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/units.h"
  44 #include "qemu/thread-context.h"
  45 #include "qemu/main-loop.h"
  46
  47 #ifdef CONFIG_LINUX
  48 #include <sys/syscall.h>
  49 #endif
  50
  51 #ifdef __FreeBSD__
  52 #include <sys/thr.h>
  53 #include <sys/user.h>
  54 #include <libutil.h>
  55 #endif
  56
  57 #ifdef __NetBSD__
  58 #include <lwp.h>
  59 #endif
  60
  61 #include "qemu/mmap-alloc.h"
  62
  63 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
  64
  65 struct MemsetThread;
  66
  67 static QLIST_HEAD(, MemsetContext) memset_contexts =
  68     QLIST_HEAD_INITIALIZER(memset_contexts);
  69
  70 typedef struct MemsetContext {
  71     bool all_threads_created;
  72     bool any_thread_failed;
  73     struct MemsetThread *threads;
  74     int num_threads;
  75     QLIST_ENTRY(MemsetContext) next;
  76 } MemsetContext;
  77
  78 struct MemsetThread {
  79     char *addr;
  80     size_t numpages;
  81     size_t hpagesize;
  82     QemuThread pgthread;
  83     sigjmp_buf env;
  84     MemsetContext *context;
  85 };
  86 typedef struct MemsetThread MemsetThread;
  87
  88 /* used by sigbus_handler() */
  89 static MemsetContext *sigbus_memset_context;
  90 struct sigaction sigbus_oldact;
  91 static QemuMutex sigbus_mutex;
  92
  93 static QemuMutex page_mutex;
  94 static QemuCond page_cond;
  95
  96 int qemu_get_thread_id(void)
  97 {
  98 #if defined(__linux__)
  99     return syscall(SYS_gettid);
 100 #elif defined(__FreeBSD__)
 101     /* thread id is up to INT_MAX */
 102     long tid;
 103     thr_self(&tid);
 104     return (int)tid;
 105 #elif defined(__NetBSD__)
 106     return _lwp_self();
 107 #elif defined(__OpenBSD__)
 108     return getthrid();
 109 #else
 110     return getpid();
 111 #endif
 112 }
 113
 114 int qemu_daemon(int nochdir, int noclose)
 115 {
 116     return daemon(nochdir, noclose);
 117 }
 118
 119 bool qemu_write_pidfile(const char *path, Error **errp)
 120 {
 121     int fd;
 122     char pidstr[32];
 123
 124     while (1) {
 125         struct stat a, b;
 126         struct flock lock = {
 127             .l_type = F_WRLCK,
 128             .l_whence = SEEK_SET,
 129             .l_len = 0,
 130         };
 131
 132         fd = qemu_create(path, O_WRONLY, S_IRUSR | S_IWUSR, errp);
 133         if (fd == -1) {
 134             return false;
 135         }
 136
 137         if (fstat(fd, &b) < 0) {
 138             error_setg_errno(errp, errno, "Cannot stat file");
 139             goto fail_close;
 140         }
 141
 142         if (fcntl(fd, F_SETLK, &lock)) {
 143             error_setg_errno(errp, errno, "Cannot lock pid file");
 144             goto fail_close;
 145         }
 146
 147         /*
 148          * Now make sure the path we locked is the same one that now
 149          * exists on the filesystem.
 150          */
 151         if (stat(path, &a) < 0) {
 152             /*
 153              * PID file disappeared, someone else must be racing with
 154              * us, so try again.
 155              */
 156             close(fd);
 157             continue;
 158         }
 159
 160         if (a.st_ino == b.st_ino) {
 161             break;
 162         }
 163
 164         /*
 165          * PID file was recreated, someone else must be racing with
 166          * us, so try again.
