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