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, sra->sysfs_array_state,
 251                              sizeof(sra->sysfs_array_state)))
 252                         goto abort;
 253         } else
 254                 sra->sysfs_array_state[0] = 0;
 255
 256         if (options & GET_CONSISTENCY_POLICY) {
 257                 strcpy(base, "consistency_policy");
 258                 if (load_sys(fname, buf, sizeof(buf))) {
 259                         sra->consistency_policy = CONSISTENCY_POLICY_UNKNOWN;
 260                 } else {
 261                         sra->consistency_policy = map_name(consistency_policies, buf);
 262                         if (sra->consistency_policy == UnSet)
 263                                 sra->consistency_policy = CONSISTENCY_POLICY_UNKNOWN;
 264                 }
 265         }
 266
 267         if (! (options & GET_DEVS))
 268                 return sra;
 269
 270         /* Get all the devices as well */
 271         *base = 0;
 272         dir = opendir(fname);
 273         if (!dir)
 274                 goto abort;
 275         sra->array.spare_disks = 0;
 276
 277         devp = &sra->devs;
 278         sra->devs = NULL;
 279         while ((de = readdir(dir)) != NULL) {
 280                 char *ep;
 281                 if (de->d_ino == 0 ||
 282                     strncmp(de->d_name, "dev-", 4) != 0)
 283                         continue;
 284                 strcpy(base, de->d_name);
 285                 dbase = base + strlen(base);
 286                 *dbase++ = '/';
 287
 288                 dev = xcalloc(1, sizeof(*dev));
 289
 290                 /* Always get slot, major, minor */
 291                 strcpy(dbase, "slot");
 292                 if (load_sys(fname, buf, sizeof(buf))) {
 293                         /* hmm... unable to read 'slot' maybe the device
 294                          * is going away?
 295                          */
 296                         strcpy(dbase, "block");
 297                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 298                             errno != ENAMETOOLONG) {
 299                                 /* ...yup device is gone */
 300                                 free(dev);
 301                                 continue;
 302                         } else {
 303                                 /* slot is unreadable but 'block' link
 304                                  * still intact... something bad is happening
 305                                  * so abort
 306                                  */
 307                                 free(dev);
 308                                 goto abort;
 309                         }
 310
 311                 }
 312                 strcpy(dev->sys_name, de->d_name);
 313                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 314                 if (*ep) dev->disk.raid_disk = -1;
 315
 316                 strcpy(dbase, "block/dev");
 317                 if (load_sys(fname, buf, sizeof(buf))) {
 318                         /* assume this is a stale reference to a hot
 319                          * removed device
 320                          */
 321                         free(dev);
 322                         continue;
 323                 }
 324                 sra->array.nr_disks++;
 325                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 326
 327                 /* special case check for block devices that can go 'offline' */
 328                 strcpy(dbase, "block/device/state");
 329                 if (load_sys(fname, buf, sizeof(buf)) == 0 &&
 330                     strncmp(buf, "offline", 7) == 0) {
 331                         free(dev);
 332                         continue;
 333                 }
 334
 335                 /* finally add this disk to the array */
 336                 *devp = dev;
 337                 devp = & dev->next;
 338                 dev->next = NULL;
 339
 340                 if (options & GET_OFFSET) {
 341                         strcpy(dbase, "offset");
 342                         if (load_sys(fname, buf, sizeof(buf)))
 343                                 goto abort;
 344                         dev->data_offset = strtoull(buf, NULL, 0);
 345                         strcpy(dbase, "new_offset");
 346                         if (load_sys(fname, buf, sizeof(buf)) == 0)
 347                                 dev->new_data_offset = strtoull(buf, NULL, 0);
 348                         else
 349                                 dev->new_data_offset = dev->data_offset;
 350                 }
 351                 if (options & GET_SIZE) {
 352                         strcpy(dbase, "size");
 353                         if (load_sys(fname, buf, sizeof(buf)))
 354                                 goto abort;
