monitor.c

   1 /*
   2  * mdmon - monitor external metadata arrays
   3  *
   4  * Copyright (C) 2007-2009 Neil Brown <neilb@suse.de>
   5  * Copyright (C) 2007-2009 Intel Corporation
   6  *
   7  * This program is free software; you can redistribute it and/or modify it
   8  * under the terms and conditions of the GNU General Public License,
   9  * version 2, as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope it will be useful, but WITHOUT
  12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  14  * more details.
  15  *
  16  * You should have received a copy of the GNU General Public License along with
  17  * this program; if not, write to the Free Software Foundation, Inc.,
  18  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  19  */
  20
  21 #include "mdadm.h"
  22 #include "mdmon.h"
  23 #include <sys/syscall.h>
  24 #include <sys/select.h>
  25 #include <signal.h>
  26
  27 static char *array_states[] = {
  28         "clear", "inactive", "suspended", "readonly", "read-auto",
  29         "clean", "active", "write-pending", "active-idle", NULL };
  30 static char *sync_actions[] = {
  31         "idle", "reshape", "resync", "recover", "check", "repair", NULL
  32 };
  33
  34 static int write_attr(char *attr, int fd)
  35 {
  36         return write(fd, attr, strlen(attr));
  37 }
  38
  39 static void add_fd(fd_set *fds, int *maxfd, int fd)
  40 {
  41         struct stat st;
  42         if (fd < 0)
  43                 return;
  44         if (fstat(fd, &st) == -1) {
  45                 dprintf("%s: Invalid fd %d\n", __func__, fd);
  46                 return;
  47         }
  48         if (st.st_nlink == 0) {
  49                 dprintf("%s: fd %d was deleted\n", __func__, fd);
  50                 return;
  51         }
  52         if (fd > *maxfd)
  53                 *maxfd = fd;
  54         FD_SET(fd, fds);
  55 }
  56
  57 static int read_attr(char *buf, int len, int fd)
  58 {
  59         int n;
  60
  61         if (fd < 0) {
  62                 buf[0] = 0;
  63                 return 0;
  64         }
  65         lseek(fd, 0, 0);
  66         n = read(fd, buf, len - 1);
  67
  68         if (n <= 0) {
  69                 buf[0] = 0;
  70                 return 0;
  71         }
  72         buf[n] = 0;
  73         if (buf[n-1] == '\n')
  74                 buf[n-1] = 0;
  75         return n;
  76 }
  77
  78 static void read_resync_start(int fd, unsigned long long *v)
  79 {
  80         char buf[30];
  81         int n;
  82
  83         n = read_attr(buf, 30, fd);
  84         if (n <= 0) {
  85                 dprintf("%s: Failed to read resync_start (%d)\n",
  86                         __func__, fd);
  87                 return;
  88         }
  89         if (strncmp(buf, "none", 4) == 0)
  90                 *v = MaxSector;
  91         else
  92                 *v = strtoull(buf, NULL, 10);
  93 }
  94
  95 static unsigned long long read_sync_completed(int fd)
  96 {
  97         unsigned long long val;
  98         char buf[50];
  99         int n;
 100         char *ep;
 101
 102         n = read_attr(buf, 50, fd);
 103
 104         if (n <= 0)
 105                 return 0;
 106         buf[n] = 0;
 107         val = strtoull(buf, &ep, 0);
 108         if (ep == buf || (*ep != 0 && *ep != '\n' && *ep != ' '))
 109                 return 0;
 110         return val;
 111 }
 112
 113 static enum array_state read_state(int fd)
 114 {
 115         char buf[20];
 116         int n = read_attr(buf, 20, fd);
 117
 118         if (n <= 0)
 119                 return bad_word;
 120         return (enum array_state) sysfs_match_word(buf, array_states);
 121 }
 122
 123 static enum sync_action read_action( int fd)
 124 {
 125         char buf[20];
 126         int n = read_attr(buf, 20, fd);
 127
 128         if (n <= 0)
 129                 return bad_action;
 130         return (enum sync_action) sysfs_match_word(buf, sync_actions);
 131 }
 132
 133 int read_dev_state(int fd)
 134 {
 135         char buf[60];
 136         int n = read_attr(buf, 60, fd);
 137         char *cp;
 138         int rv = 0;
 139
 140         if (n <= 0)
 141                 return 0;
 142
 143         cp = buf;
 144         while (cp) {
 145                 if (sysfs_attr_match(cp, "faulty"))
 146                         rv |= DS_FAULTY;
 147                 if (sysfs_attr_match(cp, "in_sync"))
 148                         rv |= DS_INSYNC;
 149                 if (sysfs_attr_match(cp, "write_mostly"))
 150                         rv |= DS_WRITE_MOSTLY;
 151                 if (sysfs_attr_match(cp, "spare"))
 152                         rv |= DS_SPARE;
 153                 if (sysfs_attr_match(cp, "blocked"))
 154                         rv |= DS_BLOCKED;
 155                 cp = strchr(cp, ',');
 156                 if (cp)
 157                         cp++;
 158         }
 159         return rv;
 160 }
 161
 162 static void signal_manager(void)
 163 {
 164         /* tgkill(getpid(), mon_tid, SIGUSR1); */
 165         int pid = getpid();
 166         syscall(SYS_tgkill, pid, mgr_tid, SIGUSR1);
 167 }
 168
 169 /* Monitor a set of active md arrays - all of which share the
 170  * same metadata - and respond to events that require
 171  * metadata update.
