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/sockets.h"
  40 #include "qemu/thread.h"
  41 #include <libgen.h>
  42 #include "qemu/cutils.h"
  43 #include "qemu/compiler.h"
  44 #include "qemu/units.h"
  45
  46 #ifdef CONFIG_LINUX
  47 #include <sys/syscall.h>
  48 #endif
  49
  50 #ifdef __FreeBSD__
  51 #include <sys/sysctl.h>
  52 #include <sys/user.h>
  53 #include <sys/thr.h>
  54 #include <libutil.h>
  55 #endif
  56
  57 #ifdef __NetBSD__
  58 #include <sys/sysctl.h>
  59 #include <lwp.h>
  60 #endif
  61
  62 #ifdef __APPLE__
  63 #include <mach-o/dyld.h>
  64 #endif
  65
  66 #ifdef __HAIKU__
  67 #include <kernel/image.h>
  68 #endif
  69
  70 #include "qemu/mmap-alloc.h"
  71
  72 #ifdef CONFIG_DEBUG_STACK_USAGE
  73 #include "qemu/error-report.h"
  74 #endif
  75
  76 #define MAX_MEM_PREALLOC_THREAD_COUNT 16
  77
  78 struct MemsetThread;
  79
  80 typedef struct MemsetContext {
  81     bool all_threads_created;
  82     bool any_thread_failed;
  83     struct MemsetThread *threads;
  84     int num_threads;
  85 } MemsetContext;
  86
  87 struct MemsetThread {
  88     char *addr;
  89     size_t numpages;
  90     size_t hpagesize;
  91     QemuThread pgthread;
  92     sigjmp_buf env;
  93     MemsetContext *context;
  94 };
  95 typedef struct MemsetThread MemsetThread;
  96
  97 /* used by sigbus_handler() */
  98 static MemsetContext *sigbus_memset_context;
  99 struct sigaction sigbus_oldact;
 100 static QemuMutex sigbus_mutex;
 101
 102 static QemuMutex page_mutex;
 103 static QemuCond page_cond;
 104
 105 int qemu_get_thread_id(void)
 106 {
 107 #if defined(__linux__)
 108     return syscall(SYS_gettid);
 109 #elif defined(__FreeBSD__)
 110     /* thread id is up to INT_MAX */
 111     long tid;
 112     thr_self(&tid);
 113     return (int)tid;
 114 #elif defined(__NetBSD__)
 115     return _lwp_self();
 116 #elif defined(__OpenBSD__)
 117     return getthrid();
 118 #else
 119     return getpid();
 120 #endif
 121 }
 122
 123 int qemu_daemon(int nochdir, int noclose)
 124 {
 125     return daemon(nochdir, noclose);
 126 }
 127
 128 bool qemu_write_pidfile(const char *path, Error **errp)
 129 {
 130     int fd;
 131     char pidstr[32];
 132
 133     while (1) {
 134         struct stat a, b;
 135         struct flock lock = {
 136             .l_type = F_WRLCK,
 137             .l_whence = SEEK_SET,
 138             .l_len = 0,
 139         };
 140
 141         fd = qemu_open_old(path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
 142         if (fd == -1) {
 143             error_setg_errno(errp, errno, "Cannot open pid file");
 144             return false;
 145         }
 146
 147         if (fstat(fd, &b) < 0) {
 148             error_setg_errno(errp, errno, "Cannot stat file");
 149             goto fail_close;
 150         }
 151
 152         if (fcntl(fd, F_SETLK, &lock)) {
 153             error_setg_errno(errp, errno, "Cannot lock pid file");
 154             goto fail_close;
 155         }
 156
 157         /*
 158          * Now make sure the path we locked is the same one that now
 159          * exists on the filesystem.
 160          */
 161         if (stat(path, &a) < 0) {
 162             /*
 163              * PID file disappeared, someone else must be racing with
 164              * us, so try again.
 165              */
 166             close(fd);
 167             continue;
 168         }
 169
 170         if (a.st_ino == b.st_ino) {
 171             break;
 172         }
 173
 174         /*
 175          * PID file was recreated, someone else must be racing with
 176          * us, so try again.
