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 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
  65         sprintf(fname, "/sys/block/%s/md/", devnum2devname(devnum));
  66         if (devname) {
  67                 strcat(fname, devname);
  68                 strcat(fname, "/");
  69         }
  70         strcat(fname, attr);
  71         fd = open(fname, O_RDWR);
  72         if (fd < 0 && errno == EACCES)
  73                 fd = open(fname, O_RDONLY);
  74         return fd;
  75 }
  76
  77 struct mdinfo *sysfs_read(int fd, int devnum, unsigned long options)
  78 {
  79         /* Longest possible name in sysfs, mounted at /sys, is
  80          *  /sys/block/md_dXXX/md/dev-XXXXX/block/dev
  81          *  /sys/block/md_dXXX/md/metadata_version
  82          * which is about 41 characters.  50 should do for now
  83          */
  84         char fname[50];
  85         char buf[1024];
  86         char *base;
  87         char *dbase;
  88         struct mdinfo *sra;
  89         struct mdinfo *dev;
  90         DIR *dir = NULL;
  91         struct dirent *de;
  92
  93         sra = malloc(sizeof(*sra));
  94         if (sra == NULL)
  95                 return sra;
  96         sra->next = NULL;
  97
  98         if (fd >= 0) {
  99                 struct stat stb;
 100                 mdu_version_t vers;
 101                 if (fstat(fd, &stb)) return NULL;
 102                 if (ioctl(fd, RAID_VERSION, &vers) != 0)
 103                         return NULL;
 104                 if (major(stb.st_rdev)==9)
 105                         sprintf(sra->sys_name, "md%d", (int)minor(stb.st_rdev));
 106                 else
 107                         sprintf(sra->sys_name, "md_d%d",
 108                                 (int)minor(stb.st_rdev)>>MdpMinorShift);
 109         } else {
 110                 if (devnum >= 0)
 111                         sprintf(sra->sys_name, "md%d", devnum);
 112                 else
 113                         sprintf(sra->sys_name, "md_d%d",
 114                                 -1-devnum);
 115         }
 116         sprintf(fname, "/sys/block/%s/md/", sra->sys_name);
 117         base = fname + strlen(fname);
 118
 119         sra->devs = NULL;
 120         if (options & GET_VERSION) {
 121                 strcpy(base, "metadata_version");
 122                 if (load_sys(fname, buf))
 123                         goto abort;
 124                 if (strncmp(buf, "none", 4) == 0) {
 125                         sra->array.major_version =
 126                                 sra->array.minor_version = -1;
 127                         strcpy(sra->text_version, "");
 128                 } else if (strncmp(buf, "external:", 9) == 0) {
 129                         sra->array.major_version = -1;
 130                         sra->array.minor_version = -2;
 131                         strcpy(sra->text_version, buf+9);
 132                 } else {
 133                         sscanf(buf, "%d.%d",
 134                                &sra->array.major_version,
 135                                &sra->array.minor_version);
 136                         strcpy(sra->text_version, buf);
 137                 }
 138         }
 139         if (options & GET_LEVEL) {
 140                 strcpy(base, "level");
 141                 if (load_sys(fname, buf))
 142                         goto abort;
 143                 sra->array.level = map_name(pers, buf);
 144         }
 145         if (options & GET_LAYOUT) {
 146                 strcpy(base, "layout");
 147                 if (load_sys(fname, buf))
 148                         goto abort;
 149                 sra->array.layout = strtoul(buf, NULL, 0);
 150         }
 151         if (options & GET_DISKS) {
 152                 strcpy(base, "raid_disks");
 153                 if (load_sys(fname, buf))
