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 int load_sys(char *path, char *buf)
  31 {
  32         int fd = open(path, O_RDONLY);
  33         int n;
  34         if (fd < 0)
  35                 return -1;
  36         n = read(fd, buf, 1024);
  37         close(fd);
  38         if (n <0 || n >= 1024)
  39                 return -1;
  40         buf[n] = 0;
  41         if (n && buf[n-1] == '\n')
  42                 buf[n-1] = 0;
  43         return 0;
  44 }
  45
  46 void sysfs_free(struct mdinfo *sra)
  47 {
  48         while (sra) {
  49                 struct mdinfo *sra2 = sra->next;
  50                 while (sra->devs) {
  51                         struct mdinfo *d = sra->devs;
  52                         sra->devs = d->next;
  53                         free(d);
  54                 }
  55                 free(sra);
  56                 sra = sra2;
  57         }
  58 }
  59
  60 int sysfs_open(int devnum, char *devname, char *attr)
  61 {
  62         char fname[50];
  63         int fd;
  64         char *mdname = devnum2devname(devnum);
  65
  66         if (!mdname)
  67                 return -1;
  68
  69         sprintf(fname, "/sys/block/%s/md/", mdname);
  70         if (devname) {
  71                 strcat(fname, devname);
  72                 strcat(fname, "/");
  73         }
  74         strcat(fname, attr);
  75         fd = open(fname, O_RDWR);
  76         if (fd < 0 && errno == EACCES)
  77                 fd = open(fname, O_RDONLY);
  78         free(mdname);
  79         return fd;
  80 }
  81
  82 void sysfs_init(struct mdinfo *mdi, int fd, int devnum)
  83 {
  84         mdi->sys_name[0] = 0;
  85         if (fd >= 0) {
  86                 mdu_version_t vers;
  87                 if (ioctl(fd, RAID_VERSION, &vers) != 0)
  88                         return;
  89                 devnum = fd2devnum(fd);
  90         }
  91         if (devnum == NoMdDev)
  92                 return;
  93         if (devnum >= 0)
  94                 sprintf(mdi->sys_name, "md%d", devnum);
  95         else
  96                 sprintf(mdi->sys_name, "md_d%d",
  97                         -1-devnum);
  98 }
  99
 100
 101 struct mdinfo *sysfs_read(int fd, int devnum, 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;
 109         DIR *dir = NULL;
 110         struct dirent *de;
 111
 112         sra = malloc(sizeof(*sra));
 113         if (sra == NULL)
 114                 return sra;
 115         memset(sra, 0, sizeof(*sra));
 116         sysfs_init(sra, fd, devnum);
 117         if (sra->sys_name[0] == 0) {
 118                 free(sra);
 119                 return NULL;
 120         }
 121
 122         sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
 123         base = fname + strlen(fname);
 124
 125         sra->devs = NULL;
 126         if (options & GET_VERSION) {
 127                 strcpy(base, "metadata_version");
 128                 if (load_sys(fname, buf))
 129                         goto abort;
 130                 if (strncmp(buf, "none", 4) == 0) {
 131                         sra->array.major_version =
 132                                 sra->array.minor_version = -1;
 133                         strcpy(sra->text_version, "");
 134                 } else if (strncmp(buf, "external:", 9) == 0) {
 135                         sra->array.major_version = -1;
 136                         sra->array.minor_version = -2;
 137                         strcpy(sra->text_version, buf+9);
 138                 } else {
 139                         sscanf(buf, "%d.%d",
 140                                &sra->array.major_version,
 141                                &sra->array.minor_version);
 142                         strcpy(sra->text_version, buf);
 143                 }
 144         }
 145         if (options & GET_LEVEL) {
 146                 strcpy(base, "level");
 147                 if (load_sys(fname, buf))
 148                         goto abort;
 149                 sra->array.level = map_name(pers, buf);
 150         }
 151         if (options & GET_LAYOUT) {
 152                 strcpy(base, "layout");
