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