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         fmt_devname(mdi->sys_name, devnum);
  94 }
  95
  96
  97 struct mdinfo *sysfs_read(int fd, int devnum, unsigned long options)
  98 {
  99         char fname[PATH_MAX];
 100         char buf[PATH_MAX];
 101         char *base;
 102         char *dbase;
 103         struct mdinfo *sra;
 104         struct mdinfo *dev;
 105         DIR *dir = NULL;
 106         struct dirent *de;
 107
 108         sra = malloc(sizeof(*sra));
 109         if (sra == NULL)
 110                 return sra;
 111         memset(sra, 0, sizeof(*sra));
 112         sysfs_init(sra, fd, devnum);
 113         if (sra->sys_name[0] == 0) {
 114                 free(sra);
 115                 return NULL;
 116         }
 117
 118         sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
 119         base = fname + strlen(fname);
 120
 121         sra->devs = NULL;
 122         if (options & GET_VERSION) {
 123                 strcpy(base, "metadata_version");
 124                 if (load_sys(fname, buf))
 125                         goto abort;
 126                 if (strncmp(buf, "none", 4) == 0) {
 127                         sra->array.major_version =
 128                                 sra->array.minor_version = -1;
 129                         strcpy(sra->text_version, "");
 130                 } else if (strncmp(buf, "external:", 9) == 0) {
 131                         sra->array.major_version = -1;
 132                         sra->array.minor_version = -2;
 133                         strcpy(sra->text_version, buf+9);
 134                 } else {
 135                         sscanf(buf, "%d.%d",
 136                                &sra->array.major_version,
 137                                &sra->array.minor_version);
 138                         strcpy(sra->text_version, buf);
 139                 }
 140         }
 141         if (options & GET_LEVEL) {
 142                 strcpy(base, "level");
 143                 if (load_sys(fname, buf))
 144                         goto abort;
 145                 sra->array.level = map_name(pers, buf);
 146         }
 147         if (options & GET_LAYOUT) {
 148                 strcpy(base, "layout");
 149                 if (load_sys(fname, buf))
 150                         goto abort;
 151                 sra->array.layout = strtoul(buf, NULL, 0);
 152         }
 153         if (options & GET_DISKS) {
 154                 strcpy(base, "raid_disks");
 155                 if (load_sys(fname, buf))
 156                         goto abort;
 157                 sra->array.raid_disks = strtoul(buf, NULL, 0);
 158         }
 159         if (options & GET_DEGRADED) {
 160                 strcpy(base, "degraded");
 161                 if (load_sys(fname, buf))
 162                         goto abort;
 163                 sra->array.failed_disks = strtoul(buf, NULL, 0);
 164         }
 165         if (options & GET_COMPONENT) {
 166                 strcpy(base, "component_size");
 167                 if (load_sys(fname, buf))
 168                         goto abort;
 169                 sra->component_size = strtoull(buf, NULL, 0);
 170                 /* sysfs reports "K", but we want sectors */
 171                 sra->component_size *= 2;
 172         }
 173         if (options & GET_CHUNK) {
 174                 strcpy(base, "chunk_size");
 175                 if (load_sys(fname, buf))
 176                         goto abort;
 177                 sra->array.chunk_size = strtoul(buf, NULL, 0);
 178         }
 179         if (options & GET_CACHE) {
 180                 strcpy(base, "stripe_cache_size");
 181                 if (load_sys(fname, buf))
 182                         goto abort;
 183                 sra->cache_size = strtoul(buf, NULL, 0);
 184         }
 185         if (options & GET_MISMATCH) {
 186                 strcpy(base, "mismatch_cnt");
 187                 if (load_sys(fname, buf))
 188                         goto abort;
 189                 sra->mismatch_cnt = strtoul(buf, NULL, 0);
 190         }
 191         if (options & GET_SAFEMODE) {
 192                 int scale = 1;
 193                 int dot = 0;
 194                 unsigned i;
 195                 unsigned long msec;
 196                 size_t len;
 197
 198                 strcpy(base, "safe_mode_delay");
