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 "qemu-common.h" 35 #include "sysemu/sysemu.h" 36 #include "trace.h" 37 #include "qapi/error.h" 38 #include "qemu/sockets.h" 39 #include "qemu/thread.h" 40 #include <libgen.h> 41 #include "qemu/cutils.h" 42 #include "qemu/compiler.h" 43 44 #ifdef CONFIG_LINUX 45 #include <sys/syscall.h> 46 #endif 47 48 #ifdef __FreeBSD__ 49 #include <sys/sysctl.h> 50 #include <sys/user.h> 51 #include <sys/thr.h> 52 #include <libutil.h> 53 #endif 54 55 #ifdef __NetBSD__ 56 #include <sys/sysctl.h> 57 #include <lwp.h> 58 #endif 59 60 #ifdef __APPLE__ 61 #include <mach-o/dyld.h> 62 #endif 63 64 #ifdef __HAIKU__ 65 #include <kernel/image.h> 66 #endif 67 68 #include "qemu/mmap-alloc.h" 69 70 #ifdef CONFIG_DEBUG_STACK_USAGE 71 #include "qemu/error-report.h" 72 #endif 73 74 #define MAX_MEM_PREALLOC_THREAD_COUNT 16 75 76 struct MemsetThread { 77 char *addr; 78 size_t numpages; 79 size_t hpagesize; 80 QemuThread pgthread; 81 sigjmp_buf env; 82 }; 83 typedef struct MemsetThread MemsetThread; 84 85 static MemsetThread *memset_thread; 86 static int memset_num_threads; 87 static bool memset_thread_failed; 88 89 static QemuMutex page_mutex; 90 static QemuCond page_cond; 91 static bool threads_created_flag; 92 93 int qemu_get_thread_id(void) 94 { 95 #if defined(__linux__) 96 return syscall(SYS_gettid); 97 #elif defined(__FreeBSD__) 98 /* thread id is up to INT_MAX */ 99 long tid; 100 thr_self(&tid); 101 return (int)tid; 102 #elif defined(__NetBSD__) 103 return _lwp_self(); 104 #elif defined(__OpenBSD__) 105 return getthrid(); 106 #else 107 return getpid(); 108 #endif 109 } 110 111 int qemu_daemon(int nochdir, int noclose) 112 { 113 return daemon(nochdir, noclose); 114 } 115 116 bool qemu_write_pidfile(const char *path, Error **errp) 117 { 118 int fd; 119 char pidstr[32]; 120 121 while (1) { 122 struct stat a, b; 123 struct flock lock = { 124 .l_type = F_WRLCK, 125 .l_whence = SEEK_SET, 126 .l_len = 0, 127 }; 128 129 fd = qemu_open_old(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR); 130 if (fd == -1) { 131 error_setg_errno(errp, errno, "Cannot open pid file"); 132 return false; 133 } 134 135 if (fstat(fd, &b) < 0) { 136 error_setg_errno(errp, errno, "Cannot stat file"); 137 goto fail_close; 138 } 139 140 if (fcntl(fd, F_SETLK, &lock)) { 141 error_setg_errno(errp, errno, "Cannot lock pid file"); 142 goto fail_close; 143 } 144 145 /* 146 * Now make sure the path we locked is the same one that now 147 * exists on the filesystem. 148 */ 149 if (stat(path, &a) < 0) { 150 /* 151 * PID file disappeared, someone else must be racing with 152 * us, so try again. 153 */ 154 close(fd); 155 continue; 156 } 157 158 if (a.st_ino == b.st_ino) { 159 break; 160 } 161 162 /* 163 * PID file was recreated, someone else must be racing with 164 * us, so try again. 165 */ 166 close(fd); 167 } 168 169 if (ftruncate(fd, 0) < 0) { 170 error_setg_errno(errp, errno, "Failed to truncate pid file"); 171 goto fail_unlink; 172 } 173 174 snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid()); 175 if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) { 176 error_setg(errp, "Failed to write pid file"); 177 goto fail_unlink; 178 } 179 180 return true; 181 182 fail_unlink: 183 unlink(path); 184 fail_close: 185 close(fd); 186 return false; 187 } 188 189 void *qemu_oom_check(void *ptr) 190 { 191 if (ptr == NULL) { 192 fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno)); 193 abort(); 194 } 195 return ptr; 196 } 197 198 