 355                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 356                 }
 357                 if (options & GET_STATE) {
 358                         dev->disk.state = 0;
 359                         strcpy(dbase, "state");
 360                         if (load_sys(fname, buf, sizeof(buf)))
 361                                 goto abort;
 362                         if (strstr(buf, "in_sync"))
 363                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 364                         if (strstr(buf, "faulty"))
 365                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 366                         if (dev->disk.state == 0)
 367                                 sra->array.spare_disks++;
 368                 }
 369                 if (options & GET_ERROR) {
 370                         strcpy(buf, "errors");
 371                         if (load_sys(fname, buf, sizeof(buf)))
 372                                 goto abort;
 373                         dev->errors = strtoul(buf, NULL, 0);
 374                 }
 375         }
 376         closedir(dir);
 377         return sra;
 378
 379  abort:
 380         if (dir)
 381                 closedir(dir);
 382         sysfs_free(sra);
 383         return NULL;
 384 }
 385
 386 int sysfs_attr_match(const char *attr, const char *str)
 387 {
 388         /* See if attr, read from a sysfs file, matches
 389          * str.  They must either be the same, or attr can
 390          * have a trailing newline or comma
 391          */
 392         while (*attr && *str && *attr == *str) {
 393                 attr++;
 394                 str++;
 395         }
 396
 397         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 398                 return 0;
 399         return 1;
 400 }
 401
 402 int sysfs_match_word(const char *word, char **list)
 403 {
 404         int n;
 405         for (n=0; list[n]; n++)
 406                 if (sysfs_attr_match(word, list[n]))
 407                         break;
 408         return n;
 409 }
 410
 411 unsigned long long get_component_size(int fd)
 412 {
 413         /* Find out the component size of the array.
 414          * We cannot trust GET_ARRAY_INFO ioctl as it's
 415          * size field is only 32bits.
 416          * So look in /sys/block/mdXXX/md/component_size
 417          *
 418          * This returns in units of sectors.
 419          */
 420         struct stat stb;
 421         char fname[MAX_SYSFS_PATH_LEN];
 422         int n;
 423         if (fstat(fd, &stb))
 424                 return 0;
 425         snprintf(fname, MAX_SYSFS_PATH_LEN,
 426                  "/sys/block/%s/md/component_size", stat2devnm(&stb));
 427         fd = open(fname, O_RDONLY);
 428         if (fd < 0)
 429                 return 0;
 430         n = read(fd, fname, sizeof(fname));
 431         close(fd);
 432         if (n < 0 || n == sizeof(fname))
 433                 return 0;
 434         fname[n] = 0;
 435         return strtoull(fname, NULL, 10) * 2;
 436 }
 437
 438 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 439                   char *name, char *val)
 440 {
 441         char fname[MAX_SYSFS_PATH_LEN];
 442         unsigned int n;
 443         int fd;
 444
 445         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 446                 sra->sys_name, dev?dev->sys_name:"", name);
 447         fd = open(fname, O_WRONLY);
 448         if (fd < 0)
 449                 return -1;
 450         n = write(fd, val, strlen(val));
 451         close(fd);
 452         if (n != strlen(val)) {
 453                 dprintf("failed to write '%s' to '%s' (%s)\n",
 454                         val, fname, strerror(errno));
 455                 return -1;
 456         }
 457         return 0;
 458 }
 459
 460 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 461                   char *name, unsigned long long val)
 462 {
 463         char valstr[50];
 464         sprintf(valstr, "%llu", val);
 465         return sysfs_set_str(sra, dev, name, valstr);
 466 }
 467
 468 int sysfs_set_num_signed(struct mdinfo *sra, struct mdinfo *dev,
 469                          char *name, long long val)
 470 {
 471         char valstr[50];
 472         sprintf(valstr, "%lli", val);
 473         return sysfs_set_str(sra, dev, name, valstr);
 474 }
 475
 476 int sysfs_uevent(struct mdinfo *sra, char *event)
 477 {
 478         char fname[MAX_SYSFS_PATH_LEN];
 479         int n;
 480         int fd;
 481
 482         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/uevent",
 483                 sra->sys_name);
 484         fd = open(fname, O_WRONLY);
 485         if (fd < 0)
 486                 return -1;
 487         n = write(fd, event, strlen(event));
 488         close(fd);
 489         if (n != (int)strlen(event)) {
