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 /* The following block of code temporarily renames the daemon() function so the
  30    compiler does not see the warning associated with it in stdlib.h on OSX */
  31 #ifdef __APPLE__
  32 #define daemon qemu_fake_daemon_function
  33 #include <stdlib.h>
  34 #undef daemon
  35 extern int daemon(int, int);
  36 #endif
  37
  38 #if defined(__linux__) && (defined(__x86_64__) || defined(__arm__))
  39    /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
  40       Valgrind does not support alignments larger than 1 MiB,
  41       therefore we need special code which handles running on Valgrind. */
  42 #  define QEMU_VMALLOC_ALIGN (512 * 4096)
  43 #elif defined(__linux__) && defined(__s390x__)
  44    /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
  45 #  define QEMU_VMALLOC_ALIGN (256 * 4096)
  46 #else
  47 #  define QEMU_VMALLOC_ALIGN getpagesize()
  48 #endif
  49
  50 #include "qemu/osdep.h"
  51 #include <termios.h>
  52 #include <termios.h>
  53
  54 #include <glib/gprintf.h>
  55
  56 #include "sysemu/sysemu.h"
  57 #include "trace.h"
  58 #include "qemu/sockets.h"
  59 #include <sys/mman.h>
  60 #include <libgen.h>
  61 #include <setjmp.h>
  62 #include <sys/signal.h>
  63
  64 #ifdef CONFIG_LINUX
  65 #include <sys/syscall.h>
  66 #endif
  67
  68 #ifdef __FreeBSD__
  69 #include <sys/sysctl.h>
  70 #endif
  71
  72 #include <qemu/mmap-alloc.h>
  73
  74 int qemu_get_thread_id(void)
  75 {
  76 #if defined(__linux__)
  77     return syscall(SYS_gettid);
  78 #else
  79     return getpid();
  80 #endif
  81 }
  82
  83 int qemu_daemon(int nochdir, int noclose)
  84 {
  85     return daemon(nochdir, noclose);
  86 }
  87
  88 void *qemu_oom_check(void *ptr)
  89 {
  90     if (ptr == NULL) {
  91         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
  92         abort();
  93     }
  94     return ptr;
  95 }
  96
  97 void *qemu_try_memalign(size_t alignment, size_t size)
  98 {
  99     void *ptr;
 100
 101     if (alignment < sizeof(void*)) {
 102         alignment = sizeof(void*);
 103     }
 104
 105 #if defined(_POSIX_C_SOURCE) && !defined(__sun__)
 106     int ret;
 107     ret = posix_memalign(&ptr, alignment, size);
 108     if (ret != 0) {
 109         errno = ret;
 110         ptr = NULL;
 111     }
 112 #elif defined(CONFIG_BSD)
 113     ptr = valloc(size);
 114 #else
 115     ptr = memalign(alignment, size);
 116 #endif
 117     trace_qemu_memalign(alignment, size, ptr);
 118     return ptr;
 119 }
 120
 121 void *qemu_memalign(size_t alignment, size_t size)
 122 {
 123     return qemu_oom_check(qemu_try_memalign(alignment, size));
 124 }
 125
 126 /* alloc shared memory pages */
 127 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment)
 128 {
 129     size_t align = QEMU_VMALLOC_ALIGN;
 130     void *ptr = qemu_ram_mmap(-1, size, align, false);
 131
 132     if (ptr == MAP_FAILED) {
 133         return NULL;
 134     }
 135
 136     if (alignment) {
 137         *alignment = align;
 138     }
 139
 140     trace_qemu_anon_ram_alloc(size, ptr);
 141     return ptr;
 142 }
 143
 144 void qemu_vfree(void *ptr)
 145 {
 146     trace_qemu_vfree(ptr);
 147     free(ptr);
 148 }
 149
 150 void qemu_anon_ram_free(void *ptr, size_t size)
 151 {
 152     trace_qemu_anon_ram_free(ptr, size);
 153     qemu_ram_munmap(ptr, size);
 154 }
 155
 156 void qemu_set_block(int fd)
 157 {
 158     int f;
 159     f = fcntl(fd, F_GETFL);
 160     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 161 }
 162
 163 void qemu_set_nonblock(int fd)
 164 {
 165     int f;
 166     f = fcntl(fd, F_GETFL);
 167     fcntl(fd, F_SETFL, f | O_NONBLOCK);
 168 }
 169
 170 int socket_set_fast_reuse(int fd)
 171 {
 172     int val = 1, ret;
 173
 174     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 175                      (const char *)&val, sizeof(val));
 176
 177     assert(ret == 0);
 178
 179     return ret;
 180 }
 181
 182 void qemu_set_cloexec(int fd)
 183 {
 184     int f;
 185     f = fcntl(fd, F_GETFD);
 186     fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 187 }