void *qemu_try_memalign(size_t alignment, size_t size) 199 { 200 void *ptr; 201 202 if (alignment < sizeof(void*)) { 203 alignment = sizeof(void*); 204 } 205 206 #if defined(CONFIG_POSIX_MEMALIGN) 207 int ret; 208 ret = posix_memalign(&ptr, alignment, size); 209 if (ret != 0) { 210 errno = ret; 211 ptr = NULL; 212 } 213 #elif defined(CONFIG_BSD) 214 ptr = valloc(size); 215 #else 216 ptr = memalign(alignment, size); 217 #endif 218 trace_qemu_memalign(alignment, size, ptr); 219 return ptr; 220 } 221 222 void *qemu_memalign(size_t alignment, size_t size) 223 { 224 return qemu_oom_check(qemu_try_memalign(alignment, size)); 225 } 226 227 /* alloc shared memory pages */ 228 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared) 229 { 230 size_t align = QEMU_VMALLOC_ALIGN; 231 void *ptr = qemu_ram_mmap(-1, size, align, shared, false); 232 233 if (ptr == MAP_FAILED) { 234 return NULL; 235 } 236 237 if (alignment) { 238 *alignment = align; 239 } 240 241 trace_qemu_anon_ram_alloc(size, ptr); 242 return ptr; 243 } 244 245 void qemu_vfree(void *ptr) 246 { 247 trace_qemu_vfree(ptr); 248 free(ptr); 249 } 250 251 void qemu_anon_ram_free(void *ptr, size_t size) 252 { 253 trace_qemu_anon_ram_free(ptr, size); 254 qemu_ram_munmap(-1, ptr, size); 255 } 256 257 void qemu_set_block(int fd) 258 { 259 int f; 260 f = fcntl(fd, F_GETFL); 261 assert(f != -1); 262 f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK); 263 assert(f != -1); 264 } 265 266 int qemu_try_set_nonblock(int fd) 267 { 268 int f; 269 f = fcntl(fd, F_GETFL); 270 if (f == -1) { 271 return -errno; 272 } 273 if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) { 274 #ifdef __OpenBSD__ 275 /* 276 * Previous to OpenBSD 6.3, fcntl(F_SETFL) is not permitted on 277 * memory devices and sets errno to ENODEV. 278 * It's OK if we fail to set O_NONBLOCK on devices like /dev/null, 279 * because they will never block anyway. 280 */ 281 if (errno == ENODEV) { 282 return 0; 283 } 284 #endif 285 return -errno; 286 } 287 return 0; 288 } 289 290 void qemu_set_nonblock(int fd) 291 { 292 int f; 293 f = qemu_try_set_nonblock(fd); 294 assert(f == 0); 295 } 296 297 int socket_set_fast_reuse(int fd) 298 { 299 int val = 1, ret; 300 301 ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, 302 (const char *)&val, sizeof(val)); 303 304 assert(ret == 0); 305 306 return ret; 307 } 308 309 void qemu_set_cloexec(int fd) 310 { 311 int f; 312 f = fcntl(fd, F_GETFD); 313 assert(f != -1); 314 f = fcntl(fd, F_SETFD, f | FD_CLOEXEC); 315 assert(f != -1); 316 } 317 318 /* 319 * Creates a pipe with FD_CLOEXEC set on both file descriptors 320 */ 321 int qemu_pipe(int pipefd[2]) 322 { 323 int ret; 324 325 #ifdef CONFIG_PIPE2 326 ret = pipe2(pipefd, O_CLOEXEC); 327 if (ret != -1 || errno != ENOSYS) { 328 return ret; 329 } 330 #endif 331 ret = pipe(pipefd); 332 if (ret == 0) { 333 qemu_set_cloexec(pipefd[0]); 334 qemu_set_cloexec(pipefd[1]); 335 } 336 337 return ret; 338 } 339 340 char * 341 qemu_get_local_state_pathname(const char *relative_pathname) 342 { 343 g_autofree char *dir = g_strdup_printf("%s/%s", 344 CONFIG_QEMU_LOCALSTATEDIR, 345 relative_pathname); 346 return get_relocated_path(dir); 347 } 348 349 void qemu_set_tty_echo(int fd, bool echo) 350 { 351 struct termios tty; 352 353 tcgetattr(fd, &tty); 354 355 if (echo) { 356 tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN; 357 } else { 358 tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN); 359 } 360 361 tcsetattr(fd, TCSANOW, &tty); 362 } 