 177          */
 178         close(fd);
 179     }
 180
 181     if (ftruncate(fd, 0) < 0) {
 182         error_setg_errno(errp, errno, "Failed to truncate pid file");
 183         goto fail_unlink;
 184     }
 185
 186     snprintf(pidstr, sizeof(pidstr), FMT_pid "\n", getpid());
 187     if (write(fd, pidstr, strlen(pidstr)) != strlen(pidstr)) {
 188         error_setg(errp, "Failed to write pid file");
 189         goto fail_unlink;
 190     }
 191
 192     return true;
 193
 194 fail_unlink:
 195     unlink(path);
 196 fail_close:
 197     close(fd);
 198     return false;
 199 }
 200
 201 void *qemu_oom_check(void *ptr)
 202 {
 203     if (ptr == NULL) {
 204         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
 205         abort();
 206     }
 207     return ptr;
 208 }
 209
 210 void *qemu_try_memalign(size_t alignment, size_t size)
 211 {
 212     void *ptr;
 213
 214     if (alignment < sizeof(void*)) {
 215         alignment = sizeof(void*);
 216     } else {
 217         g_assert(is_power_of_2(alignment));
 218     }
 219
 220 #if defined(CONFIG_POSIX_MEMALIGN)
 221     int ret;
 222     ret = posix_memalign(&ptr, alignment, size);
 223     if (ret != 0) {
 224         errno = ret;
 225         ptr = NULL;
 226     }
 227 #elif defined(CONFIG_BSD)
 228     ptr = valloc(size);
 229 #else
 230     ptr = memalign(alignment, size);
 231 #endif
 232     trace_qemu_memalign(alignment, size, ptr);
 233     return ptr;
 234 }
 235
 236 void *qemu_memalign(size_t alignment, size_t size)
 237 {
 238     return qemu_oom_check(qemu_try_memalign(alignment, size));
 239 }
 240
 241 /* alloc shared memory pages */
 242 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment, bool shared,
 243                           bool noreserve)
 244 {
 245     const uint32_t qemu_map_flags = (shared ? QEMU_MAP_SHARED : 0) |
 246                                     (noreserve ? QEMU_MAP_NORESERVE : 0);
 247     size_t align = QEMU_VMALLOC_ALIGN;
 248     void *ptr = qemu_ram_mmap(-1, size, align, qemu_map_flags, 0);
 249
 250     if (ptr == MAP_FAILED) {
 251         return NULL;
 252     }
 253
 254     if (alignment) {
 255         *alignment = align;
 256     }
 257
 258     trace_qemu_anon_ram_alloc(size, ptr);
 259     return ptr;
 260 }
 261
 262 void qemu_vfree(void *ptr)
 263 {
 264     trace_qemu_vfree(ptr);
 265     free(ptr);
 266 }
 267
 268 void qemu_anon_ram_free(void *ptr, size_t size)
 269 {
 270     trace_qemu_anon_ram_free(ptr, size);
 271     qemu_ram_munmap(-1, ptr, size);
 272 }
 273
 274 void qemu_set_block(int fd)
 275 {
 276     int f;
 277     f = fcntl(fd, F_GETFL);
 278     assert(f != -1);
 279     f = fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 280     assert(f != -1);
 281 }
 282
 283 int qemu_try_set_nonblock(int fd)
 284 {
 285     int f;
 286     f = fcntl(fd, F_GETFL);
 287     if (f == -1) {
 288         return -errno;
 289     }
 290     if (fcntl(fd, F_SETFL, f | O_NONBLOCK) == -1) {
 291         return -errno;
 292     }
 293     return 0;
 294 }
 295
 296 void qemu_set_nonblock(int fd)
 297 {
 298     int f;
 299     f = qemu_try_set_nonblock(fd);
 300     assert(f == 0);
 301 }
 302
 303 int socket_set_fast_reuse(int fd)
 304 {
 305     int val = 1, ret;
 306
 307     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 308                      (const char *)&val, sizeof(val));
 309
 310     assert(ret == 0);
 311
 312     return ret;
 313 }
 314
 315 void qemu_set_cloexec(int fd)
 316 {
 317     int f;
 318     f = fcntl(fd, F_GETFD);
 319     assert(f != -1);
 320     f = fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 321     assert(f != -1);
 322 }
 323
 324 /*
 325  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 326  */
 327 int qemu_pipe(int pipefd[2])
 328 {
 329     int ret;
 330
 331 #ifdef CONFIG_PIPE2
 332     ret = pipe2(pipefd, O_CLOEXEC);
 333     if (ret != -1 || errno != ENOSYS) {
 334         return ret;
 335     }
 336 #endif
 337     ret = pipe(pipefd);
 338     if (ret == 0) {
 339         qemu_set_cloexec(pipefd[0]);
