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