 167          */
 168         close(fd);
 169     }
 170
 171     if (ftruncate(fd, 0) < 0) {
 172         error_setg_errno(errp, errno, "Failed to truncate pid file");
 173         goto fail_unlink;
 174     }
 175
 176     snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
 177     if (qemu_write_full(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
 178         error_setg(errp, "Failed to write pid file");
 179         goto fail_unlink;
 180     }
 181
 182     return true;
 183
 184 fail_unlink:
 185     unlink(path);
 186 fail_close:
 187     close(fd);
 188     return false;
 189 }
 190
 191 /* alloc shared memory pages */
 192 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared,
 193                           bool noreserve)
 194 {
 195     const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) |
 196                                     (noreserve ? QEMU_MAP_NORESERVE : 0);
 197     size_t align = QEMU_VMALLOC_ALIGN;
 198     void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0);
 199
 200     if (ptr == MAP_FAILED) {
 201         return NULL;
 202     }
 203
 204     if (alignment) {
 205         *alignment = align;
 206     }
 207
 208     trace_qemu_anon_ram_alloc(size, ptr);
 209     return ptr;
 210 }
 211
 212 void qemu_anon_ram_free(void *ptr, size_t size)
 213 {
 214     trace_qemu_anon_ram_free(ptr, size);
 215     qemu_ram_munmap(-1, ptr, size);
 216 }
 217
 218 void qemu_socket_set_block(int fd)
 219 {
 220     g_unix_set_fd_nonblocking(fd, false, NULL);
 221 }
 222
 223 int qemu_socket_try_set_nonblock(int fd)
 224 {
 225     return g_unix_set_fd_nonblocking(fd, true, NULL) ? 0 : -errno;
 226 }
 227
 228 void qemu_socket_set_nonblock(int fd)
 229 {
 230     int f;
 231     f = qemu_socket_try_set_nonblock(fd);
 232     assert(f == 0);
 233 }
 234
 235 int socket_set_fast_reuse(int fd)
 236 {
 237     int val = 1, ret;
 238
 239     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 240                      (const char *)&val, sizeof(val));
 241
 242     assert(ret == 0);
 243
 244     return ret;
 245 }
 246
 247 void qemu_set_cloexec(int fd)
 248 {
 249     int f;
 250     f = fcntl(fd, F_GETFD);
 251     assert(f != -1);
 252     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 253     assert(f != -1);
 254 }
 255
 256 int qemu_socketpair(int domain, int type, int protocol, int sv[2])
 257 {
 258     int ret;
 259
 260 #ifdef SOCK_CLOEXEC
 261     ret = socketpair(domain, type | SOCK_CLOEXEC, protocol, sv);
 262     if (ret != -1 || errno != EINVAL) {
 263         return ret;
 264     }
 265 #endif
 266     ret = socketpair(domain, type, protocol, sv);
 267     if (ret == 0) {
 268         qemu_set_cloexec(sv[0]);
 269         qemu_set_cloexec(sv[1]);
 270     }
 271
 272     return ret;
 273 }
 274
 275 char *
 276 qemu_get_local_state_dir(void)
 277 {
 278     return get_relocated_path(CONFIG_QEMU_LOCALSTATEDIR);
 279 }
 280
 281 void qemu_set_tty_echo(int fd, bool echo)
 282 {
 283     struct termios tty;
 284
 285     tcgetattr(fd, &tty);
 286
 287     if (echo) {
 288         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 289     } else {
 290         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 291     }
 292
 293     tcsetattr(fd, TCSANOW, &tty);
 294 }
 295
 296 #ifdef CONFIG_LINUX
 297 static void sigbus_handler(int signal, siginfo_t *siginfo, void *ctx)
 298 #else /* CONFIG_LINUX */
 299 static void sigbus_handler(int signal)
 300 #endif /* CONFIG_LINUX */
 301 {
 302     int i;
 303
 304     if (sigbus_memset_context) {
 305         for (i = 0; i < sigbus_memset_context->num_threads; i++) {
 306             MemsetThread *thread = &sigbus_memset_context->threads[i];
 307
 308             if (qemu_thread_is_self(&thread->pgthread)) {
 309                 siglongjmp(thread->env, 1);
 310             }
 311         }
 312     }
 313
 314 #ifdef CONFIG_LINUX
 315     /*
 316      * We assume that the MCE SIGBUS handler could have been registered. We
 317      * should never receive BUS_MCEERR_AO on any of our threads, but only on
 318      * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not
 319      * receive BUS_MCEERR_AR triggered by action of other threads on one of
 320      * our threads. So, no need to check for unrelated SIGBUS when seeing one
 321      * for our threads.
 322      *
 323      * We will forward to the MCE handler, which will either handle the SIGBUS
 324      * or reinstall the default SIGBUS handler and reraise the SIGBUS. The
 325      * default SIGBUS handler will crash the process, so we don't care.
