sbin/hastd/primary.c

   1 /*-
   2  * Copyright (c) 2009 The FreeBSD Foundation
   3  * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net>
   4  * All rights reserved.
   5  *
   6  * This software was developed by Pawel Jakub Dawidek under sponsorship from
   7  * the FreeBSD Foundation.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
  19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
  22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  28  * SUCH DAMAGE.
  29  */
  30
  31 #include <sys/cdefs.h>
  32 __FBSDID("$FreeBSD$");
  33
  34 #include <sys/types.h>
  35 #include <sys/time.h>
  36 #include <sys/bio.h>
  37 #include <sys/disk.h>
  38 #include <sys/stat.h>
  39
  40 #include <geom/gate/g_gate.h>
  41
  42 #include <err.h>
  43 #include <errno.h>
  44 #include <fcntl.h>
  45 #include <libgeom.h>
  46 #include <pthread.h>
  47 #include <signal.h>
  48 #include <stdint.h>
  49 #include <stdio.h>
  50 #include <string.h>
  51 #include <sysexits.h>
  52 #include <unistd.h>
  53
  54 #include <activemap.h>
  55 #include <nv.h>
  56 #include <rangelock.h>
  57
  58 #include "control.h"
  59 #include "event.h"
  60 #include "hast.h"
  61 #include "hast_proto.h"
  62 #include "hastd.h"
  63 #include "hooks.h"
  64 #include "metadata.h"
  65 #include "proto.h"
  66 #include "pjdlog.h"
  67 #include "refcnt.h"
  68 #include "subr.h"
  69 #include "synch.h"
  70
  71 /* The is only one remote component for now. */
  72 #define ISREMOTE(no)    ((no) == 1)
  73
  74 struct hio {
  75         /*
  76          * Number of components we are still waiting for.
  77          * When this field goes to 0, we can send the request back to the
  78          * kernel. Each component has to decrease this counter by one
  79          * even on failure.
  80          */
  81         unsigned int             hio_countdown;
  82         /*
  83          * Each component has a place to store its own error.
  84          * Once the request is handled by all components we can decide if the
  85          * request overall is successful or not.
  86          */
  87         int                     *hio_errors;
  88         /*
  89          * Structure used to communicate with GEOM Gate class.
  90          */
  91         struct g_gate_ctl_io     hio_ggio;
  92         /*
  93          * Request was already confirmed to GEOM Gate.
  94          */
  95         bool                     hio_done;
  96         /*
  97          * Remember replication from the time the request was initiated,
  98          * so we won't get confused when replication changes on reload.
  99          */
 100         int                      hio_replication;
 101         TAILQ_ENTRY(hio)        *hio_next;
 102 };
 103 #define hio_free_next   hio_next[0]
 104 #define hio_done_next   hio_next[0]
 105
 106 /*
 107  * Free list holds unused structures. When free list is empty, we have to wait
 108  * until some in-progress requests are freed.
 109  */
 110 static TAILQ_HEAD(, hio) hio_free_list;
 111 static pthread_mutex_t hio_free_list_lock;
 112 static pthread_cond_t hio_free_list_cond;
 113 /*
 114  * There is one send list for every component. One requests is placed on all
 115  * send lists - each component gets the same request, but each component is
 116  * responsible for managing his own send list.
 117  */
 118 static TAILQ_HEAD(, hio) *hio_send_list;
 119 static pthread_mutex_t *hio_send_list_lock;
 120 static pthread_cond_t *hio_send_list_cond;
 121 /*
 122  * There is one recv list for every component, although local components don't
 123  * use recv lists as local requests are done synchronously.
 124  */
 125 static TAILQ_HEAD(, hio) *hio_recv_list;
 126 static pthread_mutex_t *hio_recv_list_lock;
 127 static pthread_cond_t *hio_recv_list_cond;
 128 /*
 129  * Request is placed on done list by the slowest component (the one that
 130  * decreased hio_countdown from 1 to 0).
 131  */
 132 static TAILQ_HEAD(, hio) hio_done_list;
 133 static pthread_mutex_t hio_done_list_lock;
 134 static pthread_cond_t hio_done_list_cond;
 135 /*
 136  * Structure below are for interaction with sync thread.
 137  */
 138 static bool sync_inprogress;
 139 static pthread_mutex_t sync_lock;
 140 static pthread_cond_t sync_cond;
 141 /*
 142  * The lock below allows to synchornize access to remote connections.
 143  */
 144 static pthread_rwlock_t *hio_remote_lock;
 145
 146 /*
 147  * Lock to synchronize metadata updates. Also synchronize access to
 148  * hr_primary_localcnt and hr_primary_remotecnt fields.
 149  */
 150 static pthread_mutex_t metadata_lock;
 151
 152 /*
 153  * Maximum number of outstanding I/O requests.
 154  */
 155 #define HAST_HIO_MAX    256
 156 /*
 157  * Number of components. At this point there are only two components: local
 158  * and remote, but in the future it might be possible to use multiple local
 159  * and remote components.
 160  */
 161 #define HAST_NCOMPONENTS        2
 162
 163 #define ISCONNECTED(res, no)    \
 164         ((res)->hr_remotein != NULL && (res)->hr_remoteout != NULL)
 165
 166 #define QUEUE_INSERT1(hio, name, ncomp) do {                            \
 167         bool _wakeup;                                                   \
 168                                                                         \
 169         mtx_lock(&hio_##name##_list_lock[(ncomp)]);                     \
 170         _wakeup = TAILQ_EMPTY(&hio_##name##_list[(ncomp)]);             \
 171         TAILQ_INSERT_TAIL(&hio_##name##_list[(ncomp)], (hio),           \
 172             hio_next[(ncomp)]);                                         \
 173         mtx_unlock(&hio_##name##_list_lock[ncomp]);                     \
 174         if (_wakeup)                                                    \
 175                 cv_signal(&hio_##name##_list_cond[(ncomp)]);            \
 176 } while (0)
 177 #define QUEUE_INSERT2(hio, name)        do {                            \
 178         bool _wakeup;                                                   \
 179                                                                         \
 180         mtx_lock(&hio_##name##_list_lock);                              \
 181         _wakeup = TAILQ_EMPTY(&hio_##name##_list);                      \
 182         TAILQ_INSERT_TAIL(&hio_##name##_list, (hio), hio_##name##_next);\
 183         mtx_unlock(&hio_##name##_list_lock);                            \
 184         if (_wakeup)                                                    \
 185                 cv_signal(&hio_##name##_list_cond);                     \
 186 } while (0)
 187 #define QUEUE_TAKE1(hio, name, ncomp, timeout)  do {                    \
 188         bool _last;                                                     \
 189                                                                         \
 190         mtx_lock(&hio_##name##_list_lock[(ncomp)]);                     \
 191         _last = false;                                                  \
 192         while (((hio) = TAILQ_FIRST(&hio_##name##_list[(ncomp)])) == NULL && !_last) { \
 193                 cv_timedwait(&hio_##name##_list_cond[(ncomp)],          \
 194                     &hio_##name##_list_lock[(ncomp)], (timeout));       \
 195                 if ((timeout) != 0)                                     \
 196                         _last = true;                                   \
 197         }                                                               \
 198         if (hio != NULL) {                                              \
 199                 TAILQ_REMOVE(&hio_##name##_list[(ncomp)], (hio),        \
 200                     hio_next[(ncomp)]);                                 \
 201         }                                                               \
 202         mtx_unlock(&hio_##name##_list_lock[(ncomp)]);                   \
 203 } while (0)
 204 #define QUEUE_TAKE2(hio, name)  do {                                    \
 205         mtx_lock(&hio_##name##_list_lock);                              \
 206         while (((hio) = TAILQ_FIRST(&hio_##name##_list)) == NULL) {     \
 207                 cv_wait(&hio_##name##_list_cond,                        \
 208                     &hio_##name##_list_lock);                           \
 209         }                                                               \
 210         TAILQ_REMOVE(&hio_##name##_list, (hio), hio_##name##_next);     \
 211         mtx_unlock(&hio_##name##_list_lock);                            \
 212 } while (0)
 213
 214 #define SYNCREQ(hio)            do {                                    \
 215         (hio)->hio_ggio.gctl_unit = -1;                                 \
 216         (hio)->hio_ggio.gctl_seq = 1;                                   \
 217 } while (0)
 218 #define ISSYNCREQ(hio)          ((hio)->hio_ggio.gctl_unit == -1)
 219 #define SYNCREQDONE(hio)        do { (hio)->hio_ggio.gctl_unit = -2; } while (0)
 220 #define ISSYNCREQDONE(hio)      ((hio)->hio_ggio.gctl_unit == -2)
 221
 222 static struct hast_resource *gres;
 223
 224 static pthread_mutex_t range_lock;
 225 static struct rangelocks *range_regular;
 226 static bool range_regular_wait;
 227 static pthread_cond_t range_regular_cond;
 228 static struct rangelocks *range_sync;
 229 static bool range_sync_wait;
 230 static pthread_cond_t range_sync_cond;
 231 static bool fullystarted;
 232
 233 static void *ggate_recv_thread(void *arg);
 234 static void *local_send_thread(void *arg);
 235 static void *remote_send_thread(void *arg);
 236 static void *remote_recv_thread(void *arg);
 237 static void *ggate_send_thread(void *arg);
 238 static void *sync_thread(void *arg);
 239 static void *guard_thread(void *arg);
 240
 241 static void
 242 cleanup(struct hast_resource *res)
 243 {
 244         int rerrno;
 245
 246         /* Remember errno. */
 247         rerrno = errno;
 248
 249         /* Destroy ggate provider if we created one. */
 250         if (res->hr_ggateunit >= 0) {
 251                 struct g_gate_ctl_destroy ggiod;
 252
 253                 bzero(&ggiod, sizeof(ggiod));
 254                 ggiod.gctl_version = G_GATE_VERSION;
 255                 ggiod.gctl_unit = res->hr_ggateunit;
 256                 ggiod.gctl_force = 1;
 257                 if (ioctl(res->hr_ggatefd, G_GATE_CMD_DESTROY, &ggiod) == -1) {
 258                         pjdlog_errno(LOG_WARNING,
 259                             "Unable to destroy hast/%s device",
 260                             res->hr_provname);
 261                 }
 262                 res->hr_ggateunit = -1;
 263         }
 264
 265         /* Restore errno. */
 266         errno = rerrno;
 267 }
 268
 269 static __dead2 void
 270 primary_exit(int exitcode, const char *fmt, ...)
 271 {
 272         va_list ap;
 273
 274         PJDLOG_ASSERT(exitcode != EX_OK);
 275         va_start(ap, fmt);
 276         pjdlogv_errno(LOG_ERR, fmt, ap);
 277         va_end(ap);
 278         cleanup(gres);
 279         exit(exitcode);
 280 }
 281
 282 static __dead2 void
 283 primary_exitx(int exitcode, const char *fmt, ...)
 284 {
 285         va_list ap;
 286
 287         va_start(ap, fmt);
 288         pjdlogv(exitcode == EX_OK ? LOG_INFO : LOG_ERR, fmt, ap);
 289         va_end(ap);
 290         cleanup(gres);
 291         exit(exitcode);
 292 }
 293
 294 static int
 295 hast_activemap_flush(struct hast_resource *res)
 296 {
 297         const unsigned char *buf;
 298         size_t size;
 299
 300         buf = activemap_bitmap(res->hr_amp, &size);
 301         PJDLOG_ASSERT(buf != NULL);
 302         PJDLOG_ASSERT((size % res->hr_local_sectorsize) == 0);
 303         if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) !=
 304             (ssize_t)size) {
 305                 pjdlog_errno(LOG_ERR, "Unable to flush activemap to disk");
 306                 res->hr_stat_activemap_write_error++;
 307                 return (-1);
 308         }
 309         if (res->hr_metaflush == 1 && g_flush(res->hr_localfd) == -1) {
 310                 if (errno == EOPNOTSUPP) {
 311                         pjdlog_warning("The %s provider doesn't support flushing write cache. Disabling it.",
 312                             res->hr_localpath);
 313                         res->hr_metaflush = 0;
 314                 } else {
 315                         pjdlog_errno(LOG_ERR,
 316                             "Unable to flush disk cache on activemap update");
 317                         res->hr_stat_activemap_flush_error++;
 318                         return (-1);
 319                 }
 320         }
 321         return (0);
 322 }
 323
 324 static bool
 325 real_remote(const struct hast_resource *res)
 326 {
 327
 328         return (strcmp(res->hr_remoteaddr, "none") != 0);
 329 }
 330
 331 static void
 332 init_environment(struct hast_resource *res __unused)
 333 {
 334         struct hio *hio;
 335         unsigned int ii, ncomps;
 336
 337         /*
 338          * In the future it might be per-resource value.
