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, buf);
 261                         if (sra->consistency_policy == UnSet)
 262                                 sra->consistency_policy = CONSISTENCY_POLICY_UNKNOWN;
 263                 }
 264         }
 265
 266         if (! (options & GET_DEVS))
 267                 return sra;
 268
 269         /* Get all the devices as well */
 270         *base = 0;
 271         dir = opendir(fname);
 272         if (!dir)
 273                 goto abort;
 274         sra->array.spare_disks = 0;
 275
 276         devp = &sra->devs;
 277         sra->devs = NULL;
 278         while ((de = readdir(dir)) != NULL) {
 279                 char *ep;
 280                 if (de->d_ino == 0 ||
 281                     strncmp(de->d_name, "dev-", 4) != 0)
 282                         continue;
 283                 strcpy(base, de->d_name);
 284                 dbase = base + strlen(base);
 285                 *dbase++ = '/';
 286
 287                 dev = xcalloc(1, sizeof(*dev));
 288
 289                 /* Always get slot, major, minor */
 290                 strcpy(dbase, "slot");
 291                 if (load_sys(fname, buf, sizeof(buf))) {
 292                         /* hmm... unable to read 'slot' maybe the device
 293                          * is going away?
 294                          */
 295                         strcpy(dbase, "block");
 296                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 297                             errno != ENAMETOOLONG) {
 298                                 /* ...yup device is gone */
 299                                 free(dev);
 300                                 continue;
 301                         } else {
 302                                 /* slot is unreadable but 'block' link
 303                                  * still intact... something bad is happening
 304                                  * so abort
 305                                  */
 306                                 free(dev);
 307                                 goto abort;
 308                         }
 309
 310                 }
 311                 strcpy(dev->sys_name, de->d_name);
 312                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 313                 if (*ep) dev->disk.raid_disk = -1;
 314
 315                 strcpy(dbase, "block/dev");
 316                 if (load_sys(fname, buf, sizeof(buf))) {
 317                         /* assume this is a stale reference to a hot
 318                          * removed device
 319                          */
 320                         free(dev);
 321                         continue;
 322                 }
 323                 sra->array.nr_disks++;
 324                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 325
 326                 /* special case check for block devices that can go 'offline' */
 327                 strcpy(dbase, "block/device/state");
 328                 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
 329                     strncmp(buf, "offline", 7) == 0) {
 330                         free(dev);
 331                         continue;
 332                 }
 333
 334                 /* finally add this disk to the array */
 335                 *devp = dev;
 336                 devp = & dev->next;
 337                 dev->next = NULL;
 338
 339                 if (options & GET_OFFSET) {
 340                         strcpy(dbase, "offset");
 341                         if (load_sys(fname, buf, sizeof(buf)))
 342                                 goto abort;
 343                         dev->data_offset = strtoull(buf, NULL, 0);
 344                         strcpy(dbase, "new_offset");
 345                         if (load_sys(fname, buf, sizeof(buf)) == 0)
 346                                 dev->new_data_offset = strtoull(buf, NULL, 0);
 347                         else
 348                                 dev->new_data_offset = dev->data_offset;
 349                 }
 350                 if (options & GET_SIZE) {
 351                         strcpy(dbase, "size");
 352                         if (load_sys(fname, buf, sizeof(buf)))
 353                                 goto abort;
 354                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 355                 }
 356                 if (options & GET_STATE) {
 357                         dev->disk.state = 0;
 358                         strcpy(dbase, "state");
 359                         if (load_sys(fname, buf, sizeof(buf)))
 360                                 goto abort;
 361                         if (strstr(buf, "in_sync"))
 362                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 363                         if (strstr(buf, "faulty"))
 364                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 365                         if (dev->disk.state == 0)
 366                                 sra->array.spare_disks++;
 367                 }
 368                 if (options & GET_ERROR) {
 369                         strcpy(buf, "errors");
 370                         if (load_sys(fname, buf, sizeof(buf)))
 371                                 goto abort;
 372                         dev->errors = strtoul(buf, NULL, 0);
 373                 }
 374         }
 375         closedir(dir);
 376         return sra;
 377
 378  abort:
 379         if (dir)
 380                 closedir(dir);
 381         sysfs_free(sra);
 382         return NULL;
 383 }
 384
 385 int sysfs_attr_match(const char *attr, const char *str)
 386 {
 387         /* See if attr, read from a sysfs file, matches
 388          * str.  They must either be the same, or attr can
 389          * have a trailing newline or comma
 390          */
 391         while (*attr && *str && *attr == *str) {
 392                 attr++;
 393                 str++;
 394         }
 395
 396         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 397                 return 0;
 398         return 1;
 399 }
 400
 401 int sysfs_match_word(const char *word, char **list)
 402 {
 403         int n;
 404         for (n=0; list[n]; n++)
 405                 if (sysfs_attr_match(word, list[n]))
 406                         break;
 407         return n;
 408 }
 409
 410 unsigned long long get_component_size(int fd)
 411 {
 412         /* Find out the component size of the array.
