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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 /* 29 * wait.c - asynchronous monitoring of "wait registered" start methods 30 * 31 * Use event ports to poll on the set of fds representing the /proc/[pid]/psinfo 32 * files. If one of these fds returns an event, then we inform the restarter 33 * that it has stopped. 34 * 35 * The wait_info_list holds the series of processes currently being monitored 36 * for exit. The wi_fd member, which contains the file descriptor of the psinfo 37 * file being polled upon ("event ported upon"), will be set to -1 if the file 38 * descriptor is inactive (already closed or not yet opened). 39 */ 40 41 #ifdef _FILE_OFFSET_BITS 42 #undef _FILE_OFFSET_BITS 43 #endif /* _FILE_OFFSET_BITS */ 44 45 #include <sys/resource.h> 46 #include <sys/stat.h> 47 #include <sys/types.h> 48 #include <sys/uio.h> 49 #include <sys/wait.h> 50 51 #include <assert.h> 52 #include <errno.h> 53 #include <fcntl.h> 54 #include <libuutil.h> 55 #include <poll.h> 56 #include <port.h> 57 #include <pthread.h> 58 #include <procfs.h> 59 #include <string.h> 60 #include <stropts.h> 61 #include <unistd.h> 62 63 #include "startd.h" 64 65 #define WAIT_FILES 262144 /* reasonably high maximum */ 66 67 static int port_fd; 68 static scf_handle_t *wait_hndl; 69 static struct rlimit init_fd_rlimit; 70 71 static uu_list_pool_t *wait_info_pool; 72 static uu_list_t *wait_info_list; 73 74 static pthread_mutex_t wait_info_lock; 75 76 /* 77 * void wait_remove(wait_info_t *, int) 78 * Remove the given wait_info structure from our list, performing various 79 * cleanup operations along the way. If the direct flag is false (meaning 80 * that we are being called with from restarter instance list context) and 81 * the instance should not be ignored, then notify the restarter that the 82 * associated instance has exited. If the wi_ignore flag is true then it 83 * means that the stop was initiated from within svc.startd, rather than 84 * from outside it. 85 * 86 * Since we may no longer be the startd that started this process, we only are 87 * concerned with a waitpid(3C) failure if the wi_parent field is non-zero. 88 */ 89 static void 90 wait_remove(wait_info_t *wi, int direct) 91 { 92 int status; 93 94 if (waitpid(wi->wi_pid, &status, 0) == -1) { 95 if (wi->wi_parent) 96 log_framework(LOG_INFO, 97 "instance %s waitpid failure: %s\n", wi->wi_fmri, 98 strerror(errno)); 99 } else { 100 if (WEXITSTATUS(status) != 0) { 101 log_framework(LOG_NOTICE, 102 "instance %s exited with status %d\n", wi->wi_fmri, 103 WEXITSTATUS(status)); 104 } 105 } 106 107 MUTEX_LOCK(&wait_info_lock); 108 if (wi->wi_fd != -1) { 109 startd_close(wi->wi_fd); 110 wi->wi_fd = -1; 111 } 112 uu_list_remove(wait_info_list, wi); 113 MUTEX_UNLOCK(&wait_info_lock); 114 115 /* 116 * Make an attempt to clear out any utmpx record associated with this 117 * PID. 118 */ 119 utmpx_mark_dead(wi->wi_pid, status, B_FALSE); 120 121 if (!direct && !wi->wi_ignore) { 122 /* 123 * Bind wait_hndl lazily. 