 199                 if (load_sys(fname, buf))
 200                         goto abort;
 201
 202                 /* remove a period, and count digits after it */
 203                 len = strlen(buf);
 204                 for (i = 0; i < len; i++) {
 205                         if (dot) {
 206                                 if (isdigit(buf[i])) {
 207                                         buf[i-1] = buf[i];
 208                                         scale *= 10;
 209                                 }
 210                                 buf[i] = 0;
 211                         } else if (buf[i] == '.') {
 212                                 dot=1;
 213                                 buf[i] = 0;
 214                         }
 215                 }
 216                 msec = strtoul(buf, NULL, 10);
 217                 msec = (msec * 1000) / scale;
 218                 sra->safe_mode_delay = msec;
 219         }
 220
 221         if (! (options & GET_DEVS))
 222                 return sra;
 223
 224         /* Get all the devices as well */
 225         *base = 0;
 226         dir = opendir(fname);
 227         if (!dir)
 228                 goto abort;
 229         sra->array.spare_disks = 0;
 230
 231         while ((de = readdir(dir)) != NULL) {
 232                 char *ep;
 233                 if (de->d_ino == 0 ||
 234                     strncmp(de->d_name, "dev-", 4) != 0)
 235                         continue;
 236                 strcpy(base, de->d_name);
 237                 dbase = base + strlen(base);
 238                 *dbase++ = '/';
 239
 240                 dev = malloc(sizeof(*dev));
 241                 if (!dev)
 242                         goto abort;
 243
 244                 /* Always get slot, major, minor */
 245                 strcpy(dbase, "slot");
 246                 if (load_sys(fname, buf)) {
 247                         /* hmm... unable to read 'slot' maybe the device
 248                          * is going away?
 249                          */
 250                         strcpy(dbase, "block");
 251                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 252                             errno != ENAMETOOLONG) {
 253                                 /* ...yup device is gone */
 254                                 free(dev);
 255                                 continue;
 256                         } else {
 257                                 /* slot is unreadable but 'block' link
 258                                  * still intact... something bad is happening
 259                                  * so abort
 260                                  */
 261                                 free(dev);
 262                                 goto abort;
 263                         }
 264
 265                 }
 266                 strcpy(dev->sys_name, de->d_name);
 267                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 268                 if (*ep) dev->disk.raid_disk = -1;
 269
 270                 strcpy(dbase, "block/dev");
 271                 if (load_sys(fname, buf)) {
 272                         /* assume this is a stale reference to a hot
 273                          * removed device
 274                          */
 275                         free(dev);
 276                         continue;
 277                 }
 278                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 279
 280                 /* special case check for block devices that can go 'offline' */
 281                 strcpy(dbase, "block/device/state");
 282                 if (load_sys(fname, buf) == 0 &&
 283                     strncmp(buf, "offline", 7) == 0) {
 284                         free(dev);
 285                         continue;
 286                 }
 287
 288                 /* finally add this disk to the array */
 289                 dev->next = sra->devs;
 290                 sra->devs = dev;
 291
 292                 if (options & GET_OFFSET) {
 293                         strcpy(dbase, "offset");
 294                         if (load_sys(fname, buf))
 295                                 goto abort;
 296                         dev->data_offset = strtoull(buf, NULL, 0);
 297                 }
 298                 if (options & GET_SIZE) {
 299                         strcpy(dbase, "size");
 300                         if (load_sys(fname, buf))
