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/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_create(path, O_WRONLY, S_IRUSR | S_IWUSR, errp);
 142         if (fd == -1) {
 143             return false;
 144         }
 145
 146         if (fstat(fd, &b) < 0) {
 147             error_setg_errno(errp, errno, "Cannot stat file");
 148             goto fail_close;
 149         }
 150
 151         if (fcntl(fd, F_SETLK, &lock)) {
 152             error_setg_errno(errp, errno, "Cannot lock pid file");
 153             goto fail_close;
 154         }
 155
 156         /*
 157          * Now make sure the path we locked is the same one that now
 158          * exists on the filesystem.
 159          */
 160         if (stat(path, &a) < 0) {
 161             /*
 162              * PID file disappeared, someone else must be racing with
 163              * us, so try again.
 164              */
 165             close(fd);
 166             continue;
 167         }
 168
 169         if (a.st_ino == b.st_ino) {
 170             break;
 171         }
 172
 173         /*
 174          * PID file was recreated, someone else must be racing with
 175          * us, so try again.
 176          */
 177         close(fd);
 178     }
 179
 180     if (ftruncate(fd, 0) < 0) {
 181         error_setg_errno(errp, errno, "Failed to truncate pid file");
 182         goto fail_unlink;
 183     }
 184
 185     snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
 186     if (qemu_write_full(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
 187         error_setg(errp, "Failed to write pid file");
 188         goto fail_unlink;
 189     }
 190
 191     return true;
 192
 193 fail_unlink:
 194     unlink(path);
 195 fail_close:
 196     close(fd);
 197     return false;
 198 }
 199
 200 /* alloc shared memory pages */
 201 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared,
 202                           bool noreserve)
 203 {
 204     const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) |
 205                                     (noreserve ? QEMU_MAP_NORESERVE : 0);
 206     size_t align = QEMU_VMALLOC_ALIGN;
 207     void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0);
 208
 209     if (ptr == MAP_FAILED) {
 210         return NULL;
 211     }
 212
 213     if (alignment) {
 214         *alignment = align;
 215     }
 216
 217     trace_qemu_anon_ram_alloc(size, ptr);
 218     return ptr;
 219 }
 220
 221 void qemu_anon_ram_free(void *ptr, size_t size)
 222 {
 223     trace_qemu_anon_ram_free(ptr, size);
 224     qemu_ram_munmap(-1, ptr, size);
 225 }
 226
 227 void qemu_socket_set_block(int fd)
 228 {
 229     g_unix_set_fd_nonblocking(fd, false, NULL);
 230 }
 231
 232 int qemu_socket_try_set_nonblock(int fd)
 233 {
 234     return g_unix_set_fd_nonblocking(fd, true, NULL) ? 0 : -errno;
 235 }
 236
 237 void qemu_socket_set_nonblock(int fd)
 238 {
 239     int f;
 240     f = qemu_socket_try_set_nonblock(fd);
 241     assert(f == 0);
 242 }
 243
 244 int socket_set_fast_reuse(int fd)
 245 {
 246     int val = 1, ret;
 247
 248     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 249                      (const char *)&val, sizeof(val));
 250
 251     assert(ret == 0);
 252
 253     return ret;
 254 }
 255
 256 void qemu_set_cloexec(int fd)
 257 {
 258     int f;
 259     f = fcntl(fd, F_GETFD);
 260     assert(f != -1);
 261     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 262     assert(f != -1);
 263 }
 264
 265 char *
 266 qemu_get_local_state_dir(void)
 267 {
 268     return get_relocated_path(CONFIG_QEMU_LOCALSTATEDIR);
 269 }
 270
 271 void qemu_set_tty_echo(int fd, bool echo)
 272 {
 273     struct termios tty;
 274
 275     tcgetattr(fd, &tty);
 276
 277     if (echo) {
 278         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 279     } else {
 280         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 281     }
 282
 283     tcsetattr(fd, TCSANOW, &tty);
 284 }
 285
 286 static const char *exec_dir;
 287
 288 void qemu_init_exec_dir(const char *argv0)
 289 {