 326      */
 327     if (sigbus_oldact.sa_flags & SA_SIGINFO) {
 328         sigbus_oldact.sa_sigaction(signal, siginfo, ctx);
 329         return;
 330     }
 331 #endif /* CONFIG_LINUX */
 332     warn_report("qemu_prealloc_mem: unrelated SIGBUS detected and ignored");
 333 }
 334
 335 static void *do_touch_pages(void *arg)
 336 {
 337     MemsetThread *memset_args = (MemsetThread *)arg;
 338     sigset_t set, oldset;
 339     int ret = 0;
 340
 341     /*
 342      * On Linux, the page faults from the loop below can cause mmap_sem
 343      * contention with allocation of the thread stacks.  Do not start
 344      * clearing until all threads have been created.
 345      */
 346     qemu_mutex_lock(&page_mutex);
 347     while (!memset_args->context->all_threads_created) {
 348         qemu_cond_wait(&page_cond, &page_mutex);
 349     }
 350     qemu_mutex_unlock(&page_mutex);
 351
 352     /* unblock SIGBUS */
 353     sigemptyset(&set);
 354     sigaddset(&set, SIGBUS);
 355     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 356
 357     if (sigsetjmp(memset_args->env, 1)) {
 358         ret = -EFAULT;
 359     } else {
 360         char *addr = memset_args->addr;
 361         size_t numpages = memset_args->numpages;
 362         size_t hpagesize = memset_args->hpagesize;
 363         size_t i;
 364         for (i = 0; i < numpages; i++) {
 365             /*
 366              * Read & write back the same value, so we don't
 367              * corrupt existing user/app data that might be
 368              * stored.
 369              *
 370              * 'volatile' to stop compiler optimizing this away
 371              * to a no-op
 372              */
 373             *(volatile char *)addr = *addr;
 374             addr += hpagesize;
 375         }
 376     }
 377     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 378     return (void *)(uintptr_t)ret;
 379 }
 380
 381 static void *do_madv_populate_write_pages(void *arg)
 382 {
 383     MemsetThread *memset_args = (MemsetThread *)arg;
 384     const size_t size = memset_args->numpages * memset_args->hpagesize;
 385     char * const addr = memset_args->addr;
 386     int ret = 0;
 387
 388     /* See do_touch_pages(). */
 389     qemu_mutex_lock(&page_mutex);
 390     while (!memset_args->context->all_threads_created) {
 391         qemu_cond_wait(&page_cond, &page_mutex);
 392     }
 393     qemu_mutex_unlock(&page_mutex);
 394
 395     if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) {
 396         ret = -errno;
 397     }
 398     return (void *)(uintptr_t)ret;
 399 }
 400
 401 static inline int get_memset_num_threads(size_t hpagesize, size_t numpages,
 402                                          int max_threads)
 403 {
 404     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 405     int ret = 1;
 406
 407     if (host_procs > 0) {
 408         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), max_threads);
 409     }
 410
 411     /* Especially with gigantic pages, don't create more threads than pages. */
 412     ret = MIN(ret, numpages);
 413     /* Don't start threads to prealloc comparatively little memory. */
 414     ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB)));
 415
 416     /* In case sysconf() fails, we fall back to single threaded */
 417     return ret;
 418 }
 419
 420 static int wait_and_free_mem_prealloc_context(MemsetContext *context)
 421 {
 422     int i, ret = 0, tmp;
 423
 424     for (i = 0; i < context->num_threads; i++) {
 425         tmp = (uintptr_t)qemu_thread_join(&context->threads[i].pgthread);
 426
 427         if (tmp) {
 428             ret = tmp;
 429         }
 430     }
 431     g_free(context->threads);
 432     g_free(context);
 433     return ret;
 434 }
 435
 436 static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 437                            int max_threads, ThreadContext *tc, bool async,
 438                            bool use_madv_populate_write)
 439 {
 440     static gsize initialized = 0;
 441     MemsetContext *context = g_malloc0(sizeof(MemsetContext));
 442     size_t numpages_per_thread, leftover;
 443     void *(*touch_fn)(void *);
 444     int ret, i = 0;
 445     char *addr = area;
 446
 447     /*
 448      * Asynchronous preallocation is only allowed when using MADV_POPULATE_WRITE
 449      * and prealloc context for thread placement.