 172  *
 173  * New arrays are detected by another thread which allocates
 174  * required memory and attaches the data structure to our list.
 175  *
 176  * Events:
 177  *  Array stops.
 178  *    This is detected by array_state going to 'clear' or 'inactive'.
 179  *    while we thought it was active.
 180  *    Response is to mark metadata as clean and 'clear' the array(??)
 181  *  write-pending
 182  *    array_state if 'write-pending'
 183  *    We mark metadata as 'dirty' then set array to 'active'.
 184  *  active_idle
 185  *    Either ignore, or mark clean, then mark metadata as clean.
 186  *
 187  *  device fails
 188  *    detected by rd-N/state reporting "faulty"
 189  *    mark device as 'failed' in metadata, let the kernel release the
 190  *    device by writing '-blocked' to rd/state, and finally write 'remove' to
 191  *    rd/state.  Before a disk can be replaced it must be failed and removed
 192  *    from all container members, this will be preemptive for the other
 193  *    arrays... safe?
 194  *
 195  *  sync completes
 196  *    sync_action was 'resync' and becomes 'idle' and resync_start becomes
 197  *    MaxSector
 198  *    Notify metadata that sync is complete.
 199  *
 200  *  recovery completes
 201  *    sync_action changes from 'recover' to 'idle'
 202  *    Check each device state and mark metadata if 'faulty' or 'in_sync'.
 203  *
 204  *  deal with resync
 205  *    This only happens on finding a new array... mdadm will have set
 206  *    'resync_start' to the correct value.  If 'resync_start' indicates that an
 207  *    resync needs to occur set the array to the 'active' state rather than the
 208  *    initial read-auto state.
 209  *
 210  *
 211  *
 212  * We wait for a change (poll/select) on array_state, sync_action, and
 213  * each rd-X/state file.
 214  * When we get any change, we check everything.  So read each state file,
 215  * then decide what to do.
 216  *
 217  * The core action is to write new metadata to all devices in the array.
 218  * This is done at most once on any wakeup.
 219  * After that we might:
 220  *   - update the array_state
 221  *   - set the role of some devices.