 339          */
 340         ncomps = HAST_NCOMPONENTS;
 341
 342         /*
 343          * Allocate memory needed by lists.
 344          */
 345         hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps);
 346         if (hio_send_list == NULL) {
 347                 primary_exitx(EX_TEMPFAIL,
 348                     "Unable to allocate %zu bytes of memory for send lists.",
 349                     sizeof(hio_send_list[0]) * ncomps);
 350         }
 351         hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps);
 352         if (hio_send_list_lock == NULL) {
 353                 primary_exitx(EX_TEMPFAIL,
 354                     "Unable to allocate %zu bytes of memory for send list locks.",
 355                     sizeof(hio_send_list_lock[0]) * ncomps);
 356         }
 357         hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps);
 358         if (hio_send_list_cond == NULL) {
 359                 primary_exitx(EX_TEMPFAIL,
 360                     "Unable to allocate %zu bytes of memory for send list condition variables.",
 361                     sizeof(hio_send_list_cond[0]) * ncomps);
 362         }
 363         hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps);
 364         if (hio_recv_list == NULL) {
 365                 primary_exitx(EX_TEMPFAIL,
 366                     "Unable to allocate %zu bytes of memory for recv lists.",
 367                     sizeof(hio_recv_list[0]) * ncomps);
 368         }
 369         hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps);
 370         if (hio_recv_list_lock == NULL) {
 371                 primary_exitx(EX_TEMPFAIL,
 372                     "Unable to allocate %zu bytes of memory for recv list locks.",
 373                     sizeof(hio_recv_list_lock[0]) * ncomps);
 374         }
 375         hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps);
 376         if (hio_recv_list_cond == NULL) {
 377                 primary_exitx(EX_TEMPFAIL,
 378                     "Unable to allocate %zu bytes of memory for recv list condition variables.",
 379                     sizeof(hio_recv_list_cond[0]) * ncomps);
 380         }
 381         hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps);
 382         if (hio_remote_lock == NULL) {
 383                 primary_exitx(EX_TEMPFAIL,
 384                     "Unable to allocate %zu bytes of memory for remote connections locks.",
 385                     sizeof(hio_remote_lock[0]) * ncomps);
 386         }
 387
 388         /*
 389          * Initialize lists, their locks and theirs condition variables.
 390          */
 391         TAILQ_INIT(&hio_free_list);
 392         mtx_init(&hio_free_list_lock);
 393         cv_init(&hio_free_list_cond);
 394         for (ii = 0; ii < HAST_NCOMPONENTS; ii++) {
 395                 TAILQ_INIT(&hio_send_list[ii]);
 396                 mtx_init(&hio_send_list_lock[ii]);
 397                 cv_init(&hio_send_list_cond[ii]);
 398                 TAILQ_INIT(&hio_recv_list[ii]);
 399                 mtx_init(&hio_recv_list_lock[ii]);
 400                 cv_init(&hio_recv_list_cond[ii]);
 401                 rw_init(&hio_remote_lock[ii]);
 402         }
 403         TAILQ_INIT(&hio_done_list);
 404         mtx_init(&hio_done_list_lock);
 405         cv_init(&hio_done_list_cond);
 406         mtx_init(&metadata_lock);
 407
 408         /*
 409          * Allocate requests pool and initialize requests.
 410          */
 411         for (ii = 0; ii < HAST_HIO_MAX; ii++) {
 412                 hio = malloc(sizeof(*hio));
 413                 if (hio == NULL) {
 414                         primary_exitx(EX_TEMPFAIL,
 415                             "Unable to allocate %zu bytes of memory for hio request.",
 416                             sizeof(*hio));
 417                 }
 418                 hio->hio_countdown = 0;
 419                 hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps);
 420                 if (hio->hio_errors == NULL) {
 421                         primary_exitx(EX_TEMPFAIL,
 422                             "Unable allocate %zu bytes of memory for hio errors.",
 423                             sizeof(hio->hio_errors[0]) * ncomps);
 424                 }
 425                 hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps);
 426                 if (hio->hio_next == NULL) {
 427                         primary_exitx(EX_TEMPFAIL,
 428                             "Unable allocate %zu bytes of memory for hio_next field.",
 429                             sizeof(hio->hio_next[0]) * ncomps);
 430                 }
 431                 hio->hio_ggio.gctl_version = G_GATE_VERSION;
 432                 hio->hio_ggio.gctl_data = malloc(MAXPHYS);
 433                 if (hio->hio_ggio.gctl_data == NULL) {
 434                         primary_exitx(EX_TEMPFAIL,
 435                             "Unable to allocate %zu bytes of memory for gctl_data.",
 436                             MAXPHYS);
 437                 }
 438                 hio->hio_ggio.gctl_length = MAXPHYS;
 439                 hio->hio_ggio.gctl_error = 0;
 440                 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next);
 441         }
 442 }
 443
 444 static bool
 445 init_resuid(struct hast_resource *res)
 446 {
 447
 448         mtx_lock(&metadata_lock);
 449         if (res->hr_resuid != 0) {
 450                 mtx_unlock(&metadata_lock);
 451                 return (false);
 452         } else {
 453                 /* Initialize unique resource identifier. */
 454                 arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid));
 455                 mtx_unlock(&metadata_lock);
 456                 if (metadata_write(res) == -1)
 457                         exit(EX_NOINPUT);
 458                 return (true);
 459         }
 460 }
 461
 462 static void
 463 init_local(struct hast_resource *res)
 464 {
 465         unsigned char *buf;
 466         size_t mapsize;
 467
 468         if (metadata_read(res, true) == -1)
 469                 exit(EX_NOINPUT);
 470         mtx_init(&res->hr_amp_lock);
 471         if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize,
 472             res->hr_local_sectorsize, res->hr_keepdirty) == -1) {
 473                 primary_exit(EX_TEMPFAIL, "Unable to create activemap");
 474         }
 475         mtx_init(&range_lock);
 476         cv_init(&range_regular_cond);
 477         if (rangelock_init(&range_regular) == -1)
 478                 primary_exit(EX_TEMPFAIL, "Unable to create regular range lock");
 479         cv_init(&range_sync_cond);
 480         if (rangelock_init(&range_sync) == -1)
 481                 primary_exit(EX_TEMPFAIL, "Unable to create sync range lock");
 482         mapsize = activemap_ondisk_size(res->hr_amp);
 483         buf = calloc(1, mapsize);
 484         if (buf == NULL) {
 485                 primary_exitx(EX_TEMPFAIL,
 486                     "Unable to allocate buffer for activemap.");
 487         }
 488         if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) !=
 489             (ssize_t)mapsize) {
 490                 primary_exit(EX_NOINPUT, "Unable to read activemap");
 491         }
 492         activemap_copyin(res->hr_amp, buf, mapsize);
 493         free(buf);
 494         if (res->hr_resuid != 0)
 495                 return;
 496         /*
 497          * We're using provider for the first time. Initialize local and remote
 498          * counters. We don't initialize resuid here, as we want to do it just
 499          * in time. The reason for this is that we want to inform secondary
 500          * that there were no writes yet, so there is no need to synchronize
 501          * anything.
 502          */
 503         res->hr_primary_localcnt = 0;
 504         res->hr_primary_remotecnt = 0;
 505         if (metadata_write(res) == -1)
 506                 exit(EX_NOINPUT);
 507 }
 508
 509 static int
 510 primary_connect(struct hast_resource *res, struct proto_conn **connp)
 511 {
 512         struct proto_conn *conn;
 513         int16_t val;
 514
 515         val = 1;
 516         if (proto_send(res->hr_conn, &val, sizeof(val)) == -1) {
 517                 primary_exit(EX_TEMPFAIL,
 518                     "Unable to send connection request to parent");
 519         }
 520         if (proto_recv(res->hr_conn, &val, sizeof(val)) == -1) {
 521                 primary_exit(EX_TEMPFAIL,
 522                     "Unable to receive reply to connection request from parent");
 523         }
 524         if (val != 0) {
 525                 errno = val;
 526                 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
 527                     res->hr_remoteaddr);
 528                 return (-1);
 529         }
 530         if (proto_connection_recv(res->hr_conn, true, &conn) == -1) {
 531                 primary_exit(EX_TEMPFAIL,
 532                     "Unable to receive connection from parent");
 533         }
 534         if (proto_connect_wait(conn, res->hr_timeout) == -1) {
 535                 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
 536                     res->hr_remoteaddr);
 537                 proto_close(conn);
 538                 return (-1);
 539         }
 540         /* Error in setting timeout is not critical, but why should it fail? */
 541         if (proto_timeout(conn, res->hr_timeout) == -1)
 542                 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
 543
 544         *connp = conn;
 545
 546         return (0);
 547 }
 548
 549 /*
 550  * Function instructs GEOM_GATE to handle reads directly from within the kernel.
 551  */
 552 static void
 553 enable_direct_reads(struct hast_resource *res)
 554 {
 555         struct g_gate_ctl_modify ggiomodify;
 556
 557         bzero(&ggiomodify, sizeof(ggiomodify));
 558         ggiomodify.gctl_version = G_GATE_VERSION;
 559         ggiomodify.gctl_unit = res->hr_ggateunit;
 560         ggiomodify.gctl_modify = GG_MODIFY_READPROV | GG_MODIFY_READOFFSET;
 561         strlcpy(ggiomodify.gctl_readprov, res->hr_localpath,
 562             sizeof(ggiomodify.gctl_readprov));
 563         ggiomodify.gctl_readoffset = res->hr_localoff;
 564         if (ioctl(res->hr_ggatefd, G_GATE_CMD_MODIFY, &ggiomodify) == 0)
 565                 pjdlog_debug(1, "Direct reads enabled.");
 566         else
 567                 pjdlog_errno(LOG_WARNING, "Failed to enable direct reads");
 568 }
 569
 570 static int
 571 init_remote(struct hast_resource *res, struct proto_conn **inp,
 572     struct proto_conn **outp)
 573 {
 574         struct proto_conn *in, *out;
 575         struct nv *nvout, *nvin;
 576         const unsigned char *token;
 577         unsigned char *map;
 578         const char *errmsg;
 579         int32_t extentsize;
 580         int64_t datasize;
 581         uint32_t mapsize;
 582         uint8_t version;
 583         size_t size;
 584         int error;
 585
 586         PJDLOG_ASSERT((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL));
 587         PJDLOG_ASSERT(real_remote(res));
 588
 589         in = out = NULL;
 590         errmsg = NULL;
 591
 592         if (primary_connect(res, &out) == -1)
 593                 return (ECONNREFUSED);
 594
 595         error = ECONNABORTED;
 596
 597         /*
 598          * First handshake step.
 599          * Setup outgoing connection with remote node.