 154                         goto abort;
 155                 sra->array.raid_disks = strtoul(buf, NULL, 0);
 156         }
 157         if (options & GET_DEGRADED) {
 158                 strcpy(base, "degraded");
 159                 if (load_sys(fname, buf))
 160                         goto abort;
 161                 sra->array.failed_disks = strtoul(buf, NULL, 0);
 162         }
 163         if (options & GET_COMPONENT) {
 164                 strcpy(base, "component_size");
 165                 if (load_sys(fname, buf))
 166                         goto abort;
 167                 sra->component_size = strtoull(buf, NULL, 0);
 168                 /* sysfs reports "K", but we want sectors */
 169                 sra->component_size *= 2;
 170         }
 171         if (options & GET_CHUNK) {
 172                 strcpy(base, "chunk_size");
 173                 if (load_sys(fname, buf))
 174                         goto abort;
 175                 sra->array.chunk_size = strtoul(buf, NULL, 0);
 176         }
 177         if (options & GET_CACHE) {
 178                 strcpy(base, "stripe_cache_size");
 179                 if (load_sys(fname, buf))
 180                         goto abort;
 181                 sra->cache_size = strtoul(buf, NULL, 0);
 182         }
 183         if (options & GET_MISMATCH) {
 184                 strcpy(base, "mismatch_cnt");
 185                 if (load_sys(fname, buf))
 186                         goto abort;
 187                 sra->mismatch_cnt = strtoul(buf, NULL, 0);
 188         }
 189         if (options & GET_SAFEMODE) {
 190                 int scale = 1;
 191                 int dot = 0;
 192                 int i;
 193                 unsigned long msec;
 194                 size_t len;
 195
 196                 strcpy(base, "safe_mode_delay");
 197                 if (load_sys(fname, buf))
 198                         goto abort;
 199
 200                 /* remove a period, and count digits after it */
 201                 len = strlen(buf);
 202                 for (i = 0; i < len; i++) {
 203                         if (dot) {
 204                                 if (isdigit(buf[i])) {
 205                                         buf[i-1] = buf[i];
 206                                         scale *= 10;
 207                                 }
 208                                 buf[i] = 0;
 209                         } else if (buf[i] == '.') {
 210                                 dot=1;
 211                                 buf[i] = 0;
 212                         }
 213                 }
 214                 msec = strtoul(buf, NULL, 10);
 215                 msec = (msec * 1000) / scale;
 216                 sra->safe_mode_delay = msec;
 217         }
 218
 219         if (! (options & GET_DEVS))
 220                 return sra;
 221
 222         /* Get all the devices as well */
 223         *base = 0;
 224         dir = opendir(fname);
 225         if (!dir)
 226                 goto abort;
 227         sra->array.spare_disks = 0;
 228
 229         while ((de = readdir(dir)) != NULL) {
 230                 char *ep;
 231                 if (de->d_ino == 0 ||
 232                     strncmp(de->d_name, "dev-", 4) != 0)
 233                         continue;
 234                 strcpy(base, de->d_name);
 235                 dbase = base + strlen(base);
 236                 *dbase++ = '/';
 237
 238                 dev = malloc(sizeof(*dev));
 239                 if (!dev)
 240                         goto abort;
 241
 242                 /* Always get slot, major, minor */
 243                 strcpy(dbase, "slot");
 244                 if (load_sys(fname, buf)) {
 245                         /* hmm... unable to read 'slot' maybe the device
 246                          * is going away?
