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                         /* assume this is a stale reference to a hot
 277                          * removed device
 278                          */
 279                         free(dev);
 280                         continue;
 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                 strcpy(dbase, "block/device/state");
 286                 if (load_sys(fname, buf) == 0 &&
 287                     strncmp(buf, "offline", 7) == 0) {
 288                         free(dev);
 289                         continue;
 290                 }
 291
 292                 /* finally add this disk to the array */
 293                 dev->next = sra->devs;
 294                 sra->devs = dev;
 295
 296                 if (options & GET_OFFSET) {
 297                         strcpy(dbase, "offset");
 298                         if (load_sys(fname, buf))
 299                                 goto abort;
 300                         dev->data_offset = strtoull(buf, NULL, 0);
 301                 }
 302                 if (options & GET_SIZE) {
 303                         strcpy(dbase, "size");
 304                         if (load_sys(fname, buf))
 305                                 goto abort;
 306                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 307                 }
 308                 if (options & GET_STATE) {
 309                         dev->disk.state = 0;
 310                         strcpy(dbase, "state");
 311                         if (load_sys(fname, buf))
 312                                 goto abort;
 313                         if (strstr(buf, "in_sync"))
 314                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 315                         if (strstr(buf, "faulty"))
 316                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 317                         if (dev->disk.state == 0)
 318                                 sra->array.spare_disks++;
 319                 }
 320                 if (options & GET_ERROR) {
 321                         strcpy(buf, "errors");
 322                         if (load_sys(fname, buf))
 323                                 goto abort;
 324                         dev->errors = strtoul(buf, NULL, 0);
 325                 }
 326         }
 327         closedir(dir);
 328         return sra;
 329
 330  abort:
 331         if (dir)
 332                 closedir(dir);
 333         sysfs_free(sra);
 334         return NULL;
 335 }
 336
 337 int sysfs_attr_match(const char *attr, const char *str)
 338 {
 339         /* See if attr, read from a sysfs file, matches
 340          * str.  They must either be the same, or attr can
 341          * have a trailing newline or comma
 342          */
 343         while (*attr && *str && *attr == *str) {
 344                 attr++;
 345                 str++;
 346         }
 347
 348         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 349                 return 0;
 350         return 1;
 351 }
 352
 353 int sysfs_match_word(const char *word, char **list)
 354 {
 355         int n;
 356         for (n=0; list[n]; n++)
 357                 if (sysfs_attr_match(word, list[n]))
 358                         break;
 359         return n;
 360 }
 361
 362 unsigned long long get_component_size(int fd)
 363 {
 364         /* Find out the component size of the array.
 365          * We cannot trust GET_ARRAY_INFO ioctl as it's
 366          * size field is only 32bits.
 367          * So look in /sys/block/mdXXX/md/component_size
 368          *
 369          * This returns in units of sectors.
 370          */
 371         struct stat stb;
 372         char fname[50];
 373         int n;
 374         if (fstat(fd, &stb)) return 0;
 375         if (major(stb.st_rdev) != get_mdp_major())
 376                 sprintf(fname, "/sys/block/md%d/md/component_size",
 377                         (int)minor(stb.st_rdev));
 378         else
 379                 sprintf(fname, "/sys/block/md_d%d/md/component_size",
 380                         (int)minor(stb.st_rdev)>>MdpMinorShift);
 381         fd = open(fname, O_RDONLY);
 382         if (fd < 0)
 383                 return 0;
 384         n = read(fd, fname, sizeof(fname));
 385         close(fd);
 386         if (n == sizeof(fname))
 387                 return 0;
 388         fname[n] = 0;
 389         return strtoull(fname, NULL, 10) * 2;
 390 }
 391
 392 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 393                   char *name, char *val)
 394 {
 395         char fname[50];
 396         int n;
 397         int fd;
 398
 399         sprintf(fname, "/sys/block/%s/md/%s/%s",
 400                 sra->sys_name, dev?dev->sys_name:"", name);
 401         fd = open(fname, O_WRONLY);
 402         if (fd < 0)
 403                 return -1;
 404         n = write(fd, val, strlen(val));
 405         close(fd);
 406         if (n != strlen(val)) {
 407                 dprintf(Name ": failed to write '%s' to '%s' (%s)\n",
 408                         val, fname, strerror(errno));
 409                 return -1;
 410         }
 411         return 0;
 412 }
 413
 414 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 415                   char *name, unsigned long long val)
 416 {
 417         char valstr[50];
 418         sprintf(valstr, "%llu", val);
 419         return sysfs_set_str(sra, dev, name, valstr);
 420 }
 421
 422 int sysfs_uevent(struct mdinfo *sra, char *event)
 423 {
 424         char fname[50];
 425         int n;
 426         int fd;
 427
 428         sprintf(fname, "/sys/block/%s/uevent",
 429                 sra->sys_name);
 430         fd = open(fname, O_WRONLY);
 431         if (fd < 0)
 432                 return -1;
 433         n = write(fd, event, strlen(event));
 434         close(fd);
