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         refcnt_t                 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_broadcast(&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_broadcast(&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 /* Expects res->hr_amp locked, returns unlocked. */
 295 static int
 296 hast_activemap_flush(struct hast_resource *res)
 297 {
 298         const unsigned char *buf;
 299         size_t size;
 300         int ret;
 301
 302         mtx_lock(&res->hr_amp_diskmap_lock);
 303         buf = activemap_bitmap(res->hr_amp, &size);
 304         mtx_unlock(&res->hr_amp_lock);
 305         PJDLOG_ASSERT(buf != NULL);
 306         PJDLOG_ASSERT((size % res->hr_local_sectorsize) == 0);
 307         ret = 0;
 308         if (pwrite(res->hr_localfd, buf, size, METADATA_SIZE) !=
 309             (ssize_t)size) {
 310                 pjdlog_errno(LOG_ERR, "Unable to flush activemap to disk");
 311                 res->hr_stat_activemap_write_error++;
 312                 ret = -1;
 313         }
 314         if (ret == 0 && res->hr_metaflush == 1 &&
 315             g_flush(res->hr_localfd) == -1) {
 316                 if (errno == EOPNOTSUPP) {
 317                         pjdlog_warning("The %s provider doesn't support flushing write cache. Disabling it.",
 318                             res->hr_localpath);
 319                         res->hr_metaflush = 0;
 320                 } else {
 321                         pjdlog_errno(LOG_ERR,
 322                             "Unable to flush disk cache on activemap update");
 323                         res->hr_stat_activemap_flush_error++;
 324                         ret = -1;
 325                 }
 326         }
 327         mtx_unlock(&res->hr_amp_diskmap_lock);
 328         return (ret);
 329 }
 330
 331 static bool
 332 real_remote(const struct hast_resource *res)
 333 {
 334
 335         return (strcmp(res->hr_remoteaddr, "none") != 0);
 336 }
 337
 338 static void
 339 init_environment(struct hast_resource *res __unused)
 340 {
 341         struct hio *hio;
 342         unsigned int ii, ncomps;
 343
 344         /*
 345          * In the future it might be per-resource value.
 346          */
 347         ncomps = HAST_NCOMPONENTS;
 348
 349         /*
 350          * Allocate memory needed by lists.
 351          */
 352         hio_send_list = malloc(sizeof(hio_send_list[0]) * ncomps);
 353         if (hio_send_list == NULL) {
 354                 primary_exitx(EX_TEMPFAIL,
 355                     "Unable to allocate %zu bytes of memory for send lists.",
 356                     sizeof(hio_send_list[0]) * ncomps);
 357         }
 358         hio_send_list_lock = malloc(sizeof(hio_send_list_lock[0]) * ncomps);
 359         if (hio_send_list_lock == NULL) {
 360                 primary_exitx(EX_TEMPFAIL,
 361                     "Unable to allocate %zu bytes of memory for send list locks.",
 362                     sizeof(hio_send_list_lock[0]) * ncomps);
 363         }
 364         hio_send_list_cond = malloc(sizeof(hio_send_list_cond[0]) * ncomps);
 365         if (hio_send_list_cond == NULL) {
 366                 primary_exitx(EX_TEMPFAIL,
 367                     "Unable to allocate %zu bytes of memory for send list condition variables.",
 368                     sizeof(hio_send_list_cond[0]) * ncomps);
 369         }
 370         hio_recv_list = malloc(sizeof(hio_recv_list[0]) * ncomps);
 371         if (hio_recv_list == NULL) {
 372                 primary_exitx(EX_TEMPFAIL,
 373                     "Unable to allocate %zu bytes of memory for recv lists.",
 374                     sizeof(hio_recv_list[0]) * ncomps);
 375         }
 376         hio_recv_list_lock = malloc(sizeof(hio_recv_list_lock[0]) * ncomps);
 377         if (hio_recv_list_lock == NULL) {
 378                 primary_exitx(EX_TEMPFAIL,
 379                     "Unable to allocate %zu bytes of memory for recv list locks.",
 380                     sizeof(hio_recv_list_lock[0]) * ncomps);
 381         }
 382         hio_recv_list_cond = malloc(sizeof(hio_recv_list_cond[0]) * ncomps);
 383         if (hio_recv_list_cond == NULL) {
 384                 primary_exitx(EX_TEMPFAIL,
 385                     "Unable to allocate %zu bytes of memory for recv list condition variables.",
 386                     sizeof(hio_recv_list_cond[0]) * ncomps);
 387         }
 388         hio_remote_lock = malloc(sizeof(hio_remote_lock[0]) * ncomps);
 389         if (hio_remote_lock == NULL) {
 390                 primary_exitx(EX_TEMPFAIL,
 391                     "Unable to allocate %zu bytes of memory for remote connections locks.",
 392                     sizeof(hio_remote_lock[0]) * ncomps);
 393         }
 394
 395         /*
 396          * Initialize lists, their locks and theirs condition variables.
 397          */
 398         TAILQ_INIT(&hio_free_list);
 399         mtx_init(&hio_free_list_lock);
 400         cv_init(&hio_free_list_cond);
 401         for (ii = 0; ii < HAST_NCOMPONENTS; ii++) {
 402                 TAILQ_INIT(&hio_send_list[ii]);
 403                 mtx_init(&hio_send_list_lock[ii]);
 404                 cv_init(&hio_send_list_cond[ii]);
 405                 TAILQ_INIT(&hio_recv_list[ii]);
 406                 mtx_init(&hio_recv_list_lock[ii]);
 407                 cv_init(&hio_recv_list_cond[ii]);
 408                 rw_init(&hio_remote_lock[ii]);
 409         }
 410         TAILQ_INIT(&hio_done_list);
 411         mtx_init(&hio_done_list_lock);
 412         cv_init(&hio_done_list_cond);
 413         mtx_init(&metadata_lock);
 414
 415         /*
 416          * Allocate requests pool and initialize requests.
 417          */
 418         for (ii = 0; ii < HAST_HIO_MAX; ii++) {
 419                 hio = malloc(sizeof(*hio));
 420                 if (hio == NULL) {
 421                         primary_exitx(EX_TEMPFAIL,
 422                             "Unable to allocate %zu bytes of memory for hio request.",
 423                             sizeof(*hio));
 424                 }
 425                 refcnt_init(&hio->hio_countdown, 0);
 426                 hio->hio_errors = malloc(sizeof(hio->hio_errors[0]) * ncomps);
 427                 if (hio->hio_errors == NULL) {
 428                         primary_exitx(EX_TEMPFAIL,
 429                             "Unable allocate %zu bytes of memory for hio errors.",
 430                             sizeof(hio->hio_errors[0]) * ncomps);
 431                 }
 432                 hio->hio_next = malloc(sizeof(hio->hio_next[0]) * ncomps);
 433                 if (hio->hio_next == NULL) {
 434                         primary_exitx(EX_TEMPFAIL,
 435                             "Unable allocate %zu bytes of memory for hio_next field.",
 436                             sizeof(hio->hio_next[0]) * ncomps);
 437                 }
 438                 hio->hio_ggio.gctl_version = G_GATE_VERSION;
 439                 hio->hio_ggio.gctl_data = malloc(MAXPHYS);
 440                 if (hio->hio_ggio.gctl_data == NULL) {
 441                         primary_exitx(EX_TEMPFAIL,
 442                             "Unable to allocate %zu bytes of memory for gctl_data.",
 443                             MAXPHYS);
 444                 }
 445                 hio->hio_ggio.gctl_length = MAXPHYS;
 446                 hio->hio_ggio.gctl_error = 0;
 447                 TAILQ_INSERT_HEAD(&hio_free_list, hio, hio_free_next);
 448         }
 449 }
 450
 451 static bool
 452 init_resuid(struct hast_resource *res)
 453 {
 454
 455         mtx_lock(&metadata_lock);
 456         if (res->hr_resuid != 0) {
 457                 mtx_unlock(&metadata_lock);
 458                 return (false);
 459         } else {
 460                 /* Initialize unique resource identifier. */
 461                 arc4random_buf(&res->hr_resuid, sizeof(res->hr_resuid));
 462                 mtx_unlock(&metadata_lock);
 463                 if (metadata_write(res) == -1)
 464                         exit(EX_NOINPUT);
 465                 return (true);
 466         }
 467 }
 468
 469 static void
 470 init_local(struct hast_resource *res)
 471 {
 472         unsigned char *buf;
 473         size_t mapsize;
 474
 475         if (metadata_read(res, true) == -1)
 476                 exit(EX_NOINPUT);
 477         mtx_init(&res->hr_amp_lock);
 478         if (activemap_init(&res->hr_amp, res->hr_datasize, res->hr_extentsize,
 479             res->hr_local_sectorsize, res->hr_keepdirty) == -1) {
 480                 primary_exit(EX_TEMPFAIL, "Unable to create activemap");
 481         }
 482         mtx_init(&range_lock);
 483         cv_init(&range_regular_cond);
 484         if (rangelock_init(&range_regular) == -1)
 485                 primary_exit(EX_TEMPFAIL, "Unable to create regular range lock");
 486         cv_init(&range_sync_cond);
 487         if (rangelock_init(&range_sync) == -1)
 488                 primary_exit(EX_TEMPFAIL, "Unable to create sync range lock");
 489         mapsize = activemap_ondisk_size(res->hr_amp);
 490         buf = calloc(1, mapsize);
 491         if (buf == NULL) {
 492                 primary_exitx(EX_TEMPFAIL,
 493                     "Unable to allocate buffer for activemap.");
 494         }
 495         if (pread(res->hr_localfd, buf, mapsize, METADATA_SIZE) !=
 496             (ssize_t)mapsize) {
 497                 primary_exit(EX_NOINPUT, "Unable to read activemap");
 498         }
 499         activemap_copyin(res->hr_amp, buf, mapsize);
 500         free(buf);
 501         if (res->hr_resuid != 0)
 502                 return;
 503         /*
 504          * We're using provider for the first time. Initialize local and remote
 505          * counters. We don't initialize resuid here, as we want to do it just
 506          * in time. The reason for this is that we want to inform secondary
 507          * that there were no writes yet, so there is no need to synchronize
 508          * anything.
 509          */
 510         res->hr_primary_localcnt = 0;
 511         res->hr_primary_remotecnt = 0;
 512         if (metadata_write(res) == -1)
 513                 exit(EX_NOINPUT);
 514 }
 515
 516 static int
 517 primary_connect(struct hast_resource *res, struct proto_conn **connp)
 518 {
 519         struct proto_conn *conn;
 520         int16_t val;
 521
 522         val = 1;
 523         if (proto_send(res->hr_conn, &val, sizeof(val)) == -1) {
 524                 primary_exit(EX_TEMPFAIL,
 525                     "Unable to send connection request to parent");
 526         }
 527         if (proto_recv(res->hr_conn, &val, sizeof(val)) == -1) {
 528                 primary_exit(EX_TEMPFAIL,
 529                     "Unable to receive reply to connection request from parent");
 530         }
 531         if (val != 0) {
 532                 errno = val;
 533                 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
 534                     res->hr_remoteaddr);
 535                 return (-1);
 536         }
 537         if (proto_connection_recv(res->hr_conn, true, &conn) == -1) {
 538                 primary_exit(EX_TEMPFAIL,
 539                     "Unable to receive connection from parent");
 540         }
 541         if (proto_connect_wait(conn, res->hr_timeout) == -1) {
 542                 pjdlog_errno(LOG_WARNING, "Unable to connect to %s",
 543                     res->hr_remoteaddr);
 544                 proto_close(conn);
 545                 return (-1);
 546         }
 547         /* Error in setting timeout is not critical, but why should it fail? */
 548         if (proto_timeout(conn, res->hr_timeout) == -1)
 549                 pjdlog_errno(LOG_WARNING, "Unable to set connection timeout");
 550
 551         *connp = conn;
 552
 553         return (0);
 554 }
 555
 556 /*
 557  * Function instructs GEOM_GATE to handle reads directly from within the kernel.
