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         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         if (fd >= 0) {
  85                 mdu_version_t vers;
  86                 if (ioctl(fd, RAID_VERSION, &vers) != 0)
  87                         return;
  88                 devnum = fd2devnum(fd);
  89         }
  90         if (devnum == NoMdDev)
  91                 return;
  92         if (devnum >= 0)
  93                 sprintf(mdi->sys_name, "md%d", devnum);
  94         else
  95                 sprintf(mdi->sys_name, "md_d%d",
  96                         -1-devnum);
  97 }
  98
  99
 100 struct mdinfo *sysfs_read(int fd, int devnum, unsigned long options)
 101 {
 102         /* Longest possible name in sysfs, mounted at /sys, is
 103          *  /sys/block/md_dXXX/md/dev-XXXXX/block/dev
 104          *  /sys/block/md_dXXX/md/metadata_version
 105          * which is about 41 characters.  50 should do for now
 106          */
 107         char fname[50];
 108         char buf[1024];
 109         char *base;
 110         char *dbase;
 111         struct mdinfo *sra;
 112         struct mdinfo *dev;
 113         DIR *dir = NULL;
 114         struct dirent *de;
 115
 116         sra = malloc(sizeof(*sra));
 117         if (sra == NULL)
 118                 return sra;
 119         memset(sra, 0, sizeof(*sra));
 120         sysfs_init(sra, fd, devnum);
 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                 dev->next = sra->devs;
 271                 sra->devs = dev;
 272
 273                 strcpy(dev->sys_name, de->d_name);
 274                 dev->disk.raid_disk = strtoul(buf, &ep, 10);
 275                 if (*ep) dev->disk.raid_disk = -1;
 276
 277                 strcpy(dbase, "block/dev");
 278                 if (load_sys(fname, buf))
 279                         goto abort;
 280                 sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 281
 282                 if (options & GET_OFFSET) {
 283                         strcpy(dbase, "offset");
 284                         if (load_sys(fname, buf))
 285                                 goto abort;
 286                         dev->data_offset = strtoull(buf, NULL, 0);
 287                 }
 288                 if (options & GET_SIZE) {
 289                         strcpy(dbase, "size");
 290                         if (load_sys(fname, buf))
 291                                 goto abort;
 292                         dev->component_size = strtoull(buf, NULL, 0) * 2;
 293                 }
 294                 if (options & GET_STATE) {
 295                         dev->disk.state = 0;
 296                         strcpy(dbase, "state");
 297                         if (load_sys(fname, buf))
 298                                 goto abort;
 299                         if (strstr(buf, "in_sync"))
 300                                 dev->disk.state |= (1<<MD_DISK_SYNC);
 301                         if (strstr(buf, "faulty"))
 302                                 dev->disk.state |= (1<<MD_DISK_FAULTY);
 303                         if (dev->disk.state == 0)
 304                                 sra->array.spare_disks++;
 305                 }
 306                 if (options & GET_ERROR) {
 307                         strcpy(buf, "errors");
 308                         if (load_sys(fname, buf))
 309                                 goto abort;
 310                         dev->errors = strtoul(buf, NULL, 0);
 311                 }
 312         }
 313         closedir(dir);
 314         return sra;
 315
 316  abort:
 317         if (dir)
 318                 closedir(dir);
 319         sysfs_free(sra);
 320         return NULL;
 321 }
 322
 323 int sysfs_attr_match(const char *attr, const char *str)
 324 {
 325         /* See if attr, read from a sysfs file, matches
 326          * str.  They must either be the same, or attr can
 327          * have a trailing newline or comma
 328          */
 329         while (*attr && *str && *attr == *str) {
 330                 attr++;
 331                 str++;
 332         }
 333
 334         if (*str || (*attr && *attr != ',' && *attr != '\n'))
 335                 return 0;
 336         return 1;
 337 }
 338
 339 int sysfs_match_word(const char *word, char **list)
 340 {
 341         int n;
 342         for (n=0; list[n]; n++)
 343                 if (sysfs_attr_match(word, list[n]))
 344                         break;
 345         return n;
 346 }
 347
 348 unsigned long long get_component_size(int fd)
 349 {
 350         /* Find out the component size of the array.