 290     char *p = NULL;
 291     char buf[PATH_MAX];
 292
 293     if (exec_dir) {
 294         return;
 295     }
 296
 297 #if defined(__linux__)
 298     {
 299         int len;
 300         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 301         if (len > 0) {
 302             buf[len] = 0;
 303             p = buf;
 304         }
 305     }
 306 #elif defined(__FreeBSD__) \
 307       || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
 308     {
 309 #if defined(__FreeBSD__)
 310         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 311 #else
 312         static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
 313 #endif
 314         size_t len = sizeof(buf) - 1;
 315
 316         *buf = '\0';
 317         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 318             *buf) {
 319             buf[sizeof(buf) - 1] = '\0';
 320             p = buf;
 321         }
 322     }
 323 #elif defined(__APPLE__)
 324     {
 325         char fpath[PATH_MAX];
 326         uint32_t len = sizeof(fpath);
 327         if (_NSGetExecutablePath(fpath, &len) == 0) {
 328             p = realpath(fpath, buf);
 329             if (!p) {
 330                 return;
 331             }
 332         }
 333     }
 334 #elif defined(__HAIKU__)
 335     {
 336         image_info ii;
 337         int32_t c = 0;
 338
 339         *buf = '\0';
 340         while (get_next_image_info(0, &c, &ii) == B_OK) {
 341             if (ii.type == B_APP_IMAGE) {
 342                 strncpy(buf, ii.name, sizeof(buf));
 343                 buf[sizeof(buf) - 1] = 0;
 344                 p = buf;
 345                 break;
 346             }
 347         }
 348     }
 349 #endif
 350     /* If we don't have any way of figuring out the actual executable
 351        location then try argv[0].  */
 352     if (!p && argv0) {
 353         p = realpath(argv0, buf);
 354     }
 355     if (p) {
 356         exec_dir = g_path_get_dirname(p);
 357     } else {
 358         exec_dir = CONFIG_BINDIR;
 359     }
 360 }
 361
 362 const char *qemu_get_exec_dir(void)
 363 {
 364     return exec_dir;
 365 }
 366
 367 #ifdef CONFIG_LINUX
 368 static void sigbus_handler(int signal, siginfo_t *siginfo, void *ctx)
 369 #else /* CONFIG_LINUX */
 370 static void sigbus_handler(int signal)
 371 #endif /* CONFIG_LINUX */
 372 {
 373     int i;
 374
 375     if (sigbus_memset_context) {
 376         for (i = 0; i < sigbus_memset_context->num_threads; i++) {
 377             MemsetThread *thread = &sigbus_memset_context->threads[i];
 378
 379             if (qemu_thread_is_self(&thread->pgthread)) {
 380                 siglongjmp(thread->env, 1);
 381             }
 382         }
 383     }
 384
 385 #ifdef CONFIG_LINUX
 386     /*
 387      * We assume that the MCE SIGBUS handler could have been registered. We
 388      * should never receive BUS_MCEERR_AO on any of our threads, but only on
 389      * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not
 390      * receive BUS_MCEERR_AR triggered by action of other threads on one of
 391      * our threads. So, no need to check for unrelated SIGBUS when seeing one
 392      * for our threads.
 393      *
 394      * We will forward to the MCE handler, which will either handle the SIGBUS
 395      * or reinstall the default SIGBUS handler and reraise the SIGBUS. The
 396      * default SIGBUS handler will crash the process, so we don't care.
 397      */
 398     if (sigbus_oldact.sa_flags & SA_SIGINFO) {
 399         sigbus_oldact.sa_sigaction(signal, siginfo, ctx);
 400         return;
 401     }
 402 #endif /* CONFIG_LINUX */
 403     warn_report("os_mem_prealloc: unrelated SIGBUS detected and ignored");
 404 }
 405
 406 static void *do_touch_pages(void *arg)
 407 {
 408     MemsetThread *memset_args = (MemsetThread *)arg;
 409     sigset_t set, oldset;
 410     int ret = 0;
 411
 412     /*
 413      * On Linux, the page faults from the loop below can cause mmap_sem
 414      * contention with allocation of the thread stacks.  Do not start
 415      * clearing until all threads have been created.