 301                                 goto abort;
 302                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 303                 }
 304                 if (options & GET_STATE) {
 305                         dev->disk.state = 0;
 306                         strcpy(dbase, "state");
 307                         if (load_sys(fname, buf))
 308                                 goto abort;
 309                         if (strstr(buf, "in_sync"))
 310                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 311                         if (strstr(buf, "faulty"))
 312                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 313                         if (dev->disk.state == 0)
 314                                 sra->array.spare_disks++;
 315                 }
 316                 if (options & GET_ERROR) {
 317                         strcpy(buf, "errors");
 318                         if (load_sys(fname, buf))
 319                                 goto abort;
 320                         dev->errors = strtoul(buf, NULL, 0);
 321                 }
 322         }
 323         closedir(dir);
 324         return sra;
 325
 326  abort:
 327         if (dir)
 328                 closedir(dir);
 329         sysfs_free(sra);
 330         return NULL;
 331 }
 332
 333 int sysfs_attr_match(const char *attr, const char *str)
 334 {
 335         /* See if attr, read from a sysfs file, matches
 336          * str.  They must either be the same, or attr can
 337          * have a trailing newline or comma
 338          */
 339         while (*attr && *str && *attr == *str) {
 340                 attr++;
 341                 str++;
 342         }
 343
 344         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 345                 return 0;
 346         return 1;
 347 }
 348
 349 int sysfs_match_word(const char *word, char **list)
 350 {
 351         int n;
 352         for (n=0; list[n]; n++)
 353                 if (sysfs_attr_match(word, list[n]))
 354                         break;
 355         return n;
 356 }
 357
 358 unsigned long long get_component_size(int fd)
 359 {
 360         /* Find out the component size of the array.
 361          * We cannot trust GET_ARRAY_INFO ioctl as it's
 362          * size field is only 32bits.
 363          * So look in /sys/block/mdXXX/md/component_size
 364          *
 365          * This returns in units of sectors.
 366          */
 367         struct stat stb;
 368         char fname[50];
 369         int n;
 370         if (fstat(fd, &stb)) return 0;
 371         if (major(stb.st_rdev) != (unsigned)get_mdp_major())
 372                 sprintf(fname, "/sys/block/md%d/md/component_size",
 373                         (int)minor(stb.st_rdev));
 374         else
 375                 sprintf(fname, "/sys/block/md_d%d/md/component_size",
 376                         (int)minor(stb.st_rdev)>>MdpMinorShift);
 377         fd = open(fname, O_RDONLY);
 378         if (fd < 0)
 379                 return 0;
 380         n = read(fd, fname, sizeof(fname));
 381         close(fd);
 382         if (n == sizeof(fname))
 383                 return 0;
 384         fname[n] = 0;
 385         return strtoull(fname, NULL, 10) * 2;
 386 }
 387
 388 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 389                   char *name, char *val)
 390 {
 391         char fname[50];
 392         unsigned int n;
 393         int fd;
 394
 395         sprintf(fname, "/sys/block/%s/md/%s/%s",
 396                 sra->sys_name, dev?dev->sys_name:"", name);
 397         fd = open(fname, O_WRONLY);
 398         if (fd < 0)
 399                 return -1;
 400         n = write(fd, val, strlen(val));
 401         close(fd);
 402         if (n != strlen(val)) {
 403                 dprintf(Name ": failed to write '%s' to '%s' (%s)\n",
 404                         val, fname, strerror(errno));
 405                 return -1;
 406         }
 407         return 0;
 408 }
 409
 410 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 411                   char *name, unsigned long long val)
 412 {
 413         char valstr[50];
 414         sprintf(valstr, "%llu", val);
 415         return sysfs_set_str(sra, dev, name, valstr);
 416 }
 417
 418 int sysfs_uevent(struct mdinfo *sra, char *event)
 419 {
 420         char fname[50];
 421         int n;