 413          * We cannot trust GET_ARRAY_INFO ioctl as it's
 414          * size field is only 32bits.
 415          * So look in /sys/block/mdXXX/md/component_size
 416          *
 417          * This returns in units of sectors.
 418          */
 419         struct stat stb;
 420         char fname[MAX_SYSFS_PATH_LEN];
 421         int n;
 422         if (fstat(fd, &stb))
 423                 return 0;
 424         snprintf(fname, MAX_SYSFS_PATH_LEN,
 425                  "/sys/block/%s/md/component_size", stat2devnm(&stb));
 426         fd = open(fname, O_RDONLY);
 427         if (fd < 0)
 428                 return 0;
 429         n = read(fd, fname, sizeof(fname));
 430         close(fd);
 431         if (n < 0 || n == sizeof(fname))
 432                 return 0;
 433         fname[n] = 0;
 434         return strtoull(fname, NULL, 10) * 2;
 435 }
 436
 437 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 438                   char *name, char *val)
 439 {
 440         char fname[MAX_SYSFS_PATH_LEN];
 441         unsigned int n;
 442         int fd;
 443
 444         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 445                 sra->sys_name, dev?dev->sys_name:"", name);
 446         fd = open(fname, O_WRONLY);
 447         if (fd < 0)
 448                 return -1;
 449         n = write(fd, val, strlen(val));
 450         close(fd);
 451         if (n != strlen(val)) {
 452                 dprintf("failed to write '%s' to '%s' (%s)\n",
 453                         val, fname, strerror(errno));
 454                 return -1;
 455         }
 456         return 0;
 457 }
 458
 459 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 460                   char *name, unsigned long long val)
 461 {
 462         char valstr[50];
 463         sprintf(valstr, "%llu", val);
 464         return sysfs_set_str(sra, dev, name, valstr);
 465 }
 466
 467 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
 468                          char *name, long long val)
 469 {
 470         char valstr[50];
 471         sprintf(valstr, "%lli", val);
 472         return sysfs_set_str(sra, dev, name, valstr);
 473 }
 474
 475 int sysfs_uevent(struct mdinfo *sra, char *event)
 476 {
 477         char fname[MAX_SYSFS_PATH_LEN];
 478         int n;
 479         int fd;
 480
 481         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/uevent",
 482                 sra->sys_name);
 483         fd = open(fname, O_WRONLY);
 484         if (fd < 0)
 485                 return -1;
 486         n = write(fd, event, strlen(event));
 487         close(fd);
 488         if (n != (int)strlen(event)) {
 489                 dprintf("failed to write '%s' to '%s' (%s)\n",
 490                         event, fname, strerror(errno));
 491                 return -1;
 492         }
 493         return 0;
 494 }
 495
 496 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
 497 {
 498         char fname[MAX_SYSFS_PATH_LEN];
 499         struct stat st;
 500
 501         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 502                 sra->sys_name, dev?dev->sys_name:"", name);
 503
 504         return stat(fname, &st) == 0;
 505 }
 506
 507 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 508                        char *name)
 509 {
 510         char fname[MAX_SYSFS_PATH_LEN];
 511         int fd;
 512
 513         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 514                 sra->sys_name, dev?dev->sys_name:"", name);
 515         fd = open(fname, O_RDWR);
 516         if (fd < 0)
 517                 fd = open(fname, O_RDONLY);
 518         return fd;
 519 }
 520
 521 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 522 {
 523         char buf[50];
 524         int n;
 525         char *ep;
 526
 527         lseek(fd, 0, 0);
 528         n = read(fd, buf, sizeof(buf));
 529         if (n <= 0 || n == sizeof(buf))
 530                 return -2;
 531         buf[n] = 0;
 532         *val = strtoull(buf, &ep, 0);
 533         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 534                 return -1;
 535         return 0;
 536 }
 537
 538 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 539                        char *name, unsigned long long *val)
 540 {
 541         int n;
 542         int fd;
 543
 544         fd = sysfs_get_fd(sra, dev, name);
 545         if (fd < 0)
 546                 return -1;
 547         n = sysfs_fd_get_ll(fd, val);
 548         close(fd);
 549         return n;
 550 }
 551
 552 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
 553 {
 554         /* two numbers in this sysfs file, either
 555          *  NNN (NNN)
 556          * or
 557          *  NNN / NNN
 558          */
 559         char buf[80];
 560         int n;
 561         char *ep, *ep2;
 562
 563         lseek(fd, 0, 0);
 564         n = read(fd, buf, sizeof(buf));
 565         if (n <= 0 || n == sizeof(buf))
 566                 return -2;
 567         buf[n] = 0;
 568         *v1 = strtoull(buf, &ep, 0);