 600          */
 601         nvout = nv_alloc();
 602         nv_add_string(nvout, res->hr_name, "resource");
 603         nv_add_uint8(nvout, HAST_PROTO_VERSION, "version");
 604         if (nv_error(nvout) != 0) {
 605                 pjdlog_common(LOG_WARNING, 0, nv_error(nvout),
 606                     "Unable to allocate header for connection with %s",
 607                     res->hr_remoteaddr);
 608                 nv_free(nvout);
 609                 goto close;
 610         }
 611         if (hast_proto_send(res, out, nvout, NULL, 0) == -1) {
 612                 pjdlog_errno(LOG_WARNING,
 613                     "Unable to send handshake header to %s",
 614                     res->hr_remoteaddr);
 615                 nv_free(nvout);
 616                 goto close;
 617         }
 618         nv_free(nvout);
 619         if (hast_proto_recv_hdr(out, &nvin) == -1) {
 620                 pjdlog_errno(LOG_WARNING,
 621                     "Unable to receive handshake header from %s",
 622                     res->hr_remoteaddr);
 623                 goto close;
 624         }
 625         errmsg = nv_get_string(nvin, "errmsg");
 626         if (errmsg != NULL) {
 627                 pjdlog_warning("%s", errmsg);
 628                 if (nv_exists(nvin, "wait"))
 629                         error = EBUSY;
 630                 nv_free(nvin);
 631                 goto close;
 632         }
 633         version = nv_get_uint8(nvin, "version");
 634         if (version == 0) {
 635                 /*
 636                  * If no version is sent, it means this is protocol version 1.
 637                  */
 638                 version = 1;
 639         }
 640         if (version > HAST_PROTO_VERSION) {
 641                 pjdlog_warning("Invalid version received (%hhu).", version);
 642                 nv_free(nvin);
 643                 goto close;
 644         }
 645         res->hr_version = version;
 646         pjdlog_debug(1, "Negotiated protocol version %d.", res->hr_version);
 647         token = nv_get_uint8_array(nvin, &size, "token");
 648         if (token == NULL) {
 649                 pjdlog_warning("Handshake header from %s has no 'token' field.",
 650                     res->hr_remoteaddr);
 651                 nv_free(nvin);
 652                 goto close;
 653         }
 654         if (size != sizeof(res->hr_token)) {
 655                 pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).",
 656                     res->hr_remoteaddr, size, sizeof(res->hr_token));
 657                 nv_free(nvin);
 658                 goto close;
 659         }
 660         bcopy(token, res->hr_token, sizeof(res->hr_token));
 661         nv_free(nvin);
 662
 663         /*
 664          * Second handshake step.
 665          * Setup incoming connection with remote node.
 666          */
 667         if (primary_connect(res, &in) == -1)
 668                 goto close;
 669
 670         nvout = nv_alloc();
 671         nv_add_string(nvout, res->hr_name, "resource");
 672         nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token),
 673             "token");
 674         if (res->hr_resuid == 0) {
 675                 /*
 676                  * The resuid field was not yet initialized.
 677                  * Because we do synchronization inside init_resuid(), it is
 678                  * possible that someone already initialized it, the function
 679                  * will return false then, but if we successfully initialized
 680                  * it, we will get true. True means that there were no writes
 681                  * to this resource yet and we want to inform secondary that
 682                  * synchronization is not needed by sending "virgin" argument.
 683                  */
 684                 if (init_resuid(res))
 685                         nv_add_int8(nvout, 1, "virgin");
 686         }
 687         nv_add_uint64(nvout, res->hr_resuid, "resuid");
 688         nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt");
 689         nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt");
 690         if (nv_error(nvout) != 0) {
 691                 pjdlog_common(LOG_WARNING, 0, nv_error(nvout),
 692                     "Unable to allocate header for connection with %s",
 693                     res->hr_remoteaddr);
 694                 nv_free(nvout);
 695                 goto close;
 696         }
 697         if (hast_proto_send(res, in, nvout, NULL, 0) == -1) {
 698                 pjdlog_errno(LOG_WARNING,
 699                     "Unable to send handshake header to %s",
 700                     res->hr_remoteaddr);
 701                 nv_free(nvout);
 702                 goto close;
 703         }
 704         nv_free(nvout);
 705         if (hast_proto_recv_hdr(out, &nvin) == -1) {
 706                 pjdlog_errno(LOG_WARNING,
 707                     "Unable to receive handshake header from %s",
 708                     res->hr_remoteaddr);
 709                 goto close;
 710         }
 711         errmsg = nv_get_string(nvin, "errmsg");
 712         if (errmsg != NULL) {
 713                 pjdlog_warning("%s", errmsg);
 714                 nv_free(nvin);
 715                 goto close;
 716         }
 717         datasize = nv_get_int64(nvin, "datasize");
 718         if (datasize != res->hr_datasize) {
 719                 pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).",
 720                     (intmax_t)res->hr_datasize, (intmax_t)datasize);
 721                 nv_free(nvin);
 722                 goto close;
 723         }
 724         extentsize = nv_get_int32(nvin, "extentsize");
 725         if (extentsize != res->hr_extentsize) {
 726                 pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).",
 727                     (ssize_t)res->hr_extentsize, (ssize_t)extentsize);
 728                 nv_free(nvin);
 729                 goto close;
 730         }
 731         res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt");
 732         res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt");
 733         res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc");
 734         if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY)
 735                 enable_direct_reads(res);
 736         if (nv_exists(nvin, "virgin")) {
 737                 /*
 738                  * Secondary was reinitialized, bump localcnt if it is 0 as
 739                  * only we have the data.
 740                  */
 741                 PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_PRIMARY);
 742                 PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
 743
 744                 if (res->hr_primary_localcnt == 0) {
 745                         PJDLOG_ASSERT(res->hr_secondary_remotecnt == 0);
 746
 747                         mtx_lock(&metadata_lock);
 748                         res->hr_primary_localcnt++;
 749                         pjdlog_debug(1, "Increasing localcnt to %ju.",
 750                             (uintmax_t)res->hr_primary_localcnt);
 751                         (void)metadata_write(res);
 752                         mtx_unlock(&metadata_lock);
 753                 }
 754         }
 755         map = NULL;
 756         mapsize = nv_get_uint32(nvin, "mapsize");
 757         if (mapsize > 0) {
 758                 map = malloc(mapsize);
 759                 if (map == NULL) {
 760                         pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).",
 761                             (uintmax_t)mapsize);
 762                         nv_free(nvin);
 763                         goto close;
 764                 }
 765                 /*
 766                  * Remote node have some dirty extents on its own, lets
 767                  * download its activemap.
 768                  */
 769                 if (hast_proto_recv_data(res, out, nvin, map,
 770                     mapsize) == -1) {
 771                         pjdlog_errno(LOG_ERR,
 772                             "Unable to receive remote activemap");
 773                         nv_free(nvin);
 774                         free(map);
 775                         goto close;
 776                 }
 777                 /*
 778                  * Merge local and remote bitmaps.
 779                  */
 780                 activemap_merge(res->hr_amp, map, mapsize);
 781                 free(map);
 782                 /*
 783                  * Now that we merged bitmaps from both nodes, flush it to the
 784                  * disk before we start to synchronize.
 785                  */
 786                 (void)hast_activemap_flush(res);
 787         }
 788         nv_free(nvin);
 789 #ifdef notyet
 790         /* Setup directions. */
 791         if (proto_send(out, NULL, 0) == -1)
 792                 pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
 793         if (proto_recv(in, NULL, 0) == -1)
 794                 pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
 795 #endif
 796         pjdlog_info("Connected to %s.", res->hr_remoteaddr);
 797         if (res->hr_original_replication == HAST_REPLICATION_MEMSYNC &&
 798             res->hr_version < 2) {
 799                 pjdlog_warning("The 'memsync' replication mode is not supported by the remote node, falling back to 'fullsync' mode.");
 800                 res->hr_replication = HAST_REPLICATION_FULLSYNC;
 801         } else if (res->hr_replication != res->hr_original_replication) {
 802                 /*
 803                  * This is in case hastd disconnected and was upgraded.
 804                  */
 805                 res->hr_replication = res->hr_original_replication;
 806         }
 807         if (inp != NULL && outp != NULL) {
 808                 *inp = in;
 809                 *outp = out;
 810         } else {
 811                 res->hr_remotein = in;
 812                 res->hr_remoteout = out;
 813         }
 814         event_send(res, EVENT_CONNECT);
 815         return (0);
 816 close:
 817         if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0)
 818                 event_send(res, EVENT_SPLITBRAIN);
 819         proto_close(out);
 820         if (in != NULL)
 821                 proto_close(in);
 822         return (error);
 823 }
 824
 825 static void
 826 sync_start(void)
 827 {
 828
 829         mtx_lock(&sync_lock);
 830         sync_inprogress = true;
 831         mtx_unlock(&sync_lock);
 832         cv_signal(&sync_cond);
 833 }
 834
 835 static void
 836 sync_stop(void)
 837 {
 838
 839         mtx_lock(&sync_lock);
 840         if (sync_inprogress)
 841                 sync_inprogress = false;
 842         mtx_unlock(&sync_lock);
 843 }
 844
 845 static void
 846 init_ggate(struct hast_resource *res)
 847 {
 848         struct g_gate_ctl_create ggiocreate;
 849         struct g_gate_ctl_cancel ggiocancel;
 850
 851         /*
 852          * We communicate with ggate via /dev/ggctl. Open it.
 853          */
 854         res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR);
 855         if (res->hr_ggatefd == -1)
 856                 primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME);
 857         /*
 858          * Create provider before trying to connect, as connection failure
 859          * is not critical, but may take some time.
 860          */
 861         bzero(&ggiocreate, sizeof(ggiocreate));
 862         ggiocreate.gctl_version = G_GATE_VERSION;
 863         ggiocreate.gctl_mediasize = res->hr_datasize;
 864         ggiocreate.gctl_sectorsize = res->hr_local_sectorsize;
 865         ggiocreate.gctl_flags = 0;
 866         ggiocreate.gctl_maxcount = 0;
 867         ggiocreate.gctl_timeout = 0;
 868         ggiocreate.gctl_unit = G_GATE_NAME_GIVEN;
 869         snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s",
 870             res->hr_provname);
 871         if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) {
 872                 pjdlog_info("Device hast/%s created.", res->hr_provname);
 873                 res->hr_ggateunit = ggiocreate.gctl_unit;
 874                 return;
 875         }
 876         if (errno != EEXIST) {
 877                 primary_exit(EX_OSERR, "Unable to create hast/%s device",
 878                     res->hr_provname);
 879         }
 880         pjdlog_debug(1,
 881             "Device hast/%s already exists, we will try to take it over.",
 882             res->hr_provname);
 883         /*
 884          * If we received EEXIST, we assume that the process who created the
 885          * provider died and didn't clean up. In that case we will start from
 886          * where he left of.
 887          */
 888         bzero(&ggiocancel, sizeof(ggiocancel));
 889         ggiocancel.gctl_version = G_GATE_VERSION;
 890         ggiocancel.gctl_unit = G_GATE_NAME_GIVEN;
 891         snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s",
 892             res->hr_provname);
 893         if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) {
 894                 pjdlog_info("Device hast/%s recovered.", res->hr_provname);
 895                 res->hr_ggateunit = ggiocancel.gctl_unit;
 896                 return;
 897         }
 898         primary_exit(EX_OSERR, "Unable to take over hast/%s device",
 899             res->hr_provname);
 900 }
 901
 902 void
 903 hastd_primary(struct hast_resource *res)
 904 {
 905         pthread_t td;
 906         pid_t pid;
 907         int error, mode, debuglevel;
 908
 909         /*
 910          * Create communication channel for sending control commands from
 911          * parent to child.
 912          */
 913         if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
 914                 /* TODO: There's no need for this to be fatal error. */
 915                 KEEP_ERRNO((void)pidfile_remove(pfh));
 916                 pjdlog_exit(EX_OSERR,
 917                     "Unable to create control sockets between parent and child");
 918         }
 919         /*
 920          * Create communication channel for sending events from child to parent.
 921          */
 922         if (proto_client(NULL, "socketpair://", &res->hr_event) == -1) {
 923                 /* TODO: There's no need for this to be fatal error. */
 924                 KEEP_ERRNO((void)pidfile_remove(pfh));
 925                 pjdlog_exit(EX_OSERR,
 926                     "Unable to create event sockets between child and parent");
 927         }
 928         /*
 929          * Create communication channel for sending connection requests from
 930          * child to parent.
