mdstat.c

   1 /*
   2  * mdstat - parse /proc/mdstat file. Part of:
   3  * mdadm - manage Linux "md" devices aka RAID arrays.
   4  *
   5  * Copyright (C) 2002-2009 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 /*
  27  * The /proc/mdstat file comes in at least 3 flavours:
  28  * In an unpatched 2.2 kernel (md 0.36.6):
  29  *  Personalities : [n raidx] ...
  30  *  read_ahead {not set|%d sectors}
  31  *  md0 : {in}active{ raidX /dev/hda...  %d blocks{ maxfault=%d}}
  32  *  md1 : .....
  33  *
  34  * Normally only 4 md lines, but all are listed.
  35  *
  36  * In a patched 2.2 kernel (md 0.90.0)
  37  *  Personalities : [raidx] ...
  38  *  read_ahead {not set|%d sectors}
  39  *  mdN : {in}active {(readonly)} raidX dev[%d]{(F)} ... %d blocks STATUS RESYNC
  40  *  ... Only initialised arrays listed
  41  *  unused devices: {dev dev ... | <none>}
  42  *
  43  * STATUS is personality dependant:
  44  *    linear:  %dk rounding
  45  *    raid0:   %dk chunks
  46  *    raid1:   [%d/%d] [U_U]   ( raid/working.  operational or not)
  47  *    raid5:   level 4/5, %dk chunk, algorithm %d [%d/%d] [U_U]
  48  *
  49  * RESYNC is  empty or:
  50  *    {resync|recovery}=%u%% finish=%u.%umin
  51  *  or
  52  *    resync=DELAYED
  53  *
  54  * In a 2.4 kernel (md 0.90.0/2.4)
  55  *  Personalities : [raidX] ...
  56  *  read_ahead {not set|%d sectors}
  57  *  mdN : {in}active {(read-only)} raidX dev[%d]{(F)} ...
  58  *       %d blocks STATUS
  59  *       RESYNC
  60  *  unused devices: {dev dev .. | <none>}
  61  *
  62  *  STATUS matches 0.90.0/2.2
  63  *  RESYNC includes [===>....],
  64  *          adds a space after {resync|recovery} and before and after '='
  65  *          adds a decimal to the recovery percent.
  66  *          adds (%d/%d) resync amount  and max_blocks, before finish.
  67  *          adds speed=%dK/sec after finish
  68  *
  69  *
  70  *
  71  * Out of this we want to extract:
  72  *   list of devices, active or not
  73  *   pattern of failed drives (so need number of drives)
  74  *   percent resync complete
  75  *
  76  * As continuation is indicated by leading space, we use
  77  *  conf_line from config.c to read logical lines
  78  *
  79  */
  80
  81 #include        "mdadm.h"
  82 #include        "dlink.h"
  83 #include        <sys/select.h>
  84 #include        <ctype.h>
  85
  86 static void free_member_devnames(struct dev_member *m)
  87 {
  88         while(m) {
  89                 struct dev_member *t = m;
  90
  91                 m = m->next;
  92                 free(t->name);
  93                 free(t);
  94         }
  95 }
  96
  97 static int add_member_devname(struct dev_member **m, char *name)
  98 {
  99         struct dev_member *new;
 100         char *t;
 101
 102         if ((t = strchr(name, '[')) == NULL)
 103                 /* not a device */
 104                 return 0;
 105
 106         new = xmalloc(sizeof(*new));
 107         new->name = strndup(name, t - name);
 108         new->next = *m;
 109         *m = new;
 110         return 1;
 111 }
 112
 113 void free_mdstat(struct mdstat_ent *ms)
 114 {
 115         while (ms) {
 116                 struct mdstat_ent *t;
 117                 free(ms->dev);
 118                 free(ms->level);
 119                 free(ms->pattern);
 120                 free(ms->metadata_version);
 121                 free_member_devnames(ms->members);
 122                 t = ms;
 123                 ms = ms->next;
 124                 free(t);
 125         }
 126 }
 127
 128 static int mdstat_fd = -1;
 129 struct mdstat_ent *mdstat_read(int hold, int start)
 130 {
 131         FILE *f;
 132         struct mdstat_ent *all, *rv, **end, **insert_here;
 133         char *line;
 134         int fd;
 135
 136         if (hold && mdstat_fd != -1) {
 137                 lseek(mdstat_fd, 0L, 0);
 138                 fd = dup(mdstat_fd);
 139                 if (fd >= 0)
 140                         f = fdopen(fd, "r");
 141                 else
 142                         return NULL;
 143         } else
 144                 f = fopen("/proc/mdstat", "r");
 145         if (f == NULL)
 146                 return NULL;
 147         else
 148                 fcntl(fileno(f), F_SETFD, FD_CLOEXEC);
 149
 150         all = NULL;
 151         end = &all;
 152         for (; (line = conf_line(f)) ; free_line(line)) {
 153                 struct mdstat_ent *ent;
 154                 char *w;
 155                 char devnm[32];
 156                 int in_devs = 0;
 157
 158                 if (strcmp(line, "Personalities")==0)
 159                         continue;
 160                 if (strcmp(line, "read_ahead")==0)
 161                         continue;
 162                 if (strcmp(line, "unused")==0)
 163                         continue;
 164                 insert_here = NULL;
 165                 /* Better be an md line.. */
 166                 if (strncmp(line, "md", 2)!= 0 || strlen(line) >= 32
 167                     || (line[2] != '_' && !isdigit(line[2])))
 168                         continue;
 169                 strcpy(devnm, line);
 170
 171                 ent = xmalloc(sizeof(*ent));
 172                 ent->dev = ent->level = ent->pattern= NULL;
 173                 ent->next = NULL;
 174                 ent->percent = RESYNC_NONE;