 435         return 0;
 436 }
 437
 438 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 439                        char *name)
 440 {
 441         char fname[50];
 442         int fd;
 443
 444         sprintf(fname, "/sys/block/%s/md/%s/%s",
 445                 sra->sys_name, dev?dev->sys_name:"", name);
 446         fd = open(fname, O_RDWR);
 447         if (fd < 0)
 448                 fd = open(fname, O_RDONLY);
 449         return fd;
 450 }
 451
 452 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 453 {
 454         char buf[50];
 455         int n;
 456         char *ep;
 457
 458         lseek(fd, 0, 0);
 459         n = read(fd, buf, sizeof(buf));
 460         if (n <= 0)
 461                 return -1;
 462         buf[n] = 0;
 463         *val = strtoull(buf, &ep, 0);
 464         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 465                 return -1;
 466         return 0;
 467 }
 468
 469 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 470                        char *name, unsigned long long *val)
 471 {
 472         int n;
 473         int fd;
 474
 475         fd = sysfs_get_fd(sra, dev, name);
 476         if (fd < 0)
 477                 return -1;
 478         n = sysfs_fd_get_ll(fd, val);
 479         close(fd);
 480         return n;
 481 }
 482
 483 int sysfs_fd_get_str(int fd, char *val, int size)
 484 {
 485         int n;
 486
 487         lseek(fd, 0, 0);
 488         n = read(fd, val, size);
 489         if (n <= 0)
 490                 return -1;
 491         val[n] = 0;
 492         return n;
 493 }
 494
 495 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 496                        char *name, char *val, int size)
 497 {
 498         int n;
 499         int fd;
 500
 501         fd = sysfs_get_fd(sra, dev, name);
 502         if (fd < 0)
 503                 return -1;
 504         n = sysfs_fd_get_str(fd, val, size);
 505         close(fd);
 506         return n;
 507 }
 508
 509 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 510 {
 511         unsigned long sec;
 512         unsigned long msec;
 513         char delay[30];
 514
 515         sec = ms / 1000;
 516         msec = ms % 1000;
 517
 518         sprintf(delay, "%ld.%03ld\n", sec, msec);
 519         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 520         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 521 }
 522
 523 int sysfs_set_array(struct mdinfo *info, int vers)
 524 {
 525         int rv = 0;
 526         char ver[100];
 527
 528         ver[0] = 0;
 529         if (info->array.major_version == -1 &&
 530             info->array.minor_version == -2) {
 531                 strcat(strcpy(ver, "external:"), info->text_version);
 532
 533                 if ((vers % 100) < 2 ||
 534                     sysfs_set_str(info, NULL, "metadata_version",
 535                                   ver) < 0) {
 536                         fprintf(stderr, Name ": This kernel does not "
 537                                 "support external metadata.\n");
 538                         return 1;
 539                 }
 540         }
 541         if (info->array.level < 0)
 542                 return 0; /* FIXME */
 543         rv |= sysfs_set_str(info, NULL, "level",
 544                             map_num(pers, info->array.level));
 545         rv |= sysfs_set_num(info, NULL, "raid_disks", info->array.raid_disks);
 546         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 547         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 548         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 549         if (info->custom_array_size) {
 550                 int rc;
 551
 552                 rc = sysfs_set_num(info, NULL, "array_size",
 553                                    info->custom_array_size/2);
 554                 if (rc && errno == ENOENT) {
 555                         fprintf(stderr, Name ": This kernel does not "
 556                                 "have the md/array_size attribute, "
 557                                 "the array may be larger than expected\n");
 558                         rc = 0;
 559                 }
 560                 rv |= rc;
 561         }
 562
 563         if (info->array.level > 0)
 564                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 565         return rv;
 566 }
 567
 568 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 569 {
 570         char dv[PATH_MAX];
 571         char nm[PATH_MAX];
 572         char *dname;
 573         int rv;
 574
 575         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 576         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 577         if (rv)
 578                 return rv;
 579
 580         memset(nm, 0, sizeof(nm));
 581         sprintf(dv, "/sys/dev/block/%d:%d", sd->disk.major, sd->disk.minor);
 582         rv = readlink(dv, nm, sizeof(nm));
 583         if (rv <= 0)
 584                 return -1;
 585         nm[rv] = '\0';
 586         dname = strrchr(nm, '/');
 587         if (dname) dname++;
 588         strcpy(sd->sys_name, "dev-");
 589         strcpy(sd->sys_name+4, dname);
 590
 591         /* test write to see if 'recovery_start' is available */
 592         if (resume && sd->recovery_start < MaxSector &&
 593             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 594                 sysfs_set_str(sra, sd, "state", "remove");
 595                 return -1;
 596         }
 597
 598         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 599         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 600         if (sra->array.level != LEVEL_CONTAINER) {
 601                 if (sd->recovery_start == MaxSector)
 602                         /* This can correctly fail if array isn't started,
 603                          * yet, so just ignore status for now.
