1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 * 25 * Copyright 2016-2018 RackTop Systems. 26 */ 27 28 29 /* 30 * transition.c - Graph State Machine 31 * 32 * The graph state machine is implemented here, with a typical approach 33 * of a function per state. Separating the implementation allows more 34 * clarity into the actions taken on notification of state change, as well 35 * as a place for future expansion including hooks for configurable actions. 36 * All functions are called with dgraph_lock held. 37 * 38 * The start action for this state machine is not explicit. The states 39 * (ONLINE and DEGRADED) which need to know when they're entering the state 40 * due to a daemon restart implement this understanding by checking for 41 * transition from uninitialized. In the future, this would likely be better 42 * as an explicit start action instead of relying on an overloaded transition. 43 * 44 * All gt_enter functions use the same set of return codes. 45 * 0 success 46 * ECONNABORTED repository connection aborted 47 */ 48 49 #include "startd.h" 50 51 static int 52 gt_running(restarter_instance_state_t state) 53 { 54 if (state == RESTARTER_STATE_ONLINE || 55 state == RESTARTER_STATE_DEGRADED) 56 return (1); 57 58 return (0); 59 } 60 61 static int 62 gt_enter_uninit(scf_handle_t *h, graph_vertex_t *v, 63 restarter_instance_state_t old_state, restarter_error_t rerr) 64 { 65 int err; 66 scf_instance_t *inst; 67 68 /* Initialize instance by refreshing it. */ 69 70 err = libscf_fmri_get_instance(h, v->gv_name, &inst); 71 switch (err) { 72 case 0: 73 break; 74 75 case ECONNABORTED: 76 return (ECONNABORTED); 77 78 case ENOENT: 79 return (0); 80 81 case EINVAL: 82 case ENOTSUP: 83 default: 84 bad_error("libscf_fmri_get_instance", err); 85 } 86 87 err = refresh_vertex(v, inst); 88 if (err == 0) 89 graph_enable_by_vertex(v, v->gv_flags & GV_ENABLED, 0); 90 91 scf_instance_destroy(inst); 92 93 /* If the service was running, propagate a stop event. */ 94 if (gt_running(old_state)) { 95 log_framework(LOG_DEBUG, "Propagating stop of %s.\n", 96 v->gv_name); 97 98 graph_transition_propagate(v, PROPAGATE_STOP, rerr); 99 } 100 101 graph_transition_sulogin(RESTARTER_STATE_UNINIT, old_state); 102 return (0); 103 } 104 105 /* ARGSUSED */ 106 static int 107 gt_enter_maint(scf_handle_t *h, graph_vertex_t *v, 108 restarter_instance_state_t old_state, restarter_error_t rerr) 109 { 110 int to_offline = v->gv_flags & GV_TOOFFLINE; 111 112 /* 113 * If the service was running, propagate a stop event. If the 114 * service was not running the maintenance transition may satisfy 115 * optional dependencies and should be propagated to determine 116 * whether new dependents are satisfiable. 117 * Instances that transition to maintenance and have the GV_TOOFFLINE 118 * flag are special because they can expose new subtree leaves so 119 * propagate the offline to the instance dependencies. 120 */ 121 122 /* instance transitioning to maintenance is considered disabled */ 123 v->gv_flags &= ~GV_TODISABLE; 124 v->gv_flags &= ~GV_TOOFFLINE; 125 126 if (gt_running(old_state)) { 127 /* 128 * Handle state change during instance disabling. 129 * Propagate offline to the new exposed leaves. 130 */ 131 if (to_offline) { 132 log_framework(LOG_DEBUG, "%s removed from subtree\n", 133 v->gv_name); 134 135 graph_offline_subtree_leaves(v, (void *)h); 136 } 137 138 log_framework(LOG_DEBUG, "Propagating maintenance (stop) of " 139 "%s.\n", v->gv_name); 140 141 graph_transition_propagate(v, PROPAGATE_STOP, rerr); 142 143 /* 144 * The maintenance transition may satisfy optional_all/restart 145 * dependencies and should be propagated to determine 146 * whether new dependents are satisfiable. 147 */ 148 graph_transition_propagate(v, PROPAGATE_SAT, rerr); 149 } else { 150 log_framework(LOG_DEBUG, "Propagating maintenance of %s.\n", 151 v->gv_name); 152 153 graph_transition_propagate(v, PROPAGATE_SAT, rerr); 154 } 155 156 graph_transition_sulogin(RESTARTER_STATE_MAINT, old_state); 157 return (0); 158 } 159 160 /* ARGSUSED */ 161 static int 162 gt_enter_offline(scf_handle_t *h, graph_vertex_t *v, 163 restarter_instance_state_t old_state, restarter_error_t rerr) 164 { 165 int to_offline = v->gv_flags & GV_TOOFFLINE; 166 167 v->gv_flags &= ~GV_TOOFFLINE; 168 169 /* 170 * If the instance should be disabled send it a disable command. 