 558  */
 559 static void
 560 enable_direct_reads(struct hast_resource *res)
 561 {
 562         struct g_gate_ctl_modify ggiomodify;
 563
 564         bzero(&ggiomodify, sizeof(ggiomodify));
 565         ggiomodify.gctl_version = G_GATE_VERSION;
 566         ggiomodify.gctl_unit = res->hr_ggateunit;
 567         ggiomodify.gctl_modify = GG_MODIFY_READPROV | GG_MODIFY_READOFFSET;
 568         strlcpy(ggiomodify.gctl_readprov, res->hr_localpath,
 569             sizeof(ggiomodify.gctl_readprov));
 570         ggiomodify.gctl_readoffset = res->hr_localoff;
 571         if (ioctl(res->hr_ggatefd, G_GATE_CMD_MODIFY, &ggiomodify) == 0)
 572                 pjdlog_debug(1, "Direct reads enabled.");
 573         else
 574                 pjdlog_errno(LOG_WARNING, "Failed to enable direct reads");
 575 }
 576
 577 static int
 578 init_remote(struct hast_resource *res, struct proto_conn **inp,
 579     struct proto_conn **outp)
 580 {
 581         struct proto_conn *in, *out;
 582         struct nv *nvout, *nvin;
 583         const unsigned char *token;
 584         unsigned char *map;
 585         const char *errmsg;
 586         int32_t extentsize;
 587         int64_t datasize;
 588         uint32_t mapsize;
 589         uint8_t version;
 590         size_t size;
 591         int error;
 592
 593         PJDLOG_ASSERT((inp == NULL && outp == NULL) || (inp != NULL && outp != NULL));
 594         PJDLOG_ASSERT(real_remote(res));
 595
 596         in = out = NULL;
 597         errmsg = NULL;
 598
 599         if (primary_connect(res, &out) == -1)
 600                 return (ECONNREFUSED);
 601
 602         error = ECONNABORTED;
 603
 604         /*
 605          * First handshake step.
 606          * Setup outgoing connection with remote node.
 607          */
 608         nvout = nv_alloc();
 609         nv_add_string(nvout, res->hr_name, "resource");
 610         nv_add_uint8(nvout, HAST_PROTO_VERSION, "version");
 611         if (nv_error(nvout) != 0) {
 612                 pjdlog_common(LOG_WARNING, 0, nv_error(nvout),
 613                     "Unable to allocate header for connection with %s",
 614                     res->hr_remoteaddr);
 615                 nv_free(nvout);
 616                 goto close;
 617         }
 618         if (hast_proto_send(res, out, nvout, NULL, 0) == -1) {
 619                 pjdlog_errno(LOG_WARNING,
 620                     "Unable to send handshake header to %s",
 621                     res->hr_remoteaddr);
 622                 nv_free(nvout);
 623                 goto close;
 624         }
 625         nv_free(nvout);
 626         if (hast_proto_recv_hdr(out, &nvin) == -1) {
 627                 pjdlog_errno(LOG_WARNING,
 628                     "Unable to receive handshake header from %s",
 629                     res->hr_remoteaddr);
 630                 goto close;
 631         }
 632         errmsg = nv_get_string(nvin, "errmsg");
 633         if (errmsg != NULL) {
 634                 pjdlog_warning("%s", errmsg);
 635                 if (nv_exists(nvin, "wait"))
 636                         error = EBUSY;
 637                 nv_free(nvin);
 638                 goto close;
 639         }
 640         version = nv_get_uint8(nvin, "version");
 641         if (version == 0) {
 642                 /*
 643                  * If no version is sent, it means this is protocol version 1.
 644                  */
 645                 version = 1;
 646         }
 647         if (version > HAST_PROTO_VERSION) {
 648                 pjdlog_warning("Invalid version received (%hhu).", version);
 649                 nv_free(nvin);
 650                 goto close;
 651         }
 652         res->hr_version = version;
 653         pjdlog_debug(1, "Negotiated protocol version %d.", res->hr_version);
 654         token = nv_get_uint8_array(nvin, &size, "token");
 655         if (token == NULL) {
 656                 pjdlog_warning("Handshake header from %s has no 'token' field.",
 657                     res->hr_remoteaddr);
 658                 nv_free(nvin);
 659                 goto close;
 660         }
 661         if (size != sizeof(res->hr_token)) {
 662                 pjdlog_warning("Handshake header from %s contains 'token' of wrong size (got %zu, expected %zu).",
 663                     res->hr_remoteaddr, size, sizeof(res->hr_token));
 664                 nv_free(nvin);
 665                 goto close;
 666         }
 667         bcopy(token, res->hr_token, sizeof(res->hr_token));
 668         nv_free(nvin);
 669
 670         /*
 671          * Second handshake step.
 672          * Setup incoming connection with remote node.
 673          */
 674         if (primary_connect(res, &in) == -1)
 675                 goto close;
 676
 677         nvout = nv_alloc();
 678         nv_add_string(nvout, res->hr_name, "resource");
 679         nv_add_uint8_array(nvout, res->hr_token, sizeof(res->hr_token),
 680             "token");
 681         if (res->hr_resuid == 0) {
 682                 /*
 683                  * The resuid field was not yet initialized.
 684                  * Because we do synchronization inside init_resuid(), it is
 685                  * possible that someone already initialized it, the function
 686                  * will return false then, but if we successfully initialized
 687                  * it, we will get true. True means that there were no writes
 688                  * to this resource yet and we want to inform secondary that
 689                  * synchronization is not needed by sending "virgin" argument.
 690                  */
 691                 if (init_resuid(res))
 692                         nv_add_int8(nvout, 1, "virgin");
 693         }
 694         nv_add_uint64(nvout, res->hr_resuid, "resuid");
 695         nv_add_uint64(nvout, res->hr_primary_localcnt, "localcnt");
 696         nv_add_uint64(nvout, res->hr_primary_remotecnt, "remotecnt");
 697         if (nv_error(nvout) != 0) {
 698                 pjdlog_common(LOG_WARNING, 0, nv_error(nvout),
 699                     "Unable to allocate header for connection with %s",
 700                     res->hr_remoteaddr);
 701                 nv_free(nvout);
 702                 goto close;
 703         }
 704         if (hast_proto_send(res, in, nvout, NULL, 0) == -1) {
 705                 pjdlog_errno(LOG_WARNING,
 706                     "Unable to send handshake header to %s",
 707                     res->hr_remoteaddr);
 708                 nv_free(nvout);
 709                 goto close;
 710         }
 711         nv_free(nvout);
 712         if (hast_proto_recv_hdr(out, &nvin) == -1) {
 713                 pjdlog_errno(LOG_WARNING,
 714                     "Unable to receive handshake header from %s",
 715                     res->hr_remoteaddr);
 716                 goto close;
 717         }
 718         errmsg = nv_get_string(nvin, "errmsg");
 719         if (errmsg != NULL) {
 720                 pjdlog_warning("%s", errmsg);
 721                 nv_free(nvin);
 722                 goto close;
 723         }
 724         datasize = nv_get_int64(nvin, "datasize");
 725         if (datasize != res->hr_datasize) {
 726                 pjdlog_warning("Data size differs between nodes (local=%jd, remote=%jd).",
 727                     (intmax_t)res->hr_datasize, (intmax_t)datasize);
 728                 nv_free(nvin);
 729                 goto close;
 730         }
 731         extentsize = nv_get_int32(nvin, "extentsize");
 732         if (extentsize != res->hr_extentsize) {
 733                 pjdlog_warning("Extent size differs between nodes (local=%zd, remote=%zd).",
 734                     (ssize_t)res->hr_extentsize, (ssize_t)extentsize);
 735                 nv_free(nvin);
 736                 goto close;
 737         }
 738         res->hr_secondary_localcnt = nv_get_uint64(nvin, "localcnt");
 739         res->hr_secondary_remotecnt = nv_get_uint64(nvin, "remotecnt");
 740         res->hr_syncsrc = nv_get_uint8(nvin, "syncsrc");
 741         if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY)
 742                 enable_direct_reads(res);
 743         if (nv_exists(nvin, "virgin")) {
 744                 /*
 745                  * Secondary was reinitialized, bump localcnt if it is 0 as
 746                  * only we have the data.
 747                  */
 748                 PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_PRIMARY);
 749                 PJDLOG_ASSERT(res->hr_secondary_localcnt == 0);
 750
 751                 if (res->hr_primary_localcnt == 0) {
 752                         PJDLOG_ASSERT(res->hr_secondary_remotecnt == 0);
 753
 754                         mtx_lock(&metadata_lock);
 755                         res->hr_primary_localcnt++;
 756                         pjdlog_debug(1, "Increasing localcnt to %ju.",
 757                             (uintmax_t)res->hr_primary_localcnt);
 758                         (void)metadata_write(res);
 759                         mtx_unlock(&metadata_lock);
 760                 }
 761         }
 762         map = NULL;
 763         mapsize = nv_get_uint32(nvin, "mapsize");
 764         if (mapsize > 0) {
 765                 map = malloc(mapsize);
 766                 if (map == NULL) {
 767                         pjdlog_error("Unable to allocate memory for remote activemap (mapsize=%ju).",
 768                             (uintmax_t)mapsize);
 769                         nv_free(nvin);
 770                         goto close;
 771                 }
 772                 /*
 773                  * Remote node have some dirty extents on its own, lets
 774                  * download its activemap.
 775                  */
 776                 if (hast_proto_recv_data(res, out, nvin, map,
 777                     mapsize) == -1) {
 778                         pjdlog_errno(LOG_ERR,
 779                             "Unable to receive remote activemap");
 780                         nv_free(nvin);
 781                         free(map);
 782                         goto close;
 783                 }
 784                 mtx_lock(&res->hr_amp_lock);
 785                 /*
 786                  * Merge local and remote bitmaps.
 787                  */
 788                 activemap_merge(res->hr_amp, map, mapsize);
 789                 free(map);
 790                 /*
 791                  * Now that we merged bitmaps from both nodes, flush it to the
 792                  * disk before we start to synchronize.
 793                  */
 794                 (void)hast_activemap_flush(res);
 795         }
 796         nv_free(nvin);
 797 #ifdef notyet
 798         /* Setup directions. */
 799         if (proto_send(out, NULL, 0) == -1)
 800                 pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
 801         if (proto_recv(in, NULL, 0) == -1)
 802                 pjdlog_errno(LOG_WARNING, "Unable to set connection direction");
 803 #endif
 804         pjdlog_info("Connected to %s.", res->hr_remoteaddr);
 805         if (res->hr_original_replication == HAST_REPLICATION_MEMSYNC &&
 806             res->hr_version < 2) {
 807                 pjdlog_warning("The 'memsync' replication mode is not supported by the remote node, falling back to 'fullsync' mode.");
 808                 res->hr_replication = HAST_REPLICATION_FULLSYNC;
 809         } else if (res->hr_replication != res->hr_original_replication) {
 810                 /*
 811                  * This is in case hastd disconnected and was upgraded.