 188
 189 /*
 190  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 191  */
 192 int qemu_pipe(int pipefd[2])
 193 {
 194     int ret;
 195
 196 #ifdef CONFIG_PIPE2
 197     ret = pipe2(pipefd, O_CLOEXEC);
 198     if (ret != -1 || errno != ENOSYS) {
 199         return ret;
 200     }
 201 #endif
 202     ret = pipe(pipefd);
 203     if (ret == 0) {
 204         qemu_set_cloexec(pipefd[0]);
 205         qemu_set_cloexec(pipefd[1]);
 206     }
 207
 208     return ret;
 209 }
 210
 211 int qemu_utimens(const char *path, const struct timespec *times)
 212 {
 213     struct timeval tv[2], tv_now;
 214     struct stat st;
 215     int i;
 216 #ifdef CONFIG_UTIMENSAT
 217     int ret;
 218
 219     ret = utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW);
 220     if (ret != -1 || errno != ENOSYS) {
 221         return ret;
 222     }
 223 #endif
 224     /* Fallback: use utimes() instead of utimensat() */
 225
 226     /* happy if special cases */
 227     if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) {
 228         return 0;
 229     }
 230     if (times[0].tv_nsec == UTIME_NOW && times[1].tv_nsec == UTIME_NOW) {
 231         return utimes(path, NULL);
 232     }
 233
 234     /* prepare for hard cases */
 235     if (times[0].tv_nsec == UTIME_NOW || times[1].tv_nsec == UTIME_NOW) {
 236         gettimeofday(&tv_now, NULL);
 237     }
 238     if (times[0].tv_nsec == UTIME_OMIT || times[1].tv_nsec == UTIME_OMIT) {
 239         stat(path, &st);
 240     }
 241
 242     for (i = 0; i < 2; i++) {
 243         if (times[i].tv_nsec == UTIME_NOW) {
 244             tv[i].tv_sec = tv_now.tv_sec;
 245             tv[i].tv_usec = tv_now.tv_usec;
 246         } else if (times[i].tv_nsec == UTIME_OMIT) {
 247             tv[i].tv_sec = (i == 0) ? st.st_atime : st.st_mtime;
 248             tv[i].tv_usec = 0;
 249         } else {
 250             tv[i].tv_sec = times[i].tv_sec;
 251             tv[i].tv_usec = times[i].tv_nsec / 1000;
 252         }
 253     }
 254
 255     return utimes(path, &tv[0]);
 256 }
 257
 258 char *
 259 qemu_get_local_state_pathname(const char *relative_pathname)
 260 {
 261     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 262                            relative_pathname);
 263 }
 264
 265 void qemu_set_tty_echo(int fd, bool echo)
 266 {
 267     struct termios tty;
 268
 269     tcgetattr(fd, &tty);
 270
 271     if (echo) {
 272         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 273     } else {
 274         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 275     }
 276
 277     tcsetattr(fd, TCSANOW, &tty);
 278 }
 279
 280 static char exec_dir[PATH_MAX];
 281
 282 void qemu_init_exec_dir(const char *argv0)
 283 {
 284     char *dir;
 285     char *p = NULL;
 286     char buf[PATH_MAX];
 287
 288     assert(!exec_dir[0]);
 289
 290 #if defined(__linux__)
 291     {
 292         int len;
 293         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 294         if (len > 0) {
 295             buf[len] = 0;
 296             p = buf;
 297         }
 298     }
 299 #elif defined(__FreeBSD__)
 300     {
 301         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 302         size_t len = sizeof(buf) - 1;
 303
 304         *buf = '\0';
 305         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 306             *buf) {
 307             buf[sizeof(buf) - 1] = '\0';
 308             p = buf;
 309         }
 310     }
 311 #endif
 312     /* If we don't have any way of figuring out the actual executable
 313        location then try argv[0].  */
 314     if (!p) {
 315         if (!argv0) {
 316             return;
 317         }
 318         p = realpath(argv0, buf);
 319         if (!p) {
 320             return;
 321         }
 322     }
 323     dir = dirname(p);
 324
 325     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 326 }
 327
 328 char *qemu_get_exec_dir(void)
 329 {
 330     return g_strdup(exec_dir);
 331 }
 332
 333 static sigjmp_buf sigjump;
 334
 335 static void sigbus_handler(int signal)
 336 {
