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