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