1 /* 2 * os-posix-lib.c 3 * 4 * Copyright (c) 2003-2008 Fabrice Bellard 5 * Copyright (c) 2010 Red Hat, Inc. 6 * 7 * QEMU library functions on POSIX which are shared between QEMU and 8 * the QEMU tools. 9 * 10 * Permission is hereby granted, free of charge, to any person obtaining a copy 11 * of this software and associated documentation files (the "Software"), to deal 12 * in the Software without restriction, including without limitation the rights 13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 14 * copies of the Software, and to permit persons to whom the Software is 15 * furnished to do so, subject to the following conditions: 16 * 17 * The above copyright notice and this permission notice shall be included in 18 * all copies or substantial portions of the Software. 19 * 20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 23 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 26 * THE SOFTWARE. 27 */ 28 29 #include "qemu/osdep.h" 30 #include <termios.h> 31 32 #include <glib/gprintf.h> 33 34 #include "sysemu/sysemu.h" 35 #include "trace.h" 36 #include "qapi/error.h" 37 #include "qemu/error-report.h" 38 #include "qemu/madvise.h" 39 #include "qemu/sockets.h" 40 #include "qemu/thread.h" 41 #include <libgen.h> 42 #include "qemu/cutils.h" 43 #include "qemu/compiler.h" 44 #include "qemu/units.h" 45 46 #ifdef CONFIG_LINUX 47 #include <sys/syscall.h> 48 #endif 49 50 #ifdef __FreeBSD__ 51 #include <sys/sysctl.h> 52 #include <sys/user.h> 53 #include <sys/thr.h> 54 #include <libutil.h> 55 #endif 56 57 #ifdef __NetBSD__ 58 #include <sys/sysctl.h> 59 #include <lwp.h> 60 #endif 61 62 #ifdef __APPLE__ 63 #include <mach-o/dyld.h> 64 #endif 65 66 #ifdef __HAIKU__ 67 #include <kernel/image.h> 68 #endif 69 70 #include "qemu/mmap-alloc.h" 71 72 #ifdef CONFIG_DEBUG_STACK_USAGE 73 #include "qemu/error-report.h" 74 #endif 75 76 #define MAX_MEM_PREALLOC_THREAD_COUNT 16 77 78 struct MemsetThread; 79 80 typedef struct MemsetContext { 81 bool all_threads_created; 82 bool any_thread_failed; 83 struct MemsetThread *threads; 84 int num_threads; 85 } MemsetContext; 86 87 struct MemsetThread { 88 char *addr; 89 size_t numpages; 90 size_t hpagesize; 91 QemuThread pgthread; 92 sigjmp_buf env; 93 MemsetContext *context; 94 }; 95 typedef struct MemsetThread MemsetThread; 96 97 /* used by sigbus_handler() */ 98 static MemsetContext *sigbus_memset_context; 99 struct sigaction sigbus_oldact; 100 static QemuMutex sigbus_mutex; 101 102 static QemuMutex page_mutex; 103 static QemuCond page_cond; 104 105 int qemu_get_thread_id(void) 106 { 107 #if defined(__linux__) 108 return syscall(SYS_gettid); 109 #elif defined(__FreeBSD__) 110 /* thread id is up to INT_MAX */ 111 long tid; 112 thr_self(&tid); 113 return (int)tid; 114 #elif defined(__NetBSD__) 115 return _lwp_self(); 116 #elif defined(__OpenBSD__) 117 return getthrid(); 118 #else 119 return getpid(); 120 #endif 121 } 122 123 int qemu_daemon(int nochdir, int noclose) 124 { 125 return daemon(nochdir, noclose); 126 } 127 128 bool qemu_write_pidfile(const char *path, Error **errp) 129 { 130 int fd; 131 char pidstr[32]; 132 133 while (1) { 134 struct stat a, b; 135 struct flock lock = { 136 .l_type = F_WRLCK, 137 .l_whence = SEEK_SET, 138 .l_len = 0, 139 }; 140 141 fd = qemu_open_old(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR); 142 if (fd == -1) { 143 error_setg_errno(errp, errno, "Cannot open pid file"); 144 return false; 145 } 146 147 if (fstat(fd, &b) < 0) { 148 error_setg_errno(errp, errno, "Cannot stat file"); 149 goto fail_close; 150 } 151 152 if (fcntl(fd, F_SETLK, &lock)) { 153 error_setg_errno(errp, errno, "Cannot lock pid file"); 154 goto fail_close; 155 } 156 157 /* 158 * Now make sure the path we locked is the same one that now 159 * exists on the filesystem. 