 422         int fd;
 423
 424         sprintf(fname, "/sys/block/%s/uevent",
 425                 sra->sys_name);
 426         fd = open(fname, O_WRONLY);
 427         if (fd < 0)
 428                 return -1;
 429         n = write(fd, event, strlen(event));
 430         close(fd);
 431         return 0;
 432 }
 433
 434 int sysfs_attribute_available(struct mdinfo *sra, struct mdinfo *dev, char *name)
 435 {
 436         char fname[50];
 437         struct stat st;
 438
 439         sprintf(fname, "/sys/block/%s/md/%s/%s",
 440                 sra->sys_name, dev?dev->sys_name:"", name);
 441
 442         return stat(fname, &st) == 0;
 443 }
 444
 445 int sysfs_get_fd(struct mdinfo *sra, struct mdinfo *dev,
 446                        char *name)
 447 {
 448         char fname[50];
 449         int fd;
 450
 451         sprintf(fname, "/sys/block/%s/md/%s/%s",
 452                 sra->sys_name, dev?dev->sys_name:"", name);
 453         fd = open(fname, O_RDWR);
 454         if (fd < 0)
 455                 fd = open(fname, O_RDONLY);
 456         return fd;
 457 }
 458
 459 int sysfs_fd_get_ll(int fd, unsigned long long *val)
 460 {
 461         char buf[50];
 462         int n;
 463         char *ep;
 464
 465         lseek(fd, 0, 0);
 466         n = read(fd, buf, sizeof(buf));
 467         if (n <= 0)
 468                 return -1;
 469         buf[n] = 0;
 470         *val = strtoull(buf, &ep, 0);
 471         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 472                 return -1;
 473         return 0;
 474 }
 475
 476 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 477                        char *name, unsigned long long *val)
 478 {
 479         int n;
 480         int fd;
 481
 482         fd = sysfs_get_fd(sra, dev, name);
 483         if (fd < 0)
 484                 return -1;
 485         n = sysfs_fd_get_ll(fd, val);
 486         close(fd);
 487         return n;
 488 }
 489
 490 int sysfs_fd_get_str(int fd, char *val, int size)
 491 {
 492         int n;
 493
 494         lseek(fd, 0, 0);
 495         n = read(fd, val, size);
 496         if (n <= 0)
 497                 return -1;
 498         val[n] = 0;
 499         return n;
 500 }
 501
 502 int sysfs_get_str(struct mdinfo *sra, struct mdinfo *dev,
 503                        char *name, char *val, int size)
 504 {
 505         int n;
 506         int fd;
 507
 508         fd = sysfs_get_fd(sra, dev, name);
 509         if (fd < 0)
 510                 return -1;
 511         n = sysfs_fd_get_str(fd, val, size);
 512         close(fd);
 513         return n;
 514 }
 515
 516 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 517 {
 518         unsigned long sec;
 519         unsigned long msec;
 520         char delay[30];
 521
 522         sec = ms / 1000;
 523         msec = ms % 1000;
 524
 525         sprintf(delay, "%ld.%03ld\n", sec, msec);
 526         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 527         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 528 }
 529
 530 int sysfs_set_array(struct mdinfo *info, int vers)
 531 {
 532         int rv = 0;
 533         char ver[100];
 534
 535         ver[0] = 0;
 536         if (info->array.major_version == -1 &&
 537             info->array.minor_version == -2) {
 538                 strcat(strcpy(ver, "external:"), info->text_version);
 539
 540                 if ((vers % 100) < 2 ||
 541                     sysfs_set_str(info, NULL, "metadata_version",
 542                                   ver) < 0) {
 543                         fprintf(stderr, Name ": This kernel does not "
 544                                 "support external metadata.\n");
 545                         return 1;
 546                 }
 547         }
 548         if (info->array.level < 0)
 549                 return 0; /* FIXME */
 550         rv |= sysfs_set_str(info, NULL, "level",
 551                             map_num(pers, info->array.level));
 552         rv |= sysfs_set_num(info, NULL, "raid_disks", info->array.raid_disks);
 553         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 554         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 555         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 556         if (info->custom_array_size) {
 557                 int rc;
 558
 559                 rc = sysfs_set_num(info, NULL, "array_size",