 812                  */
 813                 res->hr_replication = res->hr_original_replication;
 814         }
 815         if (inp != NULL && outp != NULL) {
 816                 *inp = in;
 817                 *outp = out;
 818         } else {
 819                 res->hr_remotein = in;
 820                 res->hr_remoteout = out;
 821         }
 822         event_send(res, EVENT_CONNECT);
 823         return (0);
 824 close:
 825         if (errmsg != NULL && strcmp(errmsg, "Split-brain condition!") == 0)
 826                 event_send(res, EVENT_SPLITBRAIN);
 827         proto_close(out);
 828         if (in != NULL)
 829                 proto_close(in);
 830         return (error);
 831 }
 832
 833 static void
 834 sync_start(void)
 835 {
 836
 837         mtx_lock(&sync_lock);
 838         sync_inprogress = true;
 839         mtx_unlock(&sync_lock);
 840         cv_signal(&sync_cond);
 841 }
 842
 843 static void
 844 sync_stop(void)
 845 {
 846
 847         mtx_lock(&sync_lock);
 848         if (sync_inprogress)
 849                 sync_inprogress = false;
 850         mtx_unlock(&sync_lock);
 851 }
 852
 853 static void
 854 init_ggate(struct hast_resource *res)
 855 {
 856         struct g_gate_ctl_create ggiocreate;
 857         struct g_gate_ctl_cancel ggiocancel;
 858
 859         /*
 860          * We communicate with ggate via /dev/ggctl. Open it.
 861          */
 862         res->hr_ggatefd = open("/dev/" G_GATE_CTL_NAME, O_RDWR);
 863         if (res->hr_ggatefd == -1)
 864                 primary_exit(EX_OSFILE, "Unable to open /dev/" G_GATE_CTL_NAME);
 865         /*
 866          * Create provider before trying to connect, as connection failure
 867          * is not critical, but may take some time.
 868          */
 869         bzero(&ggiocreate, sizeof(ggiocreate));
 870         ggiocreate.gctl_version = G_GATE_VERSION;
 871         ggiocreate.gctl_mediasize = res->hr_datasize;
 872         ggiocreate.gctl_sectorsize = res->hr_local_sectorsize;
 873         ggiocreate.gctl_flags = 0;
 874         ggiocreate.gctl_maxcount = 0;
 875         ggiocreate.gctl_timeout = 0;
 876         ggiocreate.gctl_unit = G_GATE_NAME_GIVEN;
 877         snprintf(ggiocreate.gctl_name, sizeof(ggiocreate.gctl_name), "hast/%s",
 878             res->hr_provname);
 879         if (ioctl(res->hr_ggatefd, G_GATE_CMD_CREATE, &ggiocreate) == 0) {
 880                 pjdlog_info("Device hast/%s created.", res->hr_provname);
 881                 res->hr_ggateunit = ggiocreate.gctl_unit;
 882                 return;
 883         }
 884         if (errno != EEXIST) {
 885                 primary_exit(EX_OSERR, "Unable to create hast/%s device",
 886                     res->hr_provname);
 887         }
 888         pjdlog_debug(1,
 889             "Device hast/%s already exists, we will try to take it over.",
 890             res->hr_provname);
 891         /*
 892          * If we received EEXIST, we assume that the process who created the
 893          * provider died and didn't clean up. In that case we will start from
 894          * where he left of.
 895          */
 896         bzero(&ggiocancel, sizeof(ggiocancel));
 897         ggiocancel.gctl_version = G_GATE_VERSION;
 898         ggiocancel.gctl_unit = G_GATE_NAME_GIVEN;
 899         snprintf(ggiocancel.gctl_name, sizeof(ggiocancel.gctl_name), "hast/%s",
 900             res->hr_provname);
 901         if (ioctl(res->hr_ggatefd, G_GATE_CMD_CANCEL, &ggiocancel) == 0) {
 902                 pjdlog_info("Device hast/%s recovered.", res->hr_provname);
 903                 res->hr_ggateunit = ggiocancel.gctl_unit;
 904                 return;
 905         }
 906         primary_exit(EX_OSERR, "Unable to take over hast/%s device",
 907             res->hr_provname);
 908 }
 909
 910 void
 911 hastd_primary(struct hast_resource *res)
 912 {
 913         pthread_t td;
 914         pid_t pid;
 915         int error, mode, debuglevel;
 916
 917         /*
 918          * Create communication channel for sending control commands from
 919          * parent to child.
 920          */
 921         if (proto_client(NULL, "socketpair://", &res->hr_ctrl) == -1) {
 922                 /* TODO: There's no need for this to be fatal error. */
 923                 KEEP_ERRNO((void)pidfile_remove(pfh));
 924                 pjdlog_exit(EX_OSERR,
 925                     "Unable to create control sockets between parent and child");
 926         }
 927         /*
 928          * Create communication channel for sending events from child to parent.
 929          */
 930         if (proto_client(NULL, "socketpair://", &res->hr_event) == -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 event sockets between child and parent");
 935         }
 936         /*
 937          * Create communication channel for sending connection requests from
 938          * child to parent.
 939          */
 940         if (proto_client(NULL, "socketpair://", &res->hr_conn) == -1) {
 941                 /* TODO: There's no need for this to be fatal error. */
 942                 KEEP_ERRNO((void)pidfile_remove(pfh));
 943                 pjdlog_exit(EX_OSERR,
 944                     "Unable to create connection sockets between child and parent");
 945         }
 946
 947         pid = fork();
 948         if (pid == -1) {
 949                 /* TODO: There's no need for this to be fatal error. */
 950                 KEEP_ERRNO((void)pidfile_remove(pfh));
 951                 pjdlog_exit(EX_TEMPFAIL, "Unable to fork");
 952         }
 953
 954         if (pid > 0) {
 955                 /* This is parent. */
 956                 /* Declare that we are receiver. */
 957                 proto_recv(res->hr_event, NULL, 0);
 958                 proto_recv(res->hr_conn, NULL, 0);
 959                 /* Declare that we are sender. */
 960                 proto_send(res->hr_ctrl, NULL, 0);
 961                 res->hr_workerpid = pid;
 962                 return;
 963         }
 964
 965         gres = res;
 966         mode = pjdlog_mode_get();
 967         debuglevel = pjdlog_debug_get();
 968
 969         /* Declare that we are sender. */
 970         proto_send(res->hr_event, NULL, 0);
 971         proto_send(res->hr_conn, NULL, 0);
 972         /* Declare that we are receiver. */
 973         proto_recv(res->hr_ctrl, NULL, 0);
 974         descriptors_cleanup(res);
 975
 976         descriptors_assert(res, mode);
 977
 978         pjdlog_init(mode);
 979         pjdlog_debug_set(debuglevel);
 980         pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
 981         setproctitle("%s (%s)", res->hr_name, role2str(res->hr_role));
 982
 983         init_local(res);
 984         init_ggate(res);
 985         init_environment(res);
 986
 987         if (drop_privs(res) != 0) {
 988                 cleanup(res);
 989                 exit(EX_CONFIG);
 990         }
 991         pjdlog_info("Privileges successfully dropped.");
 992
 993         /*
 994          * Create the guard thread first, so we can handle signals from the
 995          * very beginning.
 996          */
 997         error = pthread_create(&td, NULL, guard_thread, res);
 998         PJDLOG_ASSERT(error == 0);
 999         /*
1000          * Create the control thread before sending any event to the parent,
1001          * as we can deadlock when parent sends control request to worker,
1002          * but worker has no control thread started yet, so parent waits.
1003          * In the meantime worker sends an event to the parent, but parent
1004          * is unable to handle the event, because it waits for control
1005          * request response.
1006          */
1007         error = pthread_create(&td, NULL, ctrl_thread, res);
1008         PJDLOG_ASSERT(error == 0);
1009         if (real_remote(res)) {
1010                 error = init_remote(res, NULL, NULL);
1011                 if (error == 0) {
1012                         sync_start();
1013                 } else if (error == EBUSY) {
1014                         time_t start = time(NULL);
1015
1016                         pjdlog_warning("Waiting for remote node to become %s for %ds.",
1017                             role2str(HAST_ROLE_SECONDARY),
1018                             res->hr_timeout);
1019                         for (;;) {
1020                                 sleep(1);
1021                                 error = init_remote(res, NULL, NULL);
1022                                 if (error != EBUSY)
1023                                         break;
1024                                 if (time(NULL) > start + res->hr_timeout)
1025                                         break;
1026                         }
1027                         if (error == EBUSY) {
1028                                 pjdlog_warning("Remote node is still %s, starting anyway.",
1029                                     role2str(HAST_ROLE_PRIMARY));
1030                         }
1031                 }
1032         }
1033         error = pthread_create(&td, NULL, ggate_recv_thread, res);
1034         PJDLOG_ASSERT(error == 0);
1035         error = pthread_create(&td, NULL, local_send_thread, res);
1036         PJDLOG_ASSERT(error == 0);
1037         error = pthread_create(&td, NULL, remote_send_thread, res);
1038         PJDLOG_ASSERT(error == 0);
1039         error = pthread_create(&td, NULL, remote_recv_thread, res);
1040         PJDLOG_ASSERT(error == 0);
1041         error = pthread_create(&td, NULL, ggate_send_thread, res);
1042         PJDLOG_ASSERT(error == 0);
1043         fullystarted = true;
1044         (void)sync_thread(res);
1045 }
1046
1047 static void
1048 reqlog(int loglevel, int debuglevel, struct g_gate_ctl_io *ggio,
1049     const char *fmt, ...)
1050 {
1051         char msg[1024];
1052         va_list ap;
1053
1054         va_start(ap, fmt);
1055         (void)vsnprintf(msg, sizeof(msg), fmt, ap);
1056         va_end(ap);
1057         switch (ggio->gctl_cmd) {
1058         case BIO_READ:
1059                 (void)snprlcat(msg, sizeof(msg), "READ(%ju, %ju).",
1060                     (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1061                 break;
1062         case BIO_DELETE:
1063                 (void)snprlcat(msg, sizeof(msg), "DELETE(%ju, %ju).",
1064                     (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1065                 break;
1066         case BIO_FLUSH:
1067                 (void)snprlcat(msg, sizeof(msg), "FLUSH.");
1068                 break;
1069         case BIO_WRITE:
1070                 (void)snprlcat(msg, sizeof(msg), "WRITE(%ju, %ju).",
1071                     (uintmax_t)ggio->gctl_offset, (uintmax_t)ggio->gctl_length);
1072                 break;
1073         default:
1074                 (void)snprlcat(msg, sizeof(msg), "UNKNOWN(%u).",
1075                     (unsigned int)ggio->gctl_cmd);
1076                 break;
1077         }
1078         pjdlog_common(loglevel, debuglevel, -1, "%s", msg);
1079 }
1080
1081 static void
1082 remote_close(struct hast_resource *res, int ncomp)
1083 {
1084
1085         rw_wlock(&hio_remote_lock[ncomp]);
1086         /*
1087          * Check for a race between dropping rlock and acquiring wlock -
1088          * another thread can close connection in-between.