160 */ 161 if (stat(path, &a) < 0) { 162 /* 163 * PID file disappeared, someone else must be racing with 164 * us, so try again. 165 */ 166 close(fd); 167 continue; 168 } 169 170 if (a.st_ino == b.st_ino) { 171 break; 172 } 173 174 /* 175 * PID file was recreated, someone else must be racing with 176 * us, so try again. 177 */ 178 close(fd); 179 } 180 181 if (ftruncate(fd, 0) < 0) { 182 error_setg_errno(errp, errno, "Failed to truncate pid file"); 183 goto fail_unlink; 184 } 185 186 snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid()); 187 if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) { 188 error_setg(errp, "Failed to write pid file"); 189 goto fail_unlink; 190 } 191 192 return true; 193 194 fail_unlink: 195 unlink(path); 196 fail_close: 197 close(fd); 198 return false; 199 } 200 201 /* alloc shared memory pages */ 202 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared, 203 bool noreserve) 204 { 205 const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) | 206 (noreserve ? QEMU_MAP_NORESERVE : 0); 207 size_t align = QEMU_VMALLOC_ALIGN; 208 void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0); 209 210 if (ptr == MAP_FAILED) { 211 return NULL; 212 } 213 214 if (alignment) { 215 *alignment = align; 216 } 217 218 trace_qemu_anon_ram_alloc(size, ptr); 219 return ptr; 220 } 221 222 void qemu_anon_ram_free(void *ptr, size_t size) 223 { 224 trace_qemu_anon_ram_free(ptr, size); 225 qemu_ram_munmap(-1, ptr, size); 226 } 227 228 void qemu_set_block(int fd) 229 { 230 int f; 231 f = fcntl(fd, F_GETFL); 232 assert(f != -1); 233 f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK); 234 assert(f != -1); 235 } 236 237 int qemu_try_set_nonblock(int fd) 238 { 239 int f; 240 f = fcntl(fd, F_GETFL); 241 if (f == -1) { 242 return -errno; 243 } 244 if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) { 245 return -errno; 246 } 247 return 0; 248 } 249 250 void qemu_set_nonblock(int fd) 251 { 252 int f; 253 f = qemu_try_set_nonblock(fd); 254 assert(f == 0); 255 } 256 257 int socket_set_fast_reuse(int fd) 258 { 259 int val = 1, ret; 260 261 ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, 262 (const char *)&val, sizeof(val)); 263 264 assert(ret == 0); 265 266 return ret; 267 } 268 269 void qemu_set_cloexec(int fd) 270 { 271 int f; 272 f = fcntl(fd, F_GETFD); 273 assert(f != -1); 274 f = fcntl(fd, F_SETFD, f | FD_CLOEXEC); 275 assert(f != -1); 276 } 277 278 /* 279 * Creates a pipe with FD_CLOEXEC set on both file descriptors 280 */ 281 int qemu_pipe(int pipefd[2]) 282 { 283 int ret; 284 285 #ifdef CONFIG_PIPE2 286 ret = pipe2(pipefd, O_CLOEXEC); 287 if (ret != -1 || errno != ENOSYS) { 288 return ret; 289 } 290 #endif 291 ret = pipe(pipefd); 292 if (ret == 0) { 293 qemu_set_cloexec(pipefd[0]); 294 qemu_set_cloexec(pipefd[1]); 295 } 296 297 return ret; 298 } 299 300 char * 301 qemu_get_local_state_pathname(const char *relative_pathname) 302 { 303 g_autofree char *dir = g_strdup_printf("%s/%s", 304 CONFIG_QEMU_LOCALSTATEDIR, 305 relative_pathname); 306 return get_relocated_path(dir); 307 } 308 309 void qemu_set_tty_echo(int fd, bool echo) 310 { 311 struct termios tty; 312 313 tcgetattr(fd, &tty); 314 315 if (echo) { 316 tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN; 317 } else { 318 tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN); 319 } 320 321 