util/oslib-posix.c

   1 /*
   2  * os-posix-lib.c
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  * Copyright (c) 2010 Red Hat, Inc.
   6  *
   7  * QEMU library functions on POSIX which are shared between QEMU and
   8  * the QEMU tools.
   9  *
  10  * Permission is hereby granted, free of charge, to any person obtaining a copy
  11  * of this software and associated documentation files (the "Software"), to deal
  12  * in the Software without restriction, including without limitation the rights
  13  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  14  * copies of the Software, and to permit persons to whom the Software is
  15  * furnished to do so, subject to the following conditions:
  16  *
  17  * The above copyright notice and this permission notice shall be included in
  18  * all copies or substantial portions of the Software.
  19  *
  20  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  21  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  23  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  24  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  25  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  26  * THE SOFTWARE.
  27  */
  28
  29 #include "qemu/osdep.h"
  30 #include <termios.h>
  31
  32 #include <glib/gprintf.h>
  33
  34 #include "qemu-common.h"
  35 #include "sysemu/sysemu.h"
  36 #include "trace.h"
  37 #include "qapi/error.h"
  38 #include "qemu/sockets.h"
  39 #include "qemu/thread.h"
  40 #include <libgen.h>
  41 #include "qemu/cutils.h"
  42 #include "qemu/compiler.h"
  43
  44 #ifdef CONFIG_LINUX
  45 #include <sys/syscall.h>
  46 #endif
  47
  48 #ifdef __FreeBSD__
  49 #include <sys/sysctl.h>
  50 #include <sys/user.h>
  51 #include <sys/thr.h>
  52 #include <libutil.h>
  53 #endif
  54
  55 #ifdef __NetBSD__
  56 #include <sys/sysctl.h>
  57 #include <lwp.h>
  58 #endif
  59
  60 #ifdef __APPLE__
  61 #include <mach-o/dyld.h>
  62 #endif
  63
  64 #ifdef __HAIKU__
  65 #include <kernel/image.h>
  66 #endif
  67
  68 #include "qemu/mmap-alloc.h"
  69
  70 #ifdef CONFIG_DEBUG_STACK_USAGE
  71 #include "qemu/error-report.h"
  72 #endif
  73
  74 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
  75
  76 struct MemsetThread {
  77     char *addr;
  78     size_t numpages;
  79     size_t hpagesize;
  80     QemuThread pgthread;
  81     sigjmp_buf env;
  82 };
  83 typedef struct MemsetThread MemsetThread;
  84
  85 static MemsetThread *memset_thread;
  86 static int memset_num_threads;
  87 static bool memset_thread_failed;
  88
  89 static QemuMutex page_mutex;
  90 static QemuCond page_cond;
  91 static bool threads_created_flag;
  92
  93 int qemu_get_thread_id(void)
  94 {
  95 #if defined(__linux__)
  96     return syscall(SYS_gettid);
  97 #elif defined(__FreeBSD__)
  98     /* thread id is up to INT_MAX */
  99     long tid;
 100     thr_self(&tid);
 101     return (int)tid;
 102 #elif defined(__NetBSD__)
 103     return _lwp_self();
 104 #elif defined(__OpenBSD__)
 105     return getthrid();
 106 #else
 107     return getpid();
 108 #endif
 109 }
 110
 111 int qemu_daemon(int nochdir, int noclose)
 112 {
 113     return daemon(nochdir, noclose);
 114 }
 115
 116 bool qemu_write_pidfile(const char *path, Error **errp)
 117 {
 118     int fd;
 119     char pidstr[32];
 120
 121     while (1) {
 122         struct stat a, b;
 123         struct flock lock = {
 124             .l_type = F_WRLCK,
 125             .l_whence = SEEK_SET,
 126             .l_len = 0,
 127         };
 128
 129         fd = qemu_open_old(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
 130         if (fd == -1) {
 131             error_setg_errno(errp, errno, "Cannot open pid file");
 132             return false;
 133         }
 134
 135         if (fstat(fd, &b) < 0) {
 136             error_setg_errno(errp, errno, "Cannot stat file");
 137             goto fail_close;
 138         }
 139
 140         if (fcntl(fd, F_SETLK, &lock)) {
 141             error_setg_errno(errp, errno, "Cannot lock pid file");
 142             goto fail_close;
 143         }
 144
 145         /*
 146          * Now make sure the path we locked is the same one that now
 147          * exists on the filesystem.
