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