 340         qemu_set_cloexec(pipefd[1]);
 341     }
 342
 343     return ret;
 344 }
 345
 346 char *
 347 qemu_get_local_state_pathname(const char *relative_pathname)
 348 {
 349     g_autofree char *dir = g_strdup_printf("%s/%s",
 350                                            CONFIG_QEMU_LOCALSTATEDIR,
 351                                            relative_pathname);
 352     return get_relocated_path(dir);
 353 }
 354
 355 void qemu_set_tty_echo(int fd, bool echo)
 356 {
 357     struct termios tty;
 358
 359     tcgetattr(fd, &tty);
 360
 361     if (echo) {
 362         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 363     } else {
 364         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 365     }
 366
 367     tcsetattr(fd, TCSANOW, &tty);
 368 }
 369
 370 static const char *exec_dir;
 371
 372 void qemu_init_exec_dir(const char *argv0)
 373 {
 374     char *p = NULL;
 375     char buf[PATH_MAX];
 376
 377     if (exec_dir) {
 378         return;
 379     }
 380
 381 #if defined(__linux__)
 382     {
 383         int len;
 384         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 385         if (len > 0) {
 386             buf[len] = 0;
 387             p = buf;
 388         }
 389     }
 390 #elif defined(__FreeBSD__) \
 391       || (defined(__NetBSD__) && defined(KERN_PROC_PATHNAME))
 392     {
 393 #if defined(__FreeBSD__)
 394         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 395 #else
 396         static int mib[4] = {CTL_KERN, KERN_PROC_ARGS, -1, KERN_PROC_PATHNAME};
 397 #endif
 398         size_t len = sizeof(buf) - 1;
 399
 400         *buf = '\0';
 401         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 402             *buf) {
 403             buf[sizeof(buf) - 1] = '\0';
 404             p = buf;
 405         }
 406     }
 407 #elif defined(__APPLE__)
 408     {
 409         char fpath[PATH_MAX];
 410         uint32_t len = sizeof(fpath);
 411         if (_NSGetExecutablePath(fpath, &len) == 0) {
 412             p = realpath(fpath, buf);
 413             if (!p) {
 414                 return;
 415             }
 416         }
 417     }
 418 #elif defined(__HAIKU__)
 419     {
 420         image_info ii;
 421         int32_t c = 0;
 422
 423         *buf = '\0';
 424         while (get_next_image_info(0, &c, &ii) == B_OK) {
 425             if (ii.type == B_APP_IMAGE) {
 426                 strncpy(buf, ii.name, sizeof(buf));
 427                 buf[sizeof(buf) - 1] = 0;
 428                 p = buf;
 429                 break;
 430             }
 431         }
 432     }
 433 #endif
 434     /* If we don't have any way of figuring out the actual executable
 435        location then try argv[0].  */
 436     if (!p && argv0) {
 437         p = realpath(argv0, buf);
 438     }
 439     if (p) {
 440         exec_dir = g_path_get_dirname(p);
 441     } else {
 442         exec_dir = CONFIG_BINDIR;
 443     }
 444 }
 445
 446 const char *qemu_get_exec_dir(void)
 447 {
 448     return exec_dir;
 449 }
 450
 451 #ifdef CONFIG_LINUX
 452 static void sigbus_handler(int signal, siginfo_t *siginfo, void *ctx)
 453 #else /* CONFIG_LINUX */
 454 static void sigbus_handler(int signal)
 455 #endif /* CONFIG_LINUX */
 456 {
 457     int i;
 458
 459     if (sigbus_memset_context) {
 460         for (i = 0; i < sigbus_memset_context->num_threads; i++) {
 461             MemsetThread *thread = &sigbus_memset_context->threads[i];
 462
 463             if (qemu_thread_is_self(&thread->pgthread)) {
 464                 siglongjmp(thread->env, 1);
 465             }
 466         }
 467     }
 468
 469 #ifdef CONFIG_LINUX
 470     /*
 471      * We assume that the MCE SIGBUS handler could have been registered. We
 472      * should never receive BUS_MCEERR_AO on any of our threads, but only on
 473      * the main thread registered for PR_MCE_KILL_EARLY. Further, we should not
 474      * receive BUS_MCEERR_AR triggered by action of other threads on one of
 475      * our threads. So, no need to check for unrelated SIGBUS when seeing one
 476      * for our threads.