tcsetattr(fd, TCSANOW, &tty); 322 } 323 324 static const char *exec_dir; 325 326 void qemu_init_exec_dir(const char *argv0) 327 { 328 char *p = NULL; 329 char buf[PATH_MAX]; 330 331 if (exec_dir) { 332 return; 333 } 334 335 #if defined(__linux__) 336 { 337 int len; 338 len = readlink("/proc/self/exe", buf, sizeof(buf) - 1); 339 if (len > 0) { 340 buf[len] = 0; 341 p = buf; 342 } 343 } 344 #elif defined(__FreeBSD__) \ 345 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME)) 346 { 347 #if defined(__FreeBSD__) 348 static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1}; 349 #else 350 static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME}; 351 #endif 352 size_t len = sizeof(buf) - 1; 353 354 *buf = '\0'; 355 if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) && 356 *buf) { 357 buf[sizeof(buf) - 1] = '\0'; 358 p = buf; 359 } 360 } 361 #elif defined(__APPLE__) 362 { 363 char fpath[PATH_MAX]; 364 uint32_t len = sizeof(fpath); 365 if (_NSGetExecutablePath(fpath, &len) == 0) { 366 p = realpath(fpath, buf); 367 if (!p) { 368 return; 369 } 370 } 371 } 372 #elif defined(__HAIKU__) 373 { 374 image_info ii; 375 int32_t c = 0; 376 377 *buf = '\0'; 378 while (get_next_image_info(0, &c, &ii) == B_OK) { 379 if (ii.type == B_APP_IMAGE) { 380 strncpy(buf, ii.name, sizeof(buf)); 381 buf[sizeof(buf) - 1] = 0; 382 p = buf; 383 break; 384 } 385 } 386 } 387 #endif 388 /* If we don't have any way of figuring out the actual executable 389 location then try argv[0]. */ 390 if (!p && argv0) { 391 p = realpath(argv0, buf); 392 } 393 if (p) { 394 exec_dir = g_path_get_dirname(p); 395 } else { 396 exec_dir = CONFIG_BINDIR; 397 } 398 } 399 400 const char *qemu_get_exec_dir(void) 401 { 402 return exec_dir; 403 } 404 405 #ifdef CONFIG_LINUX 406 static void sigbus_handler(int signal, siginfo_t *siginfo, void *ctx) 407 #else /* CONFIG_LINUX */ 408 static void sigbus_handler(int signal) 409 #endif /* CONFIG_LINUX */ 410 { 411 int i; 412 413 if (sigbus_memset_context) { 414 for (i = 0; i < sigbus_memset_context->num_threads; i++) { 415 MemsetThread *thread = &sigbus_memset_context->threads[i]; 416 417 if (qemu_thread_is_self(&thread->pgthread)) { 418 siglongjmp(thread->env, 1); 419 } 420 } 421 } 422 423 #ifdef CONFIG_LINUX 424 /* 425 * We assume that the MCE SIGBUS handler could have been registered. We 426 * should never receive BUS_MCEERR_AO on any of our threads, but only on 427 * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not 428 * receive BUS_MCEERR_AR triggered by action of other threads on one of 429 * our threads. So, no need to check for unrelated SIGBUS when seeing one 430 * for our threads. 431 * 432 * We will forward to the MCE handler, which will either handle the SIGBUS 433 * or reinstall the default SIGBUS handler and reraise the SIGBUS. The 434 * default SIGBUS handler will crash the process, so we don't care. 435 */ 436 if (sigbus_oldact.sa_flags & SA_SIGINFO) { 437 sigbus_oldact.sa_sigaction(signal, siginfo, ctx); 438 return; 439 } 440 #endif /* CONFIG_LINUX */ 441 warn_report("os_mem_prealloc: unrelated SIGBUS detected and ignored"); 442 } 443 444 static void *do_touch_pages(void *arg) 445 { 446 MemsetThread *memset_args = (MemsetThread *)arg; 447 sigset_t set, oldset; 448 int ret = 0; 449 450 /* 451 * On Linux, the page faults from the loop below can cause mmap_sem 452 * contention with allocation of the thread stacks. Do not start 453 * clearing until all threads have been created. 