util/oslib-posix.c

   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     } else {
 205         g_assert(is_power_of_2(alignment));
 206     }
 207
 208 #if defined(CONFIG_POSIX_MEMALIGN)
 209     int ret;
 210     ret = posix_memalign(&ptr, alignment, size);
 211     if (ret != 0) {
 212         errno = ret;
 213         ptr = NULL;
 214     }
 215 #elif defined(CONFIG_BSD)
 216     ptr = valloc(size);
 217 #else
 218     ptr = memalign(alignment, size);
 219 #endif
 220     trace_qemu_memalign(alignment, size, ptr);
 221     return ptr;
 222 }
 223
 224 void *qemu_memalign(size_t alignment, size_t size)
 225 {
 226     return qemu_oom_check(qemu_try_memalign(alignment, size));
 227 }
 228
 229 /* alloc shared memory pages */
 230 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared,
 231                           bool noreserve)
 232 {
 233     const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) |
 234                                     (noreserve ? QEMU_MAP_NORESERVE : 0);
 235     size_t align = QEMU_VMALLOC_ALIGN;
 236     void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0);
 237
 238     if (ptr == MAP_FAILED) {
 239         return NULL;
 240     }
 241
 242     if (alignment) {
 243         *alignment = align;
 244     }
 245
 246     trace_qemu_anon_ram_alloc(size, ptr);
 247     return ptr;
 248 }
 249
 250 void qemu_vfree(void *ptr)
 251 {
 252     trace_qemu_vfree(ptr);
 253     free(ptr);
 254 }
 255
 256 void qemu_anon_ram_free(void *ptr, size_t size)
 257 {
 258     trace_qemu_anon_ram_free(ptr, size);
 259     qemu_ram_munmap(-1, ptr, size);
 260 }
 261
 262 void qemu_set_block(int fd)
 263 {
 264     int f;
 265     f = fcntl(fd, F_GETFL);
 266     assert(f != -1);
 267     f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 268     assert(f != -1);
 269 }
 270
 271 int qemu_try_set_nonblock(int fd)
 272 {
 273     int f;
 274     f = fcntl(fd, F_GETFL);
 275     if (f == -1) {
 276         return -errno;
 277     }
 278     if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) {
 279         return -errno;
 280     }
 281     return 0;
 282 }
 283
 284 void qemu_set_nonblock(int fd)
 285 {
 286     int f;
 287     f = qemu_try_set_nonblock(fd);
 288     assert(f == 0);
 289 }
 290
 291 int socket_set_fast_reuse(int fd)
 292 {
 293     int val = 1, ret;
 294
 295     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 296                      (const char *)&val, sizeof(val));
 297
 298     assert(ret == 0);
 299
 300     return ret;
 301 }
 302
 303 void qemu_set_cloexec(int fd)
 304 {
 305     int f;
 306     f = fcntl(fd, F_GETFD);
 307     assert(f != -1);
 308     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 309     assert(f != -1);
 310 }
 311
 312 /*
 313  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 314  */
 315 int qemu_pipe(int pipefd[2])
 316 {
 317     int ret;
 318
 319 #ifdef CONFIG_PIPE2
 320     ret = pipe2(pipefd, O_CLOEXEC);
 321     if (ret != -1 || errno != ENOSYS) {
 322         return ret;
 323     }
 324 #endif
 325     ret = pipe(pipefd);
 326     if (ret == 0) {
 327         qemu_set_cloexec(pipefd[0]);
 328         qemu_set_cloexec(pipefd[1]);
 329     }
 330
 331     return ret;
 332 }
 333
 334 char *
 335 qemu_get_local_state_pathname(const char *relative_pathname)
 336 {
 337     g_autofree char *dir = g_strdup_printf("%s/%s",
 338                                            CONFIG_QEMU_LOCALSTATEDIR,
 339                                            relative_pathname);
 340     return get_relocated_path(dir);
 341 }
 342
 343 void qemu_set_tty_echo(int fd, bool echo)
 344 {
 345     struct termios tty;
 346
 347     tcgetattr(fd, &tty);
 348
 349     if (echo) {
 350         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 351     } else {
 352         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 353     }
 354
 355     tcsetattr(fd, TCSANOW, &tty);
 356 }
 357
 358 static const char *exec_dir;
 359
 360 void qemu_init_exec_dir(const char *argv0)
 361 {