 490                 dprintf("failed to write '%s' to '%s' (%s)\n",
 491                         event, fname, strerror(errno));
 492                 return -1;
 493         }
 494         return 0;
 495 }
 496
 497 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
 498 {
 499         char fname[MAX_SYSFS_PATH_LEN];
 500         struct stat st;
 501
 502         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 503                 sra->sys_name, dev?dev->sys_name:"", name);
 504
 505         return stat(fname, &st) == 0;
 506 }
 507
 508 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 509                        char *name)
 510 {
 511         char fname[MAX_SYSFS_PATH_LEN];
 512         int fd;
 513
 514         snprintf(fname, MAX_SYSFS_PATH_LEN, "/sys/block/%s/md/%s/%s",
 515                 sra->sys_name, dev?dev->sys_name:"", name);
 516         fd = open(fname, O_RDWR);
 517         if (fd < 0)
 518                 fd = open(fname, O_RDONLY);
 519         return fd;
 520 }
 521
 522 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 523 {
 524         char buf[50];
 525         int n;
 526         char *ep;
 527
 528         lseek(fd, 0, 0);
 529         n = read(fd, buf, sizeof(buf));
 530         if (n <= 0 || n == sizeof(buf))
 531                 return -2;
 532         buf[n] = 0;
 533         *val = strtoull(buf, &ep, 0);
 534         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 535                 return -1;
 536         return 0;
 537 }
 538
 539 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 540                        char *name, unsigned long long *val)
 541 {
 542         int n;
 543         int fd;
 544
 545         fd = sysfs_get_fd(sra, dev, name);
 546         if (fd < 0)
 547                 return -1;
 548         n = sysfs_fd_get_ll(fd, val);
 549         close(fd);
 550         return n;
 551 }
 552
 553 int sysfs_fd_get_two(int fd, unsigned long long *v1, unsigned long long *v2)
 554 {
 555         /* two numbers in this sysfs file, either
 556          *  NNN (NNN)
 557          * or
 558          *  NNN / NNN
 559          */
 560         char buf[80];
 561         int n;
 562         char *ep, *ep2;
 563
 564         lseek(fd, 0, 0);
 565         n = read(fd, buf, sizeof(buf));
 566         if (n <= 0 || n == sizeof(buf))
 567                 return -2;
 568         buf[n] = 0;
 569         *v1 = strtoull(buf, &ep, 0);
 570         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 571                 return -1;
 572         while (*ep == ' ' || *ep == '/' || *ep == '(')
 573                 ep++;
 574         *v2 = strtoull(ep, &ep2, 0);
 575         if (ep2 == ep || (*ep2 != 0 && *ep2 != '\n' && *ep2 != ' ' && *ep2 != ')')) {
 576                 *v2 = *v1;
 577                 return 1;
 578         }
 579         return 2;
 580 }
 581
 582 int sysfs_get_two(struct mdinfo *sra, struct mdinfo *dev,
 583                   char *name, unsigned long long *v1, unsigned long long *v2)
 584 {
 585         int n;
 586         int fd;
 587
 588         fd = sysfs_get_fd(sra, dev, name);
 589         if (fd < 0)
 590                 return -1;
 591         n = sysfs_fd_get_two(fd, v1, v2);
 592         close(fd);
 593         return n;
 594 }
 595
 596 int sysfs_fd_get_str(int fd, char *val, int size)
 597 {
 598         int n;
 599
 600         lseek(fd, 0, 0);
 601         n = read(fd, val, size);
 602         if (n <= 0 || n == size)
 603                 return -1;
 604         val[n] = 0;
 605         return n;
 606 }
 607
 608 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 609                        char *name, char *val, int size)
 610 {
 611         int n;
 612         int fd;
 613
 614         fd = sysfs_get_fd(sra, dev, name);
 615         if (fd < 0)
 616                 return -1;
 617         n = sysfs_fd_get_str(fd, val, size);
 618         close(fd);
 619         return n;
 620 }
 621
 622 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 623 {
 624         unsigned long sec;
 625         unsigned long msec;
 626         char delay[30];
 627
 628         sec = ms / 1000;
 629         msec = ms % 1000;
 630
 631         sprintf(delay, "%ld.%03ld\n", sec, msec);
 632         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 633         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 634 }
 635
 636 int sysfs_set_array(struct mdinfo *info, int vers)
 637 {
 638         int rv = 0;
 639         char ver[100];
 640         int raid_disks = info->array.raid_disks;
 641
 642         ver[0] = 0;
 643         if (info->array.major_version == -1 &&
 644             info->array.minor_version == -2) {
 645                 char buf[1024];
 646
 647                 strcat(strcpy(ver, "external:"), info->text_version);
 648
 649                 /* meta version might already be set if we are setting