454 */ 455 qemu_mutex_lock(&page_mutex); 456 while (!memset_args->context->all_threads_created) { 457 qemu_cond_wait(&page_cond, &page_mutex); 458 } 459 qemu_mutex_unlock(&page_mutex); 460 461 /* unblock SIGBUS */ 462 sigemptyset(&set); 463 sigaddset(&set, SIGBUS); 464 pthread_sigmask(SIG_UNBLOCK, &set, &oldset); 465 466 if (sigsetjmp(memset_args->env, 1)) { 467 ret = -EFAULT; 468 } else { 469 char *addr = memset_args->addr; 470 size_t numpages = memset_args->numpages; 471 size_t hpagesize = memset_args->hpagesize; 472 size_t i; 473 for (i = 0; i < numpages; i++) { 474 /* 475 * Read & write back the same value, so we don't 476 * corrupt existing user/app data that might be 477 * stored. 478 * 479 * 'volatile' to stop compiler optimizing this away 480 * to a no-op 481 */ 482 *(volatile char *)addr = *addr; 483 addr += hpagesize; 484 } 485 } 486 pthread_sigmask(SIG_SETMASK, &oldset, NULL); 487 return (void *)(uintptr_t)ret; 488 } 489 490 static void *do_madv_populate_write_pages(void *arg) 491 { 492 MemsetThread *memset_args = (MemsetThread *)arg; 493 const size_t size = memset_args->numpages * memset_args->hpagesize; 494 char * const addr = memset_args->addr; 495 int ret = 0; 496 497 /* See do_touch_pages(). */ 498 qemu_mutex_lock(&page_mutex); 499 while (!memset_args->context->all_threads_created) { 500 qemu_cond_wait(&page_cond, &page_mutex); 501 } 502 qemu_mutex_unlock(&page_mutex); 503 504 if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) { 505 ret = -errno; 506 } 507 return (void *)(uintptr_t)ret; 508 } 509 510 static inline int get_memset_num_threads(size_t hpagesize, size_t numpages, 511 int smp_cpus) 512 { 513 long host_procs = sysconf(_SC_NPROCESSORS_ONLN); 514 int ret = 1; 515 516 if (host_procs > 0) { 517 ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus); 518 } 519 520 /* Especially with gigantic pages, don't create more threads than pages. */ 521 ret = MIN(ret, numpages); 522 /* Don't start threads to prealloc comparatively little memory. */ 523 ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB))); 524 525 /* In case sysconf() fails, we fall back to single threaded */ 526 return ret; 527 } 528 529 static int touch_all_pages(char *area, size_t hpagesize, size_t numpages, 530 int smp_cpus, bool use_madv_populate_write) 531 { 532 static gsize initialized = 0; 533 MemsetContext context = { 534 .num_threads = get_memset_num_threads(hpagesize, numpages, smp_cpus), 535 }; 536 size_t numpages_per_thread, leftover; 537 void *(*touch_fn)(void *); 538 int ret = 0, i = 0; 539 char *addr = area; 540 541 if (g_once_init_enter(&initialized)) { 542 qemu_mutex_init(&page_mutex); 543 qemu_cond_init(&page_cond); 544 g_once_init_leave(&initialized, 1); 545 } 546 547 if (use_madv_populate_write) { 548 /* Avoid creating a single thread for MADV_POPULATE_WRITE */ 549 if (context.num_threads == 1) { 550 if (qemu_madvise(area, hpagesize * numpages, 551 QEMU_MADV_POPULATE_WRITE)) { 552 return -errno; 553 } 554 return 0; 555 } 556 touch_fn = do_madv_populate_write_pages; 557 } else { 558 touch_fn = do_touch_pages; 559 } 560 561 context.threads = g_new0(MemsetThread, context.num_threads); 562 numpages_per_thread = numpages / context.num_threads; 563 leftover = numpages % context.num_threads; 564 for (i = 0; i < context.num_threads; i++) { 565 context.threads[i].addr = addr; 566 context.threads[i].numpages = numpages_per_thread + (i < leftover); 567 context.threads[i].hpagesize = hpagesize; 568 context.threads[i].context = &context; 569 qemu_thread_create(&context.threads[i].pgthread, "touch_pages", 570 touch_fn, &context.threads[i], 571 QEMU_THREAD_JOINABLE); 572 addr += context.threads[i].numpages * hpagesize; 573 } 574 575 if (!use_madv_populate_write) { 576 sigbus_memset_context = &context; 577 } 578 579 qemu_mutex_lock(&page_mutex); 580 context.all_threads_created = true; 581 qemu_cond_broadcast(&page_cond); 582 qemu_mutex_unlock(&page_mutex); 583 584 for (i = 0; i < context.num_threads; i++) { 585 int tmp = (uintptr_t)qemu_thread_join(&context.threads[i].pgthread); 586 587 if (tmp) { 588 ret = tmp; 589 } 590 } 591 592 if (!use_madv_populate_write) { 593 sigbus_memset_context = NULL; 594 } 595 g_free(context.threads); 596 597 return ret; 598 } 599 600 static bool madv_populate_write_possible(char *area, size_t pagesize) 601 { 602 return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) || 603 errno != EINVAL; 604 } 605 606 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus, 607 Error **errp) 608 { 609 static gsize initialized; 610 int ret; 611 size_t hpagesize = qemu_fd_getpagesize(fd); 612 size_t numpages = DIV_ROUND_UP(memory, hpagesize); 613 bool use_madv_populate_write; 614 struct sigaction act; 615 616 /* 617 * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for 618 * some special mappings, such as mapping /dev/mem. 619 */ 620 use_madv_populate_write = madv_populate_write_possible(area, hpagesize); 621 622 if (!use_madv_populate_write) { 623 if (g_once_init_enter(&initialized)) { 624 qemu_mutex_init(&sigbus_mutex); 625 g_once_init_leave(&initialized, 1); 626 } 627 628 qemu_mutex_lock(&sigbus_mutex); 629 memset(&act, 0, sizeof(act)); 630 #ifdef CONFIG_LINUX 631 act.sa_sigaction = &sigbus_handler; 632 act.sa_flags = SA_SIGINFO; 633 #else /* CONFIG_LINUX */ 634 act.sa_handler = &sigbus_handler; 635 act.sa_flags = 0; 636 #endif /* CONFIG_LINUX */ 637 638 ret = sigaction(SIGBUS, &act, &sigbus_oldact); 639 if (ret) { 640 qemu_mutex_unlock(&sigbus_mutex); 641 error_setg_errno(errp, errno, 642 "os_mem_prealloc: failed to install signal handler"); 643 return; 644 } 645 } 646 647 /* touch pages simultaneously */ 648 ret = touch_all_pages(area, hpagesize, numpages, smp_cpus, 649 use_madv_populate_write); 650 if (ret) { 651 error_setg_errno(errp, -ret, 652 "os_mem_prealloc: preallocating memory failed"); 653 } 654 655 if (!use_madv_populate_write) { 656 ret = sigaction(SIGBUS, &sigbus_oldact, NULL); 657 if (ret) { 658 /* Terminate QEMU since it can't recover from error */ 659 perror("os_mem_prealloc: failed to reinstall signal handler"); 660 exit(1); 661 } 662 qemu_mutex_unlock(&sigbus_mutex); 663 } 664 } 665 666 char *qemu_get_pid_name(pid_t pid) 667 { 668 char *name = NULL; 669 670 #if defined(__FreeBSD__) 671 /* BSDs don't have /proc, but they provide a nice substitute */ 672 struct kinfo_proc *proc = kinfo_getproc(pid); 673 674 if (proc) { 675 name = g_strdup(proc->ki_comm); 676 free(proc); 677 } 678 #else 679 /* Assume a system with reasonable procfs */ 680 char *pid_path; 681 size_t len; 682 683 pid_path = g_strdup_printf("/proc/%d/cmdline", pid); 684 g_file_get_contents(pid_path, &name, &len, NULL); 685 g_free(pid_path); 686 #endif 687 688 return name; 689 } 690 691 692 pid_t qemu_fork(Error **errp) 693 { 694 sigset_t oldmask, newmask; 695 struct sigaction sig_action; 696 int saved_errno; 697 pid_t pid; 698 699 /* 700 * Need to block signals now, so that child process can safely 701 * kill off caller's signal handlers without a race. 702 */ 703 sigfillset(&newmask); 704 if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) { 705 error_setg_errno(errp, errno, 706 "cannot block signals"); 707 return -1; 708 } 709 710 pid = fork(); 711 saved_errno = errno; 712 