 153                 if (load_sys(fname, buf))
 154                         goto abort;
 155                 sra->array.layout = strtoul(buf, NULL, 0);
 156         }
 157         if (options & GET_DISKS) {
 158                 strcpy(base, "raid_disks");
 159                 if (load_sys(fname, buf))
 160                         goto abort;
 161                 sra->array.raid_disks = strtoul(buf, NULL, 0);
 162         }
 163         if (options & GET_DEGRADED) {
 164                 strcpy(base, "degraded");
 165                 if (load_sys(fname, buf))
 166                         goto abort;
 167                 sra->array.failed_disks = strtoul(buf, NULL, 0);
 168         }
 169         if (options & GET_COMPONENT) {
 170                 strcpy(base, "component_size");
 171                 if (load_sys(fname, buf))
 172                         goto abort;
 173                 sra->component_size = strtoull(buf, NULL, 0);
 174                 /* sysfs reports "K", but we want sectors */
 175                 sra->component_size *= 2;
 176         }
 177         if (options & GET_CHUNK) {
 178                 strcpy(base, "chunk_size");
 179                 if (load_sys(fname, buf))
 180                         goto abort;
 181                 sra->array.chunk_size = strtoul(buf, NULL, 0);
 182         }
 183         if (options & GET_CACHE) {
 184                 strcpy(base, "stripe_cache_size");
 185                 if (load_sys(fname, buf))
 186                         goto abort;
 187                 sra->cache_size = strtoul(buf, NULL, 0);
 188         }
 189         if (options & GET_MISMATCH) {
 190                 strcpy(base, "mismatch_cnt");
 191                 if (load_sys(fname, buf))
 192                         goto abort;
 193                 sra->mismatch_cnt = strtoul(buf, NULL, 0);
 194         }
 195         if (options & GET_SAFEMODE) {
 196                 int scale = 1;
 197                 int dot = 0;
 198                 int i;
 199                 unsigned long msec;
 200                 size_t len;
 201
 202                 strcpy(base, "safe_mode_delay");
 203                 if (load_sys(fname, buf))
 204                         goto abort;
 205
 206                 /* remove a period, and count digits after it */
 207                 len = strlen(buf);
 208                 for (i = 0; i < len; i++) {
 209                         if (dot) {
 210                                 if (isdigit(buf[i])) {
 211                                         buf[i-1] = buf[i];
 212                                         scale *= 10;
 213                                 }
 214                                 buf[i] = 0;
 215                         } else if (buf[i] == '.') {
 216                                 dot=1;
 217                                 buf[i] = 0;
 218                         }
 219                 }
 220                 msec = strtoul(buf, NULL, 10);
 221                 msec = (msec * 1000) / scale;
 222                 sra->safe_mode_delay = msec;
 223         }
 224
 225         if (! (options & GET_DEVS))
 226                 return sra;
 227
 228         /* Get all the devices as well */
 229         *base = 0;
 230         dir = opendir(fname);
 231         if (!dir)
 232                 goto abort;
 233         sra->array.spare_disks = 0;
 234
 235         while ((de = readdir(dir)) != NULL) {
 236                 char *ep;
 237                 if (de->d_ino == 0 ||
 238                     strncmp(de->d_name, "dev-", 4) != 0)
 239                         continue;
 240                 strcpy(base, de->d_name);
 241                 dbase = base + strlen(base);
 242                 *dbase++ = '/';
 243
 244                 dev = malloc(sizeof(*dev));
 245                 if (!dev)
 246                         goto abort;
 247
 248                 /* Always get slot, major, minor */
 249                 strcpy(dbase, "slot");
 250                 if (load_sys(fname, buf)) {
 251                         /* hmm... unable to read 'slot' maybe the device
 252                          * is going away?
