sbin/hastd/control.c

   1 /*-
   2  * Copyright (c) 2009-2010 The FreeBSD Foundation
   3  * All rights reserved.
   4  *
   5  * This software was developed by Pawel Jakub Dawidek under sponsorship from
   6  * the FreeBSD Foundation.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  */
  29
  30 #include <sys/cdefs.h>
  31 __FBSDID("$FreeBSD$");
  32
  33 #include <sys/types.h>
  34 #include <sys/wait.h>
  35
  36 #include <errno.h>
  37 #include <pthread.h>
  38 #include <signal.h>
  39 #include <stdio.h>
  40 #include <string.h>
  41 #include <unistd.h>
  42
  43 #include "hast.h"
  44 #include "hastd.h"
  45 #include "hast_checksum.h"
  46 #include "hast_compression.h"
  47 #include "hast_proto.h"
  48 #include "hooks.h"
  49 #include "nv.h"
  50 #include "pjdlog.h"
  51 #include "proto.h"
  52 #include "subr.h"
  53
  54 #include "control.h"
  55
  56 void
  57 child_cleanup(struct hast_resource *res)
  58 {
  59
  60         proto_close(res->hr_ctrl);
  61         res->hr_ctrl = NULL;
  62         if (res->hr_event != NULL) {
  63                 proto_close(res->hr_event);
  64                 res->hr_event = NULL;
  65         }
  66         if (res->hr_conn != NULL) {
  67                 proto_close(res->hr_conn);
  68                 res->hr_conn = NULL;
  69         }
  70         res->hr_workerpid = 0;
  71 }
  72
  73 static void
  74 control_set_role_common(struct hastd_config *cfg, struct nv *nvout,
  75     uint8_t role, struct hast_resource *res, const char *name, unsigned int no)
  76 {
  77         int oldrole;
  78
  79         /* Name is always needed. */
  80         if (name != NULL)
  81                 nv_add_string(nvout, name, "resource%u", no);
  82
  83         if (res == NULL) {
  84                 PJDLOG_ASSERT(cfg != NULL);
  85                 PJDLOG_ASSERT(name != NULL);
  86
  87                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
  88                         if (strcmp(res->hr_name, name) == 0)
  89                                 break;
  90                 }
  91                 if (res == NULL) {
  92                         nv_add_int16(nvout, EHAST_NOENTRY, "error%u", no);
  93                         return;
  94                 }
  95         }
  96         PJDLOG_ASSERT(res != NULL);
  97
  98         /* Send previous role back. */
  99         nv_add_string(nvout, role2str(res->hr_role), "role%u", no);
 100
 101         /* Nothing changed, return here. */
 102         if (role == res->hr_role)
 103                 return;
 104
 105         pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
 106         pjdlog_info("Role changed to %s.", role2str(role));
 107
 108         /* Change role to the new one. */
 109         oldrole = res->hr_role;
 110         res->hr_role = role;
 111         pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
 112
 113         /*
 114          * If previous role was primary or secondary we have to kill process
 115          * doing that work.
 116          */
 117         if (res->hr_workerpid != 0) {
 118                 if (kill(res->hr_workerpid, SIGTERM) == -1) {
 119                         pjdlog_errno(LOG_WARNING,
 120                             "Unable to kill worker process %u",
 121                             (unsigned int)res->hr_workerpid);
 122                 } else if (waitpid(res->hr_workerpid, NULL, 0) !=
 123                     res->hr_workerpid) {
 124                         pjdlog_errno(LOG_WARNING,
 125                             "Error while waiting for worker process %u",
 126                             (unsigned int)res->hr_workerpid);
 127                 } else {
 128                         pjdlog_debug(1, "Worker process %u stopped.",
 129                             (unsigned int)res->hr_workerpid);
 130                 }
 131                 child_cleanup(res);
 132         }
 133
 134         /* Start worker process if we are changing to primary. */
 135         if (role == HAST_ROLE_PRIMARY)
 136                 hastd_primary(res);
 137         pjdlog_prefix_set("%s", "");
 138         hook_exec(res->hr_exec, "role", res->hr_name, role2str(oldrole),
 139             role2str(res->hr_role), NULL);
 140 }
 141
 142 void
 143 control_set_role(struct hast_resource *res, uint8_t role)
 144 {
 145
 146         control_set_role_common(NULL, NULL, role, res, NULL, 0);
 147 }
 148
 149 static void
 150 control_status_worker(struct hast_resource *res, struct nv *nvout,
 151     unsigned int no)
 152 {
 153         struct nv *cnvin, *cnvout;
 154         const char *str;
 155         int error;
 156
 157         cnvin = NULL;
 158
 159         /*
 160          * Prepare and send command to worker process.