 931          */
 932         if (proto_client(NULL, "socketpair://", &res->hr_conn) == -1) {
 933                 /* TODO: There's no need for this to be fatal error. */
 934                 KEEP_ERRNO((void)pidfile_remove(pfh));
 935                 pjdlog_exit(EX_OSERR,
 936                     "Unable to create connection sockets between child and parent");
 937         }
 938
 939         pid = fork();
 940         if (pid == -1) {
 941                 /* TODO: There's no need for this to be fatal error. */
 942                 KEEP_ERRNO((void)pidfile_remove(pfh));
 943                 pjdlog_exit(EX_TEMPFAIL, "Unable to fork");
 944         }
 945
 946         if (pid > 0) {
 947                 /* This is parent. */
 948                 /* Declare that we are receiver. */
 949                 proto_recv(res->hr_event, NULL, 0);
 950                 proto_recv(res->hr_conn, NULL, 0);
 951                 /* Declare that we are sender. */
 952                 proto_send(res->hr_ctrl, NULL, 0);
 953                 res->hr_workerpid = pid;
 954                 return;
 955         }
 956
 957         gres = res;
 958         mode = pjdlog_mode_get();
 959         debuglevel = pjdlog_debug_get();
 960
 961         /* Declare that we are sender. */
 962         proto_send(res->hr_event, NULL, 0);
 963         proto_send(res->hr_conn, NULL, 0);
 964         /* Declare that we are receiver. */
 965         proto_recv(res->hr_ctrl, NULL, 0);
 966         descriptors_cleanup(res);
 967
 968         descriptors_assert(res, mode);
 969
 970         pjdlog_init(mode);
 971         pjdlog_debug_set(debuglevel);
 972         pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
 973         setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
 974
 975         init_local(res);
 976         init_ggate(res);
 977         init_environment(res);
 978
 979         if (drop_privs(res) != 0) {
 980                 cleanup(res);
 981                 exit(EX_CONFIG);
 982         }
 983         pjdlog_info("Privileges successfully dropped.");
 984
 985         /*
 986          * Create the guard thread first, so we can handle signals from the
 987          * very beginning.
 988          */
 989         error = pthread_create(&td, NULL, guard_thread, res);
 990         PJDLOG_ASSERT(error == 0);
 991         /*
 992          * Create the control thread before sending any event to the parent,
 993          * as we can deadlock when parent sends control request to worker,
 994          * but worker has no control thread started yet, so parent waits.
 995          * In the meantime worker sends an event to the parent, but parent
 996          * is unable to handle the event, because it waits for control
 997          * request response.
 998          */
 999         error = pthread_create(&td, NULL, ctrl_thread, res);
1000         PJDLOG_ASSERT(error == 0);
1001         if (real_remote(res)) {
1002                 error = init_remote(res, NULL, NULL);
1003                 if (error == 0) {
1004                         sync_start();
1005                 } else if (error == EBUSY) {
1006                         time_t start = time(NULL);
1007
1008                         pjdlog_warning("Waiting for remote node to become %s for %ds.",
1009                             role2str(HAST_ROLE_SECONDARY),
1010                             res->hr_timeout);
1011                         for (;;) {
1012                                 sleep(1);
1013                                 error = init_remote(res, NULL, NULL);
1014                                 if (error != EBUSY)
1015                                         break;
1016                                 if (time(NULL) > start + res->hr_timeout)
1017                                         break;
1018                         }
1019                         if (error == EBUSY) {
1020                                 pjdlog_warning("Remote node is still %s, starting anyway.",
1021                                     role2str(HAST_ROLE_PRIMARY));
1022                         }
1023                 }
1024         }
1025         error = pthread_create(&td, NULL, ggate_recv_thread, res);
1026         PJDLOG_ASSERT(error == 0);
1027         error = pthread_create(&td, NULL, local_send_thread, res);
1028         PJDLOG_ASSERT(error == 0);
1029         error = pthread_create(&td, NULL, remote_send_thread, res);
1030         PJDLOG_ASSERT(error == 0);
1031         error = pthread_create(&td, NULL, remote_recv_thread, res);
1032         PJDLOG_ASSERT(error == 0);
1033         error = pthread_create(&td, NULL, ggate_send_thread, res);
1034         PJDLOG_ASSERT(error == 0);
1035         fullystarted = true;
1036         (void)sync_thread(res);
1037 }
1038
1039 static void
1040 reqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio,
1041     const char *fmt, ...)
1042 {
1043         char msg[1024];
1044         va_list ap;
1045
1046         va_start(ap, fmt);
1047         (void)vsnprintf(msg, sizeof(msg), fmt, ap);
1048         va_end(ap);
1049         switch (ggio->gctl_cmd) {
1050         case BIO_READ:
1051                 (void)snprlcat(msg, sizeof(msg), "READ(%ju, %ju).",
1052                     (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1053                 break;
1054         case BIO_DELETE:
1055                 (void)snprlcat(msg, sizeof(msg), "DELETE(%ju, %ju).",
1056                     (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1057                 break;
1058         case BIO_FLUSH:
1059                 (void)snprlcat(msg, sizeof(msg), "FLUSH.");
1060                 break;
1061         case BIO_WRITE:
1062                 (void)snprlcat(msg, sizeof(msg), "WRITE(%ju, %ju).",
1063                     (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1064                 break;
1065         default:
1066                 (void)snprlcat(msg, sizeof(msg), "UNKNOWN(%u).",
1067                     (unsigned int)ggio->gctl_cmd);
1068                 break;
1069         }
1070         pjdlog_common(loglevel, debuglevel, -1, "%s", msg);
1071 }
1072
1073 static void
1074 remote_close(struct hast_resource *res, int ncomp)
1075 {
1076
1077         rw_wlock(&hio_remote_lock[ncomp]);
1078         /*
1079          * Check for a race between dropping rlock and acquiring wlock -
1080          * another thread can close connection in-between.
1081          */
1082         if (!ISCONNECTED(res, ncomp)) {
1083                 PJDLOG_ASSERT(res->hr_remotein == NULL);
1084                 PJDLOG_ASSERT(res->hr_remoteout == NULL);
1085                 rw_unlock(&hio_remote_lock[ncomp]);
1086                 return;
1087         }
1088
1089         PJDLOG_ASSERT(res->hr_remotein != NULL);
1090         PJDLOG_ASSERT(res->hr_remoteout != NULL);
1091
1092         pjdlog_debug(2, "Closing incoming connection to %s.",
1093             res->hr_remoteaddr);
1094         proto_close(res->hr_remotein);
1095         res->hr_remotein = NULL;
1096         pjdlog_debug(2, "Closing outgoing connection to %s.",
1097             res->hr_remoteaddr);
1098         proto_close(res->hr_remoteout);
1099         res->hr_remoteout = NULL;
1100
1101         rw_unlock(&hio_remote_lock[ncomp]);
1102
1103         pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr);
1104
1105         /*
1106          * Stop synchronization if in-progress.
1107          */
1108         sync_stop();
1109
1110         event_send(res, EVENT_DISCONNECT);
1111 }
1112
1113 /*
1114  * Acknowledge write completion to the kernel, but don't update activemap yet.
1115  */
1116 static void
1117 write_complete(struct hast_resource *res, struct hio *hio)
1118 {
1119         struct g_gate_ctl_io *ggio;
1120         unsigned int ncomp;
1121
1122         PJDLOG_ASSERT(!hio->hio_done);
1123
1124         ggio = &hio->hio_ggio;
1125         PJDLOG_ASSERT(ggio->gctl_cmd == BIO_WRITE);
1126
1127         /*
1128          * Bump local count if this is first write after
1129          * connection failure with remote node.
1130          */
1131         ncomp = 1;
1132         rw_rlock(&hio_remote_lock[ncomp]);
1133         if (!ISCONNECTED(res, ncomp)) {
1134                 mtx_lock(&metadata_lock);
1135                 if (res->hr_primary_localcnt == res->hr_secondary_remotecnt) {
1136                         res->hr_primary_localcnt++;
1137                         pjdlog_debug(1, "Increasing localcnt to %ju.",
1138                             (uintmax_t)res->hr_primary_localcnt);
1139                         (void)metadata_write(res);
1140                 }
1141                 mtx_unlock(&metadata_lock);
1142         }
1143         rw_unlock(&hio_remote_lock[ncomp]);
1144         if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) == -1)
1145                 primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed");
1146         hio->hio_done = true;
1147 }
1148
1149 /*
1150  * Thread receives ggate I/O requests from the kernel and passes them to
1151  * appropriate threads:
1152  * WRITE - always goes to both local_send and remote_send threads
1153  * READ (when the block is up-to-date on local component) -
1154  *      only local_send thread
1155  * READ (when the block isn't up-to-date on local component) -
1156  *      only remote_send thread
1157  * DELETE - always goes to both local_send and remote_send threads
1158  * FLUSH - always goes to both local_send and remote_send threads
1159  */
1160 static void *
1161 ggate_recv_thread(void *arg)
1162 {
1163         struct hast_resource *res = arg;
1164         struct g_gate_ctl_io *ggio;
1165         struct hio *hio;
1166         unsigned int ii, ncomp, ncomps;
1167         int error;
1168
1169         for (;;) {
1170                 pjdlog_debug(2, "ggate_recv: Taking free request.");
1171                 QUEUE_TAKE2(hio, free);
1172                 pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio);
1173                 ggio = &hio->hio_ggio;
1174                 ggio->gctl_unit = res->hr_ggateunit;
1175                 ggio->gctl_length = MAXPHYS;
1176                 ggio->gctl_error = 0;
1177                 hio->hio_done = false;
1178                 hio->hio_replication = res->hr_replication;
1179                 pjdlog_debug(2,
1180                     "ggate_recv: (%p) Waiting for request from the kernel.",
1181                     hio);
1182                 if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) == -1) {
1183                         if (sigexit_received)
1184                                 pthread_exit(NULL);
1185                         primary_exit(EX_OSERR, "G_GATE_CMD_START failed");
1186                 }
1187                 error = ggio->gctl_error;
1188                 switch (error) {
1189                 case 0:
1190                         break;
1191                 case ECANCELED:
1192                         /* Exit gracefully. */
1193                         if (!sigexit_received) {
1194                                 pjdlog_debug(2,
1195                                     "ggate_recv: (%p) Received cancel from the kernel.",
1196                                     hio);
1197                                 pjdlog_info("Received cancel from the kernel, exiting.");
1198                         }
1199                         pthread_exit(NULL);
1200                 case ENOMEM:
1201                         /*
1202                          * Buffer too small? Impossible, we allocate MAXPHYS
1203                          * bytes - request can't be bigger than that.
1204                          */
1205                         /* FALLTHROUGH */
1206                 case ENXIO:
1207                 default:
1208                         primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.",
1209                             strerror(error));
1210                 }
1211
1212                 ncomp = 0;
1213                 ncomps = HAST_NCOMPONENTS;
1214
1215                 for (ii = 0; ii < ncomps; ii++)
1216                         hio->hio_errors[ii] = EINVAL;
1217                 reqlog(LOG_DEBUG, 2, ggio,
1218                     "ggate_recv: (%p) Request received from the kernel: ",
1219                     hio);
1220
1221                 /*
1222                  * Inform all components about new write request.
1223                  * For read request prefer local component unless the given
1224                  * range is out-of-date, then use remote component.
1225                  */
1226                 switch (ggio->gctl_cmd) {
1227                 case BIO_READ:
1228                         res->hr_stat_read++;
1229                         ncomps = 1;
1230                         mtx_lock(&metadata_lock);
1231                         if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF ||
1232                             res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
1233                                 /*
1234                                  * This range is up-to-date on local component,
1235                                  * so handle request locally.
1236                                  */
1237                                  /* Local component is 0 for now. */
1238                                 ncomp = 0;
1239                         } else /* if (res->hr_syncsrc ==
1240                             HAST_SYNCSRC_SECONDARY) */ {
1241                                 PJDLOG_ASSERT(res->hr_syncsrc ==
1242                                     HAST_SYNCSRC_SECONDARY);
1243                                 /*
1244                                  * This range is out-of-date on local component,
1245                                  * so send request to the remote node.