 148          */
 149         if (stat(path, &a) < 0) {
 150             /*
 151              * PID file disappeared, someone else must be racing with
 152              * us, so try again.
 153              */
 154             close(fd);
 155             continue;
 156         }
 157
 158         if (a.st_ino == b.st_ino) {
 159             break;
 160         }
 161
 162         /*
 163          * PID file was recreated, someone else must be racing with
 164          * us, so try again.
 165          */
 166         close(fd);
 167     }
 168
 169     if (ftruncate(fd, 0) < 0) {
 170         error_setg_errno(errp, errno, "Failed to truncate pid file");
 171         goto fail_unlink;
 172     }
 173
 174     snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
 175     if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
 176         error_setg(errp, "Failed to write pid file");
 177         goto fail_unlink;
 178     }
 179
 180     return true;
 181
 182 fail_unlink:
 183     unlink(path);
 184 fail_close:
 185     close(fd);
 186     return false;
 187 }
 188
 189 void *qemu_oom_check(void *ptr)
 190 {
 191     if (ptr == NULL) {
 192         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
 193         abort();
 194     }
 195     return ptr;
 196 }
 197
 198 void *qemu_try_memalign(size_t alignment, size_t size)
 199 {
 200     void *ptr;
 201
 202     if (alignment < sizeof(void*)) {
 203         alignment = sizeof(void*);
 204     } else {
 205         g_assert(is_power_of_2(alignment));
 206     }
 207
 208 #if defined(CONFIG_POSIX_MEMALIGN)
 209     int ret;
 210     ret = posix_memalign(&ptr, alignment, size);
 211     if (ret != 0) {
 212         errno = ret;
 213         ptr = NULL;
 214     }
 215 #elif defined(CONFIG_BSD)
 216     ptr = valloc(size);
 217 #else
 218     ptr = memalign(alignment, size);
 219 #endif
 220     trace_qemu_memalign(alignment, size, ptr);
 221     return ptr;
 222 }
 223
 224 void *qemu_memalign(size_t alignment, size_t size)
 225 {
 226     return qemu_oom_check(qemu_try_memalign(alignment, size));
 227 }
 228
 229 /* alloc shared memory pages */
 230 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared)
 231 {
 232     size_t align = QEMU_VMALLOC_ALIGN;
 233     void *ptr = qemu_ram_mmap(-1, size, align, false, shared, false, 0);
 234
 235     if (ptr == MAP_FAILED) {
 236         return NULL;
 237     }
 238
 239     if (alignment) {
 240         *alignment = align;
 241     }
 242
 243     trace_qemu_anon_ram_alloc(size, ptr);
 244     return ptr;
 245 }
 246
 247 void qemu_vfree(void *ptr)
 248 {
 249     trace_qemu_vfree(ptr);
 250     free(ptr);
 251 }
 252
 253 void qemu_anon_ram_free(void *ptr, size_t size)
 254 {
 255     trace_qemu_anon_ram_free(ptr, size);
 256     qemu_ram_munmap(-1, ptr, size);
 257 }
 258
 259 void qemu_set_block(int fd)
 260 {
 261     int f;
 262     f = fcntl(fd, F_GETFL);
 263     assert(f != -1);
 264     f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 265     assert(f != -1);
 266 }
 267
 268 int qemu_try_set_nonblock(int fd)
 269 {
 270     int f;
 271     f = fcntl(fd, F_GETFL);
 272     if (f == -1) {
 273         return -errno;
 274     }
 275     if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) {
 276         return -errno;
 277     }
 278     return 0;
 279 }
 280
 281 void qemu_set_nonblock(int fd)
 282 {
 283     int f;
 284     f = qemu_try_set_nonblock(fd);
 285     assert(f == 0);
 286 }
 287
 288 int socket_set_fast_reuse(int fd)
 289 {
 290     int val = 1, ret;
 291
 292     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 293                      (const char *)&val, sizeof(val));
 294
 295     assert(ret == 0);
 296
 297     return ret;
 298 }
 299
 300 void qemu_set_cloexec(int fd)
 301 {
 302     int f;
 303     f = fcntl(fd, F_GETFD);
 304     assert(f != -1);
 305     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 306     assert(f != -1);
 307 }
 308
 309 /*
 310  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 311  */
 312 int qemu_pipe(int pipefd[2])
 313 {
 314     int ret;
 315
 316 #ifdef CONFIG_PIPE2
 317     ret = pipe2(pipefd, O_CLOEXEC);
 318     if (ret != -1 || errno != ENOSYS) {
 319         return ret;
 320     }
 321 #endif
 322     ret = pipe(pipefd);
 323     if (ret == 0) {
 324         qemu_set_cloexec(pipefd[0]);
 325         qemu_set_cloexec(pipefd[1]);
 326     }
 327
 328     return ret;
 329 }
 330
 331 char *