 450      */
 451     if (!use_madv_populate_write || !tc) {
 452         async = false;
 453     }
 454
 455     context->num_threads =
 456         get_memset_num_threads(hpagesize, numpages, max_threads);
 457
 458     if (g_once_init_enter(&initialized)) {
 459         qemu_mutex_init(&page_mutex);
 460         qemu_cond_init(&page_cond);
 461         g_once_init_leave(&initialized, 1);
 462     }
 463
 464     if (use_madv_populate_write) {
 465         /*
 466          * Avoid creating a single thread for MADV_POPULATE_WRITE when
 467          * preallocating synchronously.
 468          */
 469         if (context->num_threads == 1 && !async) {
 470             ret = 0;
 471             if (qemu_madvise(area, hpagesize * numpages,
 472                              QEMU_MADV_POPULATE_WRITE)) {
 473                 ret = -errno;
 474             }
 475             g_free(context);
 476             return ret;
 477         }
 478         touch_fn = do_madv_populate_write_pages;
 479     } else {
 480         touch_fn = do_touch_pages;
 481     }
 482
 483     context->threads = g_new0(MemsetThread, context->num_threads);
 484     numpages_per_thread = numpages / context->num_threads;
 485     leftover = numpages % context->num_threads;
 486     for (i = 0; i < context->num_threads; i++) {
 487         context->threads[i].addr = addr;
 488         context->threads[i].numpages = numpages_per_thread + (i < leftover);
 489         context->threads[i].hpagesize = hpagesize;
 490         context->threads[i].context = context;
 491         if (tc) {
 492             thread_context_create_thread(tc, &context->threads[i].pgthread,
 493                                          "touch_pages",
 494                                          touch_fn, &context->threads[i],
 495                                          QEMU_THREAD_JOINABLE);
 496         } else {
 497             qemu_thread_create(&context->threads[i].pgthread, "touch_pages",
 498                                touch_fn, &context->threads[i],
 499                                QEMU_THREAD_JOINABLE);
 500         }
 501         addr += context->threads[i].numpages * hpagesize;
 502     }
 503
 504     if (async) {
 505         /*
 506          * async requests currently require the BQL. Add it to the list and kick
 507          * preallocation off during qemu_finish_async_prealloc_mem().
 508          */
 509         assert(bql_locked());
 510         QLIST_INSERT_HEAD(&memset_contexts, context, next);
 511         return 0;
 512     }
 513
 514     if (!use_madv_populate_write) {
 515         sigbus_memset_context = context;
 516     }
 517
 518     qemu_mutex_lock(&page_mutex);
 519     context->all_threads_created = true;
 520     qemu_cond_broadcast(&page_cond);
 521     qemu_mutex_unlock(&page_mutex);
 522
 523     ret = wait_and_free_mem_prealloc_context(context);
 524
 525     if (!use_madv_populate_write) {
 526         sigbus_memset_context = NULL;
 527     }
 528     return ret;
 529 }
 530
 531 bool qemu_finish_async_prealloc_mem(Error **errp)
 532 {
 533     int ret = 0, tmp;
 534     MemsetContext *context, *next_context;
 535
 536     /* Waiting for preallocation requires the BQL. */
 537     assert(bql_locked());
 538     if (QLIST_EMPTY(&memset_contexts)) {
 539         return true;
 540     }
 541
 542     qemu_mutex_lock(&page_mutex);
 543     QLIST_FOREACH(context, &memset_contexts, next) {
 544         context->all_threads_created = true;
 545     }
 546     qemu_cond_broadcast(&page_cond);
 547     qemu_mutex_unlock(&page_mutex);
 548
 549     QLIST_FOREACH_SAFE(context, &memset_contexts, next, next_context) {
 550         QLIST_REMOVE(context, next);
 551         tmp = wait_and_free_mem_prealloc_context(context);
 552         if (tmp) {
 553             ret = tmp;
 554         }
 555     }
 556
 557     if (ret) {
 558         error_setg_errno(errp, -ret,
 559                          "qemu_prealloc_mem: preallocating memory failed");
 560         return false;
 561     }
 562     return true;
 563 }
 564
 565 static bool madv_populate_write_possible(char *area, size_t pagesize)
 566 {
 567     return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) ||
 568            errno != EINVAL;
 569 }
 570
 571 bool qemu_prealloc_mem(int fd, char *area, size_t sz, int max_threads,
 572                        ThreadContext *tc, bool async, Error **errp)
 573 {
 574     static gsize initialized;
 575     int ret;
 576     size_t hpagesize = qemu_fd_getpagesize(fd);
 577     size_t numpages = DIV_ROUND_UP(sz, hpagesize);
 578     bool use_madv_populate_write;
 579     struct sigaction act;
 580     bool rv = true;
 581
 582     /*
 583      * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for
 584      * some special mappings, such as mapping /dev/mem.
