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 <assert.h>
  37 #include <errno.h>
  38 #include <pthread.h>
  39 #include <signal.h>
  40 #include <stdio.h>
  41 #include <string.h>
  42 #include <unistd.h>
  43
  44 #include "hast.h"
  45 #include "hastd.h"
  46 #include "hast_checksum.h"
  47 #include "hast_compression.h"
  48 #include "hast_proto.h"
  49 #include "hooks.h"
  50 #include "nv.h"
  51 #include "pjdlog.h"
  52 #include "proto.h"
  53 #include "subr.h"
  54
  55 #include "control.h"
  56
  57 void
  58 child_cleanup(struct hast_resource *res)
  59 {
  60
  61         proto_close(res->hr_ctrl);
  62         res->hr_ctrl = NULL;
  63         if (res->hr_event != NULL) {
  64                 proto_close(res->hr_event);
  65                 res->hr_event = NULL;
  66         }
  67         if (res->hr_conn != NULL) {
  68                 proto_close(res->hr_conn);
  69                 res->hr_conn = NULL;
  70         }
  71         res->hr_workerpid = 0;
  72 }
  73
  74 static void
  75 control_set_role_common(struct hastd_config *cfg, struct nv *nvout,
  76     uint8_t role, struct hast_resource *res, const char *name, unsigned int no)
  77 {
  78         int oldrole;
  79
  80         /* Name is always needed. */
  81         if (name != NULL)
  82                 nv_add_string(nvout, name, "resource%u", no);
  83
  84         if (res == NULL) {
  85                 assert(cfg != NULL);
  86                 assert(name != NULL);
  87
  88                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
  89                         if (strcmp(res->hr_name, name) == 0)
  90                                 break;
  91                 }
  92                 if (res == NULL) {
  93                         nv_add_int16(nvout, EHAST_NOENTRY, "error%u", no);
  94                         return;
  95                 }
  96         }
  97         assert(res != NULL);
  98
  99         /* Send previous role back. */
 100         nv_add_string(nvout, role2str(res->hr_role), "role%u", no);
 101
 102         /* Nothing changed, return here. */
 103         if (role == res->hr_role)
 104                 return;
 105
 106         pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
 107         pjdlog_info("Role changed to %s.", role2str(role));
 108
 109         /* Change role to the new one. */
 110         oldrole = res->hr_role;
 111         res->hr_role = role;
 112         pjdlog_prefix_set("[%s] (%s) ", res->hr_name, role2str(res->hr_role));
 113
 114         /*
 115          * If previous role was primary or secondary we have to kill process
 116          * doing that work.
 117          */
 118         if (res->hr_workerpid != 0) {
 119                 if (kill(res->hr_workerpid, SIGTERM) < 0) {
 120                         pjdlog_errno(LOG_WARNING,
 121                             "Unable to kill worker process %u",
 122                             (unsigned int)res->hr_workerpid);
 123                 } else if (waitpid(res->hr_workerpid, NULL, 0) !=
 124                     res->hr_workerpid) {
 125                         pjdlog_errno(LOG_WARNING,
 126                             "Error while waiting for worker process %u",
 127                             (unsigned int)res->hr_workerpid);
 128                 } else {
 129                         pjdlog_debug(1, "Worker process %u stopped.",
 130                             (unsigned int)res->hr_workerpid);
 131                 }
 132                 child_cleanup(res);
 133         }
 134
 135         /* Start worker process if we are changing to primary. */
 136         if (role == HAST_ROLE_PRIMARY)
 137                 hastd_primary(res);
 138         pjdlog_prefix_set("%s", "");
 139         hook_exec(res->hr_exec, "role", res->hr_name, role2str(oldrole),
 140             role2str(res->hr_role), NULL);
 141 }
 142
 143 void
 144 control_set_role(struct hast_resource *res, uint8_t role)
 145 {
 146
 147         control_set_role_common(NULL, NULL, role, res, NULL, 0);
 148 }
 149
 150 static void
 151 control_status_worker(struct hast_resource *res, struct nv *nvout,
 152     unsigned int no)
 153 {
 154         struct nv *cnvin, *cnvout;
 155         const char *str;
 156         int error;
 157
 158         cnvin = cnvout = NULL;
 159         error = 0;
 160
 161         /*
 162          * Prepare and send command to worker process.