 222  *   - request a sync_action
 223  *
 224  */
 225
 226 #define ARRAY_DIRTY 1
 227 #define ARRAY_BUSY 2
 228 static int read_and_act(struct active_array *a)
 229 {
 230         unsigned long long sync_completed;
 231         int check_degraded = 0;
 232         int check_reshape = 0;
 233         int deactivate = 0;
 234         struct mdinfo *mdi;
 235         int ret = 0;
 236         int count = 0;
 237         struct timeval tv;
 238
 239         a->next_state = bad_word;
 240         a->next_action = bad_action;
 241
 242         a->curr_state = read_state(a->info.state_fd);
 243         a->curr_action = read_action(a->action_fd);
 244         if (a->curr_state != clear)
 245                 /*
 246                  * In "clear" state, resync_start may wrongly be set to "0"
 247                  * when the kernel called md_clean but didn't remove the
 248                  * sysfs attributes yet
 249                  */
 250                 read_resync_start(a->resync_start_fd, &a->info.resync_start);
 251         sync_completed = read_sync_completed(a->sync_completed_fd);
 252         for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
 253                 mdi->next_state = 0;
 254                 mdi->curr_state = 0;
 255                 if (mdi->state_fd >= 0) {
 256                         read_resync_start(mdi->recovery_fd,
 257                                           &mdi->recovery_start);
 258                         mdi->curr_state = read_dev_state(mdi->state_fd);
 259                 }
 260         }
 261
 262         gettimeofday(&tv, NULL);
 263         dprintf("%s(%d): %ld.%06ld state:%s prev:%s action:%s prev: %s start:%llu\n",
 264                 __func__, a->info.container_member,
 265                 tv.tv_sec, tv.tv_usec,
 266                 array_states[a->curr_state],
 267                 array_states[a->prev_state],
 268                 sync_actions[a->curr_action],
 269                 sync_actions[a->prev_action],
 270                 a->info.resync_start
 271                 );
 272
 273         if ((a->curr_state == bad_word || a->curr_state <= inactive) &&
 274             a->prev_state > inactive) {
 275                 /* array has been stopped */
 276                 a->container->ss->set_array_state(a, 1);
 277                 a->next_state = clear;
 278                 deactivate = 1;
 279         }
 280         if (a->curr_state == write_pending) {
 281                 a->container->ss->set_array_state(a, 0);
 282                 a->next_state = active;
 283                 ret |= ARRAY_DIRTY;
 284         }
 285         if (a->curr_state == active_idle) {
 286                 /* Set array to 'clean' FIRST, then mark clean
 287                  * in the metadata
 288                  */
 289                 a->next_state = clean;
 290                 ret |= ARRAY_DIRTY;
 291         }
 292         if (a->curr_state == clean) {
 293                 a->container->ss->set_array_state(a, 1);
 294         }
 295         if (a->curr_state == active ||
 296             a->curr_state == suspended)
 297                 ret |= ARRAY_DIRTY;
 298         if (a->curr_state == readonly) {
 299                 /* Well, I'm ready to handle things.  If readonly
 300                  * wasn't requested, transition to read-auto.
 301                  */
 302                 char buf[64];
 303                 read_attr(buf, sizeof(buf), a->metadata_fd);
 304                 if (strncmp(buf, "external:-", 10) == 0) {
 305                         /* explicit request for readonly array.  Leave it alone */
 306                         ;
 307                 } else {
 308                         if (a->container->ss->set_array_state(a, 2))
 309                                 a->next_state = read_auto; /* array is clean */
 310                         else {
 311                                 a->next_state = active; /* Now active for recovery etc */
 312                                 ret |= ARRAY_DIRTY;
 313                         }
 314                 }
 315         }
 316
 317         if (!deactivate &&
 318             a->curr_action == idle &&
 319             a->prev_action == resync) {
 320                 /* A resync has finished.  The endpoint is recorded in
 321                  * 'sync_start'.  We don't update the metadata
 322                  * until the array goes inactive or readonly though.
 323                  * Just check if we need to fiddle spares.
 324                  */
 325                 a->container->ss->set_array_state(a, a->curr_state <= clean);
 326                 check_degraded = 1;
 327         }
 328
 329         if (!deactivate &&
 330             a->curr_action == idle &&
 331             a->prev_action == recover) {
 332                 /* A recovery has finished.  Some disks may be in sync now,
 333                  * and the array may no longer be degraded
 334                  */
 335                 for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
 336                         a->container->ss->set_disk(a, mdi->disk.raid_disk,
 337                                                    mdi->curr_state);
 338                         if (! (mdi->curr_state & DS_INSYNC))
 339                                 check_degraded = 1;
 340                         count++;
 341                 }
 342                 if (count != a->info.array.raid_disks)
 343                         check_degraded = 1;
 344         }
 345
 346         if (!deactivate &&
 347             a->curr_action == reshape &&
 348             a->prev_action != reshape)
 349                 /* reshape was requested by mdadm.  Need to see if
 350                  * new devices have been added.  Manager does that
 351                  * when it sees check_reshape
 352                  */
 353                 check_reshape = 1;
 354
 355         /* Check for failures and if found:
 356          * 1/ Record the failure in the metadata and unblock the device.
 357          *    FIXME update the kernel to stop notifying on failed drives when
 358          *    the array is readonly and we have cleared 'blocked'
 359          * 2/ Try to remove the device if the array is writable, or can be
 360          *    made writable.