 477      *
 478      * We will forward to the MCE handler, which will either handle the SIGBUS
 479      * or reinstall the default SIGBUS handler and reraise the SIGBUS. The
 480      * default SIGBUS handler will crash the process, so we don't care.
 481      */
 482     if (sigbus_oldact.sa_flags & SA_SIGINFO) {
 483         sigbus_oldact.sa_sigaction(signal, siginfo, ctx);
 484         return;
 485     }
 486 #endif /* CONFIG_LINUX */
 487     warn_report("os_mem_prealloc: unrelated SIGBUS detected and ignored");
 488 }
 489
 490 static void *do_touch_pages(void *arg)
 491 {
 492     MemsetThread *memset_args = (MemsetThread *)arg;
 493     sigset_t set, oldset;
 494     int ret = 0;
 495
 496     /*
 497      * On Linux, the page faults from the loop below can cause mmap_sem
 498      * contention with allocation of the thread stacks.  Do not start
 499      * clearing until all threads have been created.
 500      */
 501     qemu_mutex_lock(&page_mutex);
 502     while (!memset_args->context->all_threads_created) {
 503         qemu_cond_wait(&page_cond, &page_mutex);
 504     }
 505     qemu_mutex_unlock(&page_mutex);
 506
 507     /* unblock SIGBUS */
 508     sigemptyset(&set);
 509     sigaddset(&set, SIGBUS);
 510     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 511
 512     if (sigsetjmp(memset_args->env, 1)) {
 513         ret = -EFAULT;
 514     } else {
 515         char *addr = memset_args->addr;
 516         size_t numpages = memset_args->numpages;
 517         size_t hpagesize = memset_args->hpagesize;
 518         size_t i;
 519         for (i = 0; i < numpages; i++) {
 520             /*
 521              * Read & write back the same value, so we don't
 522              * corrupt existing user/app data that might be
 523              * stored.
 524              *
 525              * 'volatile' to stop compiler optimizing this away
 526              * to a no-op
 527              */
 528             *(volatile char *)addr = *addr;
 529             addr += hpagesize;
 530         }
 531     }
 532     pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 533     return (void *)(uintptr_t)ret;
 534 }
 535
 536 static void *do_madv_populate_write_pages(void *arg)
 537 {
 538     MemsetThread *memset_args = (MemsetThread *)arg;
 539     const size_t size = memset_args->numpages * memset_args->hpagesize;
 540     char * const addr = memset_args->addr;
 541     int ret = 0;
 542
 543     /* See do_touch_pages(). */
 544     qemu_mutex_lock(&page_mutex);
 545     while (!memset_args->context->all_threads_created) {
 546         qemu_cond_wait(&page_cond, &page_mutex);
 547     }
 548     qemu_mutex_unlock(&page_mutex);
 549
 550     if (size && qemu_madvise(addr, size, QEMU_MADV_POPULATE_WRITE)) {
 551         ret = -errno;
 552     }
 553     return (void *)(uintptr_t)ret;
 554 }
 555
 556 static inline int get_memset_num_threads(size_t hpagesize, size_t numpages,
 557                                          int smp_cpus)
 558 {
 559     long host_procs = sysconf(_SC_NPROCESSORS_ONLN);
 560     int ret = 1;
 561
 562     if (host_procs > 0) {
 563         ret = MIN(MIN(host_procs, MAX_MEM_PREALLOC_THREAD_COUNT), smp_cpus);
 564     }
 565
 566     /* Especially with gigantic pages, don't create more threads than pages. */
 567     ret = MIN(ret, numpages);
 568     /* Don't start threads to prealloc comparatively little memory. */
 569     ret = MIN(ret, MAX(1, hpagesize * numpages / (64 * MiB)));