 163          */
 164         cnvout = nv_alloc();
 165         nv_add_uint8(cnvout, CONTROL_STATUS, "cmd");
 166         error = nv_error(cnvout);
 167         if (error != 0) {
 168                 pjdlog_common(LOG_ERR, 0, error,
 169                     "Unable to prepare control header");
 170                 goto end;
 171         }
 172         if (hast_proto_send(res, res->hr_ctrl, cnvout, NULL, 0) < 0) {
 173                 error = errno;
 174                 pjdlog_errno(LOG_ERR, "Unable to send control header");
 175                 goto end;
 176         }
 177
 178         /*
 179          * Receive response.
 180          */
 181         if (hast_proto_recv_hdr(res->hr_ctrl, &cnvin) < 0) {
 182                 error = errno;
 183                 pjdlog_errno(LOG_ERR, "Unable to receive control header");
 184                 goto end;
 185         }
 186
 187         error = nv_get_int16(cnvin, "error");
 188         if (error != 0)
 189                 goto end;
 190
 191         if ((str = nv_get_string(cnvin, "status")) == NULL) {
 192                 error = ENOENT;
 193                 pjdlog_errno(LOG_ERR, "Field 'status' is missing.");
 194                 goto end;
 195         }
 196         nv_add_string(nvout, str, "status%u", no);
 197         nv_add_uint64(nvout, nv_get_uint64(cnvin, "dirty"), "dirty%u", no);
 198         nv_add_uint32(nvout, nv_get_uint32(cnvin, "extentsize"),
 199             "extentsize%u", no);
 200         nv_add_uint32(nvout, nv_get_uint32(cnvin, "keepdirty"),
 201             "keepdirty%u", no);
 202 end:
 203         if (cnvin != NULL)
 204                 nv_free(cnvin);
 205         if (cnvout != NULL)
 206                 nv_free(cnvout);
 207         if (error != 0)
 208                 nv_add_int16(nvout, error, "error");
 209 }
 210
 211 static void
 212 control_status(struct hastd_config *cfg, struct nv *nvout,
 213     struct hast_resource *res, const char *name, unsigned int no)
 214 {
 215
 216         assert(cfg != NULL);
 217         assert(nvout != NULL);
 218         assert(name != NULL);
 219
 220         /* Name is always needed. */
 221         nv_add_string(nvout, name, "resource%u", no);
 222
 223         if (res == NULL) {
 224                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 225                         if (strcmp(res->hr_name, name) == 0)
 226                                 break;
 227                 }
 228                 if (res == NULL) {
 229                         nv_add_int16(nvout, EHAST_NOENTRY, "error%u", no);
 230                         return;
 231                 }
 232         }
 233         assert(res != NULL);
 234         nv_add_string(nvout, res->hr_provname, "provname%u", no);
 235         nv_add_string(nvout, res->hr_localpath, "localpath%u", no);
 236         nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr%u", no);
 237         if (res->hr_sourceaddr[0] != '\0')
 238                 nv_add_string(nvout, res->hr_sourceaddr, "sourceaddr%u", no);
 239         switch (res->hr_replication) {
 240         case HAST_REPLICATION_FULLSYNC:
 241                 nv_add_string(nvout, "fullsync", "replication%u", no);
 242                 break;
 243         case HAST_REPLICATION_MEMSYNC:
 244                 nv_add_string(nvout, "memsync", "replication%u", no);
 245                 break;
 246         case HAST_REPLICATION_ASYNC:
 247                 nv_add_string(nvout, "async", "replication%u", no);
 248                 break;
 249         default:
 250                 nv_add_string(nvout, "unknown", "replication%u", no);
 251                 break;
 252         }
 253         nv_add_string(nvout, checksum_name(res->hr_checksum),
 254             "checksum%u", no);
 255         nv_add_string(nvout, compression_name(res->hr_compression),
 256             "compression%u", no);
 257         nv_add_string(nvout, role2str(res->hr_role), "role%u", no);
 258
 259         switch (res->hr_role) {
 260         case HAST_ROLE_PRIMARY:
 261                 assert(res->hr_workerpid != 0);
 262                 /* FALLTHROUGH */
 263         case HAST_ROLE_SECONDARY:
 264                 if (res->hr_workerpid != 0)
 265                         break;
 266                 /* FALLTHROUGH */
 267         default:
 268                 return;
 269         }
 270
 271         /*
 272          * If we are here, it means that we have a worker process, which we
 273          * want to ask some questions.