 416      */
 417     qemu_mutex_lock(&page_mutex);
 418     while (!memset_args->context->all_threads_created) {
 419         qemu_cond_wait(&page_cond, &page_mutex);
 420     }
 421     qemu_mutex_unlock(&page_mutex);
 422
 423     /* unblock SIGBUS */
 424     sigemptyset(&set);
 425     sigaddset(&set, SIGBUS);
 426     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 427
 428     if (sigsetjmp(memset_args->env, 1)) {
 429         ret = -EFAULT;
 430     } else {
 431         char *addr = memset_args->addr;
 432         size_t numpages = memset_args->numpages;
 433         size_t hpagesize = memset_args->hpagesize;
 434         size_t i;
 435         for (i = 0; i < numpages; i++) {
 436             /*
 437              * Read & write back the same value, so we don't
 438              * corrupt existing user/app data that might be
 439              * stored.
 440              *
 441              * 'volatile' to stop compiler optimizing this away
 442              * to a no-op
 443              */
 444             *(volatile char *)addr = *addr;
 445             addr += hpagesize;
 446         }
 447     }
 448     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 449     return (void *)(uintptr_t)ret;
 450 }
 451
 452 static void *do_madv_populate_write_pages(void *arg)
 453 {
 454     MemsetThread *memset_args = (MemsetThread *)arg;
 455     const size_t size = memset_args->numpages * memset_args->hpagesize;
 456     char * const addr = memset_args->addr;
 457     int ret = 0;
 458
 459     /* See do_touch_pages(). */
 460     qemu_mutex_lock(&page_mutex);
 461     while (!memset_args->context->all_threads_created) {
 462         qemu_cond_wait(&page_cond, &page_mutex);
 463     }
 464     qemu_mutex_unlock(&page_mutex);
 465
 466     if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) {
 467         ret = -errno;
 468     }
 469     return (void *)(uintptr_t)ret;
 470 }
 471
 472 static inline int get_memset_num_threads(size_t hpagesize, size_t numpages,
 473                                          int smp_cpus)
 474 {
 475     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 476     int ret = 1;
 477
 478     if (host_procs > 0) {
 479         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
 480     }
 481
 482     /* Especially with gigantic pages, don't create more threads than pages. */
 483     ret = MIN(ret, numpages);
 484     /* Don't start threads to prealloc comparatively little memory. */
 485     ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB)));
 486
 487     /* In case sysconf() fails, we fall back to single threaded */
 488     return ret;
 489 }
 490
 491 static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 492                            int smp_cpus, bool use_madv_populate_write)
 493 {
 494     static gsize initialized = 0;
 495     MemsetContext context = {
 496         .num_threads = get_memset_num_threads(hpagesize, numpages, smp_cpus),
 497     };
 498     size_t numpages_per_thread, leftover;
 499     void *(*touch_fn)(void *);
 500     int ret = 0, i = 0;
 501     char *addr = area;
 502
 503     if (g_once_init_enter(&initialized)) {
 504         qemu_mutex_init(&page_mutex);
 505         qemu_cond_init(&page_cond);
 506         g_once_init_leave(&initialized, 1);
 507     }
 508
 509     if (use_madv_populate_write) {
 510         /* Avoid creating a single thread for MADV_POPULATE_WRITE */
 511         if (context.num_threads == 1) {
 512             if (qemu_madvise(area, hpagesize * numpages,
 513                              QEMU_MADV_POPULATE_WRITE)) {
 514                 return -errno;
 515             }
 516             return 0;
 517         }
 518         touch_fn = do_madv_populate_write_pages;
 519     } else {
 520         touch_fn = do_touch_pages;
 521     }
 522
 523     context.threads = g_new0(MemsetThread, context.num_threads);
 524     numpages_per_thread = numpages / context.num_threads;
 525     leftover = numpages % context.num_threads;
 526     for (i = 0; i < context.num_threads; i++) {
 527         context.threads[i].addr = addr;
 528         context.threads[i].numpages = numpages_per_thread + (i < leftover);
 529         context.threads[i].hpagesize = hpagesize;
 530         context.threads[i].context = &context;