 570
 571     /* In case sysconf() fails, we fall back to single threaded */
 572     return ret;
 573 }
 574
 575 static int touch_all_pages(char *area, size_t hpagesize, size_t numpages,
 576                            int smp_cpus, bool use_madv_populate_write)
 577 {
 578     static gsize initialized = 0;
 579     MemsetContext context = {
 580         .num_threads = get_memset_num_threads(hpagesize, numpages, smp_cpus),
 581     };
 582     size_t numpages_per_thread, leftover;
 583     void *(*touch_fn)(void *);
 584     int ret = 0, i = 0;
 585     char *addr = area;
 586
 587     if (g_once_init_enter(&initialized)) {
 588         qemu_mutex_init(&page_mutex);
 589         qemu_cond_init(&page_cond);
 590         g_once_init_leave(&initialized, 1);
 591     }
 592
 593     if (use_madv_populate_write) {
 594         /* Avoid creating a single thread for MADV_POPULATE_WRITE */
 595         if (context.num_threads == 1) {
 596             if (qemu_madvise(area, hpagesize * numpages,
 597                              QEMU_MADV_POPULATE_WRITE)) {
 598                 return -errno;
 599             }
 600             return 0;
 601         }
 602         touch_fn = do_madv_populate_write_pages;
 603     } else {
 604         touch_fn = do_touch_pages;
 605     }
 606
 607     context.threads = g_new0(MemsetThread, context.num_threads);
 608     numpages_per_thread = numpages / context.num_threads;
 609     leftover = numpages % context.num_threads;
 610     for (i = 0; i < context.num_threads; i++) {
 611         context.threads[i].addr = addr;
 612         context.threads[i].numpages = numpages_per_thread + (i < leftover);
 613         context.threads[i].hpagesize = hpagesize;
 614         context.threads[i].context = &context;
 615         qemu_thread_create(&context.threads[i].pgthread, "touch_pages",
 616                            touch_fn, &context.threads[i],
 617                            QEMU_THREAD_JOINABLE);
 618         addr += context.threads[i].numpages * hpagesize;
 619     }
 620
 621     if (!use_madv_populate_write) {
 622         sigbus_memset_context = &context;
 623     }
 624
 625     qemu_mutex_lock(&page_mutex);
 626     context.all_threads_created = true;
 627     qemu_cond_broadcast(&page_cond);
 628     qemu_mutex_unlock(&page_mutex);
 629
 630     for (i = 0; i < context.num_threads; i++) {
 631         int tmp = (uintptr_t)qemu_thread_join(&context.threads[i].pgthread);
 632
 633         if (tmp) {
 634             ret = tmp;
 635         }
 636     }
 637
 638     if (!use_madv_populate_write) {
 639         sigbus_memset_context = NULL;
 640     }
 641     g_free(context.threads);
 642
 643     return ret;
 644 }
 645
 646 static bool madv_populate_write_possible(char *area, size_t pagesize)
 647 {
 648     return !qemu_madvise(area, pagesize, QEMU_MADV_POPULATE_WRITE) ||
 649            errno != EINVAL;
 650 }
 651
 652 void os_mem_prealloc(int fd, char *area, size_t memory, int smp_cpus,
 653                      Error **errp)
 654 {
 655     static gsize initialized;
 656     int ret;
 657     size_t hpagesize = qemu_fd_getpagesize(fd);
 658     size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 659     bool use_madv_populate_write;
 660     struct sigaction act;
 661
 662     /*
 663      * Sense on every invocation, as MADV_POPULATE_WRITE cannot be used for
 664      * some special mappings, such as mapping /dev/mem.