 650                  * new geometry for a reshape.  In that case we don't
 651                  * want to over-write the 'readonly' flag that is
 652                  * stored in the metadata version.  So read the current
 653                  * version first, and preserve the flag
 654                  */
 655                 if (sysfs_get_str(info, NULL, "metadata_version",
 656                                   buf, 1024) > 0)
 657                         if (strlen(buf) >= 9 && buf[9] == '-')
 658                                 ver[9] = '-';
 659
 660                 if ((vers % 100) < 2 ||
 661                     sysfs_set_str(info, NULL, "metadata_version",
 662                                   ver) < 0) {
 663                         pr_err("This kernel does not support external metadata.\n");
 664                         return 1;
 665                 }
 666         }
 667         if (info->array.level < 0)
 668                 return 0; /* FIXME */
 669         rv |= sysfs_set_str(info, NULL, "level",
 670                             map_num(pers, info->array.level));
 671         if (info->reshape_active && info->delta_disks != UnSet)
 672                 raid_disks -= info->delta_disks;
 673         rv |= sysfs_set_num(info, NULL, "raid_disks", raid_disks);
 674         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 675         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 676         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 677         if (info->custom_array_size) {
 678                 int rc;
 679
 680                 rc = sysfs_set_num(info, NULL, "array_size",
 681                                    info->custom_array_size/2);
 682                 if (rc && errno == ENOENT) {
 683                         pr_err("This kernel does not have the md/array_size attribute, the array may be larger than expected\n");
 684                         rc = 0;
 685                 }
 686                 rv |= rc;
 687         }
 688
 689         if (info->array.level > 0)
 690                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 691
 692         if (info->reshape_active) {
 693                 rv |= sysfs_set_num(info, NULL, "reshape_position",
 694                                     info->reshape_progress);
 695                 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
 696                 rv |= sysfs_set_num(info, NULL, "layout", info->new_layout);
 697                 rv |= sysfs_set_num(info, NULL, "raid_disks",
 698                                     info->array.raid_disks);
 699                 /* We don't set 'new_level' here.  That can only happen
 700                  * once the reshape completes.
 701                  */
 702         }
 703
 704         if (info->consistency_policy == CONSISTENCY_POLICY_PPL) {
 705                 if (sysfs_set_str(info, NULL, "consistency_policy",
 706                                   map_num(consistency_policies,
 707                                           info->consistency_policy))) {
 708                         pr_err("This kernel does not support PPL\n");
 709                         return 1;
 710                 }
 711         }
 712
 713         return rv;
 714 }
 715
 716 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 717 {
 718         char dv[PATH_MAX];
 719         char nm[PATH_MAX];
 720         char *dname;
 721         int rv;
 722         int i;
 723
 724         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 725         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 726         if (rv)
 727                 return rv;
 728
 729         memset(nm, 0, sizeof(nm));
 730         dname = devid2kname(makedev(sd->disk.major, sd->disk.minor));
 731         strcpy(sd->sys_name, "dev-");
 732         strcpy(sd->sys_name+4, dname);
 733
 734         /* test write to see if 'recovery_start' is available */
 735         if (resume && sd->recovery_start < MaxSector &&
 736             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 737                 sysfs_set_str(sra, sd, "state", "remove");
 738                 return -1;
 739         }
 740
 741         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 742         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 743         if (sra->array.level != LEVEL_CONTAINER) {
 744                 if (sd->consistency_policy == CONSISTENCY_POLICY_PPL) {
 745                         rv |= sysfs_set_num(sra, sd, "ppl_sector", sd->ppl_sector);
 746                         rv |= sysfs_set_num(sra, sd, "ppl_size", sd->ppl_size);
 747                 }
 748                 if (sd->recovery_start == MaxSector)
 749                         /* This can correctly fail if array isn't started,
 750                          * yet, so just ignore status for now.