363 364 static const char *exec_dir; 365 366 void qemu_init_exec_dir(const char *argv0) 367 { 368 char *p = NULL; 369 char buf[PATH_MAX]; 370 371 if (exec_dir) { 372 return; 373 } 374 375 #if defined(__linux__) 376 { 377 int len; 378 len = readlink("/proc/self/exe", buf, sizeof(buf) - 1); 379 if (len > 0) { 380 buf[len] = 0; 381 p = buf; 382 } 383 } 384 #elif defined(__FreeBSD__) \ 385 || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME)) 386 { 387 #if defined(__FreeBSD__) 388 static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1}; 389 #else 390 static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME}; 391 #endif 392 size_t len = sizeof(buf) - 1; 393 394 *buf = '\0'; 395 if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) && 396 *buf) { 397 buf[sizeof(buf) - 1] = '\0'; 398 p = buf; 399 } 400 } 401 #elif defined(__APPLE__) 402 { 403 char fpath[PATH_MAX]; 404 uint32_t len = sizeof(fpath); 405 if (_NSGetExecutablePath(fpath, &len) == 0) { 406 p = realpath(fpath, buf); 407 if (!p) { 408 return; 409 } 410 } 411 } 412 #elif defined(__HAIKU__) 413 { 414 image_info ii; 415 int32_t c = 0; 416 417 *buf = '\0'; 418 while (get_next_image_info(0, &c, &ii) == B_OK) { 419 if (ii.type == B_APP_IMAGE) { 420 strncpy(buf, ii.name, sizeof(buf)); 421 buf[sizeof(buf) - 1] = 0; 422 p = buf; 423 break; 424 } 425 } 426 } 427 #endif 428 /* If we don't have any way of figuring out the actual executable 429 location then try argv[0]. */ 430 if (!p && argv0) { 431 p = realpath(argv0, buf); 432 } 433 if (p) { 434 exec_dir = g_path_get_dirname(p); 435 } else { 436 exec_dir = CONFIG_BINDIR; 437 } 438 } 439 440 const char *qemu_get_exec_dir(void) 441 { 442 return exec_dir; 443 } 444 445 static void sigbus_handler(int signal) 446 { 447 int i; 448 if (memset_thread) { 449 for (i = 0; i < memset_num_threads; i++) { 450 if (qemu_thread_is_self(&memset_thread[i].pgthread)) { 451 siglongjmp(memset_thread[i].env, 1); 452 } 453 } 454 } 455 } 456 457 static void *do_touch_pages(void *arg) 458 { 459 MemsetThread *memset_args = (MemsetThread *)arg; 460 sigset_t set, oldset; 461 462 /* 463 * On Linux, the page faults from the loop below can cause mmap_sem 464 * contention with allocation of the thread stacks. Do not start 465 * clearing until all threads have been created. 466 */ 467 qemu_mutex_lock(&page_mutex); 468 while(!threads_created_flag){ 469 qemu_cond_wait(&page_cond, &page_mutex); 470 } 471 qemu_mutex_unlock(&page_mutex); 472 473 /* unblock SIGBUS */ 474 sigemptyset(&set); 475 sigaddset(&set, SIGBUS); 476 pthread_sigmask(SIG_UNBLOCK, &set, &oldset); 477 478 if (sigsetjmp(memset_args->env, 1)) { 479 memset_thread_failed = true; 480 } else { 481 char *addr = memset_args->addr; 482 size_t numpages = memset_args->numpages; 483 size_t hpagesize = memset_args->hpagesize; 484 size_t i; 485 for (i = 0; i < numpages; i++) { 486 /* 487 * Read & write back the same value, so we don't 488 * corrupt existing user/app data that might be 489 * stored. 490 * 491 * 'volatile' to stop compiler optimizing this away 492 * to a no-op 493 * 494 * TODO: get a better solution from kernel so we 495 * don't need to write at all so we don't cause 496 * wear on the storage backing the region... 