 665      */
 666     use_madv_populate_write = madv_populate_write_possible(area, hpagesize);
 667
 668     if (!use_madv_populate_write) {
 669         if (g_once_init_enter(&initialized)) {
 670             qemu_mutex_init(&sigbus_mutex);
 671             g_once_init_leave(&initialized, 1);
 672         }
 673
 674         qemu_mutex_lock(&sigbus_mutex);
 675         memset(&act, 0, sizeof(act));
 676 #ifdef CONFIG_LINUX
 677         act.sa_sigaction = &sigbus_handler;
 678         act.sa_flags = SA_SIGINFO;
 679 #else /* CONFIG_LINUX */
 680         act.sa_handler = &sigbus_handler;
 681         act.sa_flags = 0;
 682 #endif /* CONFIG_LINUX */
 683
 684         ret = sigaction(SIGBUS, &act, &sigbus_oldact);
 685         if (ret) {
 686             qemu_mutex_unlock(&sigbus_mutex);
 687             error_setg_errno(errp, errno,
 688                 "os_mem_prealloc: failed to install signal handler");
 689             return;
 690         }
 691     }
 692
 693     /* touch pages simultaneously */
 694     ret = touch_all_pages(area, hpagesize, numpages, smp_cpus,
 695                           use_madv_populate_write);
 696     if (ret) {
 697         error_setg_errno(errp, -ret,
 698                          "os_mem_prealloc: preallocating memory failed");
 699     }
 700
 701     if (!use_madv_populate_write) {
 702         ret = sigaction(SIGBUS, &sigbus_oldact, NULL);
 703         if (ret) {
 704             /* Terminate QEMU since it can't recover from error */
 705             perror("os_mem_prealloc: failed to reinstall signal handler");
 706             exit(1);
 707         }
 708         qemu_mutex_unlock(&sigbus_mutex);
 709     }
 710 }
 711
 712 char *qemu_get_pid_name(pid_t pid)
 713 {
 714     char *name = NULL;
 715
 716 #if defined(__FreeBSD__)
 717     /* BSDs don't have /proc, but they provide a nice substitute */
 718     struct kinfo_proc *proc = kinfo_getproc(pid);
 719
 720     if (proc) {
 721         name = g_strdup(proc->ki_comm);
 722         free(proc);
 723     }
 724 #else
 725     /* Assume a system with reasonable procfs */
 726     char *pid_path;
 727     size_t len;
 728
 729     pid_path = g_strdup_printf("/proc/%d/cmdline", pid);
 730     g_file_get_contents(pid_path, &name, &len, NULL);
 731     g_free(pid_path);
 732 #endif
 733
 734     return name;
 735 }
 736
 737
 738 pid_t qemu_fork(Error **errp)
 739 {
 740     sigset_t oldmask, newmask;
 741     struct sigaction sig_action;
 742     int saved_errno;
 743     pid_t pid;
 744
 745     /*
 746      * Need to block signals now, so that child process can safely
 747      * kill off caller's signal handlers without a race.