 253                          */
 254                         strcpy(dbase, "block");
 255                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 256                             errno != ENAMETOOLONG) {
 257                                 /* ...yup device is gone */
 258                                 free(dev);
 259                                 continue;
 260                         } else {
 261                                 /* slot is unreadable but 'block' link
 262                                  * still intact... something bad is happening
 263                                  * so abort
 264                                  */
 265                                 free(dev);
 266                                 goto abort;
 267                         }
 268
 269                 }
 270                 strcpy(dev->sys_name, de->d_name);
 271                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 272                 if (*ep) dev->disk.raid_disk = -1;
 273
 274                 strcpy(dbase, "block/dev");
 275                 if (load_sys(fname, buf)) {
 276                         free(dev);
 277                         if (options & SKIP_GONE_DEVS)
 278                                 continue;
 279                         else
 280                                 goto abort;
 281                 }
 282                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 283
 284                 /* special case check for block devices that can go 'offline' */
 285                 if (options & SKIP_GONE_DEVS) {
 286                         strcpy(dbase, "block/device/state");
 287                         if (load_sys(fname, buf) == 0 &&
 288                             strncmp(buf, "offline", 7) == 0) {
 289                                 free(dev);
 290                                 continue;
 291                         }
 292                 }
 293
 294                 /* finally add this disk to the array */
 295                 dev->next = sra->devs;
 296                 sra->devs = dev;
 297
 298                 if (options & GET_OFFSET) {
 299                         strcpy(dbase, "offset");
 300                         if (load_sys(fname, buf))
 301                                 goto abort;
 302                         dev->data_offset = strtoull(buf, NULL, 0);
 303                 }
 304                 if (options & GET_SIZE) {
 305                         strcpy(dbase, "size");
 306                         if (load_sys(fname, buf))
 307                                 goto abort;
 308                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 309                 }
 310                 if (options & GET_STATE) {
 311                         dev->disk.state = 0;
 312                         strcpy(dbase, "state");
 313                         if (load_sys(fname, buf))
 314                                 goto abort;
 315                         if (strstr(buf, "in_sync"))
 316                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 317                         if (strstr(buf, "faulty"))
 318                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 319                         if (dev->disk.state == 0)
 320                                 sra->array.spare_disks++;
 321                 }
 322                 if (options & GET_ERROR) {
 323                         strcpy(buf, "errors");
 324                         if (load_sys(fname, buf))
 325                                 goto abort;
 326                         dev->errors = strtoul(buf, NULL, 0);
 327                 }
 328         }
 329         closedir(dir);
 330         return sra;
 331
 332  abort:
 333         if (dir)
 334                 closedir(dir);
 335         sysfs_free(sra);
 336         return NULL;
 337 }
 338
 339 int sysfs_attr_match(const char *attr, const char *str)
 340 {
 341         /* See if attr, read from a sysfs file, matches
 342          * str.  They must either be the same, or attr can
 343          * have a trailing newline or comma
 344          */
 345         while (*attr && *str && *attr == *str) {
 346                 attr++;
 347                 str++;
 348         }
 349
 350         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 351                 return 0;
 352         return 1;
 353 }
 354
 355 int sysfs_match_word(const char *word, char **list)
 356 {
 357         int n;
 358         for (n=0; list[n]; n++)
 359                 if (sysfs_attr_match(word, list[n]))
 360                         break;
 361         return n;
 362 }
 363
 364 unsigned long long get_component_size(int fd)
 365 {
 366         /* Find out the component size of the array.
 367          * We cannot trust GET_ARRAY_INFO ioctl as it's
 368          * size field is only 32bits.
 369          * So look in /sys/block/mdXXX/md/component_size
 370          *
 371          * This returns in units of sectors.
 372          */
 373         struct stat stb;
 374         char fname[50];
 375         int n;
 376         if (fstat(fd, &stb)) return 0;
 377         if (major(stb.st_rdev) != get_mdp_major())
 378                 sprintf(fname, "/sys/block/md%d/md/component_size",
 379                         (int)minor(stb.st_rdev));
 380         else
 381                 sprintf(fname, "/sys/block/md_d%d/md/component_size",
 382                         (int)minor(stb.st_rdev)>>MdpMinorShift);
 383         fd = open(fname, O_RDONLY);
 384         if (fd < 0)
 385                 return 0;
 386         n = read(fd, fname, sizeof(fname));
 387         close(fd);
 388         if (n == sizeof(fname))
 389                 return 0;
 390         fname[n] = 0;
 391         return strtoull(fname, NULL, 10) * 2;
 392 }
 393
 394 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 395                   char *name, char *val)
 396 {
 397         char fname[50];
 398         int n;
 399         int fd;
 400
 401         sprintf(fname, "/sys/block/%s/md/%s/%s",
 402                 sra->sys_name, dev?dev->sys_name:"", name);
 403         fd = open(fname, O_WRONLY);
 404         if (fd < 0)
 405                 return -1;
 406         n = write(fd, val, strlen(val));
 407         close(fd);
 408         if (n != strlen(val)) {
 409                 dprintf(Name ": failed to write '%s' to '%s' (%s)\n",