 274          */
 275         control_status_worker(res, nvout, no);
 276 }
 277
 278 void
 279 control_handle(struct hastd_config *cfg)
 280 {
 281         struct proto_conn *conn;
 282         struct nv *nvin, *nvout;
 283         unsigned int ii;
 284         const char *str;
 285         uint8_t cmd, role;
 286         int error;
 287
 288         if (proto_accept(cfg->hc_controlconn, &conn) < 0) {
 289                 pjdlog_errno(LOG_ERR, "Unable to accept control connection");
 290                 return;
 291         }
 292
 293         cfg->hc_controlin = conn;
 294         nvin = nvout = NULL;
 295         role = HAST_ROLE_UNDEF;
 296
 297         if (hast_proto_recv_hdr(conn, &nvin) < 0) {
 298                 pjdlog_errno(LOG_ERR, "Unable to receive control header");
 299                 nvin = NULL;
 300                 goto close;
 301         }
 302
 303         /* Obtain command code. 0 means that nv_get_uint8() failed. */
 304         cmd = nv_get_uint8(nvin, "cmd");
 305         if (cmd == 0) {
 306                 pjdlog_error("Control header is missing 'cmd' field.");
 307                 error = EHAST_INVALID;
 308                 goto close;
 309         }
 310
 311         /* Allocate outgoing nv structure. */
 312         nvout = nv_alloc();
 313         if (nvout == NULL) {
 314                 pjdlog_error("Unable to allocate header for control response.");
 315                 error = EHAST_NOMEMORY;
 316                 goto close;
 317         }
 318
 319         error = 0;
 320
 321         str = nv_get_string(nvin, "resource0");
 322         if (str == NULL) {
 323                 pjdlog_error("Control header is missing 'resource0' field.");
 324                 error = EHAST_INVALID;
 325                 goto fail;
 326         }
 327         if (cmd == HASTCTL_CMD_SETROLE) {
 328                 role = nv_get_uint8(nvin, "role");
 329                 switch (role) {
 330                 case HAST_ROLE_INIT:
 331                 case HAST_ROLE_PRIMARY:
 332                 case HAST_ROLE_SECONDARY:
 333                         break;
 334                 default:
 335                         pjdlog_error("Invalid role received (%hhu).", role);
 336                         error = EHAST_INVALID;
 337                         goto fail;
 338                 }
 339         }
 340         if (strcmp(str, "all") == 0) {
 341                 struct hast_resource *res;
 342
 343                 /* All configured resources. */
 344
 345                 ii = 0;
 346                 TAILQ_FOREACH(res, &cfg->hc_resources, hr_next) {
 347                         switch (cmd) {
 348                         case HASTCTL_CMD_SETROLE:
 349                                 control_set_role_common(cfg, nvout, role, res,
 350                                     res->hr_name, ii++);
 351                                 break;
 352                         case HASTCTL_CMD_STATUS:
 353                                 control_status(cfg, nvout, res, res->hr_name,
 354                                     ii++);
 355                                 break;
 356                         default:
 357                                 pjdlog_error("Invalid command received (%hhu).",
 358                                     cmd);
 359                                 error = EHAST_UNIMPLEMENTED;
 360                                 goto fail;
 361                         }
 362                 }
 363         } else {
 364                 /* Only selected resources. */
 365
 366                 for (ii = 0; ; ii++) {
 367                         str = nv_get_string(nvin, "resource%u", ii);
 368                         if (str == NULL)
 369                                 break;
 370                         switch (cmd) {
 371                         case HASTCTL_CMD_SETROLE:
 372                                 control_set_role_common(cfg, nvout, role, NULL,
 373                                     str, ii);
 374                                 break;
 375                         case HASTCTL_CMD_STATUS:
 376                                 control_status(cfg, nvout, NULL, str, ii);
 377                                 break;
 378                         default:
 379                                 pjdlog_error("Invalid command received (%hhu).",
 380                                     cmd);
 381                                 error = EHAST_UNIMPLEMENTED;
 382                                 goto fail;
 383                         }
 384                 }
 385         }
 386         if (nv_error(nvout) != 0)
 387                 goto close;
 388 fail:
 389         if (error != 0)
 390                 nv_add_int16(nvout, error, "error");
 391
 392         if (hast_proto_send(NULL, conn, nvout, NULL, 0) < 0)
 393                 pjdlog_errno(LOG_ERR, "Unable to send control response");
 394 close:
 395         if (nvin != NULL)
 396                 nv_free(nvin);
 397         if (nvout != NULL)
 398                 nv_free(nvout);
 399         proto_close(conn);
 400         cfg->hc_controlin = NULL;
 401 }
 402
 403 /*
 404  * Thread handles control requests from the parent.