 361          */
 362         for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
 363                 if (mdi->curr_state & DS_FAULTY) {
 364                         a->container->ss->set_disk(a, mdi->disk.raid_disk,
 365                                                    mdi->curr_state);
 366                         check_degraded = 1;
 367                         if (mdi->curr_state & DS_BLOCKED)
 368                                 mdi->next_state |= DS_UNBLOCK;
 369                         if (a->curr_state == read_auto) {
 370                                 a->container->ss->set_array_state(a, 0);
 371                                 a->next_state = active;
 372                         }
 373                         if (a->curr_state > readonly)
 374                                 mdi->next_state |= DS_REMOVE;
 375                 }
 376         }
 377
 378         /* Check for recovery checkpoint notifications.  We need to be a
 379          * minimum distance away from the last checkpoint to prevent
 380          * over checkpointing.  Note reshape checkpointing is handled
 381          * in the second branch.
 382          */
 383         if (sync_completed > a->last_checkpoint &&
 384             sync_completed - a->last_checkpoint > a->info.component_size >> 4 &&
 385             a->curr_action > reshape) {
 386                 /* A (non-reshape) sync_action has reached a checkpoint.
 387                  * Record the updated position in the metadata
 388                  */
 389                 a->last_checkpoint = sync_completed;
 390                 a->container->ss->set_array_state(a, a->curr_state <= clean);
 391         } else if ((a->curr_action == idle && a->prev_action == reshape) ||
 392                    (a->curr_action == reshape
 393                     && sync_completed > a->last_checkpoint) ) {
 394                 /* Reshape has progressed or completed so we need to
 395                  * update the array state - and possibly the array size
 396                  */
 397                 if (sync_completed != 0)
 398                         a->last_checkpoint = sync_completed;
 399                 /* We might need to update last_checkpoint depending on
 400                  * the reason that reshape finished.
 401                  * if array reshape is really finished:
 402                  *        set check point to the end, this allows
 403                  *        set_array_state() to finalize reshape in metadata
 404                  * if reshape if broken: do not set checkpoint to the end
 405                  *        this allows for reshape restart from checkpoint
 406                  */
 407                 if ((a->curr_action != reshape) &&
 408                     (a->prev_action == reshape)) {
 409                         char buf[40];
 410                         if ((sysfs_get_str(&a->info, NULL,
 411                                           "reshape_position",
 412                                           buf,
 413                                           sizeof(buf)) >= 0) &&
 414                              strncmp(buf, "none", 4) == 0)
 415                                 a->last_checkpoint = a->info.component_size;
 416                 }
 417                 a->container->ss->set_array_state(a, a->curr_state <= clean);
 418                 a->last_checkpoint = sync_completed;
 419         }
 420
 421         if (sync_completed > a->last_checkpoint)
 422                 a->last_checkpoint = sync_completed;
 423
 424         a->container->ss->sync_metadata(a->container);
 425         dprintf("%s(%d): state:%s action:%s next(", __func__, a->info.container_member,
 426                 array_states[a->curr_state], sync_actions[a->curr_action]);
 427
 428         /* Effect state changes in the array */
 429         if (a->next_state != bad_word) {
 430                 dprintf(" state:%s", array_states[a->next_state]);
 431                 write_attr(array_states[a->next_state], a->info.state_fd);
 432         }
 433         if (a->next_action != bad_action) {
 434                 write_attr(sync_actions[a->next_action], a->action_fd);
 435                 dprintf(" action:%s", sync_actions[a->next_action]);
 436         }
 437         for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
 438                 if (mdi->next_state & DS_UNBLOCK) {
 439                         dprintf(" %d:-blocked", mdi->disk.raid_disk);
 440                         write_attr("-blocked", mdi->state_fd);
 441                 }
 442
 443                 if ((mdi->next_state & DS_REMOVE) && mdi->state_fd >= 0) {
 444                         int remove_result;
 445
 446                         /* The kernel may not be able to immediately remove the
 447                          * disk.  In that case we wait a little while and
 448                          * try again.