 410                         val, fname, strerror(errno));
 411                 return -1;
 412         }
 413         return 0;
 414 }
 415
 416 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 417                   char *name, unsigned long long val)
 418 {
 419         char valstr[50];
 420         sprintf(valstr, "%llu", val);
 421         return sysfs_set_str(sra, dev, name, valstr);
 422 }
 423
 424 int sysfs_uevent(struct mdinfo *sra, char *event)
 425 {
 426         char fname[50];
 427         int n;
 428         int fd;
 429
 430         sprintf(fname, "/sys/block/%s/uevent",
 431                 sra->sys_name);
 432         fd = open(fname, O_WRONLY);
 433         if (fd < 0)
 434                 return -1;
 435         n = write(fd, event, strlen(event));
 436         close(fd);
 437         return 0;
 438 }
 439
 440 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 441                        char *name)
 442 {
 443         char fname[50];
 444         int fd;
 445
 446         sprintf(fname, "/sys/block/%s/md/%s/%s",
 447                 sra->sys_name, dev?dev->sys_name:"", name);
 448         fd = open(fname, O_RDWR);
 449         if (fd < 0)
 450                 fd = open(fname, O_RDONLY);
 451         return fd;
 452 }
 453
 454 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 455 {
 456         char buf[50];
 457         int n;
 458         char *ep;
 459
 460         lseek(fd, 0, 0);
 461         n = read(fd, buf, sizeof(buf));
 462         if (n <= 0)
 463                 return -1;
 464         buf[n] = 0;
 465         *val = strtoull(buf, &ep, 0);
 466         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 467                 return -1;
 468         return 0;
 469 }
 470
 471 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 472                        char *name, unsigned long long *val)
 473 {
 474         int n;
 475         int fd;
 476
 477         fd = sysfs_get_fd(sra, dev, name);
 478         if (fd < 0)
 479                 return -1;
 480         n = sysfs_fd_get_ll(fd, val);
 481         close(fd);
 482         return n;
 483 }
 484
 485 int sysfs_fd_get_str(int fd, char *val, int size)
 486 {
 487         int n;
 488
 489         lseek(fd, 0, 0);
 490         n = read(fd, val, size);
 491         if (n <= 0)
 492                 return -1;
 493         val[n] = 0;
 494         return n;
 495 }
 496
 497 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 498                        char *name, char *val, int size)
 499 {
 500         int n;
 501         int fd;
 502
 503         fd = sysfs_get_fd(sra, dev, name);
 504         if (fd < 0)
 505                 return -1;
 506         n = sysfs_fd_get_str(fd, val, size);
 507         close(fd);
 508         return n;
 509 }
 510
 511 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 512 {
 513         unsigned long sec;
 514         unsigned long msec;
 515         char delay[30];
 516
 517         sec = ms / 1000;
 518         msec = ms % 1000;
 519
 520         sprintf(delay, "%ld.%03ld\n", sec, msec);
 521         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 522         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 523 }
 524
 525 int sysfs_set_array(struct mdinfo *info, int vers)
 526 {
 527         int rv = 0;
 528         char ver[100];
 529
 530         ver[0] = 0;
 531         if (info->array.major_version == -1 &&
 532             info->array.minor_version == -2) {
 533                 strcat(strcpy(ver, "external:"), info->text_version);
 534
 535                 if ((vers % 100) < 2 ||
 536                     sysfs_set_str(info, NULL, "metadata_version",
 537                                   ver) < 0) {
 538                         fprintf(stderr, Name ": This kernel does not "
 539                                 "support external metadata.\n");
 540                         return 1;
 541                 }
 542         }
 543         if (info->array.level < 0)
 544                 return 0; /* FIXME */
 545         rv |= sysfs_set_str(info, NULL, "level",
 546                             map_num(pers, info->array.level));
 547         rv |= sysfs_set_num(info, NULL, "raid_disks", info->array.raid_disks);
 548         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 549         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 550         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 551         if (info->custom_array_size) {
 552                 int rc;
 553
 554                 rc = sysfs_set_num(info, NULL, "array_size",
 555                                    info->custom_array_size/2);
 556                 if (rc && errno == ENOENT) {
 557                         fprintf(stderr, Name ": This kernel does not "
 558                                 "have the md/array_size attribute, "
 559                                 "the array may be larger than expected\n");
 560                         rc = 0;
 561                 }
 562                 rv |= rc;
 563         }
 564
 565         if (info->array.level > 0)
 566                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 567         return rv;
 568 }
 569
 570 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int in_sync)
 571 {
 572         char dv[PATH_MAX];
 573         char nm[PATH_MAX];
 574         char *dname;
 575         int rv;
 576
 577         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 578         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 579         if (rv)
 580                 return rv;
 581
 582         memset(nm, 0, sizeof(nm));
 583         sprintf(dv, "/sys/dev/block/%d:%d", sd->disk.major, sd->disk.minor);
 584         rv = readlink(dv, nm, sizeof(nm));
 585         if (rv <= 0)
 586                 return -1;
 587         nm[rv] = '\0';
 588         dname = strrchr(nm, '/');
 589         if (dname) dname++;
 590         strcpy(sd->sys_name, "dev-");
 591         strcpy(sd->sys_name+4, dname);
 592
 593         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 594         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 595         if (sra->array.level != LEVEL_CONTAINER) {
 596                 if (in_sync)
 597                         /* This can correctly fail if array isn't started,
 598                          * yet, so just ignore status for now.