 751                          */
 752                         sysfs_set_str(sra, sd, "state", "insync");
 753                 if (sd->disk.raid_disk >= 0)
 754                         rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 755                 if (resume)
 756                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 757         }
 758         if (sd->bb.supported) {
 759                 if (sysfs_set_str(sra, sd, "state", "external_bbl")) {
 760                         /*
 761                          * backward compatibility - if kernel doesn't support
 762                          * bad blocks for external metadata, let it continue
 763                          * as long as there are none known so far
 764                          */
 765                         if (sd->bb.count) {
 766                                 pr_err("The kernel has no support for bad blocks in external metadata\n");
 767                                 return -1;
 768                         }
 769                 }
 770
 771                 for (i = 0; i < sd->bb.count; i++) {
 772                         char s[30];
 773                         const struct md_bb_entry *entry = &sd->bb.entries[i];
 774
 775                         snprintf(s, sizeof(s) - 1, "%llu %d\n", entry->sector,
 776                                  entry->length);
 777                         rv |= sysfs_set_str(sra, sd, "bad_blocks", s);
 778                 }
 779         }
 780         return rv;
 781 }
 782
 783 #if 0
 784 int sysfs_disk_to_sg(int fd)
 785 {
 786         /* from an open block device, try find and open its corresponding
 787          * scsi_generic interface
 788          */
 789         struct stat st;
 790         char path[256];
 791         char sg_path[256];
 792         char sg_major_minor[10];
 793         char *c;
 794         DIR *dir;
 795         struct dirent *de;
 796         int major, minor, rv;
 797
 798         if (fstat(fd, &st))
 799                 return -1;
 800
 801         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 802                  major(st.st_rdev), minor(st.st_rdev));
 803
 804         dir = opendir(path);
 805         if (!dir)
 806                 return -1;
 807
 808         de = readdir(dir);
 809         while (de) {
 810                 if (strncmp("scsi_generic:", de->d_name,
 811                             strlen("scsi_generic:")) == 0)
 812                         break;
 813                 de = readdir(dir);
 814         }
 815         closedir(dir);
 816
 817         if (!de)
 818                 return -1;
 819
 820         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 821         fd = open(sg_path, O_RDONLY);
 822         if (fd < 0)
 823                 return fd;
 824
 825         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 826         close(fd);
 827         if (rv < 0 || rv == sizeof(sg_major_minor))
 828                 return -1;
 829         else
 830                 sg_major_minor[rv - 1] = '\0';
 831
 832         c = strchr(sg_major_minor, ':');
 833         *c = '\0';
 834         c++;
 835         major = strtol(sg_major_minor, NULL, 10);
 836         minor = strtol(c, NULL, 10);
 837         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 838                  (int) getpid(), major, minor);
 839         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 840                         fd = open(path, O_RDONLY);
 841                         unlink(path);
 842                         return fd;
 843         }
 844
 845         return -1;
 846 }
 847 #endif
 848
 849 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 850 {
 851         /* from an open block device, try to retrieve it scsi_id */
 852         struct stat st;
 853         char path[256];
 854         DIR *dir;
 855         struct dirent *de;
 856         int host, bus, target, lun;
 857
 858         if (fstat(fd, &st))
 859                 return 1;
 860
 861         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
 862                  major(st.st_rdev), minor(st.st_rdev));
 863
 864         dir = opendir(path);
 865         if (!dir)
 866                 return 1;
 867
 868         for (de = readdir(dir); de; de = readdir(dir)) {
 869                 int count;
 870
 871                 if (de->d_type != DT_DIR)
 872                         continue;
 873
 874                 count = sscanf(de->d_name, "%d:%d:%d:%d", &host, &bus, &target, &lun);
 875                 if (count == 4)
 876                         break;
 877         }
 878         closedir(dir);
 879
 880         if (!de)
 881                 return 1;
 882
 883         *id = (host << 24) | (bus << 16) | (target << 8) | (lun << 0);
 884         return 0;
 885 }
 886
 887 int sysfs_unique_holder(char *devnm, long rdev)
 888 {
 889         /* Check that devnm is a holder of rdev,
 890          * and is the only holder.