 362     char *p = NULL;
 363     char buf[PATH_MAX];
 364
 365     if (exec_dir) {
 366         return;
 367     }
 368
 369 #if defined(__linux__)
 370     {
 371         int len;
 372         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 373         if (len > 0) {
 374             buf[len] = 0;
 375             p = buf;
 376         }
 377     }
 378 #elif defined(__FreeBSD__) \
 379       || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
 380     {
 381 #if defined(__FreeBSD__)
 382         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 383 #else
 384         static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
 385 #endif
 386         size_t len = sizeof(buf) - 1;
 387
 388         *buf = '\0';
 389         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 390             *buf) {
 391             buf[sizeof(buf) - 1] = '\0';
 392             p = buf;
 393         }
 394     }
 395 #elif defined(__APPLE__)
 396     {
 397         char fpath[PATH_MAX];
 398         uint32_t len = sizeof(fpath);
 399         if (_NSGetExecutablePath(fpath, &len) == 0) {
 400             p = realpath(fpath, buf);
 401             if (!p) {
 402                 return;
 403             }
 404         }
 405     }
 406 #elif defined(__HAIKU__)
 407     {
 408         image_info ii;
 409         int32_t c = 0;
 410
 411         *buf = '\0';
 412         while (get_next_image_info(0, &c, &ii) == B_OK) {
 413             if (ii.type == B_APP_IMAGE) {
 414                 strncpy(buf, ii.name, sizeof(buf));
 415                 buf[sizeof(buf) - 1] = 0;
 416                 p = buf;
 417                 break;
 418             }
 419         }
 420     }
 421 #endif
 422     /* If we don't have any way of figuring out the actual executable
 423        location then try argv[0].  */
 424     if (!p && argv0) {
 425         p = realpath(argv0, buf);
 426     }
 427     if (p) {
 428         exec_dir = g_path_get_dirname(p);
 429     } else {
 430         exec_dir = CONFIG_BINDIR;
 431     }
 432 }
 433
 434 const char *qemu_get_exec_dir(void)
 435 {
 436     return exec_dir;
 437 }
 438
 439 static void sigbus_handler(int signal)
 440 {
 441     int i;
 442     if (memset_thread) {
 443         for (i = 0; i < memset_num_threads; i++) {
 444             if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
 445                 siglongjmp(memset_thread[i].env, 1);
 446             }
 447         }
 448     }
 449 }
 450
 451 static void *do_touch_pages(void *arg)
 452 {
 453     MemsetThread *memset_args = (MemsetThread *)arg;
 454     sigset_t set, oldset;
 455
 456     /*
 457      * On Linux, the page faults from the loop below can cause mmap_sem
 458      * contention with allocation of the thread stacks.  Do not start
 459      * clearing until all threads have been created.
 460      */
 461     qemu_mutex_lock(&page_mutex);
 462     while(!threads_created_flag){
 463         qemu_cond_wait(&page_cond, &page_mutex);
 464     }
 465     qemu_mutex_unlock(&page_mutex);
 466
 467     /* unblock SIGBUS */
 468     sigemptyset(&set);
 469     sigaddset(&set, SIGBUS);
 470     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 471
 472     if (sigsetjmp(memset_args->env, 1)) {
 473         memset_thread_failed = true;
 474     } else {
 475         char *addr = memset_args->addr;
 476         size_t numpages = memset_args->numpages;
 477         size_t hpagesize = memset_args->hpagesize;
 478         size_t i;
 479         for (i = 0; i < numpages; i++) {
 480             /*
 481              * Read & write back the same value, so we don't
 482              * corrupt existing user/app data that might be
 483              * stored.
 484              *
 485              * 'volatile' to stop compiler optimizing this away
 486              * to a no-op
 487              *
 488              * TODO: get a better solution from kernel so we
 489              * don't need to write at all so we don't cause
 490              * wear on the storage backing the region...