 161          */
 162         cnvout = nv_alloc();
 163         nv_add_uint8(cnvout, CONTROL_STATUS, "cmd");
 164         error = nv_error(cnvout);
 165         if (error != 0) {
 166                 pjdlog_common(LOG_ERR, 0, error,
 167                     "Unable to prepare control header");
 168                 goto end;
 169         }
 170         if (hast_proto_send(res, res->hr_ctrl, cnvout, NULL, 0) == -1) {
 171                 error = errno;
 172                 pjdlog_errno(LOG_ERR, "Unable to send control header");
 173                 goto end;
 174         }
 175
 176         /*
 177          * Receive response.
 178          */
 179         if (hast_proto_recv_hdr(res->hr_ctrl, &cnvin) == -1) {
 180                 error = errno;
 181                 pjdlog_errno(LOG_ERR, "Unable to receive control header");
 182                 goto end;
 183         }
 184
 185         error = nv_get_int16(cnvin, "error");
 186         if (error != 0)
 187                 goto end;
 188
 189         if ((str = nv_get_string(cnvin, "status")) == NULL) {
 190                 error = ENOENT;
 191                 pjdlog_errno(LOG_ERR, "Field 'status' is missing.");
 192                 goto end;
 193         }
 194         nv_add_string(nvout, str, "status%u", no);
 195         nv_add_uint64(nvout, nv_get_uint64(cnvin, "dirty"), "dirty%u", no);
 196         nv_add_uint32(nvout, nv_get_uint32(cnvin, "extentsize"),
 197             "extentsize%u", no);
 198         nv_add_uint32(nvout, nv_get_uint32(cnvin, "keepdirty"),
 199             "keepdirty%u", no);
 200         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_read"),
 201             "stat_read%u", no);
 202         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_write"),
 203             "stat_write%u", no);
 204         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_delete"),
 205             "stat_delete%u", no);
 206         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_flush"),
 207             "stat_flush%u", no);
 208         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_activemap_update"),
 209             "stat_activemap_update%u", no);
 210         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_read_error"),
 211             "stat_read_error%u", no);
 212         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_write_error"),
 213             "stat_write_error%u", no);
 214         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_delete_error"),
 215             "stat_delete_error%u", no);
 216         nv_add_uint64(nvout, nv_get_uint64(cnvin, "stat_flush_error"),
 217             "stat_flush_error%u", no);
 218 end:
 219         if (cnvin != NULL)
 220                 nv_free(cnvin);
 221         if (cnvout != NULL)
 222                 nv_free(cnvout);
 223         if (error != 0)
 224                 nv_add_int16(nvout, error, "error");
 225 }
 226
 227 static void
 228 control_status(struct hastd_config *cfg, struct nv *nvout,
 229     struct hast_resource *res, const char *name, unsigned int no)
 230 {
 231
 232         PJDLOG_ASSERT(cfg != NULL);
 233         PJDLOG_ASSERT(nvout != NULL);
 234         PJDLOG_ASSERT(name != NULL);
 235
 236         /* Name is always needed. */
 237         nv_add_string(nvout, name, "resource%u", no);
 238
 239         if (res == NULL) {
 240                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 241                         if (strcmp(res->hr_name, name) == 0)
 242                                 break;
 243                 }
 244                 if (res == NULL) {
 245                         nv_add_int16(nvout, EHAST_NOENTRY, "error%u", no);
 246                         return;
 247                 }
 248         }
 249         PJDLOG_ASSERT(res != NULL);
 250         nv_add_string(nvout, res->hr_provname, "provname%u", no);
 251         nv_add_string(nvout, res->hr_localpath, "localpath%u", no);
 252         nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr%u", no);
 253         if (res->hr_sourceaddr[0] != '\0')
 254                 nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr%u", no);
 255         switch (res->hr_replication) {
 256         case HAST_REPLICATION_FULLSYNC:
 257                 nv_add_string(nvout, "fullsync", "replication%u", no);
 258                 break;
 259         case HAST_REPLICATION_MEMSYNC:
 260                 nv_add_string(nvout, "memsync", "replication%u", no);
 261                 break;
 262         case HAST_REPLICATION_ASYNC:
 263                 nv_add_string(nvout, "async", "replication%u", no);
 264                 break;
 265         default:
 266                 nv_add_string(nvout, "unknown", "replication%u", no);
 267                 break;
 268         }
 269         nv_add_string(nvout, checksum_name(res->hr_checksum),
 270             "checksum%u", no);
 271         nv_add_string(nvout, compression_name(res->hr_compression),
 272             "compression%u", no);
 273         nv_add_string(nvout, role2str(res->hr_role), "role%u", no);
 274
 275         switch (res->hr_role) {
 276         case HAST_ROLE_PRIMARY:
 277                 PJDLOG_ASSERT(res->hr_workerpid != 0);
 278                 /* FALLTHROUGH */
 279         case HAST_ROLE_SECONDARY:
 280                 if (res->hr_workerpid != 0)
 281                         break;
 282                 /* FALLTHROUGH */
 283         default:
 284                 return;
 285         }
 286
 287         /*
 288          * If we are here, it means that we have a worker process, which we
 289          * want to ask some questions.