 531         qemu_thread_create(&context.threads[i].pgthread, "touch_pages",
 532                            touch_fn, &context.threads[i],
 533                            QEMU_THREAD_JOINABLE);
 534         addr += context.threads[i].numpages * hpagesize;
 535     }
 536
 537     if (!use_madv_populate_write) {
 538         sigbus_memset_context = &context;
 539     }
 540
 541     qemu_mutex_lock(&page_mutex);
 542     context.all_threads_created = true;
 543     qemu_cond_broadcast(&page_cond);
 544     qemu_mutex_unlock(&page_mutex);
 545
 546     for (i = 0; i < context.num_threads; i++) {
 547         int tmp = (uintptr_t)qemu_thread_join(&context.threads[i].pgthread);
 548
 549         if (tmp) {
 550             ret = tmp;
 551         }
 552     }
 553
 554     if (!use_madv_populate_write) {
 555         sigbus_memset_context = NULL;
 556     }
 557     g_free(context.threads);
 558
 559     return ret;
 560 }
 561
 562 static bool madv_populate_write_possible(char *area, size_t pagesize)
 563 {
 564     return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) ||
 565            errno != EINVAL;
 566 }
 567
 568 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
 569                      Error **errp)
 570 {
 571     static gsize initialized;
 572     int ret;
 573     size_t hpagesize = qemu_fd_getpagesize(fd);
 574     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 575     bool use_madv_populate_write;
 576     struct sigaction act;
 577
 578     /*
 579      * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for
 580      * some special mappings, such as mapping /dev/mem.
 581      */
 582     use_madv_populate_write = madv_populate_write_possible(area, hpagesize);
 583
 584     if (!use_madv_populate_write) {
 585         if (g_once_init_enter(&initialized)) {
 586             qemu_mutex_init(&sigbus_mutex);
 587             g_once_init_leave(&initialized, 1);
 588         }
 589
 590         qemu_mutex_lock(&sigbus_mutex);
 591         memset(&act, 0, sizeof(act));
 592 #ifdef CONFIG_LINUX
 593         act.sa_sigaction = &sigbus_handler;
 594         act.sa_flags = SA_SIGINFO;
 595 #else /* CONFIG_LINUX */
 596         act.sa_handler = &sigbus_handler;
 597         act.sa_flags = 0;
 598 #endif /* CONFIG_LINUX */
 599
 600         ret = sigaction(SIGBUS, &act, &sigbus_oldact);
 601         if (ret) {
 602             qemu_mutex_unlock(&sigbus_mutex);
 603             error_setg_errno(errp, errno,
 604                 "os_mem_prealloc: failed to install signal handler");
 605             return;
 606         }
 607     }
 608
 609     /* touch pages simultaneously */
 610     ret = touch_all_pages(area, hpagesize, numpages, smp_cpus,
 611                           use_madv_populate_write);
 612     if (ret) {
 613         error_setg_errno(errp, -ret,
 614                          "os_mem_prealloc: preallocating memory failed");
 615     }
 616
 617     if (!use_madv_populate_write) {
 618         ret = sigaction(SIGBUS, &sigbus_oldact, NULL);
 619         if (ret) {
 620             /* Terminate QEMU since it can't recover from error */
 621             perror("os_mem_prealloc: failed to reinstall signal handler");
 622             exit(1);
 623         }
 624         qemu_mutex_unlock(&sigbus_mutex);
 625     }
 626 }
 627
 628 char *qemu_get_pid_name(pid_t pid)
 629 {
 630     char *name = NULL;
 631
 632 #if defined(__FreeBSD__)
 633     /* BSDs don't have /proc, but they provide a nice substitute */
 634     struct kinfo_proc *proc = kinfo_getproc(pid);
 635
 636     if (proc) {
 637         name = g_strdup(proc->ki_comm);
 638         free(proc);
 639     }
 640 #else
 641     /* Assume a system with reasonable procfs */
 642     char *pid_path;
 643     size_t len;
 644
 645     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 646     g_file_get_contents(pid_path, &name, &len, NULL);
 647     g_free(pid_path);
 648 #endif
 649
 650     return name;
 651 }
 652
 653
 654 pid_t qemu_fork(Error **errp)
 655 {
 656     sigset_t oldmask, newmask;
 657     struct sigaction sig_action;
 658     int saved_errno;
 659     pid_t pid;
 660
 661     /*
 662      * Need to block signals now, so that child process can safely
 663      * kill off caller's signal handlers without a race.