 332 qemu_get_local_state_pathname(const char *relative_pathname)
 333 {
 334     g_autofree char *dir = g_strdup_printf("%s/%s",
 335                                            CONFIG_QEMU_LOCALSTATEDIR,
 336                                            relative_pathname);
 337     return get_relocated_path(dir);
 338 }
 339
 340 void qemu_set_tty_echo(int fd, bool echo)
 341 {
 342     struct termios tty;
 343
 344     tcgetattr(fd, &tty);
 345
 346     if (echo) {
 347         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 348     } else {
 349         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 350     }
 351
 352     tcsetattr(fd, TCSANOW, &tty);
 353 }
 354
 355 static const char *exec_dir;
 356
 357 void qemu_init_exec_dir(const char *argv0)
 358 {
 359     char *p = NULL;
 360     char buf[PATH_MAX];
 361
 362     if (exec_dir) {
 363         return;
 364     }
 365
 366 #if defined(__linux__)
 367     {
 368         int len;
 369         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 370         if (len > 0) {
 371             buf[len] = 0;
 372             p = buf;
 373         }
 374     }
 375 #elif defined(__FreeBSD__) \
 376       || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
 377     {
 378 #if defined(__FreeBSD__)
 379         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 380 #else
 381         static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
 382 #endif
 383         size_t len = sizeof(buf) - 1;
 384
 385         *buf = '\0';
 386         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 387             *buf) {
 388             buf[sizeof(buf) - 1] = '\0';
 389             p = buf;
 390         }
 391     }
 392 #elif defined(__APPLE__)
 393     {
 394         char fpath[PATH_MAX];
 395         uint32_t len = sizeof(fpath);
 396         if (_NSGetExecutablePath(fpath, &len) == 0) {
 397             p = realpath(fpath, buf);
 398             if (!p) {
 399                 return;
 400             }
 401         }
 402     }
 403 #elif defined(__HAIKU__)
 404     {
 405         image_info ii;
 406         int32_t c = 0;
 407
 408         *buf = '\0';
 409         while (get_next_image_info(0, &c, &ii) == B_OK) {
 410             if (ii.type == B_APP_IMAGE) {
 411                 strncpy(buf, ii.name, sizeof(buf));
 412                 buf[sizeof(buf) - 1] = 0;
 413                 p = buf;
 414                 break;
 415             }
 416         }
 417     }
 418 #endif
 419     /* If we don't have any way of figuring out the actual executable
 420        location then try argv[0].  */
 421     if (!p && argv0) {
 422         p = realpath(argv0, buf);
 423     }
 424     if (p) {
 425         exec_dir = g_path_get_dirname(p);
 426     } else {
 427         exec_dir = CONFIG_BINDIR;
 428     }
 429 }
 430
 431 const char *qemu_get_exec_dir(void)
 432 {
 433     return exec_dir;
 434 }
 435
 436 static void sigbus_handler(int signal)
 437 {
 438     int i;
 439     if (memset_thread) {
 440         for (i = 0; i < memset_num_threads; i++) {
 441             if (qemu_thread_is_self(&memset_thread[i].pgthread)) {
 442                 siglongjmp(memset_thread[i].env, 1);
 443             }
 444         }
 445     }
 446 }
 447
 448 static void *do_touch_pages(void *arg)
 449 {
 450     MemsetThread *memset_args = (MemsetThread *)arg;
 451     sigset_t set, oldset;
 452
 453     /*
 454      * On Linux, the page faults from the loop below can cause mmap_sem
 455      * contention with allocation of the thread stacks.  Do not start
 456      * clearing until all threads have been created.
 457      */
 458     qemu_mutex_lock(&page_mutex);
 459     while(!threads_created_flag){
 460         qemu_cond_wait(&page_cond, &page_mutex);
 461     }
 462     qemu_mutex_unlock(&page_mutex);
 463
 464     /* unblock SIGBUS */
 465     sigemptyset(&set);
 466     sigaddset(&set, SIGBUS);
 467     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 468
 469     if (sigsetjmp(memset_args->env, 1)) {
 470         memset_thread_failed = true;
 471     } else {
 472         char *addr = memset_args->addr;
 473         size_t numpages = memset_args->numpages;
 474         size_t hpagesize = memset_args->hpagesize;
 475         size_t i;
 476         for (i = 0; i < numpages; i++) {
 477             /*
 478              * Read & write back the same value, so we don't
 479              * corrupt existing user/app data that might be
 480              * stored.