171 * Otherwise, if GV_TOOFFLINE was not set, see if we can start it. 172 */ 173 if (v->gv_flags & GV_TODISABLE) { 174 if (gt_running(old_state) && v->gv_post_disable_f) 175 v->gv_post_disable_f(); 176 177 vertex_send_event(v, RESTARTER_EVENT_TYPE_DISABLE); 178 } else if (v->gv_flags & GV_ENABLED) { 179 if (to_offline == 0) 180 graph_start_if_satisfied(v); 181 } 182 183 /* 184 * If the service was running, propagate a stop event. If the 185 * service was not running the offline transition may satisfy 186 * optional dependencies and should be propagated to determine 187 * whether new dependents are satisfiable. 188 * Instances that transition to offline and have the GV_TOOFFLINE flag 189 * are special because they can expose new subtree leaves so propagate 190 * the offline to the instance dependencies. 191 */ 192 if (gt_running(old_state)) { 193 /* 194 * Handle state change during instance disabling. 195 * Propagate offline to the new exposed leaves. 196 */ 197 if (to_offline) { 198 log_framework(LOG_DEBUG, "%s removed from subtree\n", 199 v->gv_name); 200 201 graph_offline_subtree_leaves(v, (void *)h); 202 } 203 204 log_framework(LOG_DEBUG, "Propagating stop of %s.\n", 205 v->gv_name); 206 207 graph_transition_propagate(v, PROPAGATE_STOP, rerr); 208 209 /* 210 * The offline transition may satisfy require_any/restart 211 * dependencies and should be propagated to determine 212 * whether new dependents are satisfiable. 213 */ 214 graph_transition_propagate(v, PROPAGATE_SAT, rerr); 215 } else { 216 log_framework(LOG_DEBUG, "Propagating offline of %s.\n", 217 v->gv_name); 218 219 graph_transition_propagate(v, PROPAGATE_SAT, rerr); 220 } 221 222 graph_transition_sulogin(RESTARTER_STATE_OFFLINE, old_state); 223 return (0); 224 } 225 226 /* ARGSUSED */ 227 static int 228 gt_enter_disabled(scf_handle_t *h, graph_vertex_t *v, 229 restarter_instance_state_t old_state, restarter_error_t rerr) 230 { 231 int to_offline = v->gv_flags & GV_TOOFFLINE; 232 233 v->gv_flags &= ~GV_TODISABLE; 234 v->gv_flags &= ~GV_TOOFFLINE; 235 236 /* 237 * If the instance should be disabled, no problem. Otherwise, 238 * send an enable command, which should result in the instance 239 * moving to OFFLINE unless the instance is part of a subtree 240 * (non root) and in this case the result is unpredictable. 241 */ 242 if (v->gv_flags & GV_ENABLED) { 243 vertex_send_event(v, RESTARTER_EVENT_TYPE_ENABLE); 244 } else if (gt_running(old_state) && v->gv_post_disable_f) { 245 v->gv_post_disable_f(); 246 } 247 248 /* 249 * If the service was running, propagate this as a stop. If the 250 * service was not running the disabled transition may satisfy 251 * optional dependencies and should be propagated to determine 252 * whether new dependents are satisfiable. 253 */ 254 if (gt_running(old_state)) { 255 /* 256 * We need to propagate the offline to new exposed leaves in 257 * case we've just disabled an instance that was part of a 258 * subtree. 259 */ 260 if (to_offline) { 261 log_framework(LOG_DEBUG, "%s removed from subtree\n", 262 v->gv_name); 263 264 /* 265 * Handle state change during instance disabling. 266 * Propagate offline to the new exposed leaves. 267 */ 268 graph_offline_subtree_leaves(v, (void *)h); 269 } 270 271 272 log_framework(LOG_DEBUG, "Propagating stop of %s.\n", 273 v->gv_name); 274 275 graph_transition_propagate(v, PROPAGATE_STOP, rerr); 276 277 /* 278 * The disable transition may satisfy optional_all/restart 279 * dependencies and should be propagated to determine 280 * whether new dependents are satisfiable. 281 */ 282 graph_transition_propagate(v, PROPAGATE_SAT, rerr); 283 } else { 284 log_framework(LOG_DEBUG, "Propagating disable of %s.\n", 285 v->gv_name); 286 287 graph_transition_propagate(v, PROPAGATE_SAT, rerr); 288 } 289 290 graph_transition_sulogin(RESTARTER_STATE_DISABLED, old_state); 291 return (0); 292 } 293 294 static int 295 gt_internal_online_or_degraded(scf_handle_t *h, graph_vertex_t *v, 296 restarter_instance_state_t old_state, restarter_error_t rerr) 297 { 298 int r; 299 300 /* 301 * If the instance has just come up, update the start 302 * snapshot. 303 */ 304 if (gt_running(old_state) == 0) { 305 /* 306 * Don't fire if we're just recovering state 307 * after a restart. 