 585      */
 586     use_madv_populate_write = madv_populate_write_possible(area, hpagesize);
 587
 588     if (!use_madv_populate_write) {
 589         if (g_once_init_enter(&initialized)) {
 590             qemu_mutex_init(&sigbus_mutex);
 591             g_once_init_leave(&initialized, 1);
 592         }
 593
 594         qemu_mutex_lock(&sigbus_mutex);
 595         memset(&act, 0, sizeof(act));
 596 #ifdef CONFIG_LINUX
 597         act.sa_sigaction = &sigbus_handler;
 598         act.sa_flags = SA_SIGINFO;
 599 #else /* CONFIG_LINUX */
 600         act.sa_handler = &sigbus_handler;
 601         act.sa_flags = 0;
 602 #endif /* CONFIG_LINUX */
 603
 604         ret = sigaction(SIGBUS, &act, &sigbus_oldact);
 605         if (ret) {
 606             qemu_mutex_unlock(&sigbus_mutex);
 607             error_setg_errno(errp, errno,
 608                 "qemu_prealloc_mem: failed to install signal handler");
 609             return false;
 610         }
 611     }
 612
 613     /* touch pages simultaneously */
 614     ret = touch_all_pages(area, hpagesize, numpages, max_threads, tc, async,
 615                           use_madv_populate_write);
 616     if (ret) {
 617         error_setg_errno(errp, -ret,
 618                          "qemu_prealloc_mem: preallocating memory failed");
 619         rv = false;
 620     }
 621
 622     if (!use_madv_populate_write) {
 623         ret = sigaction(SIGBUS, &sigbus_oldact, NULL);
 624         if (ret) {
 625             /* Terminate QEMU since it can't recover from error */
 626             perror("qemu_prealloc_mem: failed to reinstall signal handler");
 627             exit(1);
 628         }
 629         qemu_mutex_unlock(&sigbus_mutex);
 630     }
 631     return rv;
 632 }
 633
 634 char *qemu_get_pid_name(pid_t pid)
 635 {
 636     char *name = NULL;
 637
 638 #if defined(__FreeBSD__)
 639     /* BSDs don't have /proc, but they provide a nice substitute */
 640     struct kinfo_proc *proc = kinfo_getproc(pid);
 641
 642     if (proc) {
 643         name = g_strdup(proc->ki_comm);
 644         free(proc);
 645     }
 646 #else
 647     /* Assume a system with reasonable procfs */
 648     char *pid_path;
 649     size_t len;
 650
 651     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 652     g_file_get_contents(pid_path, &name, &len, NULL);
 653     g_free(pid_path);
 654 #endif
 655
 656     return name;
 657 }
 658
 659
 660 void *qemu_alloc_stack(size_t *sz)
 661 {
 662     void *ptr;
 663     int flags;
 664 #ifdef CONFIG_DEBUG_STACK_USAGE
 665     void *ptr2;
 666 #endif
 667     size_t pagesz = qemu_real_host_page_size();
 668 #ifdef _SC_THREAD_STACK_MIN
 669     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 670     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 671     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 672 #endif
 673     /* adjust stack size to a multiple of the page size */
 674     *sz = ROUND_UP(*sz, pagesz);
 675     /* allocate one extra page for the guard page */
 676     *sz += pagesz;
 677
 678     flags = MAP_PRIVATE | MAP_ANONYMOUS;
 679 #if defined(MAP_STACK) && defined(__OpenBSD__)
 680     /* Only enable MAP_STACK on OpenBSD. Other OS's such as
 681      * Linux/FreeBSD/NetBSD have a flag with the same name
 682      * but have differing functionality. OpenBSD will SEGV
 683      * if it spots execution with a stack pointer pointing
 684      * at memory that was not allocated with MAP_STACK.