 481              *
 482              * 'volatile' to stop compiler optimizing this away
 483              * to a no-op
 484              *
 485              * TODO: get a better solution from kernel so we
 486              * don't need to write at all so we don't cause
 487              * wear on the storage backing the region...
 488              */
 489             *(volatile char *)addr = *addr;
 490             addr += hpagesize;
 491         }
 492     }
 493     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 494     return NULL;
 495 }
 496
 497 static inline int get_memset_num_threads(int smp_cpus)
 498 {
 499     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 500     int ret = 1;
 501
 502     if (host_procs > 0) {
 503         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
 504     }
 505     /* In case sysconf() fails, we fall back to single threaded */
 506     return ret;
 507 }
 508
 509 static bool touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 510                             int smp_cpus)
 511 {
 512     static gsize initialized = 0;
 513     size_t numpages_per_thread, leftover;
 514     char *addr = area;
 515     int i = 0;
 516
 517     if (g_once_init_enter(&initialized)) {
 518         qemu_mutex_init(&page_mutex);
 519         qemu_cond_init(&page_cond);
 520         g_once_init_leave(&initialized, 1);
 521     }
 522
 523     memset_thread_failed = false;
 524     threads_created_flag = false;
 525     memset_num_threads = get_memset_num_threads(smp_cpus);
 526     memset_thread = g_new0(MemsetThread, memset_num_threads);
 527     numpages_per_thread = numpages / memset_num_threads;
 528     leftover = numpages % memset_num_threads;
 529     for (i = 0; i < memset_num_threads; i++) {
 530         memset_thread[i].addr = addr;
 531         memset_thread[i].numpages = numpages_per_thread + (i < leftover);
 532         memset_thread[i].hpagesize = hpagesize;
 533         qemu_thread_create(&memset_thread[i].pgthread, "touch_pages",
 534                            do_touch_pages, &memset_thread[i],
 535                            QEMU_THREAD_JOINABLE);
 536         addr += memset_thread[i].numpages * hpagesize;
 537     }
 538
 539     qemu_mutex_lock(&page_mutex);
 540     threads_created_flag = true;
 541     qemu_cond_broadcast(&page_cond);
 542     qemu_mutex_unlock(&page_mutex);
 543
 544     for (i = 0; i < memset_num_threads; i++) {
 545         qemu_thread_join(&memset_thread[i].pgthread);
 546     }
 547     g_free(memset_thread);
 548     memset_thread = NULL;
 549
 550     return memset_thread_failed;
 551 }
 552
 553 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
 554                      Error **errp)
 555 {
 556     int ret;
 557     struct sigaction act, oldact;
 558     size_t hpagesize = qemu_fd_getpagesize(fd);
 559     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 560
 561     memset(&act, 0, sizeof(act));
 562     act.sa_handler = &sigbus_handler;
 563     act.sa_flags = 0;
 564
 565     ret = sigaction(SIGBUS, &act, &oldact);
 566     if (ret) {
 567         error_setg_errno(errp, errno,
 568             "os_mem_prealloc: failed to install signal handler");
 569         return;
 570     }
 571
 572     /* touch pages simultaneously */
 573     if (touch_all_pages(area, hpagesize, numpages, smp_cpus)) {
 574         error_setg(errp, "os_mem_prealloc: Insufficient free host memory "
 575             "pages available to allocate guest RAM");
 576     }
 577
 578     ret = sigaction(SIGBUS, &oldact, NULL);
 579     if (ret) {
 580         /* Terminate QEMU since it can't recover from error */
 581         perror("os_mem_prealloc: failed to reinstall signal handler");
 582         exit(1);
 583     }
 584 }
 585
 586 char *qemu_get_pid_name(pid_t pid)
 587 {
 588     char *name = NULL;
 589
 590 #if defined(__FreeBSD__)
 591     /* BSDs don't have /proc, but they provide a nice substitute */
 592     struct kinfo_proc *proc = kinfo_getproc(pid);
 593
 594     if (proc) {
 595         name = g_strdup(proc->ki_comm);
 596         free(proc);
 597     }
 598 #else
 599     /* Assume a system with reasonable procfs */
 600     char *pid_path;
 601     size_t len;
 602
 603     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 604     g_file_get_contents(pid_path, &name, &len, NULL);
 605     g_free(pid_path);
 606 #endif
 607
 608     return name;
 609 }
 610
 611
 612 pid_t qemu_fork(Error **errp)
 613 {
 614     sigset_t oldmask, newmask;
 615     struct sigaction sig_action;
 616     int saved_errno;
 617     pid_t pid;
 618
 619     /*
 620      * Need to block signals now, so that child process can safely
 621      * kill off caller's signal handlers without a race.