 891          * we should be locked against races by
 892          * an O_EXCL on devnm
 893          * Return values:
 894          *  0 - not unique, not even a holder
 895          *  1 - unique, this is the only holder.
 896          *  2/3 - not unique, there is another holder
 897          * -1 - error, cannot find the holders
 898          */
 899         DIR *dir;
 900         struct dirent *de;
 901         char dirname[100];
 902         char l;
 903         int ret = 0;
 904         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 905                 major(rdev), minor(rdev));
 906         dir = opendir(dirname);
 907         if (!dir)
 908                 return -1;
 909         l = strlen(dirname);
 910         while ((de = readdir(dir)) != NULL) {
 911                 char buf[100];
 912                 char *sl;
 913                 int n;
 914
 915                 if (de->d_ino == 0)
 916                         continue;
 917                 if (de->d_name[0] == '.')
 918                         continue;
 919                 strcpy(dirname+l, "/");
 920                 strcat(dirname+l, de->d_name);
 921                 n = readlink(dirname, buf, sizeof(buf)-1);
 922                 if (n <= 0)
 923                         continue;
 924                 buf[n] = 0;
 925                 sl = strrchr(buf, '/');
 926                 if (!sl)
 927                         continue;
 928                 sl++;
 929
 930                 if (strcmp(devnm, sl) == 0)
 931                         ret |= 1;
 932                 else
 933                         ret |= 2;
 934         }
 935         closedir(dir);
 936         return ret;
 937 }
 938
 939 int sysfs_freeze_array(struct mdinfo *sra)
 940 {
 941         /* Try to freeze resync/rebuild on this array/container.
 942          * Return -1 if the array is busy,
 943          * return 0 if this kernel doesn't support 'frozen'
 944          * return 1 if it worked.
 945          */
 946         char buf[20];
 947
 948         if (!sysfs_attribute_available(sra, NULL, "sync_action"))
 949                 return 1; /* no sync_action == frozen */
 950         if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
 951                 return 0;
 952         if (strcmp(buf, "frozen\n") == 0)
 953                 /* Already frozen */
 954                 return 0;
 955         if (strcmp(buf, "idle\n") != 0 && strcmp(buf, "recover\n") != 0)
 956                 return -1;
 957         if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
 958                 return 0;
 959         return 1;
 960 }
 961
 962 int sysfs_wait(int fd, int *msec)
 963 {
 964         /* Wait up to '*msec' for fd to have an exception condition.
 965          * if msec == NULL, wait indefinitely.
 966          */
 967         fd_set fds;
 968         int n;
 969         FD_ZERO(&fds);
 970         FD_SET(fd, &fds);
 971         if (msec == NULL)
 972                 n = select(fd+1, NULL, NULL, &fds, NULL);
 973         else if (*msec < 0)
 974                 n = 0;
 975         else {
 976                 struct timeval start, end, tv;
 977                 gettimeofday(&start, NULL);
 978                 if (*msec < 1000) {
 979                         tv.tv_sec = 0;
 980                         tv.tv_usec = (*msec)*1000;
 981                 } else {
 982                         tv.tv_sec = (*msec)/1000;
 983                         tv.tv_usec = 0;
 984                 }
 985                 n = select(fd+1, NULL, NULL, &fds, &tv);
 986                 gettimeofday(&end, NULL);
 987                 end.tv_sec -= start.tv_sec;
 988                 *msec -= (end.tv_sec * 1000 + end.tv_usec/1000
 989                           - start.tv_usec/1000) + 1;
 990         }
 991         return n;
 992 }