 560                                    info->custom_array_size/2);
 561                 if (rc && errno == ENOENT) {
 562                         fprintf(stderr, Name ": This kernel does not "
 563                                 "have the md/array_size attribute, "
 564                                 "the array may be larger than expected\n");
 565                         rc = 0;
 566                 }
 567                 rv |= rc;
 568         }
 569
 570         if (info->array.level > 0)
 571                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 572
 573         if (info->reshape_active) {
 574                 rv |= sysfs_set_num(info, NULL, "reshape_position",
 575                                     info->reshape_progress);
 576                 rv |= sysfs_set_num(info, NULL, "chunk_size", info->new_chunk);
 577                 /* Don't set layout or raid_disks here as they require some
 578                  * analysis and are set by reshape_array
 579                  */
 580         }
 581         return rv;
 582 }
 583
 584 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd, int resume)
 585 {
 586         char dv[PATH_MAX];
 587         char nm[PATH_MAX];
 588         char *dname;
 589         int rv;
 590
 591         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 592         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 593         if (rv)
 594                 return rv;
 595
 596         memset(nm, 0, sizeof(nm));
 597         sprintf(dv, "/sys/dev/block/%d:%d", sd->disk.major, sd->disk.minor);
 598         rv = readlink(dv, nm, sizeof(nm));
 599         if (rv <= 0)
 600                 return -1;
 601         nm[rv] = '\0';
 602         dname = strrchr(nm, '/');
 603         if (dname) dname++;
 604         strcpy(sd->sys_name, "dev-");
 605         strcpy(sd->sys_name+4, dname);
 606
 607         /* test write to see if 'recovery_start' is available */
 608         if (resume && sd->recovery_start < MaxSector &&
 609             sysfs_set_num(sra, sd, "recovery_start", 0)) {
 610                 sysfs_set_str(sra, sd, "state", "remove");
 611                 return -1;
 612         }
 613
 614         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 615         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 616         if (sra->array.level != LEVEL_CONTAINER) {
 617                 if (sd->recovery_start == MaxSector)
 618                         /* This can correctly fail if array isn't started,
 619                          * yet, so just ignore status for now.
 620                          */
 621                         sysfs_set_str(sra, sd, "state", "insync");
 622                 if (sd->disk.raid_disk >= 0)
 623                         rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 624                 if (resume)
 625                         sysfs_set_num(sra, sd, "recovery_start", sd->recovery_start);
 626         }
 627         return rv;
 628 }
 629
 630 #if 0
 631 int sysfs_disk_to_sg(int fd)
 632 {
 633         /* from an open block device, try find and open its corresponding
 634          * scsi_generic interface
 635          */
 636         struct stat st;
 637         char path[256];
 638         char sg_path[256];
 639         char sg_major_minor[8];
 640         char *c;
 641         DIR *dir;
 642         struct dirent *de;
 643         int major, minor, rv;
 644
 645         if (fstat(fd, &st))
 646                 return -1;
 647
 648         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 649                  major(st.st_rdev), minor(st.st_rdev));
 650
 651         dir = opendir(path);
 652         if (!dir)
 653                 return -1;
 654
 655         de = readdir(dir);
 656         while (de) {
 657                 if (strncmp("scsi_generic:", de->d_name,
 658                             strlen("scsi_generic:")) == 0)
 659                         break;
 660                 de = readdir(dir);
 661         }
 662         closedir(dir);
 663
 664         if (!de)
 665                 return -1;
 666
 667         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 668         fd = open(sg_path, O_RDONLY);
 669         if (fd < 0)
 670                 return fd;
 671
 672         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 673         close(fd);
 674         if (rv < 0)