1089          */
1090         if (!ISCONNECTED(res, ncomp)) {
1091                 PJDLOG_ASSERT(res->hr_remotein == NULL);
1092                 PJDLOG_ASSERT(res->hr_remoteout == NULL);
1093                 rw_unlock(&hio_remote_lock[ncomp]);
1094                 return;
1095         }
1096
1097         PJDLOG_ASSERT(res->hr_remotein != NULL);
1098         PJDLOG_ASSERT(res->hr_remoteout != NULL);
1099
1100         pjdlog_debug(2, "Closing incoming connection to %s.",
1101             res->hr_remoteaddr);
1102         proto_close(res->hr_remotein);
1103         res->hr_remotein = NULL;
1104         pjdlog_debug(2, "Closing outgoing connection to %s.",
1105             res->hr_remoteaddr);
1106         proto_close(res->hr_remoteout);
1107         res->hr_remoteout = NULL;
1108
1109         rw_unlock(&hio_remote_lock[ncomp]);
1110
1111         pjdlog_warning("Disconnected from %s.", res->hr_remoteaddr);
1112
1113         /*
1114          * Stop synchronization if in-progress.
1115          */
1116         sync_stop();
1117
1118         event_send(res, EVENT_DISCONNECT);
1119 }
1120
1121 /*
1122  * Acknowledge write completion to the kernel, but don't update activemap yet.
1123  */
1124 static void
1125 write_complete(struct hast_resource *res, struct hio *hio)
1126 {
1127         struct g_gate_ctl_io *ggio;
1128         unsigned int ncomp;
1129
1130         PJDLOG_ASSERT(!hio->hio_done);
1131
1132         ggio = &hio->hio_ggio;
1133         PJDLOG_ASSERT(ggio->gctl_cmd == BIO_WRITE);
1134
1135         /*
1136          * Bump local count if this is first write after
1137          * connection failure with remote node.
1138          */
1139         ncomp = 1;
1140         rw_rlock(&hio_remote_lock[ncomp]);
1141         if (!ISCONNECTED(res, ncomp)) {
1142                 mtx_lock(&metadata_lock);
1143                 if (res->hr_primary_localcnt == res->hr_secondary_remotecnt) {
1144                         res->hr_primary_localcnt++;
1145                         pjdlog_debug(1, "Increasing localcnt to %ju.",
1146                             (uintmax_t)res->hr_primary_localcnt);
1147                         (void)metadata_write(res);
1148                 }
1149                 mtx_unlock(&metadata_lock);
1150         }
1151         rw_unlock(&hio_remote_lock[ncomp]);
1152         if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) == -1)
1153                 primary_exit(EX_OSERR, "G_GATE_CMD_DONE failed");
1154         hio->hio_done = true;
1155 }
1156
1157 /*
1158  * Thread receives ggate I/O requests from the kernel and passes them to
1159  * appropriate threads:
1160  * WRITE - always goes to both local_send and remote_send threads
1161  * READ (when the block is up-to-date on local component) -
1162  *      only local_send thread
1163  * READ (when the block isn't up-to-date on local component) -
1164  *      only remote_send thread
1165  * DELETE - always goes to both local_send and remote_send threads
1166  * FLUSH - always goes to both local_send and remote_send threads
1167  */
1168 static void *
1169 ggate_recv_thread(void *arg)
1170 {
1171         struct hast_resource *res = arg;
1172         struct g_gate_ctl_io *ggio;
1173         struct hio *hio;
1174         unsigned int ii, ncomp, ncomps;
1175         int error;
1176
1177         for (;;) {
1178                 pjdlog_debug(2, "ggate_recv: Taking free request.");
1179                 QUEUE_TAKE2(hio, free);
1180                 pjdlog_debug(2, "ggate_recv: (%p) Got free request.", hio);
1181                 ggio = &hio->hio_ggio;
1182                 ggio->gctl_unit = res->hr_ggateunit;
1183                 ggio->gctl_length = MAXPHYS;
1184                 ggio->gctl_error = 0;
1185                 hio->hio_done = false;
1186                 hio->hio_replication = res->hr_replication;
1187                 pjdlog_debug(2,
1188                     "ggate_recv: (%p) Waiting for request from the kernel.",
1189                     hio);
1190                 if (ioctl(res->hr_ggatefd, G_GATE_CMD_START, ggio) == -1) {
1191                         if (sigexit_received)
1192                                 pthread_exit(NULL);
1193                         primary_exit(EX_OSERR, "G_GATE_CMD_START failed");
1194                 }
1195                 error = ggio->gctl_error;
1196                 switch (error) {
1197                 case 0:
1198                         break;
1199                 case ECANCELED:
1200                         /* Exit gracefully. */
1201                         if (!sigexit_received) {
1202                                 pjdlog_debug(2,
1203                                     "ggate_recv: (%p) Received cancel from the kernel.",
1204                                     hio);
1205                                 pjdlog_info("Received cancel from the kernel, exiting.");
1206                         }
1207                         pthread_exit(NULL);
1208                 case ENOMEM:
1209                         /*
1210                          * Buffer too small? Impossible, we allocate MAXPHYS
1211                          * bytes - request can't be bigger than that.
1212                          */
1213                         /* FALLTHROUGH */
1214                 case ENXIO:
1215                 default:
1216                         primary_exitx(EX_OSERR, "G_GATE_CMD_START failed: %s.",
1217                             strerror(error));
1218                 }
1219
1220                 ncomp = 0;
1221                 ncomps = HAST_NCOMPONENTS;
1222
1223                 for (ii = 0; ii < ncomps; ii++)
1224                         hio->hio_errors[ii] = EINVAL;
1225                 reqlog(LOG_DEBUG, 2, ggio,
1226                     "ggate_recv: (%p) Request received from the kernel: ",
1227                     hio);
1228
1229                 /*
1230                  * Inform all components about new write request.
1231                  * For read request prefer local component unless the given
1232                  * range is out-of-date, then use remote component.
1233                  */
1234                 switch (ggio->gctl_cmd) {
1235                 case BIO_READ:
1236                         res->hr_stat_read++;
1237                         ncomps = 1;
1238                         mtx_lock(&metadata_lock);
1239                         if (res->hr_syncsrc == HAST_SYNCSRC_UNDEF ||
1240                             res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
1241                                 /*
1242                                  * This range is up-to-date on local component,
1243                                  * so handle request locally.
1244                                  */
1245                                  /* Local component is 0 for now. */
1246                                 ncomp = 0;
1247                         } else /* if (res->hr_syncsrc ==
1248                             HAST_SYNCSRC_SECONDARY) */ {
1249                                 PJDLOG_ASSERT(res->hr_syncsrc ==
1250                                     HAST_SYNCSRC_SECONDARY);
1251                                 /*
1252                                  * This range is out-of-date on local component,
1253                                  * so send request to the remote node.
1254                                  */
1255                                  /* Remote component is 1 for now. */
1256                                 ncomp = 1;
1257                         }
1258                         mtx_unlock(&metadata_lock);
1259                         break;
1260                 case BIO_WRITE:
1261                         res->hr_stat_write++;
1262                         if (res->hr_resuid == 0 &&
1263                             res->hr_primary_localcnt == 0) {
1264                                 /* This is first write. */
1265                                 res->hr_primary_localcnt = 1;
1266                         }
1267                         for (;;) {
1268                                 mtx_lock(&range_lock);
1269                                 if (rangelock_islocked(range_sync,
1270                                     ggio->gctl_offset, ggio->gctl_length)) {
1271                                         pjdlog_debug(2,
1272                                             "regular: Range offset=%jd length=%zu locked.",
1273                                             (intmax_t)ggio->gctl_offset,
1274                                             (size_t)ggio->gctl_length);
1275                                         range_regular_wait = true;
1276                                         cv_wait(&range_regular_cond, &range_lock);
1277                                         range_regular_wait = false;
1278                                         mtx_unlock(&range_lock);
1279                                         continue;
1280                                 }
1281                                 if (rangelock_add(range_regular,
1282                                     ggio->gctl_offset, ggio->gctl_length) == -1) {
1283                                         mtx_unlock(&range_lock);
1284                                         pjdlog_debug(2,
1285                                             "regular: Range offset=%jd length=%zu is already locked, waiting.",
1286                                             (intmax_t)ggio->gctl_offset,
1287                                             (size_t)ggio->gctl_length);
1288                                         sleep(1);
1289                                         continue;
1290                                 }
1291                                 mtx_unlock(&range_lock);
1292                                 break;
1293                         }
1294                         mtx_lock(&res->hr_amp_lock);
1295                         if (activemap_write_start(res->hr_amp,
1296                             ggio->gctl_offset, ggio->gctl_length)) {
1297                                 res->hr_stat_activemap_update++;
1298                                 (void)hast_activemap_flush(res);
1299                         } else {
1300                                 mtx_unlock(&res->hr_amp_lock);
1301                         }
1302                         break;
1303                 case BIO_DELETE:
1304                         res->hr_stat_delete++;
1305                         break;
1306                 case BIO_FLUSH:
1307                         res->hr_stat_flush++;
1308                         break;
1309                 }
1310                 pjdlog_debug(2,
1311                     "ggate_recv: (%p) Moving request to the send queues.", hio);
1312                 if (hio->hio_replication == HAST_REPLICATION_MEMSYNC &&
1313                     ggio->gctl_cmd == BIO_WRITE) {
1314                         /* Each remote request needs two responses in memsync. */
1315                         refcnt_init(&hio->hio_countdown, ncomps + 1);
1316                 } else {
1317                         refcnt_init(&hio->hio_countdown, ncomps);
1318                 }
1319                 for (ii = ncomp; ii < ncomps; ii++)
1320                         QUEUE_INSERT1(hio, send, ii);
1321         }
1322         /* NOTREACHED */
1323         return (NULL);
1324 }
1325
1326 /*
1327  * Thread reads from or writes to local component.
1328  * If local read fails, it redirects it to remote_send thread.
1329  */
1330 static void *
1331 local_send_thread(void *arg)
1332 {
1333         struct hast_resource *res = arg;
1334         struct g_gate_ctl_io *ggio;
1335         struct hio *hio;
1336         unsigned int ncomp, rncomp;
1337         ssize_t ret;
1338
1339         /* Local component is 0 for now. */
1340         ncomp = 0;
1341         /* Remote component is 1 for now. */
1342         rncomp = 1;
1343
1344         for (;;) {
1345                 pjdlog_debug(2, "local_send: Taking request.");
1346                 QUEUE_TAKE1(hio, send, ncomp, 0);
1347                 pjdlog_debug(2, "local_send: (%p) Got request.", hio);
1348                 ggio = &hio->hio_ggio;
1349                 switch (ggio->gctl_cmd) {
1350                 case BIO_READ:
1351                         ret = pread(res->hr_localfd, ggio->gctl_data,
1352                             ggio->gctl_length,
1353                             ggio->gctl_offset + res->hr_localoff);
1354                         if (ret == ggio->gctl_length)
1355                                 hio->hio_errors[ncomp] = 0;
1356                         else if (!ISSYNCREQ(hio)) {
1357                                 /*
1358                                  * If READ failed, try to read from remote node.