 449                          */
 450                         remove_result = write_attr("remove", mdi->state_fd);
 451                         if (remove_result > 0) {
 452                                 dprintf(" %d:removed", mdi->disk.raid_disk);
 453                                 close(mdi->state_fd);
 454                                 close(mdi->recovery_fd);
 455                                 mdi->state_fd = -1;
 456                         } else
 457                                 ret |= ARRAY_BUSY;
 458                 }
 459                 if (mdi->next_state & DS_INSYNC) {
 460                         write_attr("+in_sync", mdi->state_fd);
 461                         dprintf(" %d:+in_sync", mdi->disk.raid_disk);
 462                 }
 463         }
 464         dprintf(" )\n");
 465
 466         /* move curr_ to prev_ */
 467         a->prev_state = a->curr_state;
 468
 469         a->prev_action = a->curr_action;
 470
 471         for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
 472                 mdi->prev_state = mdi->curr_state;
 473                 mdi->next_state = 0;
 474         }
 475
 476         if (check_degraded || check_reshape) {
 477                 /* manager will do the actual check */
 478                 if (check_degraded)
 479                         a->check_degraded = 1;
 480                 if (check_reshape)
 481                         a->check_reshape = 1;
 482                 signal_manager();
 483         }
 484
 485         if (deactivate)
 486                 a->container = NULL;
 487
 488         return ret;
 489 }
 490
 491 static struct mdinfo *
 492 find_device(struct active_array *a, int major, int minor)
 493 {
 494         struct mdinfo *mdi;
 495
 496         for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
 497                 if (mdi->disk.major == major && mdi->disk.minor == minor)
 498                         return mdi;
 499
 500         return NULL;
 501 }
 502
 503 static void reconcile_failed(struct active_array *aa, struct mdinfo *failed)
 504 {
 505         struct active_array *a;
 506         struct mdinfo *victim;
 507
 508         for (a = aa; a; a = a->next) {
 509                 if (!a->container || a->to_remove)
 510                         continue;
 511                 victim = find_device(a, failed->disk.major, failed->disk.minor);
 512                 if (!victim)
 513                         continue;
 514
 515                 if (!(victim->curr_state & DS_FAULTY))
 516                         write_attr("faulty", victim->state_fd);
 517         }
 518 }
 519
 520 #ifdef DEBUG
 521 static void dprint_wake_reasons(fd_set *fds)
 522 {
 523         int i;
 524         char proc_path[256];
 525         char link[256];
 526         char *basename;
 527         int rv;
 528
 529         fprintf(stderr, "monitor: wake ( ");
 530         for (i = 0; i < FD_SETSIZE; i++) {
 531                 if (FD_ISSET(i, fds)) {
 532                         sprintf(proc_path, "/proc/%d/fd/%d",
 533                                 (int) getpid(), i);
 534
 535                         rv = readlink(proc_path, link, sizeof(link) - 1);
 536                         if (rv < 0) {
 537                                 fprintf(stderr, "%d:unknown ", i);
 538                                 continue;
 539                         }
 540                         link[rv] = '\0';
 541                         basename = strrchr(link, '/');
 542                         fprintf(stderr, "%d:%s ",
 543                                 i, basename ? ++basename : link);
 544                 }
 545         }
 546         fprintf(stderr, ")\n");
 547 }
 548 #endif
 549
 550 int monitor_loop_cnt;
 551
 552 static int wait_and_act(struct supertype *container, int nowait)
 553 {
 554         fd_set rfds;
 555         int maxfd = 0;
 556         struct active_array **aap = &container->arrays;
 557         struct active_array *a, **ap;
 558         int rv;
 559         struct mdinfo *mdi;
 560         static unsigned int dirty_arrays = ~0; /* start at some non-zero value */
 561
 562         FD_ZERO(&rfds);
 563
 564         for (ap = aap ; *ap ;) {
 565                 a = *ap;
 566                 /* once an array has been deactivated we want to
 567                  * ask the manager to discard it.
 568                  */
 569                 if (!a->container || a->to_remove) {
 570                         if (discard_this) {
 571                                 ap = &(*ap)->next;
 572                                 continue;
 573                         }
 574                         *ap = a->next;
 575                         a->next = NULL;
 576                         discard_this = a;
 577                         signal_manager();
 578                         continue;
 579                 }
 580
 581                 add_fd(&rfds, &maxfd, a->info.state_fd);
 582                 add_fd(&rfds, &maxfd, a->action_fd);
 583                 add_fd(&rfds, &maxfd, a->sync_completed_fd);
 584                 for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
 585                         add_fd(&rfds, &maxfd, mdi->state_fd);
 586
 587                 ap = &(*ap)->next;
 588         }
 589
 590         if (manager_ready && (*aap == NULL || (sigterm && !dirty_arrays))) {
 591                 /* No interesting arrays, or we have been told to
 592                  * terminate and everything is clean.  Lets see about
 593                  * exiting.  Note that blocking at this point is not a
 594                  * problem as there are no active arrays, there is
 595                  * nothing that we need to be ready to do.