 405  */
 406 void *
 407 ctrl_thread(void *arg)
 408 {
 409         struct hast_resource *res = arg;
 410         struct nv *nvin, *nvout;
 411         uint8_t cmd;
 412
 413         for (;;) {
 414                 if (hast_proto_recv_hdr(res->hr_ctrl, &nvin) < 0) {
 415                         if (sigexit_received)
 416                                 pthread_exit(NULL);
 417                         pjdlog_errno(LOG_ERR,
 418                             "Unable to receive control message");
 419                         kill(getpid(), SIGTERM);
 420                         pthread_exit(NULL);
 421                 }
 422                 cmd = nv_get_uint8(nvin, "cmd");
 423                 if (cmd == 0) {
 424                         pjdlog_error("Control message is missing 'cmd' field.");
 425                         nv_free(nvin);
 426                         continue;
 427                 }
 428                 nvout = nv_alloc();
 429                 switch (cmd) {
 430                 case CONTROL_STATUS:
 431                         if (res->hr_remotein != NULL &&
 432                             res->hr_remoteout != NULL) {
 433                                 nv_add_string(nvout, "complete", "status");
 434                         } else {
 435                                 nv_add_string(nvout, "degraded", "status");
 436                         }
 437                         nv_add_uint32(nvout, (uint32_t)res->hr_extentsize,
 438                             "extentsize");
 439                         if (res->hr_role == HAST_ROLE_PRIMARY) {
 440                                 nv_add_uint32(nvout,
 441                                     (uint32_t)res->hr_keepdirty, "keepdirty");
 442                                 nv_add_uint64(nvout,
 443                                     (uint64_t)(activemap_ndirty(res->hr_amp) *
 444                                     res->hr_extentsize), "dirty");
 445                         } else {
 446                                 nv_add_uint32(nvout, (uint32_t)0, "keepdirty");
 447                                 nv_add_uint64(nvout, (uint64_t)0, "dirty");
 448                         }
 449                         nv_add_int16(nvout, 0, "error");
 450                         break;
 451                 case CONTROL_RELOAD:
 452                         /*
 453                          * When parent receives SIGHUP and discovers that
 454                          * something related to us has changes, it sends reload
 455                          * message to us.
 456                          */
 457                         assert(res->hr_role == HAST_ROLE_PRIMARY);
 458                         primary_config_reload(res, nvin);
 459                         nv_add_int16(nvout, 0, "error");
 460                         break;
 461                 default:
 462                         nv_add_int16(nvout, EINVAL, "error");
 463                         break;
 464                 }
 465                 nv_free(nvin);
 466                 if (nv_error(nvout) != 0) {
 467                         pjdlog_error("Unable to create answer on control message.");
 468                         nv_free(nvout);
 469                         continue;
 470                 }
 471                 if (hast_proto_send(NULL, res->hr_ctrl, nvout, NULL, 0) < 0) {
 472                         pjdlog_errno(LOG_ERR,
 473                             "Unable to send reply to control message");
 474                 }
 475                 nv_free(nvout);
 476         }
 477         /* NOTREACHED */
 478         return (NULL);
 479 }