1359                                  */
1360                                 if (ret == -1) {
1361                                         reqlog(LOG_WARNING, 0, ggio,
1362                                             "Local request failed (%s), trying remote node. ",
1363                                             strerror(errno));
1364                                 } else if (ret != ggio->gctl_length) {
1365                                         reqlog(LOG_WARNING, 0, ggio,
1366                                             "Local request failed (%zd != %jd), trying remote node. ",
1367                                             ret, (intmax_t)ggio->gctl_length);
1368                                 }
1369                                 QUEUE_INSERT1(hio, send, rncomp);
1370                                 continue;
1371                         }
1372                         break;
1373                 case BIO_WRITE:
1374                         ret = pwrite(res->hr_localfd, ggio->gctl_data,
1375                             ggio->gctl_length,
1376                             ggio->gctl_offset + res->hr_localoff);
1377                         if (ret == -1) {
1378                                 hio->hio_errors[ncomp] = errno;
1379                                 reqlog(LOG_WARNING, 0, ggio,
1380                                     "Local request failed (%s): ",
1381                                     strerror(errno));
1382                         } else if (ret != ggio->gctl_length) {
1383                                 hio->hio_errors[ncomp] = EIO;
1384                                 reqlog(LOG_WARNING, 0, ggio,
1385                                     "Local request failed (%zd != %jd): ",
1386                                     ret, (intmax_t)ggio->gctl_length);
1387                         } else {
1388                                 hio->hio_errors[ncomp] = 0;
1389                                 if (hio->hio_replication ==
1390                                     HAST_REPLICATION_ASYNC) {
1391                                         ggio->gctl_error = 0;
1392                                         write_complete(res, hio);
1393                                 }
1394                         }
1395                         break;
1396                 case BIO_DELETE:
1397                         ret = g_delete(res->hr_localfd,
1398                             ggio->gctl_offset + res->hr_localoff,
1399                             ggio->gctl_length);
1400                         if (ret == -1) {
1401                                 hio->hio_errors[ncomp] = errno;
1402                                 reqlog(LOG_WARNING, 0, ggio,
1403                                     "Local request failed (%s): ",
1404                                     strerror(errno));
1405                         } else {
1406                                 hio->hio_errors[ncomp] = 0;
1407                         }
1408                         break;
1409                 case BIO_FLUSH:
1410                         if (!res->hr_localflush) {
1411                                 ret = -1;
1412                                 errno = EOPNOTSUPP;
1413                                 break;
1414                         }
1415                         ret = g_flush(res->hr_localfd);
1416                         if (ret == -1) {
1417                                 if (errno == EOPNOTSUPP)
1418                                         res->hr_localflush = false;
1419                                 hio->hio_errors[ncomp] = errno;
1420                                 reqlog(LOG_WARNING, 0, ggio,
1421                                     "Local request failed (%s): ",
1422                                     strerror(errno));
1423                         } else {
1424                                 hio->hio_errors[ncomp] = 0;
1425                         }
1426                         break;
1427                 }
1428
1429                 if (hio->hio_replication != HAST_REPLICATION_MEMSYNC ||
1430                     ggio->gctl_cmd != BIO_WRITE || ISSYNCREQ(hio)) {
1431                         if (refcnt_release(&hio->hio_countdown) > 0)
1432                                 continue;
1433                 } else {
1434                         /*
1435                          * Depending on hio_countdown value, requests finished
1436                          * in the following order:
1437                          * 0: remote memsync, remote final, local write
1438                          * 1: remote memsync, local write, (remote final)
1439                          * 2: local write, (remote memsync), (remote final)
1440                          */
1441                         switch (refcnt_release(&hio->hio_countdown)) {
1442                         case 0:
1443                                 /*
1444                                  * Local write finished as last.
1445                                  */
1446                                 break;
1447                         case 1:
1448                                 /*
1449                                  * Local write finished after remote memsync
1450                                  * reply arrvied. We can complete the write now.
1451                                  */
1452                                 if (hio->hio_errors[0] == 0)
1453                                         write_complete(res, hio);
1454                                 continue;
1455                         case 2:
1456                                 /*
1457                                  * Local write finished as first.
1458                                  */
1459                                 continue;
1460                         default:
1461                                 PJDLOG_ABORT("Invalid hio_countdown.");
1462                         }
1463                 }
1464                 if (ISSYNCREQ(hio)) {
1465                         mtx_lock(&sync_lock);
1466                         SYNCREQDONE(hio);
1467                         mtx_unlock(&sync_lock);
1468                         cv_signal(&sync_cond);
1469                 } else {
1470                         pjdlog_debug(2,
1471                             "local_send: (%p) Moving request to the done queue.",
1472                             hio);
1473                         QUEUE_INSERT2(hio, done);
1474                 }
1475         }
1476         /* NOTREACHED */
1477         return (NULL);
1478 }
1479
1480 static void
1481 keepalive_send(struct hast_resource *res, unsigned int ncomp)
1482 {
1483         struct nv *nv;
1484
1485         rw_rlock(&hio_remote_lock[ncomp]);
1486
1487         if (!ISCONNECTED(res, ncomp)) {
1488                 rw_unlock(&hio_remote_lock[ncomp]);
1489                 return;
1490         }
1491
1492         PJDLOG_ASSERT(res->hr_remotein != NULL);
1493         PJDLOG_ASSERT(res->hr_remoteout != NULL);
1494
1495         nv = nv_alloc();
1496         nv_add_uint8(nv, HIO_KEEPALIVE, "cmd");
1497         if (nv_error(nv) != 0) {
1498                 rw_unlock(&hio_remote_lock[ncomp]);
1499                 nv_free(nv);
1500                 pjdlog_debug(1,
1501                     "keepalive_send: Unable to prepare header to send.");
1502                 return;
1503         }
1504         if (hast_proto_send(res, res->hr_remoteout, nv, NULL, 0) == -1) {
1505                 rw_unlock(&hio_remote_lock[ncomp]);
1506                 pjdlog_common(LOG_DEBUG, 1, errno,
1507                     "keepalive_send: Unable to send request");
1508                 nv_free(nv);
1509                 remote_close(res, ncomp);
1510                 return;
1511         }
1512
1513         rw_unlock(&hio_remote_lock[ncomp]);
1514         nv_free(nv);
1515         pjdlog_debug(2, "keepalive_send: Request sent.");
1516 }
1517
1518 /*
1519  * Thread sends request to secondary node.
1520  */
1521 static void *
1522 remote_send_thread(void *arg)
1523 {
1524         struct hast_resource *res = arg;
1525         struct g_gate_ctl_io *ggio;
1526         time_t lastcheck, now;
1527         struct hio *hio;
1528         struct nv *nv;
1529         unsigned int ncomp;
1530         bool wakeup;
1531         uint64_t offset, length;
1532         uint8_t cmd;
1533         void *data;
1534
1535         /* Remote component is 1 for now. */
1536         ncomp = 1;
1537         lastcheck = time(NULL);
1538
1539         for (;;) {
1540                 pjdlog_debug(2, "remote_send: Taking request.");
1541                 QUEUE_TAKE1(hio, send, ncomp, HAST_KEEPALIVE);
1542                 if (hio == NULL) {
1543                         now = time(NULL);
1544                         if (lastcheck + HAST_KEEPALIVE <= now) {
1545                                 keepalive_send(res, ncomp);
1546                                 lastcheck = now;
1547                         }
1548                         continue;
1549                 }
1550                 pjdlog_debug(2, "remote_send: (%p) Got request.", hio);
1551                 ggio = &hio->hio_ggio;
1552                 switch (ggio->gctl_cmd) {
1553                 case BIO_READ:
1554                         cmd = HIO_READ;
1555                         data = NULL;
1556                         offset = ggio->gctl_offset;
1557                         length = ggio->gctl_length;
1558                         break;
1559                 case BIO_WRITE:
1560                         cmd = HIO_WRITE;
1561                         data = ggio->gctl_data;
1562                         offset = ggio->gctl_offset;
1563                         length = ggio->gctl_length;
1564                         break;
1565                 case BIO_DELETE:
1566                         cmd = HIO_DELETE;
1567                         data = NULL;
1568                         offset = ggio->gctl_offset;
1569                         length = ggio->gctl_length;
1570                         break;
1571                 case BIO_FLUSH:
1572                         cmd = HIO_FLUSH;
1573                         data = NULL;
1574                         offset = 0;
1575                         length = 0;
1576                         break;
1577                 default:
1578                         PJDLOG_ABORT("invalid condition");
1579                 }
1580                 nv = nv_alloc();
1581                 nv_add_uint8(nv, cmd, "cmd");
1582                 nv_add_uint64(nv, (uint64_t)ggio->gctl_seq, "seq");
1583                 nv_add_uint64(nv, offset, "offset");
1584                 nv_add_uint64(nv, length, "length");
1585                 if (hio->hio_replication == HAST_REPLICATION_MEMSYNC &&
1586                     ggio->gctl_cmd == BIO_WRITE && !ISSYNCREQ(hio)) {
1587                         nv_add_uint8(nv, 1, "memsync");
1588                 }
1589                 if (nv_error(nv) != 0) {
1590                         hio->hio_errors[ncomp] = nv_error(nv);
1591                         pjdlog_debug(2,
1592                             "remote_send: (%p) Unable to prepare header to send.",
1593                             hio);
1594                         reqlog(LOG_ERR, 0, ggio,
1595                             "Unable to prepare header to send (%s): ",
1596                             strerror(nv_error(nv)));
1597                         /* Move failed request immediately to the done queue. */
1598                         goto done_queue;
1599                 }
1600                 /*
1601                  * Protect connection from disappearing.
1602                  */
1603                 rw_rlock(&hio_remote_lock[ncomp]);
1604                 if (!ISCONNECTED(res, ncomp)) {
1605                         rw_unlock(&hio_remote_lock[ncomp]);
1606                         hio->hio_errors[ncomp] = ENOTCONN;
1607                         goto done_queue;
1608                 }
1609                 /*
1610                  * Move the request to recv queue before sending it, because
1611                  * in different order we can get reply before we move request
1612                  * to recv queue.
1613                  */
1614                 pjdlog_debug(2,
1615                     "remote_send: (%p) Moving request to the recv queue.",
1616                     hio);
1617                 mtx_lock(&hio_recv_list_lock[ncomp]);
1618                 wakeup = TAILQ_EMPTY(&hio_recv_list[ncomp]);
1619                 TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio, hio_next[ncomp]);
1620                 mtx_unlock(&hio_recv_list_lock[ncomp]);
1621                 if (hast_proto_send(res, res->hr_remoteout, nv, data,
1622                     data != NULL ? length : 0) == -1) {
1623                         hio->hio_errors[ncomp] = errno;
1624                         rw_unlock(&hio_remote_lock[ncomp]);
1625                         pjdlog_debug(2,
1626                             "remote_send: (%p) Unable to send request.", hio);
1627                         reqlog(LOG_ERR, 0, ggio,
1628                             "Unable to send request (%s): ",
1629                             strerror(hio->hio_errors[ncomp]));
1630                         remote_close(res, ncomp);
1631                         /*
1632                          * Take request back from the receive queue and move
1633                          * it immediately to the done queue.