 491              */
 492             *(volatile char *)addr = *addr;
 493             addr += hpagesize;
 494         }
 495     }
 496     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 497     return NULL;
 498 }
 499
 500 static inline int get_memset_num_threads(int smp_cpus)
 501 {
 502     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 503     int ret = 1;
 504
 505     if (host_procs > 0) {
 506         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
 507     }
 508     /* In case sysconf() fails, we fall back to single threaded */
 509     return ret;
 510 }
 511
 512 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 513                             int smp_cpus)
 514 {
 515     static gsize initialized = 0;
 516     size_t numpages_per_thread, leftover;
 517     char *addr = area;
 518     int i = 0;
 519
 520     if (g_once_init_enter(&initialized)) {
 521         qemu_mutex_init(&page_mutex);
 522         qemu_cond_init(&page_cond);
 523         g_once_init_leave(&initialized, 1);
 524     }
 525
 526     memset_thread_failed = false;
 527     threads_created_flag = false;
 528     memset_num_threads = get_memset_num_threads(smp_cpus);
 529     memset_thread = g_new0(MemsetThread, memset_num_threads);
 530     numpages_per_thread = numpages / memset_num_threads;
 531     leftover = numpages % memset_num_threads;
 532     for (i = 0; i < memset_num_threads; i++) {
 533         memset_thread[i].addr = addr;
 534         memset_thread[i].numpages = numpages_per_thread + (i < leftover);
 535         memset_thread[i].hpagesize = hpagesize;
 536         qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
 537                            do_touch_pages, &memset_thread[i],
 538                            QEMU_THREAD_JOINABLE);
 539         addr += memset_thread[i].numpages * hpagesize;
 540     }
 541
 542     qemu_mutex_lock(&page_mutex);
 543     threads_created_flag = true;
 544     qemu_cond_broadcast(&page_cond);
 545     qemu_mutex_unlock(&page_mutex);
 546
 547     for (i = 0; i < memset_num_threads; i++) {
 548         qemu_thread_join(&memset_thread[i].pgthread);
 549     }
 550     g_free(memset_thread);
 551     memset_thread = NULL;
 552
 553     return memset_thread_failed;
 554 }
 555
 556 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
 557                      Error **errp)
 558 {
 559     int ret;
 560     struct sigaction act, oldact;
 561     size_t hpagesize = qemu_fd_getpagesize(fd);
 562     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 563
 564     memset(&act, 0, sizeof(act));
 565     act.sa_handler = &sigbus_handler;
 566     act.sa_flags = 0;
 567
 568     ret = sigaction(SIGBUS, &act, &oldact);
 569     if (ret) {
 570         error_setg_errno(errp, errno,
 571             "os_mem_prealloc: failed to install signal handler");
 572         return;
 573     }
 574
 575     /* touch pages simultaneously */
 576     if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
 577         error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
 578             "pages available to allocate guest RAM");
 579     }
 580
 581     ret = sigaction(SIGBUS, &oldact, NULL);
 582     if (ret) {
 583         /* Terminate QEMU since it can't recover from error */
 584         perror("os_mem_prealloc: failed to reinstall signal handler");
 585         exit(1);
 586     }
 587 }
 588
 589 char *qemu_get_pid_name(pid_t pid)
 590 {
 591     char *name = NULL;
 592
 593 #if defined(__FreeBSD__)
 594     /* BSDs don't have /proc, but they provide a nice substitute */
 595     struct kinfo_proc *proc = kinfo_getproc(pid);
 596
 597     if (proc) {
 598         name = g_strdup(proc->ki_comm);
 599         free(proc);
 600     }
 601 #else
 602     /* Assume a system with reasonable procfs */
 603     char *pid_path;
 604     size_t len;
 605
 606     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 607     g_file_get_contents(pid_path, &name, &len, NULL);
 608     g_free(pid_path);
 609 #endif
 610
 611     return name;
 612 }
 613
 614
 615 pid_t qemu_fork(Error **errp)
 616 {
 617     sigset_t oldmask, newmask;
 618     struct sigaction sig_action;
 619     int saved_errno;
 620     pid_t pid;
 621
 622     /*
 623      * Need to block signals now, so that child process can safely
 624      * kill off caller's signal handlers without a race.