 622      */
 623     sigfillset(&newmask);
 624     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 625         error_setg_errno(errp, errno,
 626                          "cannot block signals");
 627         return -1;
 628     }
 629
 630     pid = fork();
 631     saved_errno = errno;
 632
 633     if (pid < 0) {
 634         /* attempt to restore signal mask, but ignore failure, to
 635          * avoid obscuring the fork failure */
 636         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 637         error_setg_errno(errp, saved_errno,
 638                          "cannot fork child process");
 639         errno = saved_errno;
 640         return -1;
 641     } else if (pid) {
 642         /* parent process */
 643
 644         /* Restore our original signal mask now that the child is
 645          * safely running. Only documented failures are EFAULT (not
 646          * possible, since we are using just-grabbed mask) or EINVAL
 647          * (not possible, since we are using correct arguments).  */
 648         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 649     } else {
 650         /* child process */
 651         size_t i;
 652
 653         /* Clear out all signal handlers from parent so nothing
 654          * unexpected can happen in our child once we unblock
 655          * signals */
 656         sig_action.sa_handler = SIG_DFL;
 657         sig_action.sa_flags = 0;
 658         sigemptyset(&sig_action.sa_mask);
 659
 660         for (i = 1; i < NSIG; i++) {
 661             /* Only possible errors are EFAULT or EINVAL The former
 662              * won't happen, the latter we expect, so no need to check
 663              * return value */
 664             (void)sigaction(i, &sig_action, NULL);
 665         }
 666
 667         /* Unmask all signals in child, since we've no idea what the
 668          * caller's done with their signal mask and don't want to
 669          * propagate that to children */
 670         sigemptyset(&newmask);
 671         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 672             Error *local_err = NULL;
 673             error_setg_errno(&local_err, errno,
 674                              "cannot unblock signals");
 675             error_report_err(local_err);
 676             _exit(1);
 677         }
 678     }
 679     return pid;
 680 }
 681
 682 void *qemu_alloc_stack(size_t *sz)
 683 {
 684     void *ptr, *guardpage;
 685     int flags;
 686 #ifdef CONFIG_DEBUG_STACK_USAGE
 687     void *ptr2;
 688 #endif
 689     size_t pagesz = qemu_real_host_page_size;
 690 #ifdef _SC_THREAD_STACK_MIN
 691     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 692     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 693     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 694 #endif
 695     /* adjust stack size to a multiple of the page size */
 696     *sz = ROUND_UP(*sz, pagesz);
 697     /* allocate one extra page for the guard page */
 698     *sz += pagesz;
 699
 700     flags = MAP_PRIVATE | MAP_ANONYMOUS;
 701 #if defined(MAP_STACK) && defined(__OpenBSD__)
 702     /* Only enable MAP_STACK on OpenBSD. Other OS's such as
 703      * Linux/FreeBSD/NetBSD have a flag with the same name
 704      * but have differing functionality. OpenBSD will SEGV
 705      * if it spots execution with a stack pointer pointing
 706      * at memory that was not allocated with MAP_STACK.