1246                                  */
1247                                  /* Remote component is 1 for now. */
1248                                 ncomp = 1;
1249                         }
1250                         mtx_unlock(&metadata_lock);
1251                         break;
1252                 case BIO_WRITE:
1253                         res->hr_stat_write++;
1254                         if (res->hr_resuid == 0 &&
1255                             res->hr_primary_localcnt == 0) {
1256                                 /* This is first write. */
1257                                 res->hr_primary_localcnt = 1;
1258                         }
1259                         for (;;) {
1260                                 mtx_lock(&range_lock);
1261                                 if (rangelock_islocked(range_sync,
1262                                     ggio->gctl_offset, ggio->gctl_length)) {
1263                                         pjdlog_debug(2,
1264                                             "regular: Range offset=%jd length=%zu locked.",
1265                                             (intmax_t)ggio->gctl_offset,
1266                                             (size_t)ggio->gctl_length);
1267                                         range_regular_wait = true;
1268                                         cv_wait(&range_regular_cond, &range_lock);
1269                                         range_regular_wait = false;
1270                                         mtx_unlock(&range_lock);
1271                                         continue;
1272                                 }
1273                                 if (rangelock_add(range_regular,
1274                                     ggio->gctl_offset, ggio->gctl_length) == -1) {
1275                                         mtx_unlock(&range_lock);
1276                                         pjdlog_debug(2,
1277                                             "regular: Range offset=%jd length=%zu is already locked, waiting.",
1278                                             (intmax_t)ggio->gctl_offset,
1279                                             (size_t)ggio->gctl_length);
1280                                         sleep(1);
1281                                         continue;
1282                                 }
1283                                 mtx_unlock(&range_lock);
1284                                 break;
1285                         }
1286                         mtx_lock(&res->hr_amp_lock);
1287                         if (activemap_write_start(res->hr_amp,
1288                             ggio->gctl_offset, ggio->gctl_length)) {
1289                                 res->hr_stat_activemap_update++;
1290                                 (void)hast_activemap_flush(res);
1291                         }
1292                         mtx_unlock(&res->hr_amp_lock);
1293                         break;
1294                 case BIO_DELETE:
1295                         res->hr_stat_delete++;
1296                         break;
1297                 case BIO_FLUSH:
1298                         res->hr_stat_flush++;
1299                         break;
1300                 }
1301                 pjdlog_debug(2,
1302                     "ggate_recv: (%p) Moving request to the send queues.", hio);
1303                 hio->hio_countdown = ncomps;
1304                 if (hio->hio_replication == HAST_REPLICATION_MEMSYNC &&
1305                     ggio->gctl_cmd == BIO_WRITE) {
1306                         /* Each remote request needs two responses in memsync. */
1307                         hio->hio_countdown++;
1308                 }
1309                 for (ii = ncomp; ii < ncomps; ii++)
1310                         QUEUE_INSERT1(hio, send, ii);
1311         }
1312         /* NOTREACHED */
1313         return (NULL);
1314 }
1315
1316 /*
1317  * Thread reads from or writes to local component.
1318  * If local read fails, it redirects it to remote_send thread.
1319  */
1320 static void *
1321 local_send_thread(void *arg)
1322 {
1323         struct hast_resource *res = arg;
1324         struct g_gate_ctl_io *ggio;
1325         struct hio *hio;
1326         unsigned int ncomp, rncomp;
1327         ssize_t ret;
1328
1329         /* Local component is 0 for now. */
1330         ncomp = 0;
1331         /* Remote component is 1 for now. */
1332         rncomp = 1;
1333
1334         for (;;) {
1335                 pjdlog_debug(2, "local_send: Taking request.");
1336                 QUEUE_TAKE1(hio, send, ncomp, 0);
1337                 pjdlog_debug(2, "local_send: (%p) Got request.", hio);
1338                 ggio = &hio->hio_ggio;
1339                 switch (ggio->gctl_cmd) {
1340                 case BIO_READ:
1341                         ret = pread(res->hr_localfd, ggio->gctl_data,
1342                             ggio->gctl_length,
1343                             ggio->gctl_offset + res->hr_localoff);
1344                         if (ret == ggio->gctl_length)
1345                                 hio->hio_errors[ncomp] = 0;
1346                         else if (!ISSYNCREQ(hio)) {
1347                                 /*
1348                                  * If READ failed, try to read from remote node.
1349                                  */
1350                                 if (ret == -1) {
1351                                         reqlog(LOG_WARNING, 0, ggio,
1352                                             "Local request failed (%s), trying remote node. ",
1353                                             strerror(errno));
1354                                 } else if (ret != ggio->gctl_length) {
1355                                         reqlog(LOG_WARNING, 0, ggio,
1356                                             "Local request failed (%zd != %jd), trying remote node. ",
1357                                             ret, (intmax_t)ggio->gctl_length);
1358                                 }
1359                                 QUEUE_INSERT1(hio, send, rncomp);
1360                                 continue;
1361                         }
1362                         break;
1363                 case BIO_WRITE:
1364                         ret = pwrite(res->hr_localfd, ggio->gctl_data,
1365                             ggio->gctl_length,
1366                             ggio->gctl_offset + res->hr_localoff);
1367                         if (ret == -1) {
1368                                 hio->hio_errors[ncomp] = errno;
1369                                 reqlog(LOG_WARNING, 0, ggio,
1370                                     "Local request failed (%s): ",
1371                                     strerror(errno));
1372                         } else if (ret != ggio->gctl_length) {
1373                                 hio->hio_errors[ncomp] = EIO;
1374                                 reqlog(LOG_WARNING, 0, ggio,
1375                                     "Local request failed (%zd != %jd): ",
1376                                     ret, (intmax_t)ggio->gctl_length);
1377                         } else {
1378                                 hio->hio_errors[ncomp] = 0;
1379                                 if (hio->hio_replication ==
1380                                     HAST_REPLICATION_ASYNC) {
1381                                         ggio->gctl_error = 0;
1382                                         write_complete(res, hio);
1383                                 }
1384                         }
1385                         break;
1386                 case BIO_DELETE:
1387                         ret = g_delete(res->hr_localfd,
1388                             ggio->gctl_offset + res->hr_localoff,
1389                             ggio->gctl_length);
1390                         if (ret == -1) {
1391                                 hio->hio_errors[ncomp] = errno;
1392                                 reqlog(LOG_WARNING, 0, ggio,
1393                                     "Local request failed (%s): ",
1394                                     strerror(errno));
1395                         } else {
1396                                 hio->hio_errors[ncomp] = 0;
1397                         }
1398                         break;
1399                 case BIO_FLUSH:
1400                         if (!res->hr_localflush) {
1401                                 ret = -1;
1402                                 errno = EOPNOTSUPP;
1403                                 break;
1404                         }
1405                         ret = g_flush(res->hr_localfd);
1406                         if (ret == -1) {
1407                                 if (errno == EOPNOTSUPP)
1408                                         res->hr_localflush = false;
1409                                 hio->hio_errors[ncomp] = errno;
1410                                 reqlog(LOG_WARNING, 0, ggio,
1411                                     "Local request failed (%s): ",
1412                                     strerror(errno));
1413                         } else {
1414                                 hio->hio_errors[ncomp] = 0;
1415                         }
1416                         break;
1417                 }
1418
1419                 if (hio->hio_replication != HAST_REPLICATION_MEMSYNC ||
1420                     ggio->gctl_cmd != BIO_WRITE || ISSYNCREQ(hio)) {
1421                         if (refcnt_release(&hio->hio_countdown) > 0)
1422                                 continue;
1423                 } else {
1424                         /*
1425                          * Depending on hio_countdown value, requests finished
1426                          * in the following order:
1427                          * 0: remote memsync, remote final, local write
1428                          * 1: remote memsync, local write, (remote final)
1429                          * 2: local write, (remote memsync), (remote final)
1430                          */
1431                         switch (refcnt_release(&hio->hio_countdown)) {
1432                         case 0:
1433                                 /*
1434                                  * Local write finished as last.
1435                                  */
1436                                 break;
1437                         case 1:
1438                                 /*
1439                                  * Local write finished after remote memsync
1440                                  * reply arrvied. We can complete the write now.
1441                                  */
1442                                 if (hio->hio_errors[0] == 0)
1443                                         write_complete(res, hio);
1444                                 continue;
1445                         case 2:
1446                                 /*
1447                                  * Local write finished as first.
1448                                  */
1449                                 continue;
1450                         default:
1451                                 PJDLOG_ABORT("Invalid hio_countdown.");
1452                         }
1453                 }
1454                 if (ISSYNCREQ(hio)) {
1455                         mtx_lock(&sync_lock);
1456                         SYNCREQDONE(hio);
1457                         mtx_unlock(&sync_lock);
1458                         cv_signal(&sync_cond);
1459                 } else {
1460                         pjdlog_debug(2,
1461                             "local_send: (%p) Moving request to the done queue.",
1462                             hio);
1463                         QUEUE_INSERT2(hio, done);
1464                 }
1465         }
1466         /* NOTREACHED */
1467         return (NULL);
1468 }
1469
1470 static void
1471 keepalive_send(struct hast_resource *res, unsigned int ncomp)
1472 {
1473         struct nv *nv;
1474
1475         rw_rlock(&hio_remote_lock[ncomp]);
1476
1477         if (!ISCONNECTED(res, ncomp)) {
1478                 rw_unlock(&hio_remote_lock[ncomp]);
1479                 return;
1480         }
1481
1482         PJDLOG_ASSERT(res->hr_remotein != NULL);
1483         PJDLOG_ASSERT(res->hr_remoteout != NULL);
1484
1485         nv = nv_alloc();
1486         nv_add_uint8(nv, HIO_KEEPALIVE, "cmd");
1487         if (nv_error(nv) != 0) {
1488                 rw_unlock(&hio_remote_lock[ncomp]);
1489                 nv_free(nv);
1490                 pjdlog_debug(1,
1491                     "keepalive_send: Unable to prepare header to send.");
1492                 return;
1493         }
1494         if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) == -1) {
1495                 rw_unlock(&hio_remote_lock[ncomp]);
1496                 pjdlog_common(LOG_DEBUG, 1, errno,
1497                     "keepalive_send: Unable to send request");
1498                 nv_free(nv);
1499                 remote_close(res, ncomp);
1500                 return;
1501         }
1502
1503         rw_unlock(&hio_remote_lock[ncomp]);
1504         nv_free(nv);
1505         pjdlog_debug(2, "keepalive_send: Request sent.");
1506 }
1507
1508 /*
1509  * Thread sends request to secondary node.
1510  */
1511 static void *
1512 remote_send_thread(void *arg)
1513 {
1514         struct hast_resource *res = arg;
1515         struct g_gate_ctl_io *ggio;
1516         time_t lastcheck, now;
1517         struct hio *hio;
1518         struct nv *nv;
1519         unsigned int ncomp;
1520         bool wakeup;
1521         uint64_t offset, length;
1522         uint8_t cmd;
1523         void *data;
1524
1525         /* Remote component is 1 for now. */
1526         ncomp = 1;
1527         lastcheck = time(NULL);
1528
1529         for (;;) {
1530                 pjdlog_debug(2, "remote_send: Taking request.");
1531                 QUEUE_TAKE1(hio, send, ncomp, HAST_KEEPALIVE);
1532                 if (hio == NULL) {
1533                         now = time(NULL);
1534                         if (lastcheck + HAST_KEEPALIVE <= now) {
1535                                 keepalive_send(res, ncomp);
1536                                 lastcheck = now;
1537                         }
1538                         continue;
1539                 }
1540                 pjdlog_debug(2, "remote_send: (%p) Got request.", hio);
1541                 ggio = &hio->hio_ggio;
1542                 switch (ggio->gctl_cmd) {
1543                 case BIO_READ:
1544                         cmd = HIO_READ;
1545                         data = NULL;
1546                         offset = ggio->gctl_offset;
1547                         length = ggio->gctl_length;
1548                         break;
1549                 case BIO_WRITE:
1550                         cmd = HIO_WRITE;
1551                         data = ggio->gctl_data;
1552                         offset = ggio->gctl_offset;
1553                         length = ggio->gctl_length;
1554                         break;
1555                 case BIO_DELETE:
1556                         cmd = HIO_DELETE;
1557                         data = NULL;
1558                         offset = ggio->gctl_offset;
1559                         length = ggio->gctl_length;
1560                         break;
1561                 case BIO_FLUSH:
1562                         cmd = HIO_FLUSH;
1563                         data = NULL;
1564                         offset = 0;
1565                         length = 0;
1566                         break;
1567                 default:
1568                         PJDLOG_ABORT("invalid condition");
1569                 }
1570                 nv = nv_alloc();
1571                 nv_add_uint8(nv, cmd, "cmd");
1572                 nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq");
1573                 nv_add_uint64(nv, offset, "offset");
1574                 nv_add_uint64(nv, length, "length");
1575                 if (hio->hio_replication == HAST_REPLICATION_MEMSYNC &&
1576                     ggio->gctl_cmd == BIO_WRITE && !ISSYNCREQ(hio)) {
1577                         nv_add_uint8(nv, 1, "memsync");
1578                 }
1579                 if (nv_error(nv) != 0) {
1580                         hio->hio_errors[ncomp] = nv_error(nv);
1581                         pjdlog_debug(2,
1582                             "remote_send: (%p) Unable to prepare header to send.",
1583                             hio);
1584                         reqlog(LOG_ERR, 0, ggio,
1585                             "Unable to prepare header to send (%s): ",
1586                             strerror(nv_error(nv)));
1587                         /* Move failed request immediately to the done queue. */
1588                         goto done_queue;
1589                 }
1590                 /*
1591                  * Protect connection from disappearing.