 351          * We cannot trust GET_ARRAY_INFO ioctl as it's
 352          * size field is only 32bits.
 353          * So look in /sys/block/mdXXX/md/component_size
 354          *
 355          * This returns in units of sectors.
 356          */
 357         struct stat stb;
 358         char fname[50];
 359         int n;
 360         if (fstat(fd, &stb)) return 0;
 361         if (major(stb.st_rdev) != get_mdp_major())
 362                 sprintf(fname, "/sys/block/md%d/md/component_size",
 363                         (int)minor(stb.st_rdev));
 364         else
 365                 sprintf(fname, "/sys/block/md_d%d/md/component_size",
 366                         (int)minor(stb.st_rdev)>>MdpMinorShift);
 367         fd = open(fname, O_RDONLY);
 368         if (fd < 0)
 369                 return 0;
 370         n = read(fd, fname, sizeof(fname));
 371         close(fd);
 372         if (n == sizeof(fname))
 373                 return 0;
 374         fname[n] = 0;
 375         return strtoull(fname, NULL, 10) * 2;
 376 }
 377
 378 int sysfs_set_str(struct mdinfo *sra, struct mdinfo *dev,
 379                   char *name, char *val)
 380 {
 381         char fname[50];
 382         int n;
 383         int fd;
 384
 385         sprintf(fname, "/sys/block/%s/md/%s/%s",
 386                 sra->sys_name, dev?dev->sys_name:"", name);
 387         fd = open(fname, O_WRONLY);
 388         if (fd < 0)
 389                 return -1;
 390         n = write(fd, val, strlen(val));
 391         close(fd);
 392         if (n != strlen(val)) {
 393                 dprintf(Name ": failed to write '%s' to '%s' (%s)\n",
 394                         val, fname, strerror(errno));
 395                 return -1;
 396         }
 397         return 0;
 398 }
 399
 400 int sysfs_set_num(struct mdinfo *sra, struct mdinfo *dev,
 401                   char *name, unsigned long long val)
 402 {
 403         char valstr[50];
 404         sprintf(valstr, "%llu", val);
 405         return sysfs_set_str(sra, dev, name, valstr);
 406 }
 407
 408 int sysfs_get_ll(struct mdinfo *sra, struct mdinfo *dev,
 409                        char *name, unsigned long long *val)
 410 {
 411         char fname[50];
 412         char buf[50];
 413         int n;
 414         int fd;
 415         char *ep;
 416         sprintf(fname, "/sys/block/%s/md/%s/%s",
 417                 sra->sys_name, dev?dev->sys_name:"", name);
 418         fd = open(fname, O_RDONLY);
 419         if (fd < 0)
 420                 return -1;
 421         n = read(fd, buf, sizeof(buf));
 422         close(fd);
 423         if (n <= 0)
 424                 return -1;
 425         buf[n] = 0;
 426         *val = strtoull(buf, &ep, 0);
 427         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 428                 return -1;
 429         return 0;
 430 }
 431
 432 int sysfs_set_safemode(struct mdinfo *sra, unsigned long ms)
 433 {
 434         unsigned long sec;
 435         unsigned long msec;
 436         char delay[30];
 437
 438         sec = ms / 1000;
 439         msec = ms % 1000;
 440
 441         sprintf(delay, "%ld.%03ld\n", sec, msec);
 442         /*             this '\n' ^ needed for kernels older than 2.6.28 */
 443         return sysfs_set_str(sra, NULL, "safe_mode_delay", delay);
 444 }
 445
 446 int sysfs_set_array(struct mdinfo *info, int vers)
 447 {
 448         int rv = 0;
 449         char ver[100];
 450
 451         ver[0] = 0;
 452         if (info->array.major_version == -1 &&
 453             info->array.minor_version == -2) {
 454                 strcat(strcpy(ver, "external:"), info->text_version);
 455
 456                 if ((vers % 100) < 2 ||
 457                     sysfs_set_str(info, NULL, "metadata_version",
 458                                   ver) < 0) {
 459                         fprintf(stderr, Name ": This kernel does not "