124 */ 125 if (wait_hndl == NULL) { 126 for (wait_hndl = 127 libscf_handle_create_bound(SCF_VERSION); 128 wait_hndl == NULL; 129 wait_hndl = 130 libscf_handle_create_bound(SCF_VERSION)) { 131 log_error(LOG_INFO, "[wait_remove] Unable to " 132 "bind a new repository handle: %s\n", 133 scf_strerror(scf_error())); 134 (void) sleep(2); 135 } 136 } 137 138 log_framework(LOG_DEBUG, 139 "wait_remove requesting stop of %s\n", wi->wi_fmri); 140 (void) stop_instance_fmri(wait_hndl, wi->wi_fmri, RSTOP_EXIT); 141 } 142 143 uu_list_node_fini(wi, &wi->wi_link, wait_info_pool); 144 startd_free(wi, sizeof (wait_info_t)); 145 } 146 147 /* 148 * void wait_ignore_by_fmri(const char *) 149 * wait_ignore_by_fmri is called when svc.startd is going to stop the 150 * instance. Since we need to wait on the process and close the utmpx record, 151 * we're going to set the wi_ignore flag, so that when the process exits we 152 * clean up, but don't tell the restarter to stop it. 153 */ 154 void 155 wait_ignore_by_fmri(const char *fmri) 156 { 157 wait_info_t *wi; 158 159 MUTEX_LOCK(&wait_info_lock); 160 161 for (wi = uu_list_first(wait_info_list); wi != NULL; 162 wi = uu_list_next(wait_info_list, wi)) { 163 if (strcmp(wi->wi_fmri, fmri) == 0) 164 break; 165 } 166 167 if (wi != NULL) { 168 wi->wi_ignore = 1; 169 } 170 171 MUTEX_UNLOCK(&wait_info_lock); 172 } 173 174 /* 175 * int wait_register(pid_t, char *, int, int) 176 * wait_register is called after we have called fork(2), and know which pid we 177 * wish to monitor. However, since the child may have already exited by the 178 * time we are called, we must handle the error cases from open(2) 179 * appropriately. The am_parent flag is recorded to handle waitpid(2) 180 * behaviour on removal; similarly, the direct flag is passed through to a 181 * potential call to wait_remove() to govern its behaviour in different 182 * contexts. 183 * 184 * Returns 0 if registration successful, 1 if child pid did not exist, and -1 185 * if a different error occurred. 186 */ 187 int 188 wait_register(pid_t pid, const char *inst_fmri, int am_parent, int direct) 189 { 190 char *fname = uu_msprintf("/proc/%ld/psinfo", pid); 191 int fd; 192 wait_info_t *wi; 193 194 assert(pid != 0); 195 196 if (fname == NULL) 197 return (-1); 198 199 wi = startd_alloc(sizeof (wait_info_t)); 200 201 uu_list_node_init(wi, &wi->wi_link, wait_info_pool); 202 203 wi->wi_fd = -1; 204 wi->wi_pid = pid; 205 wi->wi_fmri = inst_fmri; 206 wi->wi_parent = am_parent; 207 wi->wi_ignore = 0; 208 209 MUTEX_LOCK(&wait_info_lock); 210 (void) uu_list_insert_before(wait_info_list, NULL, wi); 211 MUTEX_UNLOCK(&wait_info_lock); 212 213 if ((fd = open(fname, O_RDONLY)) == -1) { 214 if (errno == ENOENT) { 215 /* 216 * Child has already exited. 217 */ 218 wait_remove(wi, direct); 219 uu_free(fname); 220 return (1); 221 } else { 222 log_error(LOG_WARNING, 223 "open %s failed; not monitoring %s: %s\n", fname, 224 inst_fmri, strerror(errno)); 225 uu_free(fname); 226 return (-1); 227 } 228 } 229 230 uu_free(fname); 231 232 wi->wi_fd = fd; 233 234 if (port_associate(port_fd, PORT_SOURCE_FD, fd, 0, wi)) { 235 log_error(LOG_WARNING, 236 "initial port_association of %d / %s failed: %s\n", fd, 237 inst_fmri, strerror(errno)); 238 return (-1); 239 } 240 241 log_framework(LOG_DEBUG, "monitoring PID %ld on fd %d (%s)\n", pid, fd, 242 inst_fmri); 243 244 return (0); 245 } 246 247 /*ARGSUSED*/ 248 void * 249 wait_thread(void *args) 250 { 251 for (;;) { 252 port_event_t pe; 253 int fd; 254 wait_info_t *wi; 255 256 if (port_get(port_fd, &pe, NULL) != 0) { 257 if (errno == EINTR) 258 continue; 259 else { 260 log_error(LOG_WARNING, 261 "port_get() failed with %s\n", 262 strerror(errno)); 263 bad_error("port_get", errno); 264 } 265 } 266 267 fd = pe.portev_object; 268 wi = pe.portev_user; 269 assert(wi != NULL); 270 assert(fd == wi->wi_fd); 271 272 if ((pe.portev_events & POLLHUP) == POLLHUP) { 273 psinfo_t psi; 274 275 if (lseek(fd, 0, SEEK_SET) != 0 || 276 read(fd, &psi, sizeof (psinfo_t)) != 277 sizeof (psinfo_t)) { 278 log_framework(LOG_WARNING, 279 "couldn't get psinfo data for %s (%s); " 280 "assuming failed\n", wi->wi_fmri, 281 strerror(errno)); 282 goto err_remove; 283 } 284 285 if (psi.pr_nlwp != 0 || 286 psi.pr_nzomb != 0 || 287 psi.pr_lwp.pr_lwpid != 0) { 288 /* 289 * We have determined, in accordance with the 290 * definition in proc(4), this process is not a 291 * zombie. Reassociate. 292 */ 293 if (port_associate(port_fd, PORT_SOURCE_FD, fd, 294 0, wi)) 295 log_error(LOG_WARNING, 296 "port_association of %d / %s " 297 "failed\n", fd, wi->wi_fmri); 298 continue; 299 } 300 } else if ( 301 (pe.portev_events & POLLERR) == 0) { 302 if (port_associate(port_fd, PORT_SOURCE_FD, fd, 0, wi)) 303 log_error(LOG_WARNING, 304 "port_association of %d / %s " 305 "failed\n", fd, wi->wi_fmri); 306 continue; 307 } 308 309 err_remove: 310 wait_remove(wi, 0); 311 } 312 313 /*LINTED E_FUNC_HAS_NO_RETURN_STMT*/ 314 } 315 316 void 317 wait_prefork() 318 { 319 MUTEX_LOCK(&wait_info_lock); 320 } 321 322 void 323 wait_postfork(pid_t pid) 324 { 325 wait_info_t *wi; 326 327 MUTEX_UNLOCK(&wait_info_lock); 328 329 if (pid != 0) 330 return; 331 332 /* 333 * Close all of the child's wait-related fds. The wait_thread() is 334 * gone, so no need to worry about returning events. We always exec(2) 335 * after a fork request, so we needn't free the list elements 336 * themselves. 337 */ 338 339 for (wi = uu_list_first(wait_info_list); 340 wi != NULL; 341 wi = uu_list_next(wait_info_list, wi)) { 342 if (wi->wi_fd != -1) 343 startd_close(wi->wi_fd); 344 } 345 346 startd_close(port_fd); 347 348 (void) setrlimit(RLIMIT_NOFILE, &init_fd_rlimit); 349 } 350 351 void 352 wait_init() 353 { 354 struct rlimit fd_new; 355 356 (void) getrlimit(RLIMIT_NOFILE, &init_fd_rlimit); 357 (void) getrlimit(RLIMIT_NOFILE, &fd_new); 358 359 fd_new.rlim_max = fd_new.rlim_cur = WAIT_FILES; 360 361 (void) setrlimit(RLIMIT_NOFILE, &fd_new); 362 363 if ((port_fd = port_create()) == -1) 364 uu_die("wait_init couldn't port_create"); 365 366 wait_info_pool = uu_list_pool_create("wait_info", sizeof (wait_info_t), 367 offsetof(wait_info_t, wi_link), NULL, UU_LIST_POOL_DEBUG); 368 if (wait_info_pool == NULL) 369 uu_die("wait_init couldn't create wait_info_pool"); 370 371 wait_info_list = uu_list_create(wait_info_pool, wait_info_list, 0); 372 if (wait_info_list == NULL) 373 uu_die("wait_init couldn't create wait_info_list"); 374 375 (void) pthread_mutex_init(&wait_info_lock, &mutex_attrs); 376 } 377