 599                          */
 600                         sysfs_set_str(sra, sd, "state", "in_sync");
 601                 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 602         }
 603         return rv;
 604 }
 605
 606 #if 0
 607 int sysfs_disk_to_sg(int fd)
 608 {
 609         /* from an open block device, try find and open its corresponding
 610          * scsi_generic interface
 611          */
 612         struct stat st;
 613         char path[256];
 614         char sg_path[256];
 615         char sg_major_minor[8];
 616         char *c;
 617         DIR *dir;
 618         struct dirent *de;
 619         int major, minor, rv;
 620
 621         if (fstat(fd, &st))
 622                 return -1;
 623
 624         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 625                  major(st.st_rdev), minor(st.st_rdev));
 626
 627         dir = opendir(path);
 628         if (!dir)
 629                 return -1;
 630
 631         de = readdir(dir);
 632         while (de) {
 633                 if (strncmp("scsi_generic:", de->d_name,
 634                             strlen("scsi_generic:")) == 0)
 635                         break;
 636                 de = readdir(dir);
 637         }
 638         closedir(dir);
 639
 640         if (!de)
 641                 return -1;
 642
 643         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 644         fd = open(sg_path, O_RDONLY);
 645         if (fd < 0)
 646                 return fd;
 647
 648         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 649         close(fd);
 650         if (rv < 0)
 651                 return -1;
 652         else
 653                 sg_major_minor[rv - 1] = '\0';
 654
 655         c = strchr(sg_major_minor, ':');
 656         *c = '\0';
 657         c++;
 658         major = strtol(sg_major_minor, NULL, 10);
 659         minor = strtol(c, NULL, 10);
 660         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 661                  (int) getpid(), major, minor);
 662         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 663                         fd = open(path, O_RDONLY);
 664                         unlink(path);
 665                         return fd;
 666         }
 667
 668         return -1;
 669 }
 670 #endif
 671
 672 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 673 {
 674         /* from an open block device, try to retrieve it scsi_id */
 675         struct stat st;
 676         char path[256];
 677         char *c1, *c2;
 678         DIR *dir;
 679         struct dirent *de;
 680
 681         if (fstat(fd, &st))
 682                 return 1;
 683
 684         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 685                  major(st.st_rdev), minor(st.st_rdev));
 686
 687         dir = opendir(path);
 688         if (!dir)
 689                 return 1;
 690
 691         de = readdir(dir);
 692         while (de) {
 693                 if (strncmp("scsi_disk:", de->d_name,
 694                             strlen("scsi_disk:")) == 0)
 695                         break;
 696                 de = readdir(dir);
 697         }
 698         closedir(dir);
 699
 700         if (!de)
 701                 return 1;
 702
 703         c1 = strchr(de->d_name, ':');
 704         c1++;
 705         c2 = strchr(c1, ':');
 706         *c2 = '\0';
 707         *id = strtol(c1, NULL, 10) << 24; /* host */
 708         c1 = c2 + 1;
 709         c2 = strchr(c1, ':');
 710         *c2 = '\0';
 711         *id |= strtol(c1, NULL, 10) << 16; /* channel */
 712         c1 = c2 + 1;
 713         c2 = strchr(c1, ':');
 714         *c2 = '\0';
 715         *id |= strtol(c1, NULL, 10) << 8; /* lun */
 716         c1 = c2 + 1;
 717         *id |= strtol(c1, NULL, 10); /* id */
 718
 719         return 0;
 720 }
 721
 722
 723 int sysfs_unique_holder(int devnum, long rdev)
 724 {
 725         /* Check that devnum is a holder of rdev,
 726          * and is the only holder.
 727          * we should be locked against races by
 728          * an O_EXCL on devnum
 729          */
 730         DIR *dir;
 731         struct dirent *de;
 732         char dirname[100];
 733         char l;
 734         int found = 0;
 735         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 736                 major(rdev), minor(rdev));
 737         dir = opendir(dirname);
 738         errno = ENOENT;
 739         if (!dir)
 740                 return 0;
 741         l = strlen(dirname);
 742         while ((de = readdir(dir)) != NULL) {
 743                 char buf[10];
 744                 int n;
 745                 int mj, mn;
 746                 char c;
 747                 int fd;
 748
 749                 if (de->d_ino == 0)
 750                         continue;
 751                 if (de->d_name[0] == '.')