 460                                 "support external metadata.\n");
 461                         return 1;
 462                 }
 463         }
 464         if (info->array.level < 0)
 465                 return 0; /* FIXME */
 466         rv |= sysfs_set_str(info, NULL, "level",
 467                             map_num(pers, info->array.level));
 468         rv |= sysfs_set_num(info, NULL, "raid_disks", info->array.raid_disks);
 469         rv |= sysfs_set_num(info, NULL, "chunk_size", info->array.chunk_size);
 470         rv |= sysfs_set_num(info, NULL, "layout", info->array.layout);
 471         rv |= sysfs_set_num(info, NULL, "component_size", info->component_size/2);
 472         if (info->array.level > 0)
 473                 rv |= sysfs_set_num(info, NULL, "resync_start", info->resync_start);
 474         return rv;
 475 }
 476
 477 int sysfs_add_disk(struct mdinfo *sra, struct mdinfo *sd)
 478 {
 479         char dv[100];
 480         char nm[100];
 481         char *dname;
 482         int rv;
 483
 484         sprintf(dv, "%d:%d", sd->disk.major, sd->disk.minor);
 485         rv = sysfs_set_str(sra, NULL, "new_dev", dv);
 486         if (rv)
 487                 return rv;
 488
 489         memset(nm, 0, sizeof(nm));
 490         sprintf(dv, "/sys/dev/block/%d:%d", sd->disk.major, sd->disk.minor);
 491         rv = readlink(dv, nm, sizeof(nm));
 492         if (rv <= 0)
 493                 return -1;
 494         nm[rv] = '\0';
 495         dname = strrchr(nm, '/');
 496         if (dname) dname++;
 497         strcpy(sd->sys_name, "dev-");
 498         strcpy(sd->sys_name+4, dname);
 499
 500         rv = sysfs_set_num(sra, sd, "offset", sd->data_offset);
 501         rv |= sysfs_set_num(sra, sd, "size", (sd->component_size+1) / 2);
 502         if (sra->array.level != LEVEL_CONTAINER) {
 503                 rv |= sysfs_set_num(sra, sd, "slot", sd->disk.raid_disk);
 504 //              rv |= sysfs_set_str(sra, sd, "state", "in_sync");
 505         }
 506         return rv;
 507 }
 508
 509 #if 0
 510 int sysfs_disk_to_sg(int fd)
 511 {
 512         /* from an open block device, try find and open its corresponding
 513          * scsi_generic interface
 514          */
 515         struct stat st;
 516         char path[256];
 517         char sg_path[256];
 518         char sg_major_minor[8];
 519         char *c;
 520         DIR *dir;
 521         struct dirent *de;
 522         int major, minor, rv;
 523
 524         if (fstat(fd, &st))
 525                 return -1;
 526
 527         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 528                  major(st.st_rdev), minor(st.st_rdev));
 529
 530         dir = opendir(path);
 531         if (!dir)
 532                 return -1;
 533
 534         de = readdir(dir);
 535         while (de) {
 536                 if (strncmp("scsi_generic:", de->d_name,
 537                             strlen("scsi_generic:")) == 0)
 538                         break;
 539                 de = readdir(dir);
 540         }
 541         closedir(dir);
 542
 543         if (!de)
 544                 return -1;
 545
 546         snprintf(sg_path, sizeof(sg_path), "%s/%s/dev", path, de->d_name);
 547         fd = open(sg_path, O_RDONLY);
 548         if (fd < 0)
 549                 return fd;
 550
 551         rv = read(fd, sg_major_minor, sizeof(sg_major_minor));
 552         close(fd);
 553         if (rv < 0)
 554                 return -1;
 555         else
 556                 sg_major_minor[rv - 1] = '\0';
 557
 558         c = strchr(sg_major_minor, ':');
 559         *c = '\0';
 560         c++;
 561         major = strtol(sg_major_minor, NULL, 10);
 562         minor = strtol(c, NULL, 10);
 563         snprintf(path, sizeof(path), "/dev/.tmp.md.%d:%d:%d",
 564                  (int) getpid(), major, minor);
 565         if (mknod(path, S_IFCHR|0600, makedev(major, minor))==0) {