 247                          */
 248                         strcpy(dbase, "block");
 249                         if (readlink(fname, buf, sizeof(buf)) < 0 &&
 250                             errno != ENAMETOOLONG) {
 251                                 /* ...yup device is gone */
 252                                 free(dev);
 253                                 continue;
 254                         } else {
 255                                 /* slot is unreadable but 'block' link
 256                                  * still intact... something bad is happening
 257                                  * so abort
 258                                  */
 259                                 free(dev);
 260                                 goto abort;
 261                         }
 262
 263                 }
 264                 dev->next = sra->devs;
 265                 sra->devs = dev;
 266
 267                 strcpy(dev->sys_name, de->d_name);
 268                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 269                 if (*ep) dev->disk.raid_disk = -1;
 270
 271                 strcpy(dbase, "block/dev");
 272                 if (load_sys(fname, buf))
 273                         goto abort;
 274                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 275
 276                 if (options & GET_OFFSET) {
 277                         strcpy(dbase, "offset");
 278                         if (load_sys(fname, buf))
 279                                 goto abort;
 280                         dev->data_offset = strtoull(buf, NULL, 0);
 281                 }
 282                 if (options & GET_SIZE) {
 283                         strcpy(dbase, "size");
 284                         if (load_sys(fname, buf))
 285                                 goto abort;
 286                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 287                 }
 288                 if (options & GET_STATE) {
 289                         dev->disk.state = 0;
 290                         strcpy(dbase, "state");
 291                         if (load_sys(fname, buf))
 292                                 goto abort;
 293                         if (strstr(buf, "in_sync"))
 294                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 295                         if (strstr(buf, "faulty"))
 296                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 297                         if (dev->disk.state == 0)
 298                                 sra->array.spare_disks++;
 299                 }
 300                 if (options & GET_ERROR) {
 301                         strcpy(buf, "errors");
 302                         if (load_sys(fname, buf))
 303                                 goto abort;
 304                         dev->errors = strtoul(buf, NULL, 0);
 305                 }
 306         }
 307         closedir(dir);
 308         return sra;
 309
 310  abort:
 311         if (dir)
 312                 closedir(dir);
 313         sysfs_free(sra);
 314         return NULL;
 315 }
 316
 317 int sysfs_attr_match(const char *attr, const char *str)
 318 {
 319         /* See if attr, read from a sysfs file, matches
 320          * str.  They must either be the same, or attr can
 321          * have a trailing newline or comma
 322          */
 323         while (*attr && *str && *attr == *str) {
 324                 attr++;
 325                 str++;
 326         }
 327
 328         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 329                 return 0;
 330         return 1;
 331 }
 332
 333 int sysfs_match_word(const char *word, char **list)
 334 {
 335         int n;
 336         for (n=0; list[n]; n++)
 337                 if (sysfs_attr_match(word, list[n]))
 338                         break;
 339         return n;
 340 }
 341
 342 unsigned long long get_component_size(int fd)
 343 {
 344         /* Find out the component size of the array.
 345          * We cannot trust GET_ARRAY_INFO ioctl as it's
 346          * size field is only 32bits.
 347          * So look in /sys/block/mdXXX/md/component_size
 348          *
 349          * This returns in units of sectors.
 350          */
 351         struct stat stb;
 352         char fname[50];
 353         int n;
 354         if (fstat(fd, &stb)) return 0;
 355         if (major(stb.st_rdev) == 9)
 356                 sprintf(fname, "/sys/block/md%d/md/component_size",
 357                         (int)minor(stb.st_rdev));
 358         else
 359                 sprintf(fname, "/sys/block/md_d%d/md/component_size",
 360                         (int)minor(stb.st_rdev)>>MdpMinorShift);
 361         fd = open(fname, O_RDONLY);
 362         if (fd < 0)
 363                 return 0;
 364         n = read(fd, fname, sizeof(fname));
 365         close(fd);
 366         if (n == sizeof(fname))
 367                 return 0;
 368         fname[n] = 0;
 369         return strtoull(fname, NULL, 10) * 2;
 370 }
 371