 664      */
 665     sigfillset(&newmask);
 666     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 667         error_setg_errno(errp, errno,
 668                          "cannot block signals");
 669         return -1;
 670     }
 671
 672     pid = fork();
 673     saved_errno = errno;
 674
 675     if (pid < 0) {
 676         /* attempt to restore signal mask, but ignore failure, to
 677          * avoid obscuring the fork failure */
 678         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 679         error_setg_errno(errp, saved_errno,
 680                          "cannot fork child process");
 681         errno = saved_errno;
 682         return -1;
 683     } else if (pid) {
 684         /* parent process */
 685
 686         /* Restore our original signal mask now that the child is
 687          * safely running. Only documented failures are EFAULT (not
 688          * possible, since we are using just-grabbed mask) or EINVAL
 689          * (not possible, since we are using correct arguments).  */
 690         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 691     } else {
 692         /* child process */
 693         size_t i;
 694
 695         /* Clear out all signal handlers from parent so nothing
 696          * unexpected can happen in our child once we unblock
 697          * signals */
 698         sig_action.sa_handler = SIG_DFL;
 699         sig_action.sa_flags = 0;
 700         sigemptyset(&sig_action.sa_mask);
 701
 702         for (i = 1; i < NSIG; i++) {
 703             /* Only possible errors are EFAULT or EINVAL The former
 704              * won't happen, the latter we expect, so no need to check
 705              * return value */
 706             (void)sigaction(i, &sig_action, NULL);
 707         }
 708
 709         /* Unmask all signals in child, since we've no idea what the
 710          * caller's done with their signal mask and don't want to
 711          * propagate that to children */
 712         sigemptyset(&newmask);
 713         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 714             Error *local_err = NULL;
 715             error_setg_errno(&local_err, errno,
 716                              "cannot unblock signals");
 717             error_report_err(local_err);
 718             _exit(1);
 719         }
 720     }
 721     return pid;
 722 }
 723
 724 void *qemu_alloc_stack(size_t *sz)
 725 {
 726     void *ptr, *guardpage;
 727     int flags;
 728 #ifdef CONFIG_DEBUG_STACK_USAGE
 729     void *ptr2;
 730 #endif
 731     size_t pagesz = qemu_real_host_page_size();
 732 #ifdef _SC_THREAD_STACK_MIN
 733     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 734     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 735     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 736 #endif
 737     /* adjust stack size to a multiple of the page size */
 738     *sz = ROUND_UP(*sz, pagesz);
 739     /* allocate one extra page for the guard page */
 740     *sz += pagesz;
 741
 742     flags = MAP_PRIVATE | MAP_ANONYMOUS;
 743 #if defined(MAP_STACK) && defined(__OpenBSD__)
 744     /* Only enable MAP_STACK on OpenBSD. Other OS's such as
 745      * Linux/FreeBSD/NetBSD have a flag with the same name
 746      * but have differing functionality. OpenBSD will SEGV
 747      * if it spots execution with a stack pointer pointing
 748      * at memory that was not allocated with MAP_STACK.