 604                          */
 605                         sysfs_set_str(sra, sd, "state", "insync");
 606                 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 607                 if (resume)
 608                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 609         }
 610         return rv;
 611 }
 612
 613 #if 0
 614 int sysfs_disk_to_sg(int fd)
 615 {
 616         /* from an open block device, try find and open its corresponding
 617          * scsi_generic interface
 618          */
 619         struct stat st;
 620         char path[256];
 621         char sg_path[256];
 622         char sg_major_minor[8];
 623         char *c;
 624         DIR *dir;
 625         struct dirent *de;
 626         int major, minor, rv;
 627
 628         if (fstat(fd, &st))
 629                 return -1;
 630
 631         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 632                  major(st.st_rdev), minor(st.st_rdev));
 633
 634         dir = opendir(path);
 635         if (!dir)
 636                 return -1;
 637
 638         de = readdir(dir);
 639         while (de) {
 640                 if (strncmp("scsi_generic:", de->d_name,
 641                             strlen("scsi_generic:")) == 0)
 642                         break;
 643                 de = readdir(dir);
 644         }
 645         closedir(dir);
 646
 647         if (!de)
 648                 return -1;
 649
 650         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 651         fd = open(sg_path, O_RDONLY);
 652         if (fd < 0)
 653                 return fd;
 654
 655         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 656         close(fd);
 657         if (rv < 0)
 658                 return -1;
 659         else
 660                 sg_major_minor[rv - 1] = '\0';
 661
 662         c = strchr(sg_major_minor, ':');
 663         *c = '\0';
 664         c++;
 665         major = strtol(sg_major_minor, NULL, 10);
 666         minor = strtol(c, NULL, 10);
 667         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 668                  (int) getpid(), major, minor);
 669         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 670                         fd = open(path, O_RDONLY);
 671                         unlink(path);
 672                         return fd;
 673         }
 674
 675         return -1;
 676 }
 677 #endif
 678
 679 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 680 {
 681         /* from an open block device, try to retrieve it scsi_id */
 682         struct stat st;
 683         char path[256];
 684         char *c1, *c2;
 685         DIR *dir;
 686         struct dirent *de;
 687
 688         if (fstat(fd, &st))
 689                 return 1;
 690
 691         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 692                  major(st.st_rdev), minor(st.st_rdev));
 693
 694         dir = opendir(path);
 695         if (!dir)
 696                 return 1;
 697
 698         de = readdir(dir);
 699         while (de) {
 700                 if (strncmp("scsi_disk:", de->d_name,
 701                             strlen("scsi_disk:")) == 0)
 702                         break;
 703                 de = readdir(dir);
 704         }
 705         closedir(dir);
 706
 707         if (!de)
 708                 return 1;
 709
 710         c1 = strchr(de->d_name, ':');
 711         c1++;
 712         c2 = strchr(c1, ':');
 713         *c2 = '\0';
 714         *id = strtol(c1, NULL, 10) << 24; /* host */
 715         c1 = c2 + 1;
 716         c2 = strchr(c1, ':');
 717         *c2 = '\0';
 718         *id |= strtol(c1, NULL, 10) << 16; /* channel */
 719         c1 = c2 + 1;
 720         c2 = strchr(c1, ':');
 721         *c2 = '\0';
 722         *id |= strtol(c1, NULL, 10) << 8; /* lun */
 723         c1 = c2 + 1;
 724         *id |= strtol(c1, NULL, 10); /* id */
 725
 726         return 0;
 727 }
 728
 729
 730 int sysfs_unique_holder(int devnum, long rdev)
 731 {
 732         /* Check that devnum is a holder of rdev,
 733          * and is the only holder.