1634                          */
1635                         mtx_lock(&hio_recv_list_lock[ncomp]);
1636                         TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1637                             hio_next[ncomp]);
1638                         mtx_unlock(&hio_recv_list_lock[ncomp]);
1639                         goto done_queue;
1640                 }
1641                 rw_unlock(&hio_remote_lock[ncomp]);
1642                 nv_free(nv);
1643                 if (wakeup)
1644                         cv_signal(&hio_recv_list_cond[ncomp]);
1645                 continue;
1646 done_queue:
1647                 nv_free(nv);
1648                 if (ISSYNCREQ(hio)) {
1649                         if (refcnt_release(&hio->hio_countdown) > 0)
1650                                 continue;
1651                         mtx_lock(&sync_lock);
1652                         SYNCREQDONE(hio);
1653                         mtx_unlock(&sync_lock);
1654                         cv_signal(&sync_cond);
1655                         continue;
1656                 }
1657                 if (ggio->gctl_cmd == BIO_WRITE) {
1658                         mtx_lock(&res->hr_amp_lock);
1659                         if (activemap_need_sync(res->hr_amp, ggio->gctl_offset,
1660                             ggio->gctl_length)) {
1661                                 (void)hast_activemap_flush(res);
1662                         } else {
1663                                 mtx_unlock(&res->hr_amp_lock);
1664                         }
1665                         if (hio->hio_replication == HAST_REPLICATION_MEMSYNC)
1666                                 (void)refcnt_release(&hio->hio_countdown);
1667                 }
1668                 if (refcnt_release(&hio->hio_countdown) > 0)
1669                         continue;
1670                 pjdlog_debug(2,
1671                     "remote_send: (%p) Moving request to the done queue.",
1672                     hio);
1673                 QUEUE_INSERT2(hio, done);
1674         }
1675         /* NOTREACHED */
1676         return (NULL);
1677 }
1678
1679 /*
1680  * Thread receives answer from secondary node and passes it to ggate_send
1681  * thread.
1682  */
1683 static void *
1684 remote_recv_thread(void *arg)
1685 {
1686         struct hast_resource *res = arg;
1687         struct g_gate_ctl_io *ggio;
1688         struct hio *hio;
1689         struct nv *nv;
1690         unsigned int ncomp;
1691         uint64_t seq;
1692         bool memsyncack;
1693         int error;
1694
1695         /* Remote component is 1 for now. */
1696         ncomp = 1;
1697
1698         for (;;) {
1699                 /* Wait until there is anything to receive. */
1700                 mtx_lock(&hio_recv_list_lock[ncomp]);
1701                 while (TAILQ_EMPTY(&hio_recv_list[ncomp])) {
1702                         pjdlog_debug(2, "remote_recv: No requests, waiting.");
1703                         cv_wait(&hio_recv_list_cond[ncomp],
1704                             &hio_recv_list_lock[ncomp]);
1705                 }
1706                 mtx_unlock(&hio_recv_list_lock[ncomp]);
1707
1708                 memsyncack = false;
1709
1710                 rw_rlock(&hio_remote_lock[ncomp]);
1711                 if (!ISCONNECTED(res, ncomp)) {
1712                         rw_unlock(&hio_remote_lock[ncomp]);
1713                         /*
1714                          * Connection is dead, so move all pending requests to
1715                          * the done queue (one-by-one).
1716                          */
1717                         mtx_lock(&hio_recv_list_lock[ncomp]);
1718                         hio = TAILQ_FIRST(&hio_recv_list[ncomp]);
1719                         PJDLOG_ASSERT(hio != NULL);
1720                         TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1721                             hio_next[ncomp]);
1722                         mtx_unlock(&hio_recv_list_lock[ncomp]);
1723                         goto done_queue;
1724                 }
1725                 if (hast_proto_recv_hdr(res->hr_remotein, &nv) == -1) {
1726                         pjdlog_errno(LOG_ERR,
1727                             "Unable to receive reply header");
1728                         rw_unlock(&hio_remote_lock[ncomp]);
1729                         remote_close(res, ncomp);
1730                         continue;
1731                 }
1732                 rw_unlock(&hio_remote_lock[ncomp]);
1733                 seq = nv_get_uint64(nv, "seq");
1734                 if (seq == 0) {
1735                         pjdlog_error("Header contains no 'seq' field.");
1736                         nv_free(nv);
1737                         continue;
1738                 }
1739                 memsyncack = nv_exists(nv, "received");
1740                 mtx_lock(&hio_recv_list_lock[ncomp]);
1741                 TAILQ_FOREACH(hio, &hio_recv_list[ncomp], hio_next[ncomp]) {
1742                         if (hio->hio_ggio.gctl_seq == seq) {
1743                                 TAILQ_REMOVE(&hio_recv_list[ncomp], hio,
1744                                     hio_next[ncomp]);
1745                                 break;
1746                         }
1747                 }
1748                 mtx_unlock(&hio_recv_list_lock[ncomp]);
1749                 if (hio == NULL) {
1750                         pjdlog_error("Found no request matching received 'seq' field (%ju).",
1751                             (uintmax_t)seq);
1752                         nv_free(nv);
1753                         continue;
1754                 }
1755                 ggio = &hio->hio_ggio;
1756                 error = nv_get_int16(nv, "error");
1757                 if (error != 0) {
1758                         /* Request failed on remote side. */
1759                         hio->hio_errors[ncomp] = error;
1760                         reqlog(LOG_WARNING, 0, ggio,
1761                             "Remote request failed (%s): ", strerror(error));
1762                         nv_free(nv);
1763                         goto done_queue;
1764                 }
1765                 switch (ggio->gctl_cmd) {
1766                 case BIO_READ:
1767                         rw_rlock(&hio_remote_lock[ncomp]);
1768                         if (!ISCONNECTED(res, ncomp)) {
1769                                 rw_unlock(&hio_remote_lock[ncomp]);
1770                                 nv_free(nv);
1771                                 goto done_queue;
1772                         }
1773                         if (hast_proto_recv_data(res, res->hr_remotein, nv,
1774                             ggio->gctl_data, ggio->gctl_length) == -1) {
1775                                 hio->hio_errors[ncomp] = errno;
1776                                 pjdlog_errno(LOG_ERR,
1777                                     "Unable to receive reply data");
1778                                 rw_unlock(&hio_remote_lock[ncomp]);
1779                                 nv_free(nv);
1780                                 remote_close(res, ncomp);
1781                                 goto done_queue;
1782                         }
1783                         rw_unlock(&hio_remote_lock[ncomp]);
1784                         break;
1785                 case BIO_WRITE:
1786                 case BIO_DELETE:
1787                 case BIO_FLUSH:
1788                         break;
1789                 default:
1790                         PJDLOG_ABORT("invalid condition");
1791                 }
1792                 hio->hio_errors[ncomp] = 0;
1793                 nv_free(nv);
1794 done_queue:
1795                 if (hio->hio_replication != HAST_REPLICATION_MEMSYNC ||
1796                     hio->hio_ggio.gctl_cmd != BIO_WRITE || ISSYNCREQ(hio)) {
1797                         if (refcnt_release(&hio->hio_countdown) > 0)
1798                                 continue;
1799                 } else {
1800                         /*
1801                          * Depending on hio_countdown value, requests finished
1802                          * in the following order:
1803                          *
1804                          * 0: local write, remote memsync, remote final
1805                          * or
1806                          * 0: remote memsync, local write, remote final
1807                          *
1808                          * 1: local write, remote memsync, (remote final)
1809                          * or
1810                          * 1: remote memsync, remote final, (local write)
1811                          *
1812                          * 2: remote memsync, (local write), (remote final)
1813                          * or
1814                          * 2: remote memsync, (remote final), (local write)
1815                          */
1816                         switch (refcnt_release(&hio->hio_countdown)) {
1817                         case 0:
1818                                 /*
1819                                  * Remote final reply arrived.
1820                                  */
1821                                 PJDLOG_ASSERT(!memsyncack);
1822                                 break;
1823                         case 1:
1824                                 if (memsyncack) {
1825                                         /*
1826                                          * Local request already finished, so we
1827                                          * can complete the write.
1828                                          */
1829                                         if (hio->hio_errors[0] == 0)
1830                                                 write_complete(res, hio);
1831                                         /*
1832                                          * We still need to wait for final
1833                                          * remote reply.
1834                                          */
1835                                         pjdlog_debug(2,
1836                                             "remote_recv: (%p) Moving request back to the recv queue.",
1837                                             hio);
1838                                         mtx_lock(&hio_recv_list_lock[ncomp]);
1839                                         TAILQ_INSERT_TAIL(&hio_recv_list[ncomp],
1840                                             hio, hio_next[ncomp]);
1841                                         mtx_unlock(&hio_recv_list_lock[ncomp]);
1842                                 } else {
1843                                         /*
1844                                          * Remote final reply arrived before
1845                                          * local write finished.
1846                                          * Nothing to do in such case.
1847                                          */
1848                                 }
1849                                 continue;
1850                         case 2:
1851                                 /*
1852                                  * We received remote memsync reply even before
1853                                  * local write finished.
1854                                  */
1855                                 PJDLOG_ASSERT(memsyncack);
1856
1857                                 pjdlog_debug(2,
1858                                     "remote_recv: (%p) Moving request back to the recv queue.",
1859                                     hio);
1860                                 mtx_lock(&hio_recv_list_lock[ncomp]);
1861                                 TAILQ_INSERT_TAIL(&hio_recv_list[ncomp], hio,
1862                                     hio_next[ncomp]);
1863                                 mtx_unlock(&hio_recv_list_lock[ncomp]);
1864                                 continue;
1865                         default:
1866                                 PJDLOG_ABORT("Invalid hio_countdown.");
1867                         }
1868                 }
1869                 if (ISSYNCREQ(hio)) {
1870                         mtx_lock(&sync_lock);
1871                         SYNCREQDONE(hio);
1872                         mtx_unlock(&sync_lock);
1873                         cv_signal(&sync_cond);
1874                 } else {
1875                         pjdlog_debug(2,
1876                             "remote_recv: (%p) Moving request to the done queue.",
1877                             hio);
1878                         QUEUE_INSERT2(hio, done);
1879                 }
1880         }
1881         /* NOTREACHED */
1882         return (NULL);
1883 }
1884
1885 /*
1886  * Thread sends answer to the kernel.
1887  */
1888 static void *
1889 ggate_send_thread(void *arg)
1890 {
1891         struct hast_resource *res = arg;
1892         struct g_gate_ctl_io *ggio;
1893         struct hio *hio;
1894         unsigned int ii, ncomps;
1895
1896         ncomps = HAST_NCOMPONENTS;
1897
1898         for (;;) {
1899                 pjdlog_debug(2, "ggate_send: Taking request.");
1900                 QUEUE_TAKE2(hio, done);
1901                 pjdlog_debug(2, "ggate_send: (%p) Got request.", hio);
1902                 ggio = &hio->hio_ggio;
1903                 for (ii = 0; ii < ncomps; ii++) {
1904                         if (hio->hio_errors[ii] == 0) {
1905                                 /*
1906                                  * One successful request is enough to declare
1907                                  * success.
1908                                  */
1909                                 ggio->gctl_error = 0;
1910                                 break;
1911                         }
1912                 }
1913                 if (ii == ncomps) {
1914                         /*
1915                          * None of the requests were successful.
1916                          * Use the error from local component except the
1917                          * case when we did only remote request.