497 */ 498 *(volatile char *)addr = *addr; 499 addr += hpagesize; 500 } 501 } 502 pthread_sigmask(SIG_SETMASK, &oldset, NULL); 503 return NULL; 504 } 505 506 static inline int get_memset_num_threads(int smp_cpus) 507 { 508 long host_procs = sysconf(_SC_NPROCESSORS_ONLN); 509 int ret = 1; 510 511 if (host_procs > 0) { 512 ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus); 513 } 514 /* In case sysconf() fails, we fall back to single threaded */ 515 return ret; 516 } 517 518 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages, 519 int smp_cpus) 520 { 521 static gsize initialized = 0; 522 size_t numpages_per_thread, leftover; 523 char *addr = area; 524 int i = 0; 525 526 if (g_once_init_enter(&initialized)) { 527 qemu_mutex_init(&page_mutex); 528 qemu_cond_init(&page_cond); 529 g_once_init_leave(&initialized, 1); 530 } 531 532 memset_thread_failed = false; 533 threads_created_flag = false; 534 memset_num_threads = get_memset_num_threads(smp_cpus); 535 memset_thread = g_new0(MemsetThread, memset_num_threads); 536 numpages_per_thread = numpages / memset_num_threads; 537 leftover = numpages % memset_num_threads; 538 for (i = 0; i < memset_num_threads; i++) { 539 memset_thread[i].addr = addr; 540 memset_thread[i].numpages = numpages_per_thread + (i < leftover); 541 memset_thread[i].hpagesize = hpagesize; 542 qemu_thread_create(&memset_thread[i].pgthread, "touch_pages", 543 do_touch_pages, &memset_thread[i], 544 QEMU_THREAD_JOINABLE); 545 addr += memset_thread[i].numpages * hpagesize; 546 } 547 548 qemu_mutex_lock(&page_mutex); 549 threads_created_flag = true; 550 qemu_cond_broadcast(&page_cond); 551 qemu_mutex_unlock(&page_mutex); 552 553 for (i = 0; i < memset_num_threads; i++) { 554 qemu_thread_join(&memset_thread[i].pgthread); 555 } 556 g_free(memset_thread); 557 memset_thread = NULL; 558 559 return memset_thread_failed; 560 } 561 562 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus, 563 Error **errp) 564 { 565 int ret; 566 struct sigaction act, oldact; 567 size_t hpagesize = qemu_fd_getpagesize(fd); 568 size_t numpages = DIV_ROUND_UP(memory, hpagesize); 569 570 memset(&act, 0, sizeof(act)); 571 act.sa_handler = &sigbus_handler; 572 act.sa_flags = 0; 573 574 ret = sigaction(SIGBUS, &act, &oldact); 575 if (ret) { 576 error_setg_errno(errp, errno, 577 "os_mem_prealloc: failed to install signal handler"); 578 return; 579 } 580 581 /* touch pages simultaneously */ 582 if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) { 583 error_setg(errp, "os_mem_prealloc: Insufficient free host memory " 584 "pages available to allocate guest RAM"); 585 } 586 587 ret = sigaction(SIGBUS, &oldact, NULL); 588 if (ret) { 589 /* Terminate QEMU since it can't recover from error */ 590 perror("os_mem_prealloc: failed to reinstall signal handler"); 591 exit(1); 592 } 593 } 594 595 char *qemu_get_pid_name(pid_t pid) 596 { 597 char *name = NULL; 598 599 #if defined(__FreeBSD__) 600 /* BSDs don't have /proc, but they provide a nice substitute */ 601 struct kinfo_proc *proc = kinfo_getproc(pid); 602 603 if (proc) { 604 name = g_strdup(proc->ki_comm); 605 free(proc); 606 } 607 #else 608 /* Assume a system with reasonable procfs */ 609 char *pid_path; 610 size_t len; 611 612 pid_path = g_strdup_printf("/proc/%d/cmdline", pid); 613 g_file_get_contents(pid_path, &name, &len, NULL); 614 g_free(pid_path); 615 #endif 616 617 return name; 618 } 619 620 621 pid_t qemu_fork(Error **errp) 622 { 623 sigset_t oldmask, newmask; 624 struct sigaction sig_action; 625 int saved_errno; 626 pid_t pid; 627 628 /* 629 * Need to block signals now, so that child process can safely 630 * kill off caller's signal handlers without a race. 631 */ 632 sigfillset(&newmask); 633 if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) { 