 734          * we should be locked against races by
 735          * an O_EXCL on devnum
 736          */
 737         DIR *dir;
 738         struct dirent *de;
 739         char dirname[100];
 740         char l;
 741         int found = 0;
 742         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 743                 major(rdev), minor(rdev));
 744         dir = opendir(dirname);
 745         errno = ENOENT;
 746         if (!dir)
 747                 return 0;
 748         l = strlen(dirname);
 749         while ((de = readdir(dir)) != NULL) {
 750                 char buf[10];
 751                 int n;
 752                 int mj, mn;
 753                 char c;
 754                 int fd;
 755
 756                 if (de->d_ino == 0)
 757                         continue;
 758                 if (de->d_name[0] == '.')
 759                         continue;
 760                 strcpy(dirname+l, "/");
 761                 strcat(dirname+l, de->d_name);
 762                 strcat(dirname+l, "/dev");
 763                 fd = open(dirname, O_RDONLY);
 764                 if (fd < 0) {
 765                         errno = ENOENT;
 766                         break;
 767                 }
 768                 n = read(fd, buf, sizeof(buf)-1);
 769                 close(fd);
 770                 buf[n] = 0;
 771                 if (sscanf(buf, "%d:%d%c", &mj, &mn, &c) != 3 ||
 772                     c != '\n') {
 773                         errno = ENOENT;
 774                         break;
 775                 }
 776                 if (mj != MD_MAJOR)
 777                         mn = -1-(mn>>6);
 778
 779                 if (devnum != mn) {
 780                         errno = EEXIST;
 781                         break;
 782                 }
 783                 found = 1;
 784         }
 785         closedir(dir);
 786         if (de)
 787                 return 0;
 788         else
 789                 return found;
 790 }
 791
 792 #ifndef MDASSEMBLE
 793
 794 static char *clean_states[] = {
 795         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
 796
 797 int WaitClean(char *dev, int sock, int verbose)
 798 {
 799         int fd;
 800         struct mdinfo *mdi;
 801         int rv = 1;
 802         int devnum;
 803
 804         fd = open(dev, O_RDONLY);
 805         if (fd < 0) {
 806                 if (verbose)
 807                         fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
 808                 return 1;
 809         }
 810
 811         devnum = fd2devnum(fd);
 812         mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
 813         if (!mdi) {
 814                 if (verbose)
 815                         fprintf(stderr, Name ": Failed to read sysfs attributes for "
 816                                 "%s\n", dev);
 817                 close(fd);
 818                 return 0;
 819         }
 820
 821         switch(mdi->array.level) {
 822         case LEVEL_LINEAR:
 823         case LEVEL_MULTIPATH:
 824         case 0:
 825                 /* safemode delay is irrelevant for these levels */
 826                 rv = 0;
 827
 828         }
 829
 830         /* for internal metadata the kernel handles the final clean
 831          * transition, containers can never be dirty
 832          */
 833         if (!is_subarray(mdi->text_version))
 834                 rv = 0;
 835
 836         /* safemode disabled ? */
 837         if (mdi->safe_mode_delay == 0)
 838                 rv = 0;
 839
 840         if (rv) {
 841                 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
 842                 char buf[20];
 843                 fd_set fds;
 844                 struct timeval tm;
 845
 846                 /* minimize the safe_mode_delay and prepare to wait up to 5s
 847                  * for writes to quiesce
 848                  */
 849                 sysfs_set_safemode(mdi, 1);
 850                 tm.tv_sec = 5;
 851                 tm.tv_usec = 0;
 852
 853                 FD_ZERO(&fds);
 854
 855                 /* wait for array_state to be clean */
 856                 while (1) {
 857                         rv = read(state_fd, buf, sizeof(buf));
 858                         if (rv < 0)
 859                                 break;
 860                         if (sysfs_match_word(buf, clean_states) <= 4)
 861                                 break;
 862                         FD_SET(state_fd, &fds);
 863                         rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
 864                         if (rv < 0 && errno != EINTR)
 865                                 break;
 866                         lseek(state_fd, 0, SEEK_SET);
 867                 }
 868                 if (rv < 0)
 869                         rv = 1;
 870                 else if (fping_monitor(sock) == 0 ||
 871                          ping_monitor(mdi->text_version) == 0) {
 872                         /* we need to ping to close the window between array
 873                          * state transitioning to clean and the metadata being
 874                          * marked clean
 875                          */
 876                         rv = 0;
 877                 } else
 878                         rv = 1;
 879                 if (rv && verbose)
 880                         fprintf(stderr, Name ": Error waiting for %s to be clean\n",
 881                                 dev);
 882
 883                 /* restore the original safe_mode_delay */
 884                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
 885                 close(state_fd);
 886         }
 887
 888         sysfs_free(mdi);
 889         close(fd);
 890
 891         return rv;
 892 }
 893 #endif /* MDASSEMBLE */