 625      */
 626     sigfillset(&newmask);
 627     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 628         error_setg_errno(errp, errno,
 629                          "cannot block signals");
 630         return -1;
 631     }
 632
 633     pid = fork();
 634     saved_errno = errno;
 635
 636     if (pid < 0) {
 637         /* attempt to restore signal mask, but ignore failure, to
 638          * avoid obscuring the fork failure */
 639         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 640         error_setg_errno(errp, saved_errno,
 641                          "cannot fork child process");
 642         errno = saved_errno;
 643         return -1;
 644     } else if (pid) {
 645         /* parent process */
 646
 647         /* Restore our original signal mask now that the child is
 648          * safely running. Only documented failures are EFAULT (not
 649          * possible, since we are using just-grabbed mask) or EINVAL
 650          * (not possible, since we are using correct arguments).  */
 651         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 652     } else {
 653         /* child process */
 654         size_t i;
 655
 656         /* Clear out all signal handlers from parent so nothing
 657          * unexpected can happen in our child once we unblock
 658          * signals */
 659         sig_action.sa_handler = SIG_DFL;
 660         sig_action.sa_flags = 0;
 661         sigemptyset(&sig_action.sa_mask);
 662
 663         for (i = 1; i < NSIG; i++) {
 664             /* Only possible errors are EFAULT or EINVAL The former
 665              * won't happen, the latter we expect, so no need to check
 666              * return value */
 667             (void)sigaction(i, &sig_action, NULL);
 668         }
 669
 670         /* Unmask all signals in child, since we've no idea what the
 671          * caller's done with their signal mask and don't want to
 672          * propagate that to children */
 673         sigemptyset(&newmask);
 674         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 675             Error *local_err = NULL;
 676             error_setg_errno(&local_err, errno,
 677                              "cannot unblock signals");
 678             error_report_err(local_err);
 679             _exit(1);
 680         }
 681     }
 682     return pid;
 683 }
 684
 685 void *qemu_alloc_stack(size_t *sz)
 686 {
 687     void *ptr, *guardpage;
 688     int flags;
 689 #ifdef CONFIG_DEBUG_STACK_USAGE
 690     void *ptr2;
 691 #endif
 692     size_t pagesz = qemu_real_host_page_size;
 693 #ifdef _SC_THREAD_STACK_MIN
 694     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 695     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 696     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 697 #endif
 698     /* adjust stack size to a multiple of the page size */
 699     *sz = ROUND_UP(*sz, pagesz);
 700     /* allocate one extra page for the guard page */
 701     *sz += pagesz;
 702
 703     flags = MAP_PRIVATE | MAP_ANONYMOUS;
 704 #if defined(MAP_STACK) && defined(__OpenBSD__)
 705     /* Only enable MAP_STACK on OpenBSD. Other OS's such as
 706      * Linux/FreeBSD/NetBSD have a flag with the same name
 707      * but have differing functionality. OpenBSD will SEGV
 708      * if it spots execution with a stack pointer pointing
 709      * at memory that was not allocated with MAP_STACK.
 710      */
 711     flags |= MAP_STACK;
 712 #endif
 713
 714     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
 715     if (ptr == MAP_FAILED) {
 716         perror("failed to allocate memory for stack");
 717         abort();
 718     }
 719
 720 #if defined(HOST_IA64)
 721     /* separate register stack */
 722     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
 723 #elif defined(HOST_HPPA)
 724     /* stack grows up */
 725     guardpage = ptr + *sz - pagesz;
 726 #else
 727     /* stack grows down */
 728     guardpage = ptr;
 729 #endif
 730     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
 731         perror("failed to set up stack guard page");
 732         abort();
 733     }
 734
 735 #ifdef CONFIG_DEBUG_STACK_USAGE
 736     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 737         *(uint32_t *)ptr2 = 0xdeadbeaf;
 738     }
 739 #endif
 740
 741     return ptr;
 742 }
 743
 744 #ifdef CONFIG_DEBUG_STACK_USAGE
 745 static __thread unsigned int max_stack_usage;
 746 #endif
 747
 748 void qemu_free_stack(void *stack, size_t sz)
 749 {
 750 #ifdef CONFIG_DEBUG_STACK_USAGE
 751     unsigned int usage;
 752     void *ptr;
 753
 754     for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz;
 755          ptr += sizeof(uint32_t)) {
 756         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 757             break;
 758         }
 759     }
 760     usage = sz - (uintptr_t) (ptr - stack);
 761     if (usage > max_stack_usage) {
 762         error_report("thread %d max stack usage increased from %u to %u",
 763                      qemu_get_thread_id(), max_stack_usage, usage);
 764         max_stack_usage = usage;
 765     }
 766 #endif
 767
 768     munmap(stack, sz);
 769 }
 770
 771 /*
 772  * Disable CFI checks.