 748      */
 749     sigfillset(&newmask);
 750     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 751         error_setg_errno(errp, errno,
 752                          "cannot block signals");
 753         return -1;
 754     }
 755
 756     pid = fork();
 757     saved_errno = errno;
 758
 759     if (pid < 0) {
 760         /* attempt to restore signal mask, but ignore failure, to
 761          * avoid obscuring the fork failure */
 762         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 763         error_setg_errno(errp, saved_errno,
 764                          "cannot fork child process");
 765         errno = saved_errno;
 766         return -1;
 767     } else if (pid) {
 768         /* parent process */
 769
 770         /* Restore our original signal mask now that the child is
 771          * safely running. Only documented failures are EFAULT (not
 772          * possible, since we are using just-grabbed mask) or EINVAL
 773          * (not possible, since we are using correct arguments).  */
 774         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 775     } else {
 776         /* child process */
 777         size_t i;
 778
 779         /* Clear out all signal handlers from parent so nothing
 780          * unexpected can happen in our child once we unblock
 781          * signals */
 782         sig_action.sa_handler = SIG_DFL;
 783         sig_action.sa_flags = 0;
 784         sigemptyset(&sig_action.sa_mask);
 785
 786         for (i = 1; i < NSIG; i++) {
 787             /* Only possible errors are EFAULT or EINVAL The former
 788              * won't happen, the latter we expect, so no need to check
 789              * return value */
 790             (void)sigaction(i, &sig_action, NULL);
 791         }
 792
 793         /* Unmask all signals in child, since we've no idea what the
 794          * caller's done with their signal mask and don't want to
 795          * propagate that to children */
 796         sigemptyset(&newmask);
 797         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 798             Error *local_err = NULL;
 799             error_setg_errno(&local_err, errno,
 800                              "cannot unblock signals");
 801             error_report_err(local_err);
 802             _exit(1);
 803         }
 804     }
 805     return pid;
 806 }
 807
 808 void *qemu_alloc_stack(size_t *sz)
 809 {
 810     void *ptr, *guardpage;
 811     int flags;
 812 #ifdef CONFIG_DEBUG_STACK_USAGE
 813     void *ptr2;
 814 #endif
 815     size_t pagesz = qemu_real_host_page_size;
 816 #ifdef _SC_THREAD_STACK_MIN
 817     /* avoid stacks smaller than _SC_THREAD_STACK_MIN */
 818     long min_stack_sz = sysconf(_SC_THREAD_STACK_MIN);
 819     *sz = MAX(MAX(min_stack_sz, 0), *sz);
 820 #endif
 821     /* adjust stack size to a multiple of the page size */
 822     *sz = ROUND_UP(*sz, pagesz);
 823     /* allocate one extra page for the guard page */
 824     *sz += pagesz;
 825
 826     flags = MAP_PRIVATE | MAP_ANONYMOUS;
 827 #if defined(MAP_STACK) && defined(__OpenBSD__)
 828     /* Only enable MAP_STACK on OpenBSD. Other OS's such as
 829      * Linux/FreeBSD/NetBSD have a flag with the same name
 830      * but have differing functionality. OpenBSD will SEGV
 831      * if it spots execution with a stack pointer pointing
 832      * at memory that was not allocated with MAP_STACK.
 833      */
 834     flags |= MAP_STACK;
 835 #endif
 836
 837     ptr = mmap(NULL, *sz, PROT_READ | PROT_WRITE, flags, -1, 0);
 838     if (ptr == MAP_FAILED) {
 839         perror("failed to allocate memory for stack");
 840         abort();
 841     }
 842
 843 #if defined(HOST_IA64)
 844     /* separate register stack */
 845     guardpage = ptr + (((*sz - pagesz) / 2) & ~pagesz);
 846 #elif defined(HOST_HPPA)
 847     /* stack grows up */
 848     guardpage = ptr + *sz - pagesz;
 849 #else
 850     /* stack grows down */
 851     guardpage = ptr;
 852 #endif
 853     if (mprotect(guardpage, pagesz, PROT_NONE) != 0) {
 854         perror("failed to set up stack guard page");
 855         abort();
 856     }
 857
 858 #ifdef CONFIG_DEBUG_STACK_USAGE
 859     for (ptr2 = ptr + pagesz; ptr2 < ptr + *sz; ptr2 += sizeof(uint32_t)) {
 860         *(uint32_t *)ptr2 = 0xdeadbeaf;
 861     }
 862 #endif
 863
 864     return ptr;
 865 }
 866
 867 #ifdef CONFIG_DEBUG_STACK_USAGE
 868 static __thread unsigned int max_stack_usage;
 869 #endif
 870
 871 void qemu_free_stack(void *stack, size_t sz)
 872 {
 873 #ifdef CONFIG_DEBUG_STACK_USAGE
 874     unsigned int usage;
 875     void *ptr;
 876
 877     for (ptr = stack + qemu_real_host_page_size; ptr < stack + sz;
 878          ptr += sizeof(uint32_t)) {
 879         if (*(uint32_t *)ptr != 0xdeadbeaf) {
 880             break;
 881         }
 882     }
 883     usage = sz - (uintptr_t) (ptr - stack);
 884     if (usage > max_stack_usage) {
 885         error_report("thread %d max stack usage increased from %u to %u",
 886                      qemu_get_thread_id(), max_stack_usage, usage);
 887         max_stack_usage = usage;
 888     }
 889 #endif
 890
 891     munmap(stack, sz);
 892 }
 893
 894 /*
 895  * Disable CFI checks.