 175                 ent->active = -1;
 176                 ent->resync = 0;
 177                 ent->metadata_version = NULL;
 178                 ent->raid_disks = 0;
 179                 ent->devcnt = 0;
 180                 ent->members = NULL;
 181
 182                 ent->dev = xstrdup(line);
 183                 strcpy(ent->devnm, devnm);
 184
 185                 for (w=dl_next(line); w!= line ; w=dl_next(w)) {
 186                         int l = strlen(w);
 187                         char *eq;
 188                         if (strcmp(w, "active")==0)
 189                                 ent->active = 1;
 190                         else if (strcmp(w, "inactive")==0) {
 191                                 ent->active = 0;
 192                                 in_devs = 1;
 193                         } else if (ent->active > 0 &&
 194                                  ent->level == NULL &&
 195                                  w[0] != '(' /*readonly*/) {
 196                                 ent->level = xstrdup(w);
 197                                 in_devs = 1;
 198                         } else if (in_devs && strcmp(w, "blocks")==0)
 199                                 in_devs = 0;
 200                         else if (in_devs) {
 201                                 char *ep = strchr(w, '[');
 202                                 ent->devcnt +=
 203                                         add_member_devname(&ent->members, w);
 204                                 if (ep && strncmp(w, "md", 2)==0) {
 205                                         /* This has an md device as a component.
 206                                          * If that device is already in the
 207                                          * list, make sure we insert before
 208                                          * there.
 209                                          */
 210                                         struct mdstat_ent **ih;
 211                                         ih = &all;
 212                                         while (ih != insert_here && *ih &&
 213                                                ((int)strlen((*ih)->devnm) != ep-w
 214                                                 || strncmp((*ih)->devnm, w, ep-w) != 0))
 215                                                 ih = & (*ih)->next;
 216                                         insert_here = ih;
 217                                 }
 218                         } else if (strcmp(w, "super") == 0 &&
 219                                    dl_next(w) != line) {
 220                                 w = dl_next(w);
 221                                 ent->metadata_version = xstrdup(w);
 222                         } else if (w[0] == '[' && isdigit(w[1])) {
 223                                 ent->raid_disks = atoi(w+1);
 224                         } else if (!ent->pattern &&
 225                                  w[0] == '[' &&
 226                                  (w[1] == 'U' || w[1] == '_')) {
 227                                 ent->pattern = xstrdup(w+1);
 228                                 if (ent->pattern[l-2]==']')
 229                                         ent->pattern[l-2] = '\0';
 230                         } else if (ent->percent == RESYNC_NONE &&
 231                                    strncmp(w, "re", 2)== 0 &&
 232                                    w[l-1] == '%' &&
 233                                    (eq=strchr(w, '=')) != NULL ) {
 234                                 ent->percent = atoi(eq+1);
 235                                 if (strncmp(w,"resync", 6)==0)
 236                                         ent->resync = 1;
 237                                 else if (strncmp(w, "reshape", 7)==0)
 238                                         ent->resync = 2;
 239                                 else
 240                                         ent->resync = 0;
 241                         } else if (ent->percent == RESYNC_NONE &&
 242                                    (w[0] == 'r' || w[0] == 'c')) {
 243                                 if (strncmp(w, "resync", 4)==0)
 244                                         ent->resync = 1;
 245                                 if (strncmp(w, "reshape", 7)==0)
 246                                         ent->resync = 2;
 247                                 if (strncmp(w, "recovery", 8)==0)
 248                                         ent->resync = 0;
 249                                 if (strncmp(w, "check", 5)==0)
 250                                         ent->resync = 3;
 251
 252                                 if (l > 8 && strcmp(w+l-8, "=DELAYED") == 0)
 253                                         ent->percent = RESYNC_DELAYED;
 254                                 if (l > 8 && strcmp(w+l-8, "=PENDING") == 0)
 255                                         ent->percent = RESYNC_PENDING;
 256                         } else if (ent->percent == RESYNC_NONE &&
 257                                    w[0] >= '0' &&
 258                                    w[0] <= '9' &&
 259                                    w[l-1] == '%') {
 260                                 ent->percent = atoi(w);
 261                         }
 262                 }
 263                 if (insert_here && (*insert_here)) {