 675                 return -1;
 676         else
 677                 sg_major_minor[rv - 1] = '\0';
 678
 679         c = strchr(sg_major_minor, ':');
 680         *c = '\0';
 681         c++;
 682         major = strtol(sg_major_minor, NULL, 10);
 683         minor = strtol(c, NULL, 10);
 684         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 685                  (int) getpid(), major, minor);
 686         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 687                         fd = open(path, O_RDONLY);
 688                         unlink(path);
 689                         return fd;
 690         }
 691
 692         return -1;
 693 }
 694 #endif
 695
 696 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 697 {
 698         /* from an open block device, try to retrieve it scsi_id */
 699         struct stat st;
 700         char path[256];
 701         char *c1, *c2;
 702         DIR *dir;
 703         struct dirent *de;
 704
 705         if (fstat(fd, &st))
 706                 return 1;
 707
 708         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device/scsi_device",
 709                  major(st.st_rdev), minor(st.st_rdev));
 710
 711         dir = opendir(path);
 712         if (!dir)
 713                 return 1;
 714
 715         de = readdir(dir);
 716         while (de) {
 717                 if (strchr(de->d_name, ':'))
 718                         break;
 719                 de = readdir(dir);
 720         }
 721         closedir(dir);
 722
 723         if (!de)
 724                 return 1;
 725
 726         c1 = de->d_name;
 727         c2 = strchr(c1, ':');
 728         *c2 = '\0';
 729         *id = strtol(c1, NULL, 10) << 24; /* host */
 730         c1 = c2 + 1;
 731         c2 = strchr(c1, ':');
 732         *c2 = '\0';
 733         *id |= strtol(c1, NULL, 10) << 16; /* bus */
 734         c1 = c2 + 1;
 735         c2 = strchr(c1, ':');
 736         *c2 = '\0';
 737         *id |= strtol(c1, NULL, 10) << 8; /* target */
 738         c1 = c2 + 1;
 739         *id |= strtol(c1, NULL, 10); /* lun */
 740
 741         return 0;
 742 }
 743
 744
 745 int sysfs_unique_holder(int devnum, long rdev)
 746 {
 747         /* Check that devnum is a holder of rdev,
 748          * and is the only holder.
 749          * we should be locked against races by
 750          * an O_EXCL on devnum
 751          */
 752         DIR *dir;
 753         struct dirent *de;
 754         char dirname[100];
 755         char l;
 756         int found = 0;
 757         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 758                 major(rdev), minor(rdev));
 759         dir = opendir(dirname);
 760         errno = ENOENT;
 761         if (!dir)
 762                 return 0;
 763         l = strlen(dirname);
 764         while ((de = readdir(dir)) != NULL) {
 765                 char buf[10];
 766                 int n;
 767                 int mj, mn;
 768                 char c;
 769                 int fd;
 770
 771                 if (de->d_ino == 0)
 772                         continue;
 773                 if (de->d_name[0] == '.')
 774                         continue;
 775                 strcpy(dirname+l, "/");
 776                 strcat(dirname+l, de->d_name);
 777                 strcat(dirname+l, "/dev");
 778                 fd = open(dirname, O_RDONLY);
 779                 if (fd < 0) {
 780                         errno = ENOENT;
 781                         break;
 782                 }
 783                 n = read(fd, buf, sizeof(buf)-1);
 784                 close(fd);
 785                 buf[n] = 0;
 786                 if (sscanf(buf, "%d:%d%c", &mj, &mn, &c) != 3 ||
 787                     c != '\n') {
 788                         errno = ENOENT;
 789                         break;
 790                 }
 791                 if (mj != MD_MAJOR)
 792                         mn = -1-(mn>>6);
 793
 794                 if (devnum != mn) {
 795                         errno = EEXIST;
 796                         break;
 797                 }
 798                 found = 1;
 799         }
 800         closedir(dir);
 801         if (de)
 802                 return 0;
 803         else
 804                 return found;
 805 }
 806
 807 int sysfs_freeze_array(struct mdinfo *sra)
 808 {
 809         /* Try to freeze resync/rebuild on this array/container.