 896  * We are going to call a signal hander directly. Such handler may or may not
 897  * have been defined in our binary, so there's no guarantee that the pointer
 898  * used to set the handler is a cfi-valid pointer. Since the handlers are
 899  * stored in kernel memory, changing the handler to an attacker-defined
 900  * function requires being able to call a sigaction() syscall,
 901  * which is not as easy as overwriting a pointer in memory.
 902  */
 903 QEMU_DISABLE_CFI
 904 void sigaction_invoke(struct sigaction *action,
 905                       struct qemu_signalfd_siginfo *info)
 906 {
 907     siginfo_t si = {};
 908     si.si_signo = info->ssi_signo;
 909     si.si_errno = info->ssi_errno;
 910     si.si_code = info->ssi_code;
 911
 912     /* Convert the minimal set of fields defined by POSIX.
 913      * Positive si_code values are reserved for kernel-generated
 914      * signals, where the valid siginfo fields are determined by
 915      * the signal number.  But according to POSIX, it is unspecified
 916      * whether SI_USER and SI_QUEUE have values less than or equal to
 917      * zero.
 918      */
 919     if (info->ssi_code == SI_USER || info->ssi_code == SI_QUEUE ||
 920         info->ssi_code <= 0) {
 921         /* SIGTERM, etc.  */
 922         si.si_pid = info->ssi_pid;
 923         si.si_uid = info->ssi_uid;
 924     } else if (info->ssi_signo == SIGILL || info->ssi_signo == SIGFPE ||
 925                info->ssi_signo == SIGSEGV || info->ssi_signo == SIGBUS) {
 926         si.si_addr = (void *)(uintptr_t)info->ssi_addr;
 927     } else if (info->ssi_signo == SIGCHLD) {
 928         si.si_pid = info->ssi_pid;
 929         si.si_status = info->ssi_status;
 930         si.si_uid = info->ssi_uid;
 931     }
 932     action->sa_sigaction(info->ssi_signo, &si, NULL);
 933 }
 934
 935 #ifndef HOST_NAME_MAX
 936 # ifdef _POSIX_HOST_NAME_MAX
 937 #  define HOST_NAME_MAX _POSIX_HOST_NAME_MAX
 938 # else
 939 #  define HOST_NAME_MAX 255
 940 # endif
 941 #endif
 942
 943 char *qemu_get_host_name(Error **errp)
 944 {
 945     long len = -1;
 946     g_autofree char *hostname = NULL;
 947
 948 #ifdef _SC_HOST_NAME_MAX
 949     len = sysconf(_SC_HOST_NAME_MAX);
 950 #endif /* _SC_HOST_NAME_MAX */
 951
 952     if (len < 0) {
 953         len = HOST_NAME_MAX;
 954     }
 955
 956     /* Unfortunately, gethostname() below does not guarantee a
 957      * NULL terminated string. Therefore, allocate one byte more
 958      * to be sure. */
 959     hostname = g_new0(char, len + 1);
 960
 961     if (gethostname(hostname, len) < 0) {
 962         error_setg_errno(errp, errno,
 963                          "cannot get hostname");
 964         return NULL;
 965     }
 966
 967     return g_steal_pointer(&hostname);
 968 }
 969
 970 size_t qemu_get_host_physmem(void)
 971 {
 972 #ifdef _SC_PHYS_PAGES
 973     long pages = sysconf(_SC_PHYS_PAGES);
 974     if (pages > 0) {
 975         if (pages > SIZE_MAX / qemu_real_host_page_size) {
 976             return SIZE_MAX;
 977         } else {
 978             return pages * qemu_real_host_page_size;
 979         }
 980     }
 981 #endif
 982     return 0;
 983 }