 290          */
 291         control_status_worker(res, nvout, no);
 292 }
 293
 294 void
 295 control_handle(struct hastd_config *cfg)
 296 {
 297         struct proto_conn *conn;
 298         struct nv *nvin, *nvout;
 299         unsigned int ii;
 300         const char *str;
 301         uint8_t cmd, role;
 302         int error;
 303
 304         if (proto_accept(cfg->hc_controlconn, &conn) == -1) {
 305                 pjdlog_errno(LOG_ERR, "Unable to accept control connection");
 306                 return;
 307         }
 308
 309         cfg->hc_controlin = conn;
 310         nvin = nvout = NULL;
 311         role = HAST_ROLE_UNDEF;
 312
 313         if (hast_proto_recv_hdr(conn, &nvin) == -1) {
 314                 pjdlog_errno(LOG_ERR, "Unable to receive control header");
 315                 nvin = NULL;
 316                 goto close;
 317         }
 318
 319         /* Obtain command code. 0 means that nv_get_uint8() failed. */
 320         cmd = nv_get_uint8(nvin, "cmd");
 321         if (cmd == 0) {
 322                 pjdlog_error("Control header is missing 'cmd' field.");
 323                 goto close;
 324         }
 325
 326         /* Allocate outgoing nv structure. */
 327         nvout = nv_alloc();
 328         if (nvout == NULL) {
 329                 pjdlog_error("Unable to allocate header for control response.");
 330                 goto close;
 331         }
 332
 333         error = 0;
 334
 335         str = nv_get_string(nvin, "resource0");
 336         if (str == NULL) {
 337                 pjdlog_error("Control header is missing 'resource0' field.");
 338                 error = EHAST_INVALID;
 339                 goto fail;
 340         }
 341         if (cmd == HASTCTL_CMD_SETROLE) {
 342                 role = nv_get_uint8(nvin, "role");
 343                 switch (role) {
 344                 case HAST_ROLE_INIT:
 345                 case HAST_ROLE_PRIMARY:
 346                 case HAST_ROLE_SECONDARY:
 347                         break;
 348                 default:
 349                         pjdlog_error("Invalid role received (%hhu).", role);
 350                         error = EHAST_INVALID;
 351                         goto fail;
 352                 }
 353         }
 354         if (strcmp(str, "all") == 0) {
 355                 struct hast_resource *res;
 356
 357                 /* All configured resources. */
 358
 359                 ii = 0;
 360                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 361                         switch (cmd) {
 362                         case HASTCTL_CMD_SETROLE:
 363                                 control_set_role_common(cfg, nvout, role, res,
 364                                     res->hr_name, ii++);
 365                                 break;
 366                         case HASTCTL_CMD_STATUS:
 367                                 control_status(cfg, nvout, res, res->hr_name,
 368                                     ii++);
 369                                 break;
 370                         default:
 371                                 pjdlog_error("Invalid command received (%hhu).",
 372                                     cmd);
 373                                 error = EHAST_UNIMPLEMENTED;
 374                                 goto fail;
 375                         }
 376                 }
 377         } else {
 378                 /* Only selected resources. */
 379
 380                 for (ii = 0; ; ii++) {
 381                         str = nv_get_string(nvin, "resource%u", ii);
 382                         if (str == NULL)
 383                                 break;
 384                         switch (cmd) {
 385                         case HASTCTL_CMD_SETROLE:
 386                                 control_set_role_common(cfg, nvout, role, NULL,
 387                                     str, ii);
 388                                 break;
 389                         case HASTCTL_CMD_STATUS:
 390                                 control_status(cfg, nvout, NULL, str, ii);
 391                                 break;
 392                         default:
 393                                 pjdlog_error("Invalid command received (%hhu).",
 394                                     cmd);
 395                                 error = EHAST_UNIMPLEMENTED;
 396                                 goto fail;
 397                         }
 398                 }
 399         }
 400         if (nv_error(nvout) != 0)
 401                 goto close;
 402 fail:
 403         if (error != 0)
 404                 nv_add_int16(nvout, error, "error");
 405
 406         if (hast_proto_send(NULL, conn, nvout, NULL, 0) == -1)
 407                 pjdlog_errno(LOG_ERR, "Unable to send control response");
 408 close:
 409         if (nvin != NULL)
 410                 nv_free(nvin);
 411         if (nvout != NULL)
 412                 nv_free(nvout);
 413         proto_close(conn);
 414         cfg->hc_controlin = NULL;
 415 }
 416
 417 /*
 418  * Thread handles control requests from the parent.
