 822         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 823         close(fd);
 824         if (rv < 0 || rv == sizeof(sg_major_minor))
 825                 return -1;
 826         else
 827                 sg_major_minor[rv - 1] = '\0';
 828
 829         c = strchr(sg_major_minor, ':');
 830         *c = '\0';
 831         c++;
 832         major = strtol(sg_major_minor, NULL, 10);
 833         minor = strtol(c, NULL, 10);
 834         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 835                  (int) getpid(), major, minor);
 836         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 837                         fd = open(path, O_RDONLY);
 838                         unlink(path);
 839                         return fd;
 840         }
 841
 842         return -1;
 843 }
 844 #endif
 845
 846 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 847 {
 848         /* from an open block device, try to retrieve it scsi_id */
 849         struct stat st;
 850         char path[256];
 851         DIR *dir;
 852         struct dirent *de;
 853         int host, bus, target, lun;
 854
 855         if (fstat(fd, &st))
 856                 return 1;
 857
 858         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
 859                  major(st.st_rdev), minor(st.st_rdev));
 860
 861         dir = opendir(path);
 862         if (!dir)
 863                 return 1;
 864
 865         for (de = readdir(dir); de; de = readdir(dir)) {
 866                 int count;
 867
 868                 if (de->d_type != DT_DIR)
 869                         continue;
 870
 871                 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
 872                 if (count == 4)
 873                         break;
 874         }
 875         closedir(dir);
 876
 877         if (!de)
 878                 return 1;
 879
 880         *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
 881         return 0;
 882 }
 883
 884 int sysfs_unique_holder(char *devnm, long rdev)
 885 {
 886         /* Check that devnm is a holder of rdev,
 887          * and is the only holder.
 888          * we should be locked against races by
 889          * an O_EXCL on devnm
 890          * Return values:
 891          *  0 - not unique, not even a holder
 892          *  1 - unique, this is the only holder.
 893          *  2/3 - not unique, there is another holder
 894          * -1 - error, cannot find the holders
 895          */
 896         DIR *dir;
 897         struct dirent *de;
 898         char dirname[100];
 899         char l;
 900         int ret = 0;
 901         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 902                 major(rdev), minor(rdev));
 903         dir = opendir(dirname);
 904         if (!dir)
 905                 return -1;
 906         l = strlen(dirname);
 907         while ((de = readdir(dir)) != NULL) {
 908                 char buf[100];
 909                 char *sl;
 910                 int n;
 911
 912                 if (de->d_ino == 0)
 913                         continue;
 914                 if (de->d_name[0] == '.')
 915                         continue;
 916                 strcpy(dirname+l, "/");
 917                 strcat(dirname+l, de->d_name);
 918                 n = readlink(dirname, buf, sizeof(buf)-1);
 919                 if (n <= 0)
 920                         continue;
 921                 buf[n] = 0;
 922                 sl = strrchr(buf, '/');
 923                 if (!sl)
 924                         continue;
 925                 sl++;
 926
 927                 if (strcmp(devnm, sl) == 0)
 928                         ret |= 1;
 929                 else
 930                         ret |= 2;
 931         }
 932         closedir(dir);
 933         return ret;
 934 }
 935
 936 int sysfs_freeze_array(struct mdinfo *sra)
 937 {
 938         /* Try to freeze resync/rebuild on this array/container.
 939          * Return -1 if the array is busy,
 940          * return 0 if this kernel doesn't support 'frozen'
 941          * return 1 if it worked.
 942          */
 943         char buf[20];
 944
 945         if (!sysfs_attribute_available(sra, NULL, "sync_action"))
 946                 return 1; /* no sync_action == frozen */
 947         if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
 948                 return 0;
 949         if (strcmp(buf, "frozen\n") == 0)
 950                 /* Already frozen */
 951                 return 0;
 952         if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
 953                 return -1;
 954         if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
 955                 return 0;
 956         return 1;
 957 }
 958
 959 int sysfs_wait(int fd, int *msec)
 960 {
 961         /* Wait up to '*msec' for fd to have an exception condition.
 962          * if msec == NULL, wait indefinitely.
 963          */
 964         fd_set fds;
 965         int n;
 966         FD_ZERO(&fds);
 967         FD_SET(fd, &fds);
 968         if (msec == NULL)
 969                 n = select(fd+1, NULL, NULL, &fds, NULL);
 970         else if (*msec < 0)
 971                 n = 0;
 972         else {
 973                 struct timeval start, end, tv;
 974                 gettimeofday(&start, NULL);
 975                 if (*msec < 1000) {
 976                         tv.tv_sec = 0;
 977                         tv.tv_usec = (*msec)*1000;
 978                 } else {
 979                         tv.tv_sec = (*msec)/1000;
 980                         tv.tv_usec = 0;
 981                 }
 982                 n = select(fd+1, NULL, NULL, &fds, &tv);
 983                 gettimeofday(&end, NULL);
 984                 end.tv_sec -= start.tv_sec;
 985                 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
 986                           - start.tv_usec/1000) + 1;
 987         }
 988         return n;
 989 }