1918                          */
1919                         if (ggio->gctl_cmd == BIO_READ &&
1920                             res->hr_syncsrc == HAST_SYNCSRC_SECONDARY)
1921                                 ggio->gctl_error = hio->hio_errors[1];
1922                         else
1923                                 ggio->gctl_error = hio->hio_errors[0];
1924                 }
1925                 if (ggio->gctl_error == 0 && ggio->gctl_cmd == BIO_WRITE) {
1926                         mtx_lock(&res->hr_amp_lock);
1927                         if (activemap_write_complete(res->hr_amp,
1928                             ggio->gctl_offset, ggio->gctl_length)) {
1929                                 res->hr_stat_activemap_update++;
1930                                 (void)hast_activemap_flush(res);
1931                         } else {
1932                                 mtx_unlock(&res->hr_amp_lock);
1933                         }
1934                 }
1935                 if (ggio->gctl_cmd == BIO_WRITE) {
1936                         /*
1937                          * Unlock range we locked.
1938                          */
1939                         mtx_lock(&range_lock);
1940                         rangelock_del(range_regular, ggio->gctl_offset,
1941                             ggio->gctl_length);
1942                         if (range_sync_wait)
1943                                 cv_signal(&range_sync_cond);
1944                         mtx_unlock(&range_lock);
1945                         if (!hio->hio_done)
1946                                 write_complete(res, hio);
1947                 } else {
1948                         if (ioctl(res->hr_ggatefd, G_GATE_CMD_DONE, ggio) == -1) {
1949                                 primary_exit(EX_OSERR,
1950                                     "G_GATE_CMD_DONE failed");
1951                         }
1952                 }
1953                 if (hio->hio_errors[0]) {
1954                         switch (ggio->gctl_cmd) {
1955                         case BIO_READ:
1956                                 res->hr_stat_read_error++;
1957                                 break;
1958                         case BIO_WRITE:
1959                                 res->hr_stat_write_error++;
1960                                 break;
1961                         case BIO_DELETE:
1962                                 res->hr_stat_delete_error++;
1963                                 break;
1964                         case BIO_FLUSH:
1965                                 res->hr_stat_flush_error++;
1966                                 break;
1967                         }
1968                 }
1969                 pjdlog_debug(2,
1970                     "ggate_send: (%p) Moving request to the free queue.", hio);
1971                 QUEUE_INSERT2(hio, free);
1972         }
1973         /* NOTREACHED */
1974         return (NULL);
1975 }
1976
1977 /*
1978  * Thread synchronize local and remote components.
1979  */
1980 static void *
1981 sync_thread(void *arg __unused)
1982 {
1983         struct hast_resource *res = arg;
1984         struct hio *hio;
1985         struct g_gate_ctl_io *ggio;
1986         struct timeval tstart, tend, tdiff;
1987         unsigned int ii, ncomp, ncomps;
1988         off_t offset, length, synced;
1989         bool dorewind, directreads;
1990         int syncext;
1991
1992         ncomps = HAST_NCOMPONENTS;
1993         dorewind = true;
1994         synced = 0;
1995         offset = -1;
1996         directreads = false;
1997
1998         for (;;) {
1999                 mtx_lock(&sync_lock);
2000                 if (offset >= 0 && !sync_inprogress) {
2001                         gettimeofday(&tend, NULL);
2002                         timersub(&tend, &tstart, &tdiff);
2003                         pjdlog_info("Synchronization interrupted after %#.0T. "
2004                             "%NB synchronized so far.", &tdiff,
2005                             (intmax_t)synced);
2006                         event_send(res, EVENT_SYNCINTR);
2007                 }
2008                 while (!sync_inprogress) {
2009                         dorewind = true;
2010                         synced = 0;
2011                         cv_wait(&sync_cond, &sync_lock);
2012                 }
2013                 mtx_unlock(&sync_lock);
2014                 /*
2015                  * Obtain offset at which we should synchronize.
2016                  * Rewind synchronization if needed.
2017                  */
2018                 mtx_lock(&res->hr_amp_lock);
2019                 if (dorewind)
2020                         activemap_sync_rewind(res->hr_amp);
2021                 offset = activemap_sync_offset(res->hr_amp, &length, &syncext);
2022                 if (syncext != -1) {
2023                         /*
2024                          * We synchronized entire syncext extent, we can mark
2025                          * it as clean now.
2026                          */
2027                         if (activemap_extent_complete(res->hr_amp, syncext))
2028                                 (void)hast_activemap_flush(res);
2029                         else
2030                                 mtx_unlock(&res->hr_amp_lock);
2031                 } else {
2032                         mtx_unlock(&res->hr_amp_lock);
2033                 }
2034                 if (dorewind) {
2035                         dorewind = false;
2036                         if (offset == -1)
2037                                 pjdlog_info("Nodes are in sync.");
2038                         else {
2039                                 pjdlog_info("Synchronization started. %NB to go.",
2040                                     (intmax_t)(res->hr_extentsize *
2041                                     activemap_ndirty(res->hr_amp)));
2042                                 event_send(res, EVENT_SYNCSTART);
2043                                 gettimeofday(&tstart, NULL);
2044                         }
2045                 }
2046                 if (offset == -1) {
2047                         sync_stop();
2048                         pjdlog_debug(1, "Nothing to synchronize.");
2049                         /*
2050                          * Synchronization complete, make both localcnt and
2051                          * remotecnt equal.
2052                          */
2053                         ncomp = 1;
2054                         rw_rlock(&hio_remote_lock[ncomp]);
2055                         if (ISCONNECTED(res, ncomp)) {
2056                                 if (synced > 0) {
2057                                         int64_t bps;
2058
2059                                         gettimeofday(&tend, NULL);
2060                                         timersub(&tend, &tstart, &tdiff);
2061                                         bps = (int64_t)((double)synced /
2062                                             ((double)tdiff.tv_sec +
2063                                             (double)tdiff.tv_usec / 1000000));
2064                                         pjdlog_info("Synchronization complete. "
2065                                             "%NB synchronized in %#.0lT (%NB/sec).",
2066                                             (intmax_t)synced, &tdiff,
2067                                             (intmax_t)bps);
2068                                         event_send(res, EVENT_SYNCDONE);
2069                                 }
2070                                 mtx_lock(&metadata_lock);
2071                                 if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY)
2072                                         directreads = true;
2073                                 res->hr_syncsrc = HAST_SYNCSRC_UNDEF;
2074                                 res->hr_primary_localcnt =
2075                                     res->hr_secondary_remotecnt;
2076                                 res->hr_primary_remotecnt =
2077                                     res->hr_secondary_localcnt;
2078                                 pjdlog_debug(1,
2079                                     "Setting localcnt to %ju and remotecnt to %ju.",
2080                                     (uintmax_t)res->hr_primary_localcnt,
2081                                     (uintmax_t)res->hr_primary_remotecnt);
2082                                 (void)metadata_write(res);
2083                                 mtx_unlock(&metadata_lock);
2084                         }
2085                         rw_unlock(&hio_remote_lock[ncomp]);
2086                         if (directreads) {
2087                                 directreads = false;
2088                                 enable_direct_reads(res);
2089                         }
2090                         continue;
2091                 }
2092                 pjdlog_debug(2, "sync: Taking free request.");
2093                 QUEUE_TAKE2(hio, free);
2094                 pjdlog_debug(2, "sync: (%p) Got free request.", hio);
2095                 /*
2096                  * Lock the range we are going to synchronize. We don't want
2097                  * race where someone writes between our read and write.
2098                  */
2099                 for (;;) {
2100                         mtx_lock(&range_lock);
2101                         if (rangelock_islocked(range_regular, offset, length)) {
2102                                 pjdlog_debug(2,
2103                                     "sync: Range offset=%jd length=%jd locked.",
2104                                     (intmax_t)offset, (intmax_t)length);
2105                                 range_sync_wait = true;
2106                                 cv_wait(&range_sync_cond, &range_lock);
2107                                 range_sync_wait = false;
2108                                 mtx_unlock(&range_lock);
2109                                 continue;
2110                         }
2111                         if (rangelock_add(range_sync, offset, length) == -1) {
2112                                 mtx_unlock(&range_lock);
2113                                 pjdlog_debug(2,
2114                                     "sync: Range offset=%jd length=%jd is already locked, waiting.",
2115                                     (intmax_t)offset, (intmax_t)length);
2116                                 sleep(1);
2117                                 continue;
2118                         }
2119                         mtx_unlock(&range_lock);
2120                         break;
2121                 }
2122                 /*
2123                  * First read the data from synchronization source.
2124                  */
2125                 SYNCREQ(hio);
2126                 ggio = &hio->hio_ggio;
2127                 ggio->gctl_cmd = BIO_READ;
2128                 ggio->gctl_offset = offset;
2129                 ggio->gctl_length = length;
2130                 ggio->gctl_error = 0;
2131                 hio->hio_done = false;
2132                 hio->hio_replication = res->hr_replication;
2133                 for (ii = 0; ii < ncomps; ii++)
2134                         hio->hio_errors[ii] = EINVAL;
2135                 reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ",
2136                     hio);
2137                 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2138                     hio);
2139                 mtx_lock(&metadata_lock);
2140                 if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
2141                         /*
2142                          * This range is up-to-date on local component,
2143                          * so handle request locally.
2144                          */
2145                          /* Local component is 0 for now. */
2146                         ncomp = 0;
2147                 } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ {
2148                         PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY);
2149                         /*
2150                          * This range is out-of-date on local component,
2151                          * so send request to the remote node.
2152                          */
2153                          /* Remote component is 1 for now. */
2154                         ncomp = 1;
2155                 }
2156                 mtx_unlock(&metadata_lock);
2157                 refcnt_init(&hio->hio_countdown, 1);
2158                 QUEUE_INSERT1(hio, send, ncomp);
2159
2160                 /*
2161                  * Let's wait for READ to finish.
2162                  */
2163                 mtx_lock(&sync_lock);
2164                 while (!ISSYNCREQDONE(hio))
2165                         cv_wait(&sync_cond, &sync_lock);
2166                 mtx_unlock(&sync_lock);
2167
2168                 if (hio->hio_errors[ncomp] != 0) {
2169                         pjdlog_error("Unable to read synchronization data: %s.",
2170                             strerror(hio->hio_errors[ncomp]));
2171                         goto free_queue;
2172                 }
2173
2174                 /*
2175                  * We read the data from synchronization source, now write it
2176                  * to synchronization target.
2177                  */
2178                 SYNCREQ(hio);
2179                 ggio->gctl_cmd = BIO_WRITE;
2180                 for (ii = 0; ii < ncomps; ii++)
2181                         hio->hio_errors[ii] = EINVAL;
2182                 reqlog(LOG_DEBUG, 2, ggio, "sync: (%p) Sending sync request: ",
2183                     hio);
2184                 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2185                     hio);
2186                 mtx_lock(&metadata_lock);
2187                 if (res->hr_syncsrc == HAST_SYNCSRC_PRIMARY) {
2188                         /*
2189                          * This range is up-to-date on local component,
2190                          * so we update remote component.
2191                          */
2192                          /* Remote component is 1 for now. */
2193                         ncomp = 1;
2194                 } else /* if (res->hr_syncsrc == HAST_SYNCSRC_SECONDARY) */ {
2195                         PJDLOG_ASSERT(res->hr_syncsrc == HAST_SYNCSRC_SECONDARY);
2196                         /*
2197                          * This range is out-of-date on local component,
2198                          * so we update it.