713 if (pid < 0) { 714 /* attempt to restore signal mask, but ignore failure, to 715 * avoid obscuring the fork failure */ 716 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); 717 error_setg_errno(errp, saved_errno, 718 "cannot fork child process"); 719 errno = saved_errno; 720 return -1; 721 } else if (pid) { 722 /* parent process */ 723 724 /* Restore our original signal mask now that the child is 725 * safely running. Only documented failures are EFAULT (not 726 * possible, since we are using just-grabbed mask) or EINVAL 727 * (not possible, since we are using correct arguments). */ 728 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); 729 } else { 730 /* child process */ 731 size_t i; 732 733 /* Clear out all signal handlers from parent so nothing 734 * unexpected can happen in our child once we unblock 735 * signals */ 736 sig_action.sa_handler = SIG_DFL; 737 sig_action.sa_flags = 0; 738 sigemptyset(&sig_action.sa_mask); 739 740 for (i = 1; i < NSIG; i++) { 741 /* Only possible errors are EFAULT or EINVAL The former 742 * won't happen, the latter we expect, so no need to check 743 * return value */ 744 (void)sigaction(i, &sig_action, NULL); 745 } 746 747 /* Unmask all signals in child, since we've no idea what the 748 * caller's done with their signal mask and don't want to 749 * propagate that to children */ 750 sigemptyset(&newmask); 751 if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) { 752 Error *local_err = NULL; 753 error_setg_errno(&local_err, errno, 754 "cannot unblock signals"); 755 error_report_err(local_err); 756 _exit(1); 757 } 758 } 759 return pid; 760 } 761 762 void *qemu_alloc_stack(size_t *sz) 763 { 764 void *ptr, *guardpage; 765 int flags; 766 #ifdef CONFIG_DEBUG_STACK_USAGE 767 void *ptr2; 768 #endif 769 size_t pagesz = qemu_real_host_page_size(); 770 #ifdef _SC_THREAD_STACK_MIN 771 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */ 772 long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN); 773 *sz = MAX(MAX(min_stack_sz, 0), *sz); 774 #endif 775 /* adjust stack size to a multiple of the page size */ 776 *sz = ROUND_UP(*sz, pagesz); 777 /* allocate one extra page for the guard page */ 778 *sz += pagesz; 779 780 flags = MAP_PRIVATE | MAP_ANONYMOUS; 781 #if defined(MAP_STACK) && defined(__OpenBSD__) 782 /* Only enable MAP_STACK on OpenBSD. Other OS's such as 783 * Linux/FreeBSD/NetBSD have a flag with the same name 784 * but have differing functionality. OpenBSD will SEGV 785 * if it spots execution with a stack pointer pointing 786 * at memory that was not allocated with MAP_STACK. 787 */ 788 flags |= MAP_STACK; 789 #endif 790 791 ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0); 792 if (ptr == MAP_FAILED) { 793 perror("failed to allocate memory for stack"); 794 abort(); 795 } 796 797 #if defined(HOST_IA64) 798 /* separate register stack */ 799 guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz); 800 #elif defined(HOST_HPPA) 801 /* stack grows up */ 802 guardpage = ptr + *sz - pagesz; 803 #else 804 /* stack grows down */ 805 guardpage = ptr; 806 #endif 807 if (mprotect(guardpage, pagesz, PROT_NONE) != 0) { 808 perror("failed to set up stack guard page"); 809 abort(); 810 } 811 812 #ifdef CONFIG_DEBUG_STACK_USAGE 813 for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) { 814 *(uint32_t *)ptr2 = 0xdeadbeaf; 815 } 816 #endif 817 818 return ptr; 819 } 820 821 #ifdef CONFIG_DEBUG_STACK_USAGE 822 static __thread unsigned int max_stack_usage; 823 #endif 824 825 void qemu_free_stack(void *stack, size_t sz) 826 { 827 #ifdef