 685      */
 686     flags |= MAP_STACK;
 687 #endif
 688
 689     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
 690     if (ptr == MAP_FAILED) {
 691         perror("failed to allocate memory for stack");
 692         abort();
 693     }
 694
 695     /* Stack grows down -- guard page at the bottom. */
 696     if (mprotect(ptr, pagesz, PROT_NONE) != 0) {
 697         perror("failed to set up stack guard page");
 698         abort();
 699     }
 700
 701 #ifdef CONFIG_DEBUG_STACK_USAGE
 702     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 703         *(uint32_t *)ptr2 = 0xdeadbeaf;
 704     }
 705 #endif
 706
 707     return ptr;
 708 }
 709
 710 #ifdef CONFIG_DEBUG_STACK_USAGE
 711 static __thread unsigned int max_stack_usage;
 712 #endif
 713
 714 void qemu_free_stack(void *stack, size_t sz)
 715 {
 716 #ifdef CONFIG_DEBUG_STACK_USAGE
 717     unsigned int usage;
 718     void *ptr;
 719
 720     for (ptr = stack + qemu_real_host_page_size(); ptr < stack + sz;
 721          ptr += sizeof(uint32_t)) {
 722         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 723             break;
 724         }
 725     }
 726     usage = sz - (uintptr_t) (ptr - stack);
 727     if (usage > max_stack_usage) {
 728         error_report("thread %d max stack usage increased from %u to %u",
 729                      qemu_get_thread_id(), max_stack_usage, usage);
 730         max_stack_usage = usage;
 731     }
 732 #endif
 733
 734     munmap(stack, sz);
 735 }
 736
 737 /*
 738  * Disable CFI checks.
 739  * We are going to call a signal handler directly. Such handler may or may not
 740  * have been defined in our binary, so there's no guarantee that the pointer
 741  * used to set the handler is a cfi-valid pointer. Since the handlers are
 742  * stored in kernel memory, changing the handler to an attacker-defined
 743  * function requires being able to call a sigaction() syscall,
 744  * which is not as easy as overwriting a pointer in memory.
 745  */
 746 QEMU_DISABLE_CFI
 747 void sigaction_invoke(struct sigaction *action,
 748                       struct qemu_signalfd_siginfo *info)
 749 {
 750     siginfo_t si = {};
 751     si.si_signo = info->ssi_signo;
 752     si.si_errno = info->ssi_errno;
 753     si.si_code = info->ssi_code;
 754
 755     /* Convert the minimal set of fields defined by POSIX.
 756      * Positive si_code values are reserved for kernel-generated
 757      * signals, where the valid siginfo fields are determined by
 758      * the signal number.  But according to POSIX, it is unspecified
 759      * whether SI_USER and SI_QUEUE have values less than or equal to
 760      * zero.