 264                         ent->next = *insert_here;
 265                         *insert_here = ent;
 266                 } else {
 267                         *end = ent;
 268                         end = &ent->next;
 269                 }
 270         }
 271         if (hold && mdstat_fd == -1) {
 272                 mdstat_fd = dup(fileno(f));
 273                 fcntl(mdstat_fd, F_SETFD, FD_CLOEXEC);
 274         }
 275         fclose(f);
 276
 277         /* If we might want to start array,
 278          * reverse the order, so that components comes before composites
 279          */
 280         if (start) {
 281                 rv = NULL;
 282                 while (all) {
 283                         struct mdstat_ent *e = all;
 284                         all = all->next;
 285                         e->next = rv;
 286                         rv = e;
 287                 }
 288         } else rv = all;
 289         return rv;
 290 }
 291
 292 void mdstat_close(void)
 293 {
 294         if (mdstat_fd >= 0)
 295                 close(mdstat_fd);
 296         mdstat_fd = -1;
 297 }
 298
 299 void mdstat_wait(int seconds)
 300 {
 301         fd_set fds;
 302         struct timeval tm;
 303         int maxfd = 0;
 304         FD_ZERO(&fds);
 305         if (mdstat_fd >= 0) {
 306                 FD_SET(mdstat_fd, &fds);
 307                 maxfd = mdstat_fd;
 308         }
 309         tm.tv_sec = seconds;
 310         tm.tv_usec = 0;
 311         select(maxfd + 1, NULL, NULL, &fds, &tm);
 312 }
 313
 314 void mdstat_wait_fd(int fd, const sigset_t *sigmask)
 315 {
 316         fd_set fds, rfds;
 317         int maxfd = 0;
 318
 319         FD_ZERO(&fds);
 320         FD_ZERO(&rfds);
 321         if (mdstat_fd >= 0)
 322                 FD_SET(mdstat_fd, &fds);
 323
 324         if (fd >= 0) {
 325                 struct stat stb;
 326                 fstat(fd, &stb);
 327                 if ((stb.st_mode & S_IFMT) == S_IFREG)
 328                         /* Must be a /proc or /sys fd, so expect
 329                          * POLLPRI
 330                          * i.e. an 'exceptional' event.
 331                          */
 332                         FD_SET(fd, &fds);
 333                 else
 334                         FD_SET(fd, &rfds);
 335
 336                 if (fd > maxfd)
 337                         maxfd = fd;
 338
 339         }
 340         if (mdstat_fd > maxfd)
 341                 maxfd = mdstat_fd;
 342
 343         pselect(maxfd + 1, &rfds, NULL, &fds,
 344                 NULL, sigmask);
 345 }
 346
 347 int mddev_busy(char *devnm)
 348 {
 349         struct mdstat_ent *mdstat = mdstat_read(0, 0);
 350         struct mdstat_ent *me;
 351
 352         for (me = mdstat ; me ; me = me->next)
 353                 if (strcmp(me->devnm, devnm) == 0)
 354                         break;
 355         free_mdstat(mdstat);
 356         return me != NULL;
 357 }
 358
 359 struct mdstat_ent *mdstat_by_component(char *name)
 360 {
 361         struct mdstat_ent *mdstat = mdstat_read(0, 0);
 362
 363         while (mdstat) {
 364                 struct dev_member *m;
 365                 struct mdstat_ent *ent;
 366                 if (mdstat->metadata_version &&
 367                     strncmp(mdstat->metadata_version, "external:", 9) == 0 &&
 368                     is_subarray(mdstat->metadata_version+9))
 369                         /* don't return subarrays, only containers */
 370                         ;
 371                 else for (m = mdstat->members; m; m = m->next) {
 372                                 if (strcmp(m->name, name) == 0) {
 373                                         free_mdstat(mdstat->next);
 374                                         mdstat->next = NULL;
 375                                         return mdstat;
 376                                 }
 377                         }
 378                 ent = mdstat;
 379                 mdstat = mdstat->next;
 380                 ent->next = NULL;
 381                 free_mdstat(ent);
 382         }
 383         return NULL;
 384 }
 385
 386 struct mdstat_ent *mdstat_by_subdev(char *subdev, char *container)
 387 {
 388         struct mdstat_ent *mdstat = mdstat_read(0, 0);
 389         struct mdstat_ent *ent = NULL;
 390
 391         while (mdstat) {
 392                 /* metadata version must match:
 393                  *   external:[/-]%s/%s
 394                  * where first %s is 'container' and second %s is 'subdev'
 395                  */
 396                 if (ent)
 397                         free_mdstat(ent);
 398                 ent = mdstat;
 399                 mdstat = mdstat->next;
 400                 ent->next = NULL;
 401
 402                 if (ent->metadata_version == NULL ||
 403                     strncmp(ent->metadata_version, "external:", 9) != 0)
 404                         continue;
 405
 406                 if (!metadata_container_matches(ent->metadata_version+9,
 407                                                container) ||
 408                     !metadata_subdev_matches(ent->metadata_version+9,
 409                                              subdev))
 410                         continue;
 411
 412                 free_mdstat(mdstat);
 413                 return ent;
 414         }
 415         return NULL;
 416 }