2199                          */
2200                          /* Local component is 0 for now. */
2201                         ncomp = 0;
2202                 }
2203                 mtx_unlock(&metadata_lock);
2204
2205                 pjdlog_debug(2, "sync: (%p) Moving request to the send queue.",
2206                     hio);
2207                 refcnt_init(&hio->hio_countdown, 1);
2208                 QUEUE_INSERT1(hio, send, ncomp);
2209
2210                 /*
2211                  * Let's wait for WRITE to finish.
2212                  */
2213                 mtx_lock(&sync_lock);
2214                 while (!ISSYNCREQDONE(hio))
2215                         cv_wait(&sync_cond, &sync_lock);
2216                 mtx_unlock(&sync_lock);
2217
2218                 if (hio->hio_errors[ncomp] != 0) {
2219                         pjdlog_error("Unable to write synchronization data: %s.",
2220                             strerror(hio->hio_errors[ncomp]));
2221                         goto free_queue;
2222                 }
2223
2224                 synced += length;
2225 free_queue:
2226                 mtx_lock(&range_lock);
2227                 rangelock_del(range_sync, offset, length);
2228                 if (range_regular_wait)
2229                         cv_signal(&range_regular_cond);
2230                 mtx_unlock(&range_lock);
2231                 pjdlog_debug(2, "sync: (%p) Moving request to the free queue.",
2232                     hio);
2233                 QUEUE_INSERT2(hio, free);
2234         }
2235         /* NOTREACHED */
2236         return (NULL);
2237 }
2238
2239 void
2240 primary_config_reload(struct hast_resource *res, struct nv *nv)
2241 {
2242         unsigned int ii, ncomps;
2243         int modified, vint;
2244         const char *vstr;
2245
2246         pjdlog_info("Reloading configuration...");
2247
2248         PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
2249         PJDLOG_ASSERT(gres == res);
2250         nv_assert(nv, "remoteaddr");
2251         nv_assert(nv, "sourceaddr");
2252         nv_assert(nv, "replication");
2253         nv_assert(nv, "checksum");
2254         nv_assert(nv, "compression");
2255         nv_assert(nv, "timeout");
2256         nv_assert(nv, "exec");
2257         nv_assert(nv, "metaflush");
2258
2259         ncomps = HAST_NCOMPONENTS;
2260
2261 #define MODIFIED_REMOTEADDR     0x01
2262 #define MODIFIED_SOURCEADDR     0x02
2263 #define MODIFIED_REPLICATION    0x04
2264 #define MODIFIED_CHECKSUM       0x08
2265 #define MODIFIED_COMPRESSION    0x10
2266 #define MODIFIED_TIMEOUT        0x20
2267 #define MODIFIED_EXEC           0x40
2268 #define MODIFIED_METAFLUSH      0x80
2269         modified = 0;
2270
2271         vstr = nv_get_string(nv, "remoteaddr");
2272         if (strcmp(gres->hr_remoteaddr, vstr) != 0) {
2273                 /*
2274                  * Don't copy res->hr_remoteaddr to gres just yet.
2275                  * We want remote_close() to log disconnect from the old
2276                  * addresses, not from the new ones.
2277                  */
2278                 modified |= MODIFIED_REMOTEADDR;
2279         }
2280         vstr = nv_get_string(nv, "sourceaddr");
2281         if (strcmp(gres->hr_sourceaddr, vstr) != 0) {
2282                 strlcpy(gres->hr_sourceaddr, vstr, sizeof(gres->hr_sourceaddr));
2283                 modified |= MODIFIED_SOURCEADDR;
2284         }
2285         vint = nv_get_int32(nv, "replication");
2286         if (gres->hr_replication != vint) {
2287                 gres->hr_replication = vint;
2288                 modified |= MODIFIED_REPLICATION;
2289         }
2290         vint = nv_get_int32(nv, "checksum");
2291         if (gres->hr_checksum != vint) {
2292                 gres->hr_checksum = vint;
2293                 modified |= MODIFIED_CHECKSUM;
2294         }
2295         vint = nv_get_int32(nv, "compression");
2296         if (gres->hr_compression != vint) {
2297                 gres->hr_compression = vint;
2298                 modified |= MODIFIED_COMPRESSION;
2299         }
2300         vint = nv_get_int32(nv, "timeout");
2301         if (gres->hr_timeout != vint) {
2302                 gres->hr_timeout = vint;
2303                 modified |= MODIFIED_TIMEOUT;
2304         }
2305         vstr = nv_get_string(nv, "exec");
2306         if (strcmp(gres->hr_exec, vstr) != 0) {
2307                 strlcpy(gres->hr_exec, vstr, sizeof(gres->hr_exec));
2308                 modified |= MODIFIED_EXEC;
2309         }
2310         vint = nv_get_int32(nv, "metaflush");
2311         if (gres->hr_metaflush != vint) {
2312                 gres->hr_metaflush = vint;
2313                 modified |= MODIFIED_METAFLUSH;
2314         }
2315
2316         /*
2317          * Change timeout for connected sockets.
2318          * Don't bother if we need to reconnect.
2319          */
2320         if ((modified & MODIFIED_TIMEOUT) != 0 &&
2321             (modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR)) == 0) {
2322                 for (ii = 0; ii < ncomps; ii++) {
2323                         if (!ISREMOTE(ii))
2324                                 continue;
2325                         rw_rlock(&hio_remote_lock[ii]);
2326                         if (!ISCONNECTED(gres, ii)) {
2327                                 rw_unlock(&hio_remote_lock[ii]);
2328                                 continue;
2329                         }
2330                         rw_unlock(&hio_remote_lock[ii]);
2331                         if (proto_timeout(gres->hr_remotein,
2332                             gres->hr_timeout) == -1) {
2333                                 pjdlog_errno(LOG_WARNING,
2334                                     "Unable to set connection timeout");
2335                         }
2336                         if (proto_timeout(gres->hr_remoteout,
2337                             gres->hr_timeout) == -1) {
2338                                 pjdlog_errno(LOG_WARNING,
2339                                     "Unable to set connection timeout");
2340                         }
2341                 }
2342         }
2343         if ((modified & (MODIFIED_REMOTEADDR | MODIFIED_SOURCEADDR)) != 0) {
2344                 for (ii = 0; ii < ncomps; ii++) {
2345                         if (!ISREMOTE(ii))
2346                                 continue;
2347                         remote_close(gres, ii);
2348                 }
2349                 if (modified & MODIFIED_REMOTEADDR) {
2350                         vstr = nv_get_string(nv, "remoteaddr");
2351                         strlcpy(gres->hr_remoteaddr, vstr,
2352                             sizeof(gres->hr_remoteaddr));
2353                 }
2354         }
2355 #undef  MODIFIED_REMOTEADDR
2356 #undef  MODIFIED_SOURCEADDR
2357 #undef  MODIFIED_REPLICATION
2358 #undef  MODIFIED_CHECKSUM
2359 #undef  MODIFIED_COMPRESSION
2360 #undef  MODIFIED_TIMEOUT
2361 #undef  MODIFIED_EXEC
2362 #undef  MODIFIED_METAFLUSH
2363
2364         pjdlog_info("Configuration reloaded successfully.");
2365 }
2366
2367 static void
2368 guard_one(struct hast_resource *res, unsigned int ncomp)
2369 {
2370         struct proto_conn *in, *out;
2371
2372         if (!ISREMOTE(ncomp))
2373                 return;
2374
2375         rw_rlock(&hio_remote_lock[ncomp]);
2376
2377         if (!real_remote(res)) {
2378                 rw_unlock(&hio_remote_lock[ncomp]);
2379                 return;
2380         }
2381
2382         if (ISCONNECTED(res, ncomp)) {
2383                 PJDLOG_ASSERT(res->hr_remotein != NULL);
2384                 PJDLOG_ASSERT(res->hr_remoteout != NULL);
2385                 rw_unlock(&hio_remote_lock[ncomp]);
2386                 pjdlog_debug(2, "remote_guard: Connection to %s is ok.",
2387                     res->hr_remoteaddr);
2388                 return;
2389         }
2390
2391         PJDLOG_ASSERT(res->hr_remotein == NULL);
2392         PJDLOG_ASSERT(res->hr_remoteout == NULL);
2393         /*
2394          * Upgrade the lock. It doesn't have to be atomic as no other thread
2395          * can change connection status from disconnected to connected.
2396          */
2397         rw_unlock(&hio_remote_lock[ncomp]);
2398         pjdlog_debug(2, "remote_guard: Reconnecting to %s.",
2399             res->hr_remoteaddr);
2400         in = out = NULL;
2401         if (init_remote(res, &in, &out) == 0) {
2402                 rw_wlock(&hio_remote_lock[ncomp]);
2403                 PJDLOG_ASSERT(res->hr_remotein == NULL);
2404                 PJDLOG_ASSERT(res->hr_remoteout == NULL);
2405                 PJDLOG_ASSERT(in != NULL && out != NULL);
2406                 res->hr_remotein = in;
2407                 res->hr_remoteout = out;
2408                 rw_unlock(&hio_remote_lock[ncomp]);
2409                 pjdlog_info("Successfully reconnected to %s.",
2410                     res->hr_remoteaddr);
2411                 sync_start();
2412         } else {
2413                 /* Both connections should be NULL. */
2414                 PJDLOG_ASSERT(res->hr_remotein == NULL);
2415                 PJDLOG_ASSERT(res->hr_remoteout == NULL);
2416                 PJDLOG_ASSERT(in == NULL && out == NULL);
2417                 pjdlog_debug(2, "remote_guard: Reconnect to %s failed.",
2418                     res->hr_remoteaddr);
2419         }
2420 }
2421
2422 /*
2423  * Thread guards remote connections and reconnects when needed, handles
2424  * signals, etc.
2425  */
2426 static void *
2427 guard_thread(void *arg)
2428 {
2429         struct hast_resource *res = arg;
2430         unsigned int ii, ncomps;
2431         struct timespec timeout;
2432         time_t lastcheck, now;
2433         sigset_t mask;
2434         int signo;
2435
2436         ncomps = HAST_NCOMPONENTS;
2437         lastcheck = time(NULL);
2438
2439         PJDLOG_VERIFY(sigemptyset(&mask) == 0);
2440         PJDLOG_VERIFY(sigaddset(&mask, SIGINT) == 0);
2441         PJDLOG_VERIFY(sigaddset(&mask, SIGTERM) == 0);
2442
2443         timeout.tv_sec = HAST_KEEPALIVE;
2444         timeout.tv_nsec = 0;
2445         signo = -1;
2446
2447         for (;;) {
2448                 switch (signo) {
2449                 case SIGINT:
2450                 case SIGTERM:
2451                         sigexit_received = true;
2452                         primary_exitx(EX_OK,
2453                             "Termination signal received, exiting.");
2454                         break;
2455                 default:
2456                         break;
2457                 }
2458
2459                 /*
2460                  * Don't check connections until we fully started,
2461                  * as we may still be looping, waiting for remote node
2462                  * to switch from primary to secondary.
2463                  */
2464                 if (fullystarted) {
2465                         pjdlog_debug(2, "remote_guard: Checking connections.");
2466                         now = time(NULL);
2467                         if (lastcheck + HAST_KEEPALIVE <= now) {
2468                                 for (ii = 0; ii < ncomps; ii++)
2469                                         guard_one(res, ii);
2470                                 lastcheck = now;
2471                         }
2472                 }
2473                 signo = sigtimedwait(&mask, NULL, &timeout);
2474         }
2475         /* NOTREACHED */
2476         return (NULL);
2477 }