 773  * We are going to call a signal hander directly. Such handler may or may not
 774  * have been defined in our binary, so there's no guarantee that the pointer
 775  * used to set the handler is a cfi-valid pointer. Since the handlers are
 776  * stored in kernel memory, changing the handler to an attacker-defined
 777  * function requires being able to call a sigaction() syscall,
 778  * which is not as easy as overwriting a pointer in memory.
 779  */
 780 QEMU_DISABLE_CFI
 781 void sigaction_invoke(struct sigaction *action,
 782                       struct qemu_signalfd_siginfo *info)
 783 {
 784     siginfo_t si = {};
 785     si.si_signo = info->ssi_signo;
 786     si.si_errno = info->ssi_errno;
 787     si.si_code = info->ssi_code;
 788
 789     /* Convert the minimal set of fields defined by POSIX.
 790      * Positive si_code values are reserved for kernel-generated
 791      * signals, where the valid siginfo fields are determined by
 792      * the signal number.  But according to POSIX, it is unspecified
 793      * whether SI_USER and SI_QUEUE have values less than or equal to
 794      * zero.
 795      */
 796     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 797         info->ssi_code <= 0) {
 798         /* SIGTERM, etc.  */
 799         si.si_pid = info->ssi_pid;
 800         si.si_uid = info->ssi_uid;
 801     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 802                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 803         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 804     } else if (info->ssi_signo == SIGCHLD) {
 805         si.si_pid = info->ssi_pid;
 806         si.si_status = info->ssi_status;
 807         si.si_uid = info->ssi_uid;
 808     }
 809     action->sa_sigaction(info->ssi_signo, &si, NULL);
 810 }
 811
 812 #ifndef HOST_NAME_MAX
 813 # ifdef _POSIX_HOST_NAME_MAX
 814 #  define HOST_NAME_MAX _POSIX_HOST_NAME_MAX
 815 # else
 816 #  define HOST_NAME_MAX 255
 817 # endif
 818 #endif
 819
 820 char *qemu_get_host_name(Error **errp)
 821 {
 822     long len = -1;
 823     g_autofree char *hostname = NULL;
 824
 825 #ifdef _SC_HOST_NAME_MAX
 826     len = sysconf(_SC_HOST_NAME_MAX);
 827 #endif /* _SC_HOST_NAME_MAX */
 828
 829     if (len < 0) {
 830         len = HOST_NAME_MAX;
 831     }
 832
 833     /* Unfortunately, gethostname() below does not guarantee a
 834      * NULL terminated string. Therefore, allocate one byte more
 835      * to be sure. */
 836     hostname = g_new0(char, len + 1);
 837
 838     if (gethostname(hostname, len) < 0) {
 839         error_setg_errno(errp, errno,
 840                          "cannot get hostname");
 841         return NULL;
 842     }
 843
 844     return g_steal_pointer(&hostname);
 845 }
 846
 847 size_t qemu_get_host_physmem(void)
 848 {
 849 #ifdef _SC_PHYS_PAGES
 850     long pages = sysconf(_SC_PHYS_PAGES);
 851     if (pages > 0) {
 852         if (pages > SIZE_MAX / qemu_real_host_page_size) {
 853             return SIZE_MAX;
 854         } else {
 855             return pages * qemu_real_host_page_size;
 856         }
 857     }
 858 #endif
 859     return 0;
 860 }