1592                  */
1593                 rw_rlock(&hio_remote_lock[ncomp]);
1594                 if (!ISCONNECTED(res, ncomp)) {
1595                         rw_unlock(&hio_remote_lock[ncomp]);
1596                         hio->hio_errors[ncomp] = ENOTCONN;
1597                         goto done_queue;
1598                 }
1599                 /*
1600                  * Move the request to recv queue before sending it, because
1601                  * in different order we can get reply before we move request
1602                  * to recv queue.
1603                  */
1604                 pjdlog_debug(2,
1605                     "remote_send: (%p) Moving request to the recv queue.",
1606                     hio);
1607                 mtx_lock(&hio_recv_list_lock[ncomp]);
1608                 wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]);
1609                 TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]);
1610                 mtx_unlock(&hio_recv_list_lock[ncomp]);
1611                 if (hast_proto_send(res, res->hr_remoteout, nv, data,
1612                     data != NULL ? length : 0) == -1) {
1613                         hio->hio_errors[ncomp] = errno;
1614                         rw_unlock(&hio_remote_lock[ncomp]);
1615                         pjdlog_debug(2,
1616                             "remote_send: (%p) Unable to send request.", hio);
1617                         reqlog(LOG_ERR, 0, ggio,
1618                             "Unable to send request (%s): ",
1619                             strerror(hio->hio_errors[ncomp]));
1620                         remote_close(res, ncomp);
1621                         /*
1622                          * Take request back from the receive queue and move
1623                          * it immediately to the done queue.
1624                          */
1625                         mtx_lock(&hio_recv_list_lock[ncomp]);
1626                         TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1627                             hio_next[ncomp]);
1628                         mtx_unlock(&hio_recv_list_lock[ncomp]);
1629                         goto done_queue;
1630                 }
1631                 rw_unlock(&hio_remote_lock[ncomp]);
1632                 nv_free(nv);
1633                 if (wakeup)
1634                         cv_signal(&hio_recv_list_cond[ncomp]);
1635                 continue;
1636 done_queue:
1637                 nv_free(nv);
1638                 if (ISSYNCREQ(hio)) {
1639                         if (refcnt_release(&hio->hio_countdown) > 0)
1640                                 continue;
1641                         mtx_lock(&sync_lock);
1642                         SYNCREQDONE(hio);
1643                         mtx_unlock(&sync_lock);
1644                         cv_signal(&sync_cond);
1645                         continue;
1646                 }
1647                 if (ggio->gctl_cmd == BIO_WRITE) {
1648                         mtx_lock(&res->hr_amp_lock);
1649                         if (activemap_need_sync(res->hr_amp, ggio->gctl_offset,
1650                             ggio->gctl_length)) {
1651                                 (void)hast_activemap_flush(res);
1652                         }
1653                         mtx_unlock(&res->hr_amp_lock);
1654                         if (hio->hio_replication == HAST_REPLICATION_MEMSYNC)
1655                                 (void)refcnt_release(&hio->hio_countdown);
1656                 }
1657                 if (refcnt_release(&hio->hio_countdown) > 0)
1658                         continue;
1659                 pjdlog_debug(2,
1660                     "remote_send: (%p) Moving request to the done queue.",
1661                     hio);
1662                 QUEUE_INSERT2(hio, done);
1663         }
1664         /* NOTREACHED */
1665         return (NULL);
1666 }
1667
1668 /*
1669  * Thread receives answer from secondary node and passes it to ggate_send
1670  * thread.
1671  */
1672 static void *
1673 remote_recv_thread(void *arg)
1674 {
1675         struct hast_resource *res = arg;
1676         struct g_gate_ctl_io *ggio;
1677         struct hio *hio;
1678         struct nv *nv;
1679         unsigned int ncomp;
1680         uint64_t seq;
1681         bool memsyncack;
1682         int error;
1683
1684         /* Remote component is 1 for now. */
1685         ncomp = 1;
1686
1687         for (;;) {
1688                 /* Wait until there is anything to receive. */
1689                 mtx_lock(&hio_recv_list_lock[ncomp]);
1690                 while (TAILQ_EMPTY(&hio_recv_list[ncomp])) {
1691                         pjdlog_debug(2, "remote_recv: No requests, waiting.");
1692                         cv_wait(&hio_recv_list_cond[ncomp],
1693                             &hio_recv_list_lock[ncomp]);
1694                 }
1695                 mtx_unlock(&hio_recv_list_lock[ncomp]);
1696
1697                 memsyncack = false;
1698
1699                 rw_rlock(&hio_remote_lock[ncomp]);
1700                 if (!ISCONNECTED(res, ncomp)) {
1701                         rw_unlock(&hio_remote_lock[ncomp]);
1702                         /*
1703                          * Connection is dead, so move all pending requests to
1704                          * the done queue (one-by-one).
1705                          */
1706                         mtx_lock(&hio_recv_list_lock[ncomp]);
1707                         hio = TAILQ_FIRST(&hio_recv_list[ncomp]);
1708                         PJDLOG_ASSERT(hio != NULL);
1709                         TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1710                             hio_next[ncomp]);
1711                         mtx_unlock(&hio_recv_list_lock[ncomp]);
1712                         goto done_queue;
1713                 }
1714                 if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
1715                         pjdlog_errno(LOG_ERR,
1716                             "Unable to receive reply header");
1717                         rw_unlock(&hio_remote_lock[ncomp]);
1718                         remote_close(res, ncomp);
1719                         continue;
1720                 }
1721                 rw_unlock(&hio_remote_lock[ncomp]);
1722                 seq = nv_get_uint64(nv, "seq");
1723                 if (seq == 0) {
1724                         pjdlog_error("Header contains no 'seq' field.");
1725                         nv_free(nv);
1726                         continue;
1727                 }
1728                 memsyncack = nv_exists(nv, "received");
1729                 mtx_lock(&hio_recv_list_lock[ncomp]);
1730                 TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) {
1731                         if (hio->hio_ggio.gctl_seq == seq) {
1732                                 TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1733                                     hio_next[ncomp]);
1734                                 break;
1735                         }
1736                 }
1737                 mtx_unlock(&hio_recv_list_lock[ncomp]);
1738                 if (hio == NULL) {
1739                         pjdlog_error("Found no request matching received 'seq' field (%ju).",
1740                             (uintmax_t)seq);
1741                         nv_free(nv);
1742                         continue;
1743                 }
1744                 ggio = &hio->hio_ggio;
1745                 error = nv_get_int16(nv, "error");
1746                 if (error != 0) {
1747                         /* Request failed on remote side. */
1748                         hio->hio_errors[ncomp] = error;
1749                         reqlog(LOG_WARNING, 0, ggio,
1750                             "Remote request failed (%s): ", strerror(error));
1751                         nv_free(nv);
1752                         goto done_queue;
1753                 }
1754                 switch (ggio->gctl_cmd) {
1755                 case BIO_READ:
1756                         rw_rlock(&hio_remote_lock[ncomp]);
1757                         if (!ISCONNECTED(res, ncomp)) {
1758                                 rw_unlock(&hio_remote_lock[ncomp]);
1759                                 nv_free(nv);
1760                                 goto done_queue;
1761                         }
1762                         if (hast_proto_recv_data(res, res->hr_remotein, nv,
1763                             ggio->gctl_data, ggio->gctl_length) == -1) {
1764                                 hio->hio_errors[ncomp] = errno;
1765                                 pjdlog_errno(LOG_ERR,
1766                                     "Unable to receive reply data");
1767                                 rw_unlock(&hio_remote_lock[ncomp]);
1768                                 nv_free(nv);
1769                                 remote_close(res, ncomp);
1770                                 goto done_queue;
1771                         }
1772                         rw_unlock(&hio_remote_lock[ncomp]);
1773                         break;
1774                 case BIO_WRITE:
1775                 case BIO_DELETE:
1776                 case BIO_FLUSH:
1777                         break;
1778                 default:
1779                         PJDLOG_ABORT("invalid condition");
1780                 }
1781                 hio->hio_errors[ncomp] = 0;
1782                 nv_free(nv);
1783 done_queue:
1784                 if (hio->hio_replication != HAST_REPLICATION_MEMSYNC ||
1785                     hio->hio_ggio.gctl_cmd != BIO_WRITE || ISSYNCREQ(hio)) {
1786                         if (refcnt_release(&hio->hio_countdown) > 0)
1787                                 continue;
1788                 } else {
1789                         /*
1790                          * Depending on hio_countdown value, requests finished
1791                          * in the following order:
1792                          *
1793                          * 0: local write, remote memsync, remote final
1794                          * or
1795                          * 0: remote memsync, local write, remote final
1796                          *
1797                          * 1: local write, remote memsync, (remote final)
1798                          * or
1799                          * 1: remote memsync, remote final, (local write)
1800                          *
1801                          * 2: remote memsync, (local write), (remote final)
1802                          * or
1803                          * 2: remote memsync, (remote final), (local write)
1804                          */
1805                         switch (refcnt_release(&hio->hio_countdown)) {
1806                         case 0:
1807                                 /*
1808                                  * Remote final reply arrived.
1809                                  */
1810                                 PJDLOG_ASSERT(!memsyncack);
1811                                 break;
1812                         case 1:
1813                                 if (memsyncack) {
1814                                         /*
1815                                          * Local request already finished, so we
1816                                          * can complete the write.
1817                                          */
1818                                         if (hio->hio_errors[0] == 0)
1819                                                 write_complete(res, hio);
1820                                         /*
1821                                          * We still need to wait for final
1822                                          * remote reply.
1823                                          */
1824                                         pjdlog_debug(2,
1825                                             "remote_recv: (%p) Moving request back to the recv queue.",
1826                                             hio);
1827                                         mtx_lock(&hio_recv_list_lock[ncomp]);
1828                                         TAILQ_INSERT_TAIL(&hio_recv_list[ncomp],
1829                                             hio, hio_next[ncomp]);
1830                                         mtx_unlock(&hio_recv_list_lock[ncomp]);
1831                                 } else {
1832                                         /*
1833                                          * Remote final reply arrived before
1834                                          * local write finished.
1835                                          * Nothing to do in such case.
1836                                          */
1837                                 }
1838                                 continue;
1839                         case 2:
1840                                 /*
1841                                  * We received remote memsync reply even before
1842                                  * local write finished.