 707      */
 708     flags |= MAP_STACK;
 709 #endif
 710
 711     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
 712     if (ptr == MAP_FAILED) {
 713         perror("failed to allocate memory for stack");
 714         abort();
 715     }
 716
 717 #if defined(HOST_IA64)
 718     /* separate register stack */
 719     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
 720 #elif defined(HOST_HPPA)
 721     /* stack grows up */
 722     guardpage = ptr + *sz - pagesz;
 723 #else
 724     /* stack grows down */
 725     guardpage = ptr;
 726 #endif
 727     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
 728         perror("failed to set up stack guard page");
 729         abort();
 730     }
 731
 732 #ifdef CONFIG_DEBUG_STACK_USAGE
 733     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 734         *(uint32_t *)ptr2 = 0xdeadbeaf;
 735     }
 736 #endif
 737
 738     return ptr;
 739 }
 740
 741 #ifdef CONFIG_DEBUG_STACK_USAGE
 742 static __thread unsigned int max_stack_usage;
 743 #endif
 744
 745 void qemu_free_stack(void *stack, size_t sz)
 746 {
 747 #ifdef CONFIG_DEBUG_STACK_USAGE
 748     unsigned int usage;
 749     void *ptr;
 750
 751     for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz;
 752          ptr += sizeof(uint32_t)) {
 753         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 754             break;
 755         }
 756     }
 757     usage = sz - (uintptr_t) (ptr - stack);
 758     if (usage > max_stack_usage) {
 759         error_report("thread %d max stack usage increased from %u to %u",
 760                      qemu_get_thread_id(), max_stack_usage, usage);
 761         max_stack_usage = usage;
 762     }
 763 #endif
 764
 765     munmap(stack, sz);
 766 }
 767
 768 /*
 769  * Disable CFI checks.
 770  * We are going to call a signal hander directly. Such handler may or may not
 771  * have been defined in our binary, so there's no guarantee that the pointer
 772  * used to set the handler is a cfi-valid pointer. Since the handlers are
 773  * stored in kernel memory, changing the handler to an attacker-defined
 774  * function requires being able to call a sigaction() syscall,
 775  * which is not as easy as overwriting a pointer in memory.
 776  */
 777 QEMU_DISABLE_CFI
 778 void sigaction_invoke(struct sigaction *action,
 779                       struct qemu_signalfd_siginfo *info)
 780 {
 781     siginfo_t si = {};
 782     si.si_signo = info->ssi_signo;
 783     si.si_errno = info->ssi_errno;
 784     si.si_code = info->ssi_code;
 785
 786     /* Convert the minimal set of fields defined by POSIX.
 787      * Positive si_code values are reserved for kernel-generated
 788      * signals, where the valid siginfo fields are determined by
 789      * the signal number.  But according to POSIX, it is unspecified
 790      * whether SI_USER and SI_QUEUE have values less than or equal to
 791      * zero.
 792      */
 793     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 794         info->ssi_code <= 0) {
 795         /* SIGTERM, etc.  */
 796         si.si_pid = info->ssi_pid;
 797         si.si_uid = info->ssi_uid;
 798     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 799                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 800         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 801     } else if (info->ssi_signo == SIGCHLD) {
 802         si.si_pid = info->ssi_pid;
 803         si.si_status = info->ssi_status;
 804         si.si_uid = info->ssi_uid;
 805     }
 806     action->sa_sigaction(info->ssi_signo, &si, NULL);
 807 }
 808
 809 #ifndef HOST_NAME_MAX
 810 # ifdef _POSIX_HOST_NAME_MAX
 811 #  define HOST_NAME_MAX _POSIX_HOST_NAME_MAX
 812 # else
 813 #  define HOST_NAME_MAX 255
 814 # endif
 815 #endif
 816
 817 char *qemu_get_host_name(Error **errp)
 818 {
 819     long len = -1;
 820     g_autofree char *hostname = NULL;
 821
 822 #ifdef _SC_HOST_NAME_MAX
 823     len = sysconf(_SC_HOST_NAME_MAX);
 824 #endif /* _SC_HOST_NAME_MAX */
 825
 826     if (len < 0) {
 827         len = HOST_NAME_MAX;
 828     }
 829
 830     /* Unfortunately, gethostname() below does not guarantee a
 831      * NULL terminated string. Therefore, allocate one byte more
 832      * to be sure. */
 833     hostname = g_new0(char, len + 1);
 834
 835     if (gethostname(hostname, len) < 0) {
 836         error_setg_errno(errp, errno,
 837                          "cannot get hostname");
 838         return NULL;
 839     }
 840
 841     return g_steal_pointer(&hostname);
 842 }
 843
 844 size_t qemu_get_host_physmem(void)
 845 {
 846 #ifdef _SC_PHYS_PAGES
 847     long pages = sysconf(_SC_PHYS_PAGES);
 848     if (pages > 0) {
 849         if (pages > SIZE_MAX / qemu_real_host_page_size) {
 850             return SIZE_MAX;
 851         } else {
 852             return pages * qemu_real_host_page_size;
 853         }
 854     }
 855 #endif
 856     return 0;
 857 }