 372 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 373                   char *name, char *val)
 374 {
 375         char fname[50];
 376         int n;
 377         int fd;
 378
 379         sprintf(fname, "/sys/block/%s/md/%s/%s",
 380                 sra->sys_name, dev?dev->sys_name:"", name);
 381         fd = open(fname, O_WRONLY);
 382         if (fd < 0)
 383                 return -1;
 384         n = write(fd, val, strlen(val));
 385         close(fd);
 386         if (n != strlen(val))
 387                 return -1;
 388         return 0;
 389 }
 390
 391 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 392                   char *name, unsigned long long val)
 393 {
 394         char valstr[50];
 395         sprintf(valstr, "%llu", val);
 396         return sysfs_set_str(sra, dev, name, valstr);
 397 }
 398
 399 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 400                        char *name, unsigned long long *val)
 401 {
 402         char fname[50];
 403         char buf[50];
 404         int n;
 405         int fd;
 406         char *ep;
 407         sprintf(fname, "/sys/block/%s/md/%s/%s",
 408                 sra->sys_name, dev?dev->sys_name:"", name);
 409         fd = open(fname, O_RDONLY);
 410         if (fd < 0)
 411                 return -1;
 412         n = read(fd, buf, sizeof(buf));
 413         close(fd);
 414         if (n <= 0)
 415                 return -1;
 416         buf[n] = 0;
 417         *val = strtoull(buf, &ep, 0);
 418         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 419                 return -1;
 420         return 0;
 421 }
 422
 423 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 424 {
 425         unsigned long sec;
 426         unsigned long msec;
 427         char delay[30];
 428
 429         sec = ms / 1000;
 430         msec = ms - (sec * 1000);
 431
 432         sprintf(delay, "%ld.%ld", sec, msec);
 433         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 434 }
 435
 436 int sysfs_set_array(struct mdinfo *sra,
 437                     struct mdinfo *info)
 438 {
 439         int rv = 0;
 440         sra->array = info->array;
 441
 442         if (info->array.level < 0)
 443                 return 0; /* FIXME */
 444         rv |= sysfs_set_str(sra, NULL, "level",
 445                             map_num(pers, info->array.level));
 446         rv |= sysfs_set_num(sra, NULL, "raid_disks", info->array.raid_disks);
 447         rv |= sysfs_set_num(sra, NULL, "chunk_size", info->array.chunk_size);
 448         rv |= sysfs_set_num(sra, NULL, "layout", info->array.layout);
 449         rv |= sysfs_set_num(sra, NULL, "component_size", info->component_size/2);
 450         rv |= sysfs_set_num(sra, NULL, "resync_start", info->resync_start);
 451         sra->array = info->array;
 452         return rv;
 453 }
 454
 455 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd)
 456 {
 457         char dv[100];
 458         char nm[100];
 459         struct mdinfo *sd2;
 460         char *dname;
 461         int rv;
 462
 463         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 464         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 465         if (rv)
 466                 return rv;
 467
 468         memset(nm, 0, sizeof(nm));
 469         sprintf(dv, "/sys/dev/block/%d:%d", sd->disk.major, sd->disk.minor);
 470         rv = readlink(dv, nm, sizeof(nm));
 471         if (rv <= 0)
 472                 return -1;
 473         nm[rv] = '\0';
 474         dname = strrchr(nm, '/');
 475         if (dname) dname++;
 476         strcpy(sd->sys_name, "dev-");
 477         strcpy(sd->sys_name+4, dname);
 478
 479         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 480         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 481         if (sra->array.level != LEVEL_CONTAINER) {
 482                 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 483 //              rv |= sysfs_set_str(sra, sd, "state", "in_sync");
 484         }
 485         if (! rv) {
 486                 sd2 = malloc(sizeof(*sd2));
 487                 *sd2 = *sd;
 488                 sd2->next = sra->devs;
 489                 sra->devs = sd2;
 490         }
 491         return rv;
 492 }
 493
 494 #if 0
 495 int sysfs_disk_to_sg(int fd)
 496 {
 497         /* from an open block device, try find and open its corresponding
 498          * scsi_generic interface
 499          */
 500         struct stat st;
 501         char path[256];
 502         char sg_path[256];
 503         char sg_major_minor[8];
 504         char *c;
 505         DIR *dir;
 506         struct dirent *de;
 507         int major, minor, rv;
 508
 509         if (fstat(fd, &st))
 510                 return -1;
 511
 512         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 513                  major(st.st_rdev), minor(st.st_rdev));
 514
 515         dir = opendir(path);
 516         if (!dir)
 517                 return -1;
 518