634 error_setg_errno(errp, errno, 635 "cannot block signals"); 636 return -1; 637 } 638 639 pid = fork(); 640 saved_errno = errno; 641 642 if (pid < 0) { 643 /* attempt to restore signal mask, but ignore failure, to 644 * avoid obscuring the fork failure */ 645 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); 646 error_setg_errno(errp, saved_errno, 647 "cannot fork child process"); 648 errno = saved_errno; 649 return -1; 650 } else if (pid) { 651 /* parent process */ 652 653 /* Restore our original signal mask now that the child is 654 * safely running. Only documented failures are EFAULT (not 655 * possible, since we are using just-grabbed mask) or EINVAL 656 * (not possible, since we are using correct arguments). */ 657 (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL); 658 } else { 659 /* child process */ 660 size_t i; 661 662 /* Clear out all signal handlers from parent so nothing 663 * unexpected can happen in our child once we unblock 664 * signals */ 665 sig_action.sa_handler = SIG_DFL; 666 sig_action.sa_flags = 0; 667 sigemptyset(&sig_action.sa_mask); 668 669 for (i = 1; i < NSIG; i++) { 670 /* Only possible errors are EFAULT or EINVAL The former 671 * won't happen, the latter we expect, so no need to check 672 * return value */ 673 (void)sigaction(i, &sig_action, NULL); 674 } 675 676 /* Unmask all signals in child, since we've no idea what the 677 * caller's done with their signal mask and don't want to 678 * propagate that to children */ 679 sigemptyset(&newmask); 680 if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) { 681 Error *local_err = NULL; 682 error_setg_errno(&local_err, errno, 683 "cannot unblock signals"); 684 error_report_err(local_err); 685 _exit(1); 686 } 687 } 688 return pid; 689 } 690 691 void *qemu_alloc_stack(size_t *sz) 692 { 693 void *ptr, *guardpage; 694 int flags; 695 #ifdef CONFIG_DEBUG_STACK_USAGE 696 void *ptr2; 697 #endif 698 size_t pagesz = qemu_real_host_page_size; 699 #ifdef _SC_THREAD_STACK_MIN 700 /* avoid stacks smaller than _SC_THREAD_STACK_MIN */ 701 long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN); 702 *sz = MAX(MAX(min_stack_sz, 0), *sz); 703 #endif 704 /* adjust stack size to a multiple of the page size */ 705 *sz = ROUND_UP(*sz, pagesz); 706 /* allocate one extra page for the guard page */ 707 *sz += pagesz; 708 709 flags = MAP_PRIVATE | MAP_ANONYMOUS; 710 #if defined(MAP_STACK) && defined(__OpenBSD__) 711 /* Only enable MAP_STACK on OpenBSD. Other OS's such as 712 * Linux/FreeBSD/NetBSD have a flag with the same name 713 * but have differing functionality. OpenBSD will SEGV 714 * if it spots execution with a stack pointer pointing 715 * at memory that was not allocated with MAP_STACK. 716 */ 717 flags |= MAP_STACK; 718 #endif 719 720 ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0); 721 if (ptr == MAP_FAILED) { 722 perror("failed to allocate memory for stack"); 723 abort(); 724 } 725 726 #if defined(HOST_IA64) 727 /* separate register stack */ 728 guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz); 729 #elif defined(HOST_HPPA) 730 /* stack grows up */ 731 guardpage = ptr + *sz - pagesz; 732 #else 733 /* stack grows down */ 734 guardpage = ptr; 735 #endif 736 if (mprotect(guardpage, pagesz, PROT_NONE) != 0) { 737 perror("failed to set up stack guard page"); 738 abort(); 739 } 740 741 #ifdef CONFIG_DEBUG_STACK_USAGE 742 for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) { 743 *(uint32_t *)ptr2 = 0xdeadbeaf; 744 } 745 #endif 746 747 return ptr; 748 } 749 750 #ifdef CONFIG_DEBUG_STACK_USAGE 751 static __thread unsigned int max_stack_usage; 752 #endif 753 754 void qemu_free_stack(void *stack, size_t sz) 755 { 756 #ifdef CONFIG_DEBUG_STACK_USAGE 757 unsigned int usage; 758 void *ptr; 759 760 for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz; 761 ptr += sizeof(uint32_t)) { 762 if (*(uint32_t *)ptr != 0xdeadbeaf) { 763 break; 764 } 765 } 766 usage = sz - (uintptr_t) (ptr - stack); 767 if (usage > max_stack_usage) { 768 error_report("thread %d max stack usage increased from %u to %u", 769 qemu_get_thread_id(), max_stack_usage, usage); 770 max_stack_usage = usage; 771 } 772 #endif 773 774 munmap(stack, sz); 775 } 776 777 /* 778 * Disable CFI checks. 779 * We are going to call a signal hander directly. Such handler may or may not 780 * have been defined in our binary, so there's no guarantee that the pointer 781 * used to set the handler is a cfi-valid pointer. Since the handlers are 782 * stored in kernel memory, changing the handler to an attacker-defined 783 * function requires being able to call a sigaction() syscall, 784 * which is not as easy as overwriting a pointer in memory. 785 */ 786 QEMU_DISABLE_CFI 787 void sigaction_invoke(struct sigaction *action, 788 struct qemu_signalfd_siginfo *info) 789 { 790 siginfo_t si = {}; 791 si.si_signo = info->ssi_signo; 792 si.si_errno = info->ssi_errno; 793 si.si_code = info->ssi_code; 794 795 /* Convert the minimal set of fields defined by POSIX. 796 * Positive si_code values are reserved for kernel-generated 797 * signals, where the valid siginfo fields are determined by 798 * the signal number. But according to POSIX, it is unspecified 799 * whether SI_USER and SI_QUEUE have values less than or equal to 800 * zero. 801 */ 802 if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE || 803 info->ssi_code <= 0) { 804 /* SIGTERM, etc. */ 805 si.si_pid = info->ssi_pid; 806 si.si_uid = info->ssi_uid; 807 } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE || 808 info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) { 809 si.si_addr = (void *)(uintptr_t)info->ssi_addr; 810 } else if (info->ssi_signo == SIGCHLD) { 811 si.si_pid = info->ssi_pid; 812 si.si_status = info->ssi_status; 813 si.si_uid = info->ssi_uid; 814 } 815 action->sa_sigaction(info->ssi_signo, &si, NULL); 816 } 817 818 #ifndef HOST_NAME_MAX 819 # ifdef _POSIX_HOST_NAME_MAX 820 # define HOST_NAME_MAX _POSIX_HOST_NAME_MAX 821 # else 822 # define HOST_NAME_MAX 255 823 # endif 824 #endif 825 826 char *qemu_get_host_name(Error **errp) 827 { 828 long len = -1; 829 g_autofree char *hostname = NULL; 830 831 #ifdef _SC_HOST_NAME_MAX 832 len = sysconf(_SC_HOST_NAME_MAX); 833 #endif /* _SC_HOST_NAME_MAX */ 834 835 if (len < 0) { 836 len = HOST_NAME_MAX; 837 } 838 839 /* Unfortunately, gethostname() below does not guarantee a 840 * NULL terminated string. Therefore, allocate one byte more 841 * to be sure. */ 842 hostname = g_new0(char, len + 1); 843 844 if (gethostname(hostname, len) < 0) { 845 error_setg_errno(errp, errno, 846 "cannot get hostname"); 847 return NULL; 848 } 849 850 return g_steal_pointer(&hostname); 851 } 852 853 size_t qemu_get_host_physmem(void) 854 { 855 #ifdef _SC_PHYS_PAGES 856 long pages = sysconf(_SC_PHYS_PAGES); 857 if (pages > 0) { 858 if (pages > SIZE_MAX / qemu_real_host_page_size) { 859 return SIZE_MAX; 860 } else { 861 return pages * qemu_real_host_page_size; 862 } 863 } 864 #endif 865 return 0; 866 } 867