 569         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 570                 return -1;
 571         while (*ep == ' ' || *ep == '/' || *ep == '(')
 572                 ep++;
 573         *v2 = strtoull(ep, &ep2, 0);
 574         if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
 575                 *v2 = *v1;
 576                 return 1;
 577         }
 578         return 2;
 579 }
 580
 581 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
 582                   char *name, unsigned long long *v1, unsigned long long *v2)
 583 {
 584         int n;
 585         int fd;
 586
 587         fd = sysfs_get_fd(sra, dev, name);
 588         if (fd < 0)
 589                 return -1;
 590         n = sysfs_fd_get_two(fd, v1, v2);
 591         close(fd);
 592         return n;
 593 }
 594
 595 int sysfs_fd_get_str(int fd, char *val, int size)
 596 {
 597         int n;
 598
 599         lseek(fd, 0, 0);
 600         n = read(fd, val, size);
 601         if (n <= 0 || n == size)
 602                 return -1;
 603         val[n] = 0;
 604         return n;
 605 }
 606
 607 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 608                        char *name, char *val, int size)
 609 {
 610         int n;
 611         int fd;
 612
 613         fd = sysfs_get_fd(sra, dev, name);
 614         if (fd < 0)
 615                 return -1;
 616         n = sysfs_fd_get_str(fd, val, size);
 617         close(fd);
 618         return n;
 619 }
 620
 621 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 622 {
 623         unsigned long sec;
 624         unsigned long msec;
 625         char delay[30];
 626
 627         sec = ms / 1000;
 628         msec = ms % 1000;
 629
 630         sprintf(delay, "%ld.%03ld\n", sec, msec);
 631         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 632         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 633 }
 634
 635 int sysfs_set_array(struct mdinfo *info, int vers)
 636 {
 637         int rv = 0;
 638         char ver[100];
 639         int raid_disks = info->array.raid_disks;
 640
 641         ver[0] = 0;
 642         if (info->array.major_version == -1 &&
 643             info->array.minor_version == -2) {
 644                 char buf[1024];
 645
 646                 strcat(strcpy(ver, "external:"), info->text_version);
 647
 648                 /* meta version might already be set if we are setting
 649                  * new geometry for a reshape.  In that case we don't
 650                  * want to over-write the 'readonly' flag that is
 651                  * stored in the metadata version.  So read the current
 652                  * version first, and preserve the flag
 653                  */
 654                 if (sysfs_get_str(info, NULL, "metadata_version",
 655                                   buf, 1024) > 0)
 656                         if (strlen(buf) >= 9 && buf[9] == '-')
 657                                 ver[9] = '-';
 658
 659                 if ((vers % 100) < 2 ||
 660                     sysfs_set_str(info, NULL, "metadata_version",
 661                                   ver) < 0) {
 662                         pr_err("This kernel does not support external metadata.\n");
 663                         return 1;
 664                 }
 665         }
 666         if (info->array.level < 0)
 667                 return 0; /* FIXME */
 668         rv |= sysfs_set_str(info, NULL, "level",
 669                             map_num(pers, info->array.level));
 670         if (info->reshape_active && info->delta_disks != UnSet)
 671                 raid_disks -= info->delta_disks;
 672         rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
 673         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 674         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 675         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 676         if (info->custom_array_size) {
 677                 int rc;
 678
 679                 rc = sysfs_set_num(info, NULL, "array_size",
 680                                    info->custom_array_size/2);
 681                 if (rc && errno == ENOENT) {
 682                         pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
 683                         rc = 0;
 684                 }
 685                 rv |= rc;
 686         }
 687
 688         if (info->array.level > 0)
 689                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 690
 691         if (info->reshape_active) {
 692                 rv |= sysfs_set_num(info, NULL, "reshape_position",
 693                                     info->reshape_progress);
 694                 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
 695                 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
 696                 rv |= sysfs_set_num(info, NULL, "raid_disks",
 697                                     info->array.raid_disks);
 698                 /* We don't set 'new_level' here.  That can only happen
 699                  * once the reshape completes.