 596                  */
 597                 int fd;
 598                 if (sigterm)
 599                         fd = open_dev_excl(container->devnm);
 600                 else
 601                         fd = open_dev_flags(container->devnm, O_RDONLY|O_EXCL);
 602                 if (fd >= 0 || errno != EBUSY) {
 603                         /* OK, we are safe to leave */
 604                         if (sigterm && !dirty_arrays)
 605                                 dprintf("caught sigterm, all clean... exiting\n");
 606                         else
 607                                 dprintf("no arrays to monitor... exiting\n");
 608                         if (!sigterm)
 609                                 /* On SIGTERM, someone (the take-over mdmon) will
 610                                  * clean up
 611                                  */
 612                                 remove_pidfile(container->devnm);
 613                         exit_now = 1;
 614                         signal_manager();
 615                         close(fd);
 616                         exit(0);
 617                 }
 618         }
 619
 620         if (!nowait) {
 621                 sigset_t set;
 622                 struct timespec ts;
 623                 ts.tv_sec = 24*3600;
 624                 ts.tv_nsec = 0;
 625                 if (*aap == NULL || container->retry_soon) {
 626                         /* just waiting to get O_EXCL access */
 627                         ts.tv_sec = 0;
 628                         ts.tv_nsec = 20000000ULL;
 629                 }
 630                 sigprocmask(SIG_UNBLOCK, NULL, &set);
 631                 sigdelset(&set, SIGUSR1);
 632                 monitor_loop_cnt |= 1;
 633                 rv = pselect(maxfd+1, NULL, NULL, &rfds, &ts, &set);
 634                 monitor_loop_cnt += 1;
 635                 if (rv == -1) {
 636                         if (errno == EINTR) {
 637                                 rv = 0;
 638                                 dprintf("monitor: caught signal\n");
 639                         } else
 640                                 dprintf("monitor: error %d in pselect\n",
 641                                         errno);
 642                 }
 643                 #ifdef DEBUG
 644                 else
 645                         dprint_wake_reasons(&rfds);
 646                 #endif
 647                 container->retry_soon = 0;
 648         }
 649
 650         if (update_queue) {
 651                 struct metadata_update *this;
 652
 653                 for (this = update_queue; this ; this = this->next)
 654                         container->ss->process_update(container, this);
 655
 656                 update_queue_handled = update_queue;
 657                 update_queue = NULL;
 658                 signal_manager();
 659                 container->ss->sync_metadata(container);
 660         }
 661
 662         rv = 0;
 663         dirty_arrays = 0;
 664         for (a = *aap; a ; a = a->next) {
 665
 666                 if (a->replaces && !discard_this) {
 667                         struct active_array **ap;
 668                         for (ap = &a->next; *ap && *ap != a->replaces;
 669                              ap = & (*ap)->next)
 670                                 ;
 671                         if (*ap)
 672                                 *ap = (*ap)->next;
 673                         discard_this = a->replaces;
 674                         a->replaces = NULL;
 675                         /* FIXME check if device->state_fd need to be cleared?*/
 676                         signal_manager();
 677                 }
 678                 if (a->container && !a->to_remove) {
 679                         int ret = read_and_act(a);
 680                         rv |= 1;
 681                         dirty_arrays += !!(ret & ARRAY_DIRTY);
 682                         /* when terminating stop manipulating the array after it
 683                          * is clean, but make sure read_and_act() is given a
 684                          * chance to handle 'active_idle'
 685                          */
 686                         if (sigterm && !(ret & ARRAY_DIRTY))
 687                                 a->container = NULL; /* stop touching this array */
 688                         if (ret & ARRAY_BUSY)
 689                                 container->retry_soon = 1;
 690                 }
 691         }
 692
 693         /* propagate failures across container members */
 694         for (a = *aap; a ; a = a->next) {
 695                 if (!a->container || a->to_remove)
 696                         continue;
 697                 for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
 698                         if (mdi->curr_state & DS_FAULTY)
 699                                 reconcile_failed(*aap, mdi);
 700         }
 701
 702         return rv;
 703 }
 704
 705 void do_monitor(struct supertype *container)
 706 {
 707         int rv;
 708         int first = 1;
 709         do {
 710                 rv = wait_and_act(container, first);
 711                 first = 0;
 712         } while (rv >= 0);
 713 }