 749      */
 750     flags |= MAP_STACK;
 751 #endif
 752
 753     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
 754     if (ptr == MAP_FAILED) {
 755         perror("failed to allocate memory for stack");
 756         abort();
 757     }
 758
 759 #if defined(HOST_IA64)
 760     /* separate register stack */
 761     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
 762 #elif defined(HOST_HPPA)
 763     /* stack grows up */
 764     guardpage = ptr + *sz - pagesz;
 765 #else
 766     /* stack grows down */
 767     guardpage = ptr;
 768 #endif
 769     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
 770         perror("failed to set up stack guard page");
 771         abort();
 772     }
 773
 774 #ifdef CONFIG_DEBUG_STACK_USAGE
 775     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 776         *(uint32_t *)ptr2 = 0xdeadbeaf;
 777     }
 778 #endif
 779
 780     return ptr;
 781 }
 782
 783 #ifdef CONFIG_DEBUG_STACK_USAGE
 784 static __thread unsigned int max_stack_usage;
 785 #endif
 786
 787 void qemu_free_stack(void *stack, size_t sz)
 788 {
 789 #ifdef CONFIG_DEBUG_STACK_USAGE
 790     unsigned int usage;
 791     void *ptr;
 792
 793     for (ptr = stack + qemu_real_host_page_size(); ptr < stack + sz;
 794          ptr += sizeof(uint32_t)) {
 795         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 796             break;
 797         }
 798     }
 799     usage = sz - (uintptr_t) (ptr - stack);
 800     if (usage > max_stack_usage) {
 801         error_report("thread %d max stack usage increased from %u to %u",
 802                      qemu_get_thread_id(), max_stack_usage, usage);
 803         max_stack_usage = usage;
 804     }
 805 #endif
 806
 807     munmap(stack, sz);
 808 }
 809
 810 /*
 811  * Disable CFI checks.
 812  * We are going to call a signal hander directly. Such handler may or may not
 813  * have been defined in our binary, so there's no guarantee that the pointer
 814  * used to set the handler is a cfi-valid pointer. Since the handlers are
 815  * stored in kernel memory, changing the handler to an attacker-defined
 816  * function requires being able to call a sigaction() syscall,
 817  * which is not as easy as overwriting a pointer in memory.
 818  */
 819 QEMU_DISABLE_CFI
 820 void sigaction_invoke(struct sigaction *action,
 821                       struct qemu_signalfd_siginfo *info)
 822 {
 823     siginfo_t si = {};
 824     si.si_signo = info->ssi_signo;
 825     si.si_errno = info->ssi_errno;
 826     si.si_code = info->ssi_code;
 827
 828     /* Convert the minimal set of fields defined by POSIX.
 829      * Positive si_code values are reserved for kernel-generated
 830      * signals, where the valid siginfo fields are determined by
 831      * the signal number.  But according to POSIX, it is unspecified
 832      * whether SI_USER and SI_QUEUE have values less than or equal to
 833      * zero.
 834      */
 835     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 836         info->ssi_code <= 0) {
 837         /* SIGTERM, etc.  */
 838         si.si_pid = info->ssi_pid;
 839         si.si_uid = info->ssi_uid;
 840     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 841                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 842         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 843     } else if (info->ssi_signo == SIGCHLD) {
 844         si.si_pid = info->ssi_pid;
 845         si.si_status = info->ssi_status;
 846         si.si_uid = info->ssi_uid;
 847     }
 848     action->sa_sigaction(info->ssi_signo, &si, NULL);
 849 }
 850
 851 size_t qemu_get_host_physmem(void)
 852 {
 853 #ifdef _SC_PHYS_PAGES
 854     long pages = sysconf(_SC_PHYS_PAGES);
 855     if (pages > 0) {
 856         if (pages > SIZE_MAX / qemu_real_host_page_size()) {
 857             return SIZE_MAX;
 858         } else {
 859             return pages * qemu_real_host_page_size();
 860         }
 861     }
 862 #endif
 863     return 0;
 864 }
 865
 866 int qemu_msync(void *addr, size_t length, int fd)
 867 {
 868     size_t align_mask = ~(qemu_real_host_page_size() - 1);
 869
 870     /**
 871      * There are no strict reqs as per the length of mapping
 872      * to be synced. Still the length needs to follow the address
 873      * alignment changes. Additionally - round the size to the multiple
 874      * of PAGE_SIZE
 875      */
 876     length += ((uintptr_t)addr & (qemu_real_host_page_size() - 1));
 877     length = (length + ~align_mask) & align_mask;
 878
 879     addr = (void *)((uintptr_t)addr & align_mask);
 880
 881     return msync(addr, length, MS_SYNC);
 882 }