 752                         continue;
 753                 strcpy(dirname+l, "/");
 754                 strcat(dirname+l, de->d_name);
 755                 strcat(dirname+l, "/dev");
 756                 fd = open(dirname, O_RDONLY);
 757                 if (fd < 0) {
 758                         errno = ENOENT;
 759                         break;
 760                 }
 761                 n = read(fd, buf, sizeof(buf)-1);
 762                 close(fd);
 763                 buf[n] = 0;
 764                 if (sscanf(buf, "%d:%d%c", &mj, &mn, &c) != 3 ||
 765                     c != '\n') {
 766                         errno = ENOENT;
 767                         break;
 768                 }
 769                 if (mj != MD_MAJOR)
 770                         mn = -1-(mn>>6);
 771
 772                 if (devnum != mn) {
 773                         errno = EEXIST;
 774                         break;
 775                 }
 776                 found = 1;
 777         }
 778         closedir(dir);
 779         if (de)
 780                 return 0;
 781         else
 782                 return found;
 783 }
 784
 785 #ifndef MDASSEMBLE
 786
 787 static char *clean_states[] = {
 788         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
 789
 790 int WaitClean(char *dev, int sock, int verbose)
 791 {
 792         int fd;
 793         struct mdinfo *mdi;
 794         int rv = 1;
 795         int devnum;
 796
 797         fd = open(dev, O_RDONLY);
 798         if (fd < 0) {
 799                 if (verbose)
 800                         fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
 801                 return 1;
 802         }
 803
 804         devnum = fd2devnum(fd);
 805         mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
 806         if (!mdi) {
 807                 if (verbose)
 808                         fprintf(stderr, Name ": Failed to read sysfs attributes for "
 809                                 "%s\n", dev);
 810                 close(fd);
 811                 return 0;
 812         }
 813
 814         switch(mdi->array.level) {
 815         case LEVEL_LINEAR:
 816         case LEVEL_MULTIPATH:
 817         case 0:
 818                 /* safemode delay is irrelevant for these levels */
 819                 rv = 0;
 820
 821         }
 822
 823         /* for internal metadata the kernel handles the final clean
 824          * transition, containers can never be dirty
 825          */
 826         if (!is_subarray(mdi->text_version))
 827                 rv = 0;
 828
 829         /* safemode disabled ? */
 830         if (mdi->safe_mode_delay == 0)
 831                 rv = 0;
 832
 833         if (rv) {
 834                 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
 835                 char buf[20];
 836                 fd_set fds;
 837                 struct timeval tm;
 838
 839                 /* minimize the safe_mode_delay and prepare to wait up to 5s
 840                  * for writes to quiesce
 841                  */
 842                 sysfs_set_safemode(mdi, 1);
 843                 tm.tv_sec = 5;
 844                 tm.tv_usec = 0;
 845
 846                 /* give mdmon a chance to checkpoint resync */
 847                 sysfs_set_str(mdi, NULL, "sync_action", "idle");
 848
 849                 FD_ZERO(&fds);
 850
 851                 /* wait for array_state to be clean */
 852                 while (1) {
 853                         rv = read(state_fd, buf, sizeof(buf));
 854                         if (rv < 0)
 855                                 break;
 856                         if (sysfs_match_word(buf, clean_states) <= 4)
 857                                 break;
 858                         FD_SET(state_fd, &fds);
 859                         rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
 860                         if (rv < 0 && errno != EINTR)
 861                                 break;
 862                         lseek(state_fd, 0, SEEK_SET);
 863                 }
 864                 if (rv < 0)
 865                         rv = 1;
 866                 else if (fping_monitor(sock) == 0 ||
 867                          ping_monitor(mdi->text_version) == 0) {
 868                         /* we need to ping to close the window between array
 869                          * state transitioning to clean and the metadata being
 870                          * marked clean
 871                          */
 872                         rv = 0;
 873                 } else
 874                         rv = 1;
 875                 if (rv && verbose)
 876                         fprintf(stderr, Name ": Error waiting for %s to be clean\n",
 877                                 dev);
 878
 879                 /* restore the original safe_mode_delay */
 880                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
 881                 close(state_fd);
 882         }
 883
 884         sysfs_free(mdi);
 885         close(fd);
 886
 887         return rv;
 888 }
 889 #endif /* MDASSEMBLE */