 700                  */
 701         }
 702
 703         if (info->consistency_policy == CONSISTENCY_POLICY_PPL) {
 704                 if (sysfs_set_str(info, NULL, "consistency_policy",
 705                                   map_num(consistency_policies,
 706                                           info->consistency_policy))) {
 707                         pr_err("This kernel does not support PPL\n");
 708                         return 1;
 709                 }
 710         }
 711
 712         return rv;
 713 }
 714
 715 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 716 {
 717         char dv[PATH_MAX];
 718         char nm[PATH_MAX];
 719         char *dname;
 720         int rv;
 721         int i;
 722
 723         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 724         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 725         if (rv)
 726                 return rv;
 727
 728         memset(nm, 0, sizeof(nm));
 729         dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
 730         strcpy(sd->sys_name, "dev-");
 731         strcpy(sd->sys_name+4, dname);
 732
 733         /* test write to see if 'recovery_start' is available */
 734         if (resume && sd->recovery_start < MaxSector &&
 735             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 736                 sysfs_set_str(sra, sd, "state", "remove");
 737                 return -1;
 738         }
 739
 740         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 741         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 742         if (sra->array.level != LEVEL_CONTAINER) {
 743                 if (sd->consistency_policy == CONSISTENCY_POLICY_PPL) {
 744                         rv |= sysfs_set_num(sra, sd, "ppl_sector", sd->ppl_sector);
 745                         rv |= sysfs_set_num(sra, sd, "ppl_size", sd->ppl_size);
 746                 }
 747                 if (sd->recovery_start == MaxSector)
 748                         /* This can correctly fail if array isn't started,
 749                          * yet, so just ignore status for now.
 750                          */
 751                         sysfs_set_str(sra, sd, "state", "insync");
 752                 if (sd->disk.raid_disk >= 0)
 753                         rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 754                 if (resume)
 755                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 756         }
 757         if (sd->bb.supported) {
 758                 if (sysfs_set_str(sra, sd, "state", "external_bbl")) {
 759                         /*
 760                          * backward compatibility - if kernel doesn't support
 761                          * bad blocks for external metadata, let it continue
 762                          * as long as there are none known so far
 763                          */
 764                         if (sd->bb.count) {
 765                                 pr_err("The kernel has no support for bad blocks in external metadata\n");
 766                                 return -1;
 767                         }
 768                 }
 769
 770                 for (i = 0; i < sd->bb.count; i++) {
 771                         char s[30];
 772                         const struct md_bb_entry *entry = &sd->bb.entries[i];
 773
 774                         snprintf(s, sizeof(s) - 1, "%llu %d\n", entry->sector,
 775                                  entry->length);
 776                         rv |= sysfs_set_str(sra, sd, "bad_blocks", s);
 777                 }
 778         }
 779         return rv;
 780 }
 781
 782 #if 0
 783 int sysfs_disk_to_sg(int fd)
 784 {
 785         /* from an open block device, try find and open its corresponding
 786          * scsi_generic interface
 787          */
 788         struct stat st;
 789         char path[256];
 790         char sg_path[256];
 791         char sg_major_minor[10];
 792         char *c;
 793         DIR *dir;
 794         struct dirent *de;
 795         int major, minor, rv;
 796
 797         if (fstat(fd, &st))
 798                 return -1;
 799
 800         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 801                  major(st.st_rdev), minor(st.st_rdev));
 802
 803         dir = opendir(path);
 804         if (!dir)
 805                 return -1;
 806
 807         de = readdir(dir);
 808         while (de) {
 809                 if (strncmp("scsi_generic:", de->d_name,
 810                             strlen("scsi_generic:")) == 0)
 811                         break;
 812                 de = readdir(dir);
 813         }
 814         closedir(dir);
 815
 816         if (!de)
 817                 return -1;
 818
 819         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 820         fd = open(sg_path, O_RDONLY);
 821         if (fd < 0)
 822                 return fd;
 823
 824         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 825         close(fd);
 826         if (rv < 0 || rv == sizeof(sg_major_minor))
 827                 return -1;
 828         else
 829                 sg_major_minor[rv - 1] = '\0';
 830
 831         c = strchr(sg_major_minor, ':');
 832         *c = '\0';
 833         c++;
 834         major = strtol(sg_major_minor, NULL, 10);
 835         minor = strtol(c, NULL, 10);
 836         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 837                  (int) getpid(), major, minor);
 838         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 839                         fd = open(path, O_RDONLY);
 840                         unlink(path);
 841                         return fd;
 842         }
 843
 844         return -1;
 845 }
 846 #endif
 847
 848 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 849 {
 850         /* from an open block device, try to retrieve it scsi_id */
 851         struct stat st;
 852         char path[256];
 853         DIR *dir;
 854         struct dirent *de;
 855         int host, bus, target, lun;
 856
 857         if (fstat(fd, &st))
 858                 return 1;
 859
 860         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
 861                  major(st.st_rdev), minor(st.st_rdev));
 862
 863         dir = opendir(path);
 864         if (!dir)
 865                 return 1;
 866
 867         for (de = readdir(dir); de; de = readdir(dir)) {
 868                 int count;
 869
 870                 if (de->d_type != DT_DIR)
 871                         continue;
 872
 873                 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
 874                 if (count == 4)
 875                         break;
 876         }
 877         closedir(dir);
 878
 879         if (!de)
 880                 return 1;
 881
 882         *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
 883         return 0;
 884 }
 885
 886 int sysfs_unique_holder(char *devnm, long rdev)
 887 {
 888         /* Check that devnm is a holder of rdev,
 889          * and is the only holder.