308 */ 309 if (old_state != RESTARTER_STATE_UNINIT && 310 v->gv_post_online_f) 311 v->gv_post_online_f(); 312 313 r = libscf_snapshots_poststart(h, v->gv_name, B_TRUE); 314 switch (r) { 315 case 0: 316 case ENOENT: 317 /* 318 * If ENOENT, the instance must have been 319 * deleted. Pretend we were successful since 320 * we should get a delete event later. 321 */ 322 break; 323 324 case ECONNABORTED: 325 return (ECONNABORTED); 326 327 case EACCES: 328 case ENOTSUP: 329 default: 330 bad_error("libscf_snapshots_poststart", r); 331 } 332 } 333 334 if (!(v->gv_flags & GV_ENABLED)) { 335 vertex_send_event(v, RESTARTER_EVENT_TYPE_DISABLE); 336 } else if (v->gv_flags & GV_TOOFFLINE) { 337 /* 338 * If the vertex has the GV_TOOFFLINE flag set then that's 339 * because the instance was transitioning from offline to 340 * online and the reverse disable algorithm doesn't offline 341 * those instances because it was already appearing offline. 342 * So do it now. 343 */ 344 offline_vertex(v); 345 } 346 347 if (gt_running(old_state) == 0) { 348 log_framework(LOG_DEBUG, "Propagating start of %s.\n", 349 v->gv_name); 350 351 graph_transition_propagate(v, PROPAGATE_START, rerr); 352 } else if (rerr == RERR_REFRESH) { 353 /* For refresh we'll get a message sans state change */ 354 355 log_framework(LOG_DEBUG, "Propagating refresh of %s.\n", 356 v->gv_name); 357 358 graph_transition_propagate(v, PROPAGATE_STOP, rerr); 359 } 360 361 return (0); 362 } 363 364 static int 365 gt_enter_online(scf_handle_t *h, graph_vertex_t *v, 366 restarter_instance_state_t old_state, restarter_error_t rerr) 367 { 368 int r; 369 370 r = gt_internal_online_or_degraded(h, v, old_state, rerr); 371 if (r != 0) 372 return (r); 373 374 graph_transition_sulogin(RESTARTER_STATE_ONLINE, old_state); 375 return (0); 376 } 377 378 static int 379 gt_enter_degraded(scf_handle_t *h, graph_vertex_t *v, 380 restarter_instance_state_t old_state, restarter_error_t rerr) 381 { 382 int r; 383 384 r = gt_internal_online_or_degraded(h, v, old_state, rerr); 385 if (r != 0) 386 return (r); 387 388 graph_transition_sulogin(RESTARTER_STATE_DEGRADED, old_state); 389 return (0); 390 } 391 392 /* 393 * gt_transition() implements the state transition for the graph 394 * state machine. It can return: 395 * 0 success 396 * ECONNABORTED repository connection aborted 397 * 398 * v->gv_state should be set to the state we're transitioning to before 399 * calling this function. 400 */ 401 int 402 gt_transition(scf_handle_t *h, graph_vertex_t *v, restarter_error_t rerr, 403 restarter_instance_state_t old_state) 404 { 405 int err; 406 int lost_repository = 0; 407 408 /* 409 * If there's a common set of work to be done on exit from the 410 * old_state, include it as a separate set of functions here. For 411 * now there's no such work, so there are no gt_exit functions. 412 */ 413 414 err = vertex_subgraph_dependencies_shutdown(h, v, old_state); 415 switch (err) { 416 case 0: 417 break; 418 419 case ECONNABORTED: 420 lost_repository = 1; 421 break; 422 423 default: 424 bad_error("vertex_subgraph_dependencies_shutdown", err); 425 } 426 427 /* 428 * Now call the appropriate gt_enter function for the new state. 429 */ 430 switch (v->gv_state) { 431 case RESTARTER_STATE_UNINIT: 432 err = gt_enter_uninit(h, v, old_state, rerr); 433 break; 434 435 case RESTARTER_STATE_DISABLED: 436 err = gt_enter_disabled(h, v, old_state, rerr); 437 break; 438 439 case RESTARTER_STATE_OFFLINE: 440 err = gt_enter_offline(h, v, old_state, rerr); 441 break; 442 443 case RESTARTER_STATE_ONLINE: 444 err = gt_enter_online(h, v, old_state, rerr); 445 break; 446 447 case RESTARTER_STATE_DEGRADED: 448 err = gt_enter_degraded(h, v, old_state, rerr); 449 break; 450 451 case RESTARTER_STATE_MAINT: 452 err = gt_enter_maint(h, v, old_state, rerr); 453 break; 454 455 default: 456 /* Shouldn't be in an invalid state. */ 457 #ifndef NDEBUG 458 uu_warn("%s:%d: Invalid state %d.\n", __FILE__, __LINE__, 459 v->gv_state); 460 #endif 461 abort(); 462 } 463 464 switch (err) { 465 case 0: 466 break; 467 468 case ECONNABORTED: 469 lost_repository = 1; 470 break; 471 472 default: 473 #ifndef NDEBUG 474 uu_warn("%s:%d: " 475 "gt_enter_%s() failed with unexpected error %d.\n", 476 __FILE__, __LINE__, instance_state_str[v->gv_state], err); 477 #endif 478 abort(); 479 } 480 481 return (lost_repository ? ECONNABORTED : 0); 482 } 483