 761      */
 762     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 763         info->ssi_code <= 0) {
 764         /* SIGTERM, etc.  */
 765         si.si_pid = info->ssi_pid;
 766         si.si_uid = info->ssi_uid;
 767     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 768                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 769         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 770     } else if (info->ssi_signo == SIGCHLD) {
 771         si.si_pid = info->ssi_pid;
 772         si.si_status = info->ssi_status;
 773         si.si_uid = info->ssi_uid;
 774     }
 775     action->sa_sigaction(info->ssi_signo, &si, NULL);
 776 }
 777
 778 size_t qemu_get_host_physmem(void)
 779 {
 780 #ifdef _SC_PHYS_PAGES
 781     long pages = sysconf(_SC_PHYS_PAGES);
 782     if (pages > 0) {
 783         if (pages > SIZE_MAX / qemu_real_host_page_size()) {
 784             return SIZE_MAX;
 785         } else {
 786             return pages * qemu_real_host_page_size();
 787         }
 788     }
 789 #endif
 790     return 0;
 791 }
 792
 793 int qemu_msync(void *addr, size_t length, int fd)
 794 {
 795     size_t align_mask = ~(qemu_real_host_page_size() - 1);
 796
 797     /**
 798      * There are no strict reqs as per the length of mapping
 799      * to be synced. Still the length needs to follow the address
 800      * alignment changes. Additionally - round the size to the multiple
 801      * of PAGE_SIZE
 802      */
 803     length += ((uintptr_t)addr & (qemu_real_host_page_size() - 1));
 804     length = (length + ~align_mask) & align_mask;
 805
 806     addr = (void *)((uintptr_t)addr & align_mask);
 807
 808     return msync(addr, length, MS_SYNC);
 809 }
 810
 811 static bool qemu_close_all_open_fd_proc(const int *skip, unsigned int nskip)
 812 {
 813     struct dirent *de;
 814     int fd, dfd;
 815     DIR *dir;
 816     unsigned int skip_start = 0, skip_end = nskip;
 817
 818     dir = opendir("/proc/self/fd");
 819     if (!dir) {
 820         /* If /proc is not mounted, there is nothing that can be done. */
 821         return false;
 822     }
 823     /* Avoid closing the directory. */
 824     dfd = dirfd(dir);
 825
 826     for (de = readdir(dir); de; de = readdir(dir)) {
 827         bool close_fd = true;
 828
 829         if (de->d_name[0] == '.') {
 830             continue;
 831         }
 832         fd = atoi(de->d_name);
 833         if (fd == dfd) {
 834             continue;
 835         }
 836
 837         for (unsigned int i = skip_start; i < skip_end; i++) {
 838             if (fd < skip[i]) {
 839                 /* We are below the next skipped fd, break */
 840                 break;
 841             } else if (fd == skip[i]) {
 842                 close_fd = false;
 843                 /* Restrict the range as we found fds matching start/end */
 844                 if (i == skip_start) {
 845                     skip_start++;
 846                 } else if (i == skip_end) {
 847                     skip_end--;
 848                 }
 849                 break;
 850             }
 851         }
 852
 853         if (close_fd) {
 854             close(fd);
 855         }
 856     }
 857     closedir(dir);
 858
 859     return true;
 860 }
 861
 862 static bool qemu_close_all_open_fd_close_range(const int *skip,
 863                                                unsigned int nskip,
 864                                                int open_max)
 865 {
 866 #ifdef CONFIG_CLOSE_RANGE
 867     int max_fd = open_max - 1;
 868     int first = 0, last;
 869     unsigned int cur_skip = 0;
 870     int ret;
 871
 872     do {
 873         /* Find the start boundary of the range to close */
 874         while (cur_skip < nskip && first == skip[cur_skip]) {
 875             cur_skip++;
 876             first++;
 877         }
 878
 879         /* Find the upper boundary of the range to close */
 880         last = max_fd;
 881         if (cur_skip < nskip) {
 882             last = skip[cur_skip] - 1;
 883             last = MIN(last, max_fd);
 884         }
 885
 886         /* With the adjustments to the range, we might be done. */
 887         if (first > last) {
 888             break;
 889         }
 890
 891         ret = close_range(first, last, 0);
 892         if (ret < 0) {
 893             return false;
 894         }
 895
 896         first = last + 1;
 897     } while (last < max_fd);
 898
 899     return true;
 900 #else
 901     return false;
 902 #endif
 903 }
 904
 905 static void qemu_close_all_open_fd_fallback(const int *skip, unsigned int nskip,
 906                                             int open_max)
 907 {
 908     unsigned int cur_skip = 0;
 909
 910     /* Fallback */
 911     for (int i = 0; i < open_max; i++) {
 912         if (cur_skip < nskip && i == skip[cur_skip]) {
 913             cur_skip++;
 914             continue;
 915         }
 916         close(i);
 917     }
 918 }
 919
 920 /*
 921  * Close all open file descriptors.
 922  */
 923 void qemu_close_all_open_fd(const int *skip, unsigned int nskip)
 924 {
 925     int open_max = sysconf(_SC_OPEN_MAX);
 926
 927     assert(skip != NULL || nskip == 0);
 928
 929     if (!qemu_close_all_open_fd_close_range(skip, nskip, open_max) &&
 930         !qemu_close_all_open_fd_proc(skip, nskip)) {
 931         qemu_close_all_open_fd_fallback(skip, nskip, open_max);
 932     }
 933 }