1843                                  */
1844                                 PJDLOG_ASSERT(memsyncack);
1845
1846                                 pjdlog_debug(2,
1847                                     "remote_recv: (%p) Moving request back to the recv queue.",
1848                                     hio);
1849                                 mtx_lock(&hio_recv_list_lock[ncomp]);
1850                                 TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio,
1851                                     hio_next[ncomp]);
1852                                 mtx_unlock(&hio_recv_list_lock[ncomp]);
1853                                 continue;
1854                         default:
1855                                 PJDLOG_ABORT("Invalid hio_countdown.");
1856                         }
1857                 }
1858                 if (ISSYNCREQ(hio)) {
1859                         mtx_lock(&sync_lock);
1860                         SYNCREQDONE(hio);
1861                         mtx_unlock(&sync_lock);
1862                         cv_signal(&sync_cond);
1863                 } else {
1864                         pjdlog_debug(2,
1865                             "remote_recv: (%p) Moving request to the done queue.",
1866                             hio);
1867                         QUEUE_INSERT2(hio, done);
1868                 }
1869         }
1870         /* NOTREACHED */
1871         return (NULL);
1872 }
1873
1874 /*
1875  * Thread sends answer to the kernel.
1876  */
1877 static void *
1878 ggate_send_thread(void *arg)
1879 {
1880         struct hast_resource *res = arg;
1881         struct g_gate_ctl_io *ggio;
1882         struct hio *hio;
1883         unsigned int ii, ncomps;
1884
1885         ncomps = HAST_NCOMPONENTS;
1886
1887         for (;;) {
1888                 pjdlog_debug(2, "ggate_send: Taking request.");
1889                 QUEUE_TAKE2(hio, done);
1890                 pjdlog_debug(2, "ggate_send: (%p) Got request.", hio);
1891                 ggio = &hio->hio_ggio;
1892                 for (ii = 0; ii < ncomps; ii++) {
1893                         if (hio->hio_errors[ii] == 0) {
1894                                 /*
1895                                  * One successful request is enough to declare
1896                                  * success.
1897                                  */
1898                                 ggio->gctl_error = 0;
1899                                 break;
1900                         }
1901                 }
1902                 if (ii == ncomps) {
1903                         /*
1904                          * None of the requests were successful.
1905                          * Use the error from local component except the
1906                          * case when we did only remote request.
1907                          */
1908                         if (ggio->gctl_cmd == BIO_READ &&
1909                             res->hr_syncsrc == HAST_SYNCSRC_SECONDARY)
1910                                 ggio->gctl_error = hio->hio_errors[1];
1911                         else
1912                                 ggio->gctl_error = hio->hio_errors[0];
1913                 }
1914                 if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) {
1915                         mtx_lock(&res->hr_amp_lock);
1916                         if (activemap_write_complete(res->hr_amp,
1917                             ggio->gctl_offset, ggio->gctl_length)) {
1918                                 res->hr_stat_activemap_update++;
1919                                 (void)hast_activemap_flush(res);
1920                         }
1921                         mtx_unlock(&res->hr_amp_lock);
1922                 }
1923                 if (ggio->gctl_cmd == BIO_WRITE) {
1924                         /*
1925                          * Unlock range we locked.
1926                          */
1927                         mtx_lock(&range_lock);
1928                         rangelock_del(range_regular, ggio->gctl_offset,
1929                             ggio->gctl_length);
1930                         if (range_sync_wait)
1931                                 cv_signal(&range_sync_cond);
1932                         mtx_unlock(&range_lock);
1933                         if (!hio->hio_done)
1934                                 write_complete(res, hio);
1935                 } else {
1936                         if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) == -1) {
1937                                 primary_exit(EX_OSERR,
1938                                     "G_GATE_CMD_DONE failed");
1939                         }
1940                 }
1941                 if (hio->hio_errors[0]) {
1942                         switch (ggio->gctl_cmd) {
1943                         case BIO_READ:
1944                                 res->hr_stat_read_error++;
1945                                 break;
1946                         case BIO_WRITE:
1947                                 res->hr_stat_write_error++;
1948                                 break;
1949                         case BIO_DELETE:
1950                                 res->hr_stat_delete_error++;
1951                                 break;
1952                         case BIO_FLUSH:
1953                                 res->hr_stat_flush_error++;
1954                                 break;
1955                         }
1956                 }
1957                 pjdlog_debug(2,
1958                     "ggate_send: (%p) Moving request to the free queue.", hio);
1959                 QUEUE_INSERT2(hio, free);
1960         }
1961         /* NOTREACHED */
1962         return (NULL);
1963 }
1964
1965 /*
1966  * Thread synchronize local and remote components.
1967  */
1968 static void *
1969 sync_thread(void *arg __unused)
1970 {
1971         struct hast_resource *res = arg;
1972         struct hio *hio;
1973         struct g_gate_ctl_io *ggio;
1974         struct timeval tstart, tend, tdiff;
1975         unsigned int ii, ncomp, ncomps;
1976         off_t offset, length, synced;
1977         bool dorewind, directreads;
1978         int syncext;
1979
1980         ncomps = HAST_NCOMPONENTS;
1981         dorewind = true;
1982         synced = 0;
1983         offset = -1;
1984         directreads = false;
1985
1986         for (;;) {
1987                 mtx_lock(&sync_lock);
1988                 if (offset >= 0 && !sync_inprogress) {
1989                         gettimeofday(&tend, NULL);
1990                         timersub(&tend, &tstart, &tdiff);
1991                         pjdlog_info("Synchronization interrupted after %#.0T. "
1992                             "%NB synchronized so far.", &tdiff,
1993                             (intmax_t)synced);
1994                         event_send(res, EVENT_SYNCINTR);
1995                 }
1996                 while (!sync_inprogress) {
1997                         dorewind = true;
1998                         synced = 0;
1999                         cv_wait(&sync_cond, &sync_lock);
2000                 }
2001                 mtx_unlock(&sync_lock);
2002                 /*
2003                  * Obtain offset at which we should synchronize.
2004                  * Rewind synchronization if needed.
2005                  */
2006                 mtx_lock(&res->hr_amp_lock);
2007                 if (dorewind)
2008                         activemap_sync_rewind(res->hr_amp);
2009                 offset = activemap_sync_offset(res->hr_amp, &length, &syncext);
2010                 if (syncext != -1) {
2011                         /*
2012                          * We synchronized entire syncext extent, we can mark
2013                          * it as clean now.
2014                          */
2015                         if (activemap_extent_complete(res->hr_amp, syncext))
2016                                 (void)hast_activemap_flush(res);
2017                 }
2018                 mtx_unlock(&res->hr_amp_lock);
2019                 if (dorewind) {
2020                         dorewind = false;
2021                         if (offset == -1)
2022                                 pjdlog_info("Nodes are in sync.");
2023                         else {
2024                                 pjdlog_info("Synchronization started. %NB to go.",
2025                                     (intmax_t)(res->hr_extentsize *
2026                                     activemap_ndirty(res->hr_amp)));
2027                                 event_send(res, EVENT_SYNCSTART);
2028                                 gettimeofday(&tstart, NULL);
2029                         }
2030                 }
2031                 if (offset == -1) {
2032                         sync_stop();
2033                         pjdlog_debug(1, "Nothing to synchronize.");
2034                         /*
2035                          * Synchronization complete, make both localcnt and
2036                          * remotecnt equal.
2037                          */
2038                         ncomp = 1;
2039                         rw_rlock(&hio_remote_lock[ncomp]);
2040                         if (ISCONNECTED(res, ncomp)) {
2041                                 if (synced > 0) {
2042                                         int64_t bps;
2043
2044                                         gettimeofday(&tend, NULL);
2045                                         timersub(&tend, &tstart, &tdiff);
2046                                         bps = (int64_t)((double)synced /
2047                                             ((double)tdiff.tv_sec +
2048                                             (double)tdiff.tv_usec / 1000000));
2049                                         pjdlog_info("Synchronization complete. "
2050                                             "%NB synchronized in %#.0lT (%NB/sec).",
2051                                             (intmax_t)synced, &tdiff,
2052                                             (intmax_t)bps);
2053                                         event_send(res, EVENT_SYNCDONE);
2054                                 }
2055                                 mtx_lock(&metadata_lock);
2056                                 if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY)
2057                                         directreads = true;
2058                                 res->hr_syncsrc = HAST_SYNCSRC_UNDEF;
2059                                 res->hr_primary_localcnt =
2060                                     res->hr_secondary_remotecnt;
2061                                 res->hr_primary_remotecnt =
2062                                     res->hr_secondary_localcnt;
2063                                 pjdlog_debug(1,
2064                                     "Setting localcnt to %ju and remotecnt to %ju.",
2065                                     (uintmax_t)res->hr_primary_localcnt,
2066                                     (uintmax_t)res->hr_primary_remotecnt);
2067                                 (void)metadata_write(res);
2068                                 mtx_unlock(&metadata_lock);
2069                         }
2070                         rw_unlock(&hio_remote_lock[ncomp]);
2071                         if (directreads) {
2072                                 directreads = false;
2073                                 enable_direct_reads(res);
2074                         }
2075                         continue;
2076                 }
2077                 pjdlog_debug(2, "sync: Taking free request.");
2078                 QUEUE_TAKE2(hio, free);
2079                 pjdlog_debug(2, "sync: (%p) Got free request.", hio);
2080                 /*
2081                  * Lock the range we are going to synchronize. We don't want
2082                  * race where someone writes between our read and write.
2083                  */
2084                 for (;;) {
2085                         mtx_lock(&range_lock);
2086                         if (rangelock_islocked(range_regular, offset, length)) {
2087                                 pjdlog_debug(2,
2088                                     "sync: Range offset=%jd length=%jd locked.",
2089                                     (intmax_t)offset, (intmax_t)length);
2090                                 range_sync_wait = true;
2091                                 cv_wait(&range_sync_cond, &range_lock);
2092                                 range_sync_wait = false;
2093                                 mtx_unlock(&range_lock);
2094                                 continue;
2095                         }
2096                         if (rangelock_add(range_sync, offset, length) == -1) {
2097                                 mtx_unlock(&range_lock);
2098                                 pjdlog_debug(2,
2099                                     "sync: Range offset=%jd length=%jd is already locked, waiting.",
2100                                     (intmax_t)offset, (intmax_t)length);
2101                                 sleep(1);
2102                                 continue;
2103                         }
2104                         mtx_unlock(&range_lock);
2105                         break;
2106                 }
2107                 /*
2108                  * First read the data from synchronization source.
2109                  */
2110                 SYNCREQ(hio);
2111                 ggio = &hio->hio_ggio;
2112                 ggio->gctl_cmd = BIO_READ;
2113                 ggio->gctl_offset = offset;
2114                 ggio->gctl_length = length;
2115                 ggio->gctl_error = 0;
2116                 hio->hio_done = false;
2117                 hio->hio_replication = res->hr_replication;
2118                 for (ii = 0; ii < ncomps; ii++)
2119                         hio->hio_errors[ii] = EINVAL;
2120                 reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ",
2121                     hio);
2122                 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2123                     hio);
2124                 mtx_lock(&metadata_lock);
2125                 if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
2126                         /*
2127                          * This range is up-to-date on local component,
2128                          * so handle request locally.
2129                          */
2130                          /* Local component is 0 for now. */
2131                         ncomp = 0;
2132                 } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ {
2133                         PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY);
2134                         /*
2135                          * This range is out-of-date on local component,
2136                          * so send request to the remote node.
2137                          */
2138                          /* Remote component is 1 for now. */
2139                         ncomp = 1;
2140                 }
2141                 mtx_unlock(&metadata_lock);
2142                 hio->hio_countdown = 1;
2143                 QUEUE_INSERT1(hio, send, ncomp);
2144
2145                 /*
2146                  * Let's wait for READ to finish.
2147                  */
2148                 mtx_lock(&sync_lock);
2149                 while (!ISSYNCREQDONE(hio))
2150                         cv_wait(&sync_cond, &sync_lock);
2151                 mtx_unlock(&sync_lock);
2152
2153                 if (hio->hio_errors[ncomp] != 0) {
2154                         pjdlog_error("Unable to read synchronization data: %s.",
2155                             strerror(hio->hio_errors[ncomp]));
2156                         goto free_queue;
2157                 }
2158
2159                 /*
2160                  * We read the data from synchronization source, now write it
2161                  * to synchronization target.