 419  */
 420 void *
 421 ctrl_thread(void *arg)
 422 {
 423         struct hast_resource *res = arg;
 424         struct nv *nvin, *nvout;
 425         uint8_t cmd;
 426
 427         for (;;) {
 428                 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) == -1) {
 429                         if (sigexit_received)
 430                                 pthread_exit(NULL);
 431                         pjdlog_errno(LOG_ERR,
 432                             "Unable to receive control message");
 433                         kill(getpid(), SIGTERM);
 434                         pthread_exit(NULL);
 435                 }
 436                 cmd = nv_get_uint8(nvin, "cmd");
 437                 if (cmd == 0) {
 438                         pjdlog_error("Control message is missing 'cmd' field.");
 439                         nv_free(nvin);
 440                         continue;
 441                 }
 442                 nvout = nv_alloc();
 443                 switch (cmd) {
 444                 case CONTROL_STATUS:
 445                         if (res->hr_remotein != NULL &&
 446                             res->hr_remoteout != NULL) {
 447                                 nv_add_string(nvout, "complete", "status");
 448                         } else {
 449                                 nv_add_string(nvout, "degraded", "status");
 450                         }
 451                         nv_add_uint32(nvout, (uint32_t)res->hr_extentsize,
 452                             "extentsize");
 453                         if (res->hr_role == HAST_ROLE_PRIMARY) {
 454                                 nv_add_uint32(nvout,
 455                                     (uint32_t)res->hr_keepdirty, "keepdirty");
 456                                 nv_add_uint64(nvout,
 457                                     (uint64_t)(activemap_ndirty(res->hr_amp) *
 458                                     res->hr_extentsize), "dirty");
 459                         } else {
 460                                 nv_add_uint32(nvout, (uint32_t)0, "keepdirty");
 461                                 nv_add_uint64(nvout, (uint64_t)0, "dirty");
 462                         }
 463                         nv_add_uint64(nvout, res->hr_stat_read, "stat_read");
 464                         nv_add_uint64(nvout, res->hr_stat_write, "stat_write");
 465                         nv_add_uint64(nvout, res->hr_stat_delete,
 466                             "stat_delete");
 467                         nv_add_uint64(nvout, res->hr_stat_flush, "stat_flush");
 468                         nv_add_uint64(nvout, res->hr_stat_activemap_update,
 469                             "stat_activemap_update");
 470                         nv_add_uint64(nvout, res->hr_stat_read_error,
 471                             "stat_read_error");
 472                         nv_add_uint64(nvout, res->hr_stat_write_error +
 473                             res->hr_stat_activemap_write_error,
 474                             "stat_write_error");
 475                         nv_add_uint64(nvout, res->hr_stat_delete_error,
 476                             "stat_delete_error");
 477                         nv_add_uint64(nvout, res->hr_stat_flush_error +
 478                             res->hr_stat_activemap_flush_error,
 479                             "stat_flush_error");
 480                         nv_add_int16(nvout, 0, "error");
 481                         break;
 482                 case CONTROL_RELOAD:
 483                         /*
 484                          * When parent receives SIGHUP and discovers that
 485                          * something related to us has changes, it sends reload
 486                          * message to us.
 487                          */
 488                         PJDLOG_ASSERT(res->hr_role == HAST_ROLE_PRIMARY);
 489                         primary_config_reload(res, nvin);
 490                         nv_add_int16(nvout, 0, "error");
 491                         break;
 492                 default:
 493                         nv_add_int16(nvout, EINVAL, "error");
 494                         break;
 495                 }
 496                 nv_free(nvin);
 497                 if (nv_error(nvout) != 0) {
 498                         pjdlog_error("Unable to create answer on control message.");
 499                         nv_free(nvout);
 500                         continue;
 501                 }
 502                 if (hast_proto_send(NULL, res->hr_ctrl, nvout, NULL, 0) == -1) {
 503                         pjdlog_errno(LOG_ERR,
 504                             "Unable to send reply to control message");
 505                 }
 506                 nv_free(nvout);
 507         }
 508         /* NOTREACHED */
 509         return (NULL);
 510 }