 519         de = readdir(dir);
 520         while (de) {
 521                 if (strncmp("scsi_generic:", de->d_name,
 522                             strlen("scsi_generic:")) == 0)
 523                         break;
 524                 de = readdir(dir);
 525         }
 526         closedir(dir);
 527
 528         if (!de)
 529                 return -1;
 530
 531         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 532         fd = open(sg_path, O_RDONLY);
 533         if (fd < 0)
 534                 return fd;
 535
 536         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 537         close(fd);
 538         if (rv < 0)
 539                 return -1;
 540         else
 541                 sg_major_minor[rv - 1] = '\0';
 542
 543         c = strchr(sg_major_minor, ':');
 544         *c = '\0';
 545         c++;
 546         major = strtol(sg_major_minor, NULL, 10);
 547         minor = strtol(c, NULL, 10);
 548         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 549                  (int) getpid(), major, minor);
 550         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 551                         fd = open(path, O_RDONLY);
 552                         unlink(path);
 553                         return fd;
 554         }
 555
 556         return -1;
 557 }
 558 #endif
 559
 560 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 561 {
 562         /* from an open block device, try to retrieve it scsi_id */
 563         struct stat st;
 564         char path[256];
 565         char *c1, *c2;
 566         DIR *dir;
 567         struct dirent *de;
 568
 569         if (fstat(fd, &st))
 570                 return 1;
 571
 572         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 573                  major(st.st_rdev), minor(st.st_rdev));
 574
 575         dir = opendir(path);
 576         if (!dir)
 577                 return 1;
 578
 579         de = readdir(dir);
 580         while (de) {
 581                 if (strncmp("scsi_disk:", de->d_name,
 582                             strlen("scsi_disk:")) == 0)
 583                         break;
 584                 de = readdir(dir);
 585         }
 586         closedir(dir);
 587
 588         if (!de)
 589                 return 1;
 590
 591         c1 = strchr(de->d_name, ':');
 592         c1++;
 593         c2 = strchr(c1, ':');
 594         *c2 = '\0';
 595         *id = strtol(c1, NULL, 10) << 24; /* host */
 596         c1 = c2 + 1;
 597         c2 = strchr(c1, ':');
 598         *c2 = '\0';
 599         *id |= strtol(c1, NULL, 10) << 16; /* channel */
 600         c1 = c2 + 1;
 601         c2 = strchr(c1, ':');
 602         *c2 = '\0';
 603         *id |= strtol(c1, NULL, 10) << 8; /* lun */
 604         c1 = c2 + 1;
 605         *id |= strtol(c1, NULL, 10); /* id */
 606
 607         return 0;
 608 }
 609
 610
 611 int sysfs_unique_holder(int devnum, long rdev)
 612 {
 613         /* Check that devnum is a holder of rdev,
 614          * and is the only holder.
 615          * we should be locked against races by
 616          * an O_EXCL on devnum
 617          */
 618         DIR *dir;
 619         struct dirent *de;
 620         char dirname[100];
 621         char l;
 622         int found = 0;
 623         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 624                 major(rdev), minor(rdev));
 625         dir = opendir(dirname);
 626         errno = ENOENT;
 627         if (!dir)
 628                 return 0;
 629         l = strlen(dirname);
 630         while ((de = readdir(dir)) != NULL) {
 631                 char buf[10];
 632                 int n;
 633                 int mj, mn;
 634                 char c;
 635                 int fd;
 636
 637                 if (de->d_ino == 0)
 638                         continue;
 639                 if (de->d_name[0] == '.')
 640                         continue;
 641                 strcpy(dirname+l, "/");
 642                 strcat(dirname+l, de->d_name);
 643                 strcat(dirname+l, "/dev");
 644                 fd = open(dirname, O_RDONLY);
 645                 if (fd < 0) {
 646                         errno = ENOENT;
 647                         break;
 648                 }
 649                 n = read(fd, buf, sizeof(buf)-1);
 650                 close(fd);
 651                 buf[n] = 0;
 652                 if (sscanf(buf, "%d:%d%c", &mj, &mn, &c) != 3 ||
 653                     c != '\n') {
 654                         errno = ENOENT;
 655                         break;
 656                 }
 657                 if (mj != MD_MAJOR)
 658                         mn = -1-(mn>>6);
 659
 660                 if (devnum != mn) {
 661                         errno = EEXIST;
 662                         break;
 663                 }
 664                 found = 1;
 665         }
 666         closedir(dir);
 667         if (de)
 668                 return 0;
 669         else
 670                 return found;
 671 }