2162                  */
2163                 SYNCREQ(hio);
2164                 ggio->gctl_cmd = BIO_WRITE;
2165                 for (ii = 0; ii < ncomps; ii++)
2166                         hio->hio_errors[ii] = EINVAL;
2167                 reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ",
2168                     hio);
2169                 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2170                     hio);
2171                 mtx_lock(&metadata_lock);
2172                 if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
2173                         /*
2174                          * This range is up-to-date on local component,
2175                          * so we update remote component.
2176                          */
2177                          /* Remote component is 1 for now. */
2178                         ncomp = 1;
2179                 } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ {
2180                         PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY);
2181                         /*
2182                          * This range is out-of-date on local component,
2183                          * so we update it.
2184                          */
2185                          /* Local component is 0 for now. */
2186                         ncomp = 0;
2187                 }
2188                 mtx_unlock(&metadata_lock);
2189
2190                 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2191                     hio);
2192                 hio->hio_countdown = 1;
2193                 QUEUE_INSERT1(hio, send, ncomp);
2194
2195                 /*
2196                  * Let's wait for WRITE to finish.
2197                  */
2198                 mtx_lock(&sync_lock);
2199                 while (!ISSYNCREQDONE(hio))
2200                         cv_wait(&sync_cond, &sync_lock);
2201                 mtx_unlock(&sync_lock);
2202
2203                 if (hio->hio_errors[ncomp] != 0) {
2204                         pjdlog_error("Unable to write synchronization data: %s.",
2205                             strerror(hio->hio_errors[ncomp]));
2206                         goto free_queue;
2207                 }
2208
2209                 synced += length;
2210 free_queue:
2211                 mtx_lock(&range_lock);
2212                 rangelock_del(range_sync, offset, length);
2213                 if (range_regular_wait)
2214                         cv_signal(&range_regular_cond);
2215                 mtx_unlock(&range_lock);
2216                 pjdlog_debug(2, "sync: (%p) Moving request to the free queue.",
2217                     hio);
2218                 QUEUE_INSERT2(hio, free);
2219         }
2220         /* NOTREACHED */
2221         return (NULL);
2222 }
2223
2224 void
2225 primary_config_reload(struct hast_resource *res, struct nv *nv)
2226 {
2227         unsigned int ii, ncomps;
2228         int modified, vint;
2229         const char *vstr;
2230
2231         pjdlog_info("Reloading configuration...");
2232
2233         PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
2234         PJDLOG_ASSERT(gres == res);
2235         nv_assert(nv, "remoteaddr");
2236         nv_assert(nv, "sourceaddr");
2237         nv_assert(nv, "replication");
2238         nv_assert(nv, "checksum");
2239         nv_assert(nv, "compression");
2240         nv_assert(nv, "timeout");
2241         nv_assert(nv, "exec");
2242         nv_assert(nv, "metaflush");
2243
2244         ncomps = HAST_NCOMPONENTS;
2245
2246 #define MODIFIED_REMOTEADDR     0x01
2247 #define MODIFIED_SOURCEADDR     0x02
2248 #define MODIFIED_REPLICATION    0x04
2249 #define MODIFIED_CHECKSUM       0x08
2250 #define MODIFIED_COMPRESSION    0x10
2251 #define MODIFIED_TIMEOUT        0x20
2252 #define MODIFIED_EXEC           0x40
2253 #define MODIFIED_METAFLUSH      0x80
2254         modified = 0;
2255
2256         vstr = nv_get_string(nv, "remoteaddr");
2257         if (strcmp(gres->hr_remoteaddr, vstr) != 0) {
2258                 /*
2259                  * Don't copy res->hr_remoteaddr to gres just yet.
2260                  * We want remote_close() to log disconnect from the old
2261                  * addresses, not from the new ones.
2262                  */
2263                 modified |= MODIFIED_REMOTEADDR;
2264         }
2265         vstr = nv_get_string(nv, "sourceaddr");
2266         if (strcmp(gres->hr_sourceaddr, vstr) != 0) {
2267                 strlcpy(gres->hr_sourceaddr, vstr, sizeof(gres->hr_sourceaddr));
2268                 modified |= MODIFIED_SOURCEADDR;
2269         }
2270         vint = nv_get_int32(nv, "replication");
2271         if (gres->hr_replication != vint) {
2272                 gres->hr_replication = vint;
2273                 modified |= MODIFIED_REPLICATION;
2274         }
2275         vint = nv_get_int32(nv, "checksum");
2276         if (gres->hr_checksum != vint) {
2277                 gres->hr_checksum = vint;
2278                 modified |= MODIFIED_CHECKSUM;
2279         }
2280         vint = nv_get_int32(nv, "compression");
2281         if (gres->hr_compression != vint) {
2282                 gres->hr_compression = vint;
2283                 modified |= MODIFIED_COMPRESSION;
2284         }
2285         vint = nv_get_int32(nv, "timeout");
2286         if (gres->hr_timeout != vint) {
2287                 gres->hr_timeout = vint;
2288                 modified |= MODIFIED_TIMEOUT;
2289         }
2290         vstr = nv_get_string(nv, "exec");
2291         if (strcmp(gres->hr_exec, vstr) != 0) {
2292                 strlcpy(gres->hr_exec, vstr, sizeof(gres->hr_exec));
2293                 modified |= MODIFIED_EXEC;
2294         }
2295         vint = nv_get_int32(nv, "metaflush");
2296         if (gres->hr_metaflush != vint) {
2297                 gres->hr_metaflush = vint;
2298                 modified |= MODIFIED_METAFLUSH;
2299         }
2300
2301         /*
2302          * Change timeout for connected sockets.
2303          * Don't bother if we need to reconnect.
2304          */
2305         if ((modified & MODIFIED_TIMEOUT) != 0 &&
2306             (modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR)) == 0) {
2307                 for (ii = 0; ii < ncomps; ii++) {
2308                         if (!ISREMOTE(ii))
2309                                 continue;
2310                         rw_rlock(&hio_remote_lock[ii]);
2311                         if (!ISCONNECTED(gres, ii)) {
2312                                 rw_unlock(&hio_remote_lock[ii]);
2313                                 continue;
2314                         }
2315                         rw_unlock(&hio_remote_lock[ii]);
2316                         if (proto_timeout(gres->hr_remotein,
2317                             gres->hr_timeout) == -1) {
2318                                 pjdlog_errno(LOG_WARNING,
2319                                     "Unable to set connection timeout");
2320                         }
2321                         if (proto_timeout(gres->hr_remoteout,
2322                             gres->hr_timeout) == -1) {
2323                                 pjdlog_errno(LOG_WARNING,
2324                                     "Unable to set connection timeout");
2325                         }
2326                 }
2327         }
2328         if ((modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR)) != 0) {
2329                 for (ii = 0; ii < ncomps; ii++) {
2330                         if (!ISREMOTE(ii))
2331                                 continue;
2332                         remote_close(gres, ii);
2333                 }
2334                 if (modified & MODIFIED_REMOTEADDR) {
2335                         vstr = nv_get_string(nv, "remoteaddr");
2336                         strlcpy(gres->hr_remoteaddr, vstr,
2337                             sizeof(gres->hr_remoteaddr));
2338                 }
2339         }
2340 #undef  MODIFIED_REMOTEADDR
2341 #undef  MODIFIED_SOURCEADDR
2342 #undef  MODIFIED_REPLICATION
2343 #undef  MODIFIED_CHECKSUM
2344 #undef  MODIFIED_COMPRESSION
2345 #undef  MODIFIED_TIMEOUT
2346 #undef  MODIFIED_EXEC
2347 #undef  MODIFIED_METAFLUSH
2348
2349         pjdlog_info("Configuration reloaded successfully.");
2350 }
2351
2352 static void
2353 guard_one(struct hast_resource *res, unsigned int ncomp)
2354 {
2355         struct proto_conn *in, *out;
2356
2357         if (!ISREMOTE(ncomp))
2358                 return;
2359
2360         rw_rlock(&hio_remote_lock[ncomp]);
2361
2362         if (!real_remote(res)) {
2363                 rw_unlock(&hio_remote_lock[ncomp]);
2364                 return;
2365         }
2366
2367         if (ISCONNECTED(res, ncomp)) {
2368                 PJDLOG_ASSERT(res->hr_remotein != NULL);
2369                 PJDLOG_ASSERT(res->hr_remoteout != NULL);
2370                 rw_unlock(&hio_remote_lock[ncomp]);
2371                 pjdlog_debug(2, "remote_guard: Connection to %s is ok.",
2372                     res->hr_remoteaddr);
2373                 return;
2374         }
2375
2376         PJDLOG_ASSERT(res->hr_remotein == NULL);
2377         PJDLOG_ASSERT(res->hr_remoteout == NULL);
2378         /*
2379          * Upgrade the lock. It doesn't have to be atomic as no other thread
2380          * can change connection status from disconnected to connected.
2381          */
2382         rw_unlock(&hio_remote_lock[ncomp]);
2383         pjdlog_debug(2, "remote_guard: Reconnecting to %s.",
2384             res->hr_remoteaddr);
2385         in = out = NULL;
2386         if (init_remote(res, &in, &out) == 0) {
2387                 rw_wlock(&hio_remote_lock[ncomp]);
2388                 PJDLOG_ASSERT(res->hr_remotein == NULL);
2389                 PJDLOG_ASSERT(res->hr_remoteout == NULL);
2390                 PJDLOG_ASSERT(in != NULL && out != NULL);
2391                 res->hr_remotein = in;
2392                 res->hr_remoteout = out;
2393                 rw_unlock(&hio_remote_lock[ncomp]);
2394                 pjdlog_info("Successfully reconnected to %s.",
2395                     res->hr_remoteaddr);
2396                 sync_start();
2397         } else {
2398                 /* Both connections should be NULL. */
2399                 PJDLOG_ASSERT(res->hr_remotein == NULL);
2400                 PJDLOG_ASSERT(res->hr_remoteout == NULL);
2401                 PJDLOG_ASSERT(in == NULL && out == NULL);
2402                 pjdlog_debug(2, "remote_guard: Reconnect to %s failed.",
2403                     res->hr_remoteaddr);
2404         }
2405 }
2406
2407 /*
2408  * Thread guards remote connections and reconnects when needed, handles
2409  * signals, etc.
2410  */
2411 static void *
2412 guard_thread(void *arg)
2413 {
2414         struct hast_resource *res = arg;
2415         unsigned int ii, ncomps;
2416         struct timespec timeout;
2417         time_t lastcheck, now;
2418         sigset_t mask;
2419         int signo;
2420
2421         ncomps = HAST_NCOMPONENTS;
2422         lastcheck = time(NULL);
2423
2424         PJDLOG_VERIFY(sigemptyset(&mask) == 0);
2425         PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
2426         PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
2427
2428         timeout.tv_sec = HAST_KEEPALIVE;
2429         timeout.tv_nsec = 0;
2430         signo = -1;
2431
2432         for (;;) {
2433                 switch (signo) {
2434                 case SIGINT:
2435                 case SIGTERM:
2436                         sigexit_received = true;
2437                         primary_exitx(EX_OK,
2438                             "Termination signal received, exiting.");
2439                         break;
2440                 default:
2441                         break;
2442                 }
2443
2444                 /*
2445                  * Don't check connections until we fully started,
2446                  * as we may still be looping, waiting for remote node
2447                  * to switch from primary to secondary.
2448                  */
2449                 if (fullystarted) {
2450                         pjdlog_debug(2, "remote_guard: Checking connections.");
2451                         now = time(NULL);
2452                         if (lastcheck + HAST_KEEPALIVE <= now) {
2453                                 for (ii = 0; ii < ncomps; ii++)
2454                                         guard_one(res, ii);
2455                                 lastcheck = now;
2456                         }
2457                 }
2458                 signo = sigtimedwait(&mask, NULL, &timeout);
2459         }
2460         /* NOTREACHED */
2461         return (NULL);
2462 }