 566                         fd = open(path, O_RDONLY);
 567                         unlink(path);
 568                         return fd;
 569         }
 570
 571         return -1;
 572 }
 573 #endif
 574
 575 int sysfs_disk_to_scsi_id(int fd, __u32 *id)
 576 {
 577         /* from an open block device, try to retrieve it scsi_id */
 578         struct stat st;
 579         char path[256];
 580         char *c1, *c2;
 581         DIR *dir;
 582         struct dirent *de;
 583
 584         if (fstat(fd, &st))
 585                 return 1;
 586
 587         snprintf(path, sizeof(path), "/sys/dev/block/%d:%d/device",
 588                  major(st.st_rdev), minor(st.st_rdev));
 589
 590         dir = opendir(path);
 591         if (!dir)
 592                 return 1;
 593
 594         de = readdir(dir);
 595         while (de) {
 596                 if (strncmp("scsi_disk:", de->d_name,
 597                             strlen("scsi_disk:")) == 0)
 598                         break;
 599                 de = readdir(dir);
 600         }
 601         closedir(dir);
 602
 603         if (!de)
 604                 return 1;
 605
 606         c1 = strchr(de->d_name, ':');
 607         c1++;
 608         c2 = strchr(c1, ':');
 609         *c2 = '\0';
 610         *id = strtol(c1, NULL, 10) << 24; /* host */
 611         c1 = c2 + 1;
 612         c2 = strchr(c1, ':');
 613         *c2 = '\0';
 614         *id |= strtol(c1, NULL, 10) << 16; /* channel */
 615         c1 = c2 + 1;
 616         c2 = strchr(c1, ':');
 617         *c2 = '\0';
 618         *id |= strtol(c1, NULL, 10) << 8; /* lun */
 619         c1 = c2 + 1;
 620         *id |= strtol(c1, NULL, 10); /* id */
 621
 622         return 0;
 623 }
 624
 625
 626 int sysfs_unique_holder(int devnum, long rdev)
 627 {
 628         /* Check that devnum is a holder of rdev,
 629          * and is the only holder.
 630          * we should be locked against races by
 631          * an O_EXCL on devnum
 632          */
 633         DIR *dir;
 634         struct dirent *de;
 635         char dirname[100];
 636         char l;
 637         int found = 0;
 638         sprintf(dirname, "/sys/dev/block/%d:%d/holders",
 639                 major(rdev), minor(rdev));
 640         dir = opendir(dirname);
 641         errno = ENOENT;
 642         if (!dir)
 643                 return 0;
 644         l = strlen(dirname);
 645         while ((de = readdir(dir)) != NULL) {
 646                 char buf[10];
 647                 int n;
 648                 int mj, mn;
 649                 char c;
 650                 int fd;
 651
 652                 if (de->d_ino == 0)
 653                         continue;
 654                 if (de->d_name[0] == '.')
 655                         continue;
 656                 strcpy(dirname+l, "/");
 657                 strcat(dirname+l, de->d_name);
 658                 strcat(dirname+l, "/dev");
 659                 fd = open(dirname, O_RDONLY);
 660                 if (fd < 0) {
 661                         errno = ENOENT;
 662                         break;
 663                 }
 664                 n = read(fd, buf, sizeof(buf)-1);
 665                 close(fd);
 666                 buf[n] = 0;
 667                 if (sscanf(buf, "%d:%d%c", &mj, &mn, &c) != 3 ||
 668                     c != '\n') {
 669                         errno = ENOENT;
 670                         break;
 671                 }
 672                 if (mj != MD_MAJOR)
 673                         mn = -1-(mn>>6);
 674
 675                 if (devnum != mn) {
 676                         errno = EEXIST;
 677                         break;
 678                 }
 679                 found = 1;
 680         }
 681         closedir(dir);
 682         if (de)
 683                 return 0;
 684         else
 685                 return found;
 686 }