CONFIG_DEBUG_STACK_USAGE 828 unsigned int usage; 829 void *ptr; 830 831 for (ptr = stack + qemu_real_host_page_size(); ptr < stack + sz; 832 ptr += sizeof(uint32_t)) { 833 if (*(uint32_t *)ptr != 0xdeadbeaf) { 834 break; 835 } 836 } 837 usage = sz - (uintptr_t) (ptr - stack); 838 if (usage > max_stack_usage) { 839 error_report("thread %d max stack usage increased from %u to %u", 840 qemu_get_thread_id(), max_stack_usage, usage); 841 max_stack_usage = usage; 842 } 843 #endif 844 845 munmap(stack, sz); 846 } 847 848 /* 849 * Disable CFI checks. 850 * We are going to call a signal hander directly. Such handler may or may not 851 * have been defined in our binary, so there's no guarantee that the pointer 852 * used to set the handler is a cfi-valid pointer. Since the handlers are 853 * stored in kernel memory, changing the handler to an attacker-defined 854 * function requires being able to call a sigaction() syscall, 855 * which is not as easy as overwriting a pointer in memory. 856 */ 857 QEMU_DISABLE_CFI 858 void sigaction_invoke(struct sigaction *action, 859 struct qemu_signalfd_siginfo *info) 860 { 861 siginfo_t si = {}; 862 si.si_signo = info->ssi_signo; 863 si.si_errno = info->ssi_errno; 864 si.si_code = info->ssi_code; 865 866 /* Convert the minimal set of fields defined by POSIX. 867 * Positive si_code values are reserved for kernel-generated 868 * signals, where the valid siginfo fields are determined by 869 * the signal number. But according to POSIX, it is unspecified 870 * whether SI_USER and SI_QUEUE have values less than or equal to 871 * zero. 872 */ 873 if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE || 874 info->ssi_code <= 0) { 875 /* SIGTERM, etc. */ 876 si.si_pid = info->ssi_pid; 877 si.si_uid = info->ssi_uid; 878 } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE || 879 info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) { 880 si.si_addr = (void *)(uintptr_t)info->ssi_addr; 881 } else if (info->ssi_signo == SIGCHLD) { 882 si.si_pid = info->ssi_pid; 883 si.si_status = info->ssi_status; 884 si.si_uid = info->ssi_uid; 885 } 886 action->sa_sigaction(info->ssi_signo, &si, NULL); 887 } 888 889 #ifndef HOST_NAME_MAX 890 # ifdef _POSIX_HOST_NAME_MAX 891 # define HOST_NAME_MAX _POSIX_HOST_NAME_MAX 892 # else 893 # define HOST_NAME_MAX 255 894 # endif 895 #endif 896 897 char *qemu_get_host_name(Error **errp) 898 { 899 long len = -1; 900 g_autofree char *hostname = NULL; 901 902 #ifdef _SC_HOST_NAME_MAX 903 len = sysconf(_SC_HOST_NAME_MAX); 904 #endif /* _SC_HOST_NAME_MAX */ 905 906 if (len < 0) { 907 len = HOST_NAME_MAX; 908 } 909 910 /* Unfortunately, gethostname() below does not guarantee a 911 * NULL terminated string. Therefore, allocate one byte more 912 * to be sure. */ 913 hostname = g_new0(char, len + 1); 914 915 if (gethostname(hostname, len) < 0) { 916 error_setg_errno(errp, errno, 917 "cannot get hostname"); 918 return NULL; 919 } 920 921 return g_steal_pointer(&hostname); 922 } 923 924 size_t qemu_get_host_physmem(void) 925 { 926 #ifdef _SC_PHYS_PAGES 927 long pages = sysconf(_SC_PHYS_PAGES); 928 if (pages > 0) { 929 if (pages > SIZE_MAX / qemu_real_host_page_size()) { 930 return SIZE_MAX; 931 } else { 932 return pages * qemu_real_host_page_size(); 933 } 934 } 935 #endif 936 return 0; 937 } 938 939 /* Sets a specific flag */ 940 int fcntl_setfl(int fd, int flag) 941 { 942 int flags; 943 944 flags = fcntl(fd, F_GETFL); 945 if (flags == -1) { 946 return -errno; 947 } 948 if (fcntl(fd, F_SETFL, flags | flag) == -1) { 949 return -errno; 950 } 951 return 0; 952 } 953