 810          * Return -1 if the array is busy,
 811          * return -2 container cannot be frozen,
 812          * return 0 if this kernel doesn't support 'frozen'
 813          * return 1 if it worked.
 814          */
 815         char buf[20];
 816
 817         if (!sysfs_attribute_available(sra, NULL, "sync_action"))
 818                 return 1; /* no sync_action == frozen */
 819         if (sysfs_get_str(sra, NULL, "sync_action", buf, 20) <= 0)
 820                 return 0;
 821         if (strcmp(buf, "idle\n") != 0 &&
 822             strcmp(buf, "frozen\n") != 0)
 823                 return -1;
 824         if (sysfs_set_str(sra, NULL, "sync_action", "frozen") < 0)
 825                 return 0;
 826         return 1;
 827 }
 828
 829 #ifndef MDASSEMBLE
 830
 831 static char *clean_states[] = {
 832         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
 833
 834 int WaitClean(char *dev, int sock, int verbose)
 835 {
 836         int fd;
 837         struct mdinfo *mdi;
 838         int rv = 1;
 839         int devnum;
 840
 841         fd = open(dev, O_RDONLY);
 842         if (fd < 0) {
 843                 if (verbose)
 844                         fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
 845                 return 1;
 846         }
 847
 848         devnum = fd2devnum(fd);
 849         mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
 850         if (!mdi) {
 851                 if (verbose)
 852                         fprintf(stderr, Name ": Failed to read sysfs attributes for "
 853                                 "%s\n", dev);
 854                 close(fd);
 855                 return 0;
 856         }
 857
 858         switch(mdi->array.level) {
 859         case LEVEL_LINEAR:
 860         case LEVEL_MULTIPATH:
 861         case 0:
 862                 /* safemode delay is irrelevant for these levels */
 863                 rv = 0;
 864
 865         }
 866
 867         /* for internal metadata the kernel handles the final clean
 868          * transition, containers can never be dirty
 869          */
 870         if (!is_subarray(mdi->text_version))
 871                 rv = 0;
 872
 873         /* safemode disabled ? */
 874         if (mdi->safe_mode_delay == 0)
 875                 rv = 0;
 876
 877         if (rv) {
 878                 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
 879                 char buf[20];
 880                 fd_set fds;
 881                 struct timeval tm;
 882
 883                 /* minimize the safe_mode_delay and prepare to wait up to 5s
 884                  * for writes to quiesce
 885                  */
 886                 sysfs_set_safemode(mdi, 1);
 887                 tm.tv_sec = 5;
 888                 tm.tv_usec = 0;
 889
 890                 FD_ZERO(&fds);
 891
 892                 /* wait for array_state to be clean */
 893                 while (1) {
 894                         rv = read(state_fd, buf, sizeof(buf));
 895                         if (rv < 0)
 896                                 break;
 897                         if (sysfs_match_word(buf, clean_states) <= 4)
 898                                 break;
 899                         FD_SET(state_fd, &fds);
 900                         rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
 901                         if (rv < 0 && errno != EINTR)
 902                                 break;
 903                         lseek(state_fd, 0, SEEK_SET);
 904                 }
 905                 if (rv < 0)
 906                         rv = 1;
 907                 else if (fping_monitor(sock) == 0 ||
 908                          ping_monitor(mdi->text_version) == 0) {
 909                         /* we need to ping to close the window between array
 910                          * state transitioning to clean and the metadata being
 911                          * marked clean
 912                          */
 913                         rv = 0;
 914                 } else
 915                         rv = 1;
 916                 if (rv && verbose)
 917                         fprintf(stderr, Name ": Error waiting for %s to be clean\n",
 918                                 dev);
 919
 920                 /* restore the original safe_mode_delay */
 921                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
 922                 close(state_fd);
 923         }
 924
 925         sysfs_free(mdi);
 926         close(fd);
 927
 928         return rv;
 929 }
 930 #endif /* MDASSEMBLE */