 890          * we should be locked against races by
 891          * an O_EXCL on devnm
 892          * Return values:
 893          *  0 - not unique, not even a holder
 894          *  1 - unique, this is the only holder.
 895          *  2/3 - not unique, there is another holder
 896          * -1 - error, cannot find the holders
 897          */
 898         DIR *dir;
 899         struct dirent *de;
 900         char dirname[100];
 901         char l;
 902         int ret = 0;
 903         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 904                 major(rdev), minor(rdev));
 905         dir = opendir(dirname);
 906         if (!dir)
 907                 return -1;
 908         l = strlen(dirname);
 909         while ((de = readdir(dir)) != NULL) {
 910                 char buf[100];
 911                 char *sl;
 912                 int n;
 913
 914                 if (de->d_ino == 0)
 915                         continue;
 916                 if (de->d_name[0] == '.')
 917                         continue;
 918                 strcpy(dirname+l, "/");
 919                 strcat(dirname+l, de->d_name);
 920                 n = readlink(dirname, buf, sizeof(buf)-1);
 921                 if (n <= 0)
 922                         continue;
 923                 buf[n] = 0;
 924                 sl = strrchr(buf, '/');
 925                 if (!sl)
 926                         continue;
 927                 sl++;
 928
 929                 if (strcmp(devnm, sl) == 0)
 930                         ret |= 1;
 931                 else
 932                         ret |= 2;
 933         }
 934         closedir(dir);
 935         return ret;
 936 }
 937
 938 int sysfs_freeze_array(struct mdinfo *sra)
 939 {
 940         /* Try to freeze resync/rebuild on this array/container.
 941          * Return -1 if the array is busy,
 942          * return 0 if this kernel doesn't support 'frozen'
 943          * return 1 if it worked.
 944          */
 945         char buf[20];
 946
 947         if (!sysfs_attribute_available(sra, NULL, "sync_action"))
 948                 return 1; /* no sync_action == frozen */
 949         if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
 950                 return 0;
 951         if (strcmp(buf, "frozen\n") == 0)
 952                 /* Already frozen */
 953                 return 0;
 954         if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
 955                 return -1;
 956         if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
 957                 return 0;
 958         return 1;
 959 }
 960
 961 int sysfs_wait(int fd, int *msec)
 962 {
 963         /* Wait up to '*msec' for fd to have an exception condition.
 964          * if msec == NULL, wait indefinitely.
 965          */
 966         fd_set fds;
 967         int n;
 968         FD_ZERO(&fds);
 969         FD_SET(fd, &fds);
 970         if (msec == NULL)
 971                 n = select(fd+1, NULL, NULL, &fds, NULL);
 972         else if (*msec < 0)
 973                 n = 0;
 974         else {
 975                 struct timeval start, end, tv;
 976                 gettimeofday(&start, NULL);
 977                 if (*msec < 1000) {
 978                         tv.tv_sec = 0;
 979                         tv.tv_usec = (*msec)*1000;
 980                 } else {
 981                         tv.tv_sec = (*msec)/1000;
 982                         tv.tv_usec = 0;
 983                 }
 984                 n = select(fd+1, NULL, NULL, &fds, &tv);
 985                 gettimeofday(&end, NULL);
 986                 end.tv_sec -= start.tv_sec;
 987                 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
 988                           - start.tv_usec/1000) + 1;
 989         }
 990         return n;
 991 }