 337     siglongjmp(sigjump, 1);
 338 }
 339
 340 void os_mem_prealloc(int fd, char *area, size_t memory)
 341 {
 342     int ret;
 343     struct sigaction act, oldact;
 344     sigset_t set, oldset;
 345
 346     memset(&act, 0, sizeof(act));
 347     act.sa_handler = &sigbus_handler;
 348     act.sa_flags = 0;
 349
 350     ret = sigaction(SIGBUS, &act, &oldact);
 351     if (ret) {
 352         perror("os_mem_prealloc: failed to install signal handler");
 353         exit(1);
 354     }
 355
 356     /* unblock SIGBUS */
 357     sigemptyset(&set);
 358     sigaddset(&set, SIGBUS);
 359     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 360
 361     if (sigsetjmp(sigjump, 1)) {
 362         fprintf(stderr, "os_mem_prealloc: Insufficient free host memory "
 363                         "pages available to allocate guest RAM\n");
 364         exit(1);
 365     } else {
 366         int i;
 367         size_t hpagesize = qemu_fd_getpagesize(fd);
 368         size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 369
 370         /* MAP_POPULATE silently ignores failures */
 371         for (i = 0; i < numpages; i++) {
 372             memset(area + (hpagesize * i), 0, 1);
 373         }
 374
 375         ret = sigaction(SIGBUS, &oldact, NULL);
 376         if (ret) {
 377             perror("os_mem_prealloc: failed to reinstall signal handler");
 378             exit(1);
 379         }
 380
 381         pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 382     }
 383 }
 384
 385
 386 static struct termios oldtty;
 387
 388 static void term_exit(void)
 389 {
 390     tcsetattr(0, TCSANOW, &oldtty);
 391 }
 392
 393 static void term_init(void)
 394 {
 395     struct termios tty;
 396
 397     tcgetattr(0, &tty);
 398     oldtty = tty;
 399
 400     tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
 401                           |INLCR|IGNCR|ICRNL|IXON);
 402     tty.c_oflag |= OPOST;
 403     tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
 404     tty.c_cflag &= ~(CSIZE|PARENB);
 405     tty.c_cflag |= CS8;
 406     tty.c_cc[VMIN] = 1;
 407     tty.c_cc[VTIME] = 0;
 408
 409     tcsetattr(0, TCSANOW, &tty);
 410
 411     atexit(term_exit);
 412 }
 413
 414 int qemu_read_password(char *buf, int buf_size)
 415 {
 416     uint8_t ch;
 417     int i, ret;
 418
 419     printf("password: ");
 420     fflush(stdout);
 421     term_init();
 422     i = 0;
 423     for (;;) {
 424         ret = read(0, &ch, 1);
 425         if (ret == -1) {
 426             if (errno == EAGAIN || errno == EINTR) {
 427                 continue;
 428             } else {
 429                 break;
 430             }
 431         } else if (ret == 0) {
 432             ret = -1;
 433             break;
 434         } else {
 435             if (ch == '\r' ||
 436                 ch == '\n') {
 437                 ret = 0;
 438                 break;
 439             }
 440             if (i < (buf_size - 1)) {
 441                 buf[i++] = ch;
 442             }
 443         }
 444     }
 445     term_exit();
 446     buf[i] = '\0';
 447     printf("\n");
 448     return ret;
 449 }
 450
 451
 452 pid_t qemu_fork(Error **errp)
 453 {
 454     sigset_t oldmask, newmask;
 455     struct sigaction sig_action;
 456     int saved_errno;
 457     pid_t pid;
 458
 459     /*
 460      * Need to block signals now, so that child process can safely
 461      * kill off caller's signal handlers without a race.
 462      */
 463     sigfillset(&newmask);
 464     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 465         error_setg_errno(errp, errno,
 466                          "cannot block signals");
 467         return -1;
 468     }
 469
 470     pid = fork();
 471     saved_errno = errno;
 472
 473     if (pid < 0) {
 474         /* attempt to restore signal mask, but ignore failure, to
 475          * avoid obscuring the fork failure */
 476         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 477         error_setg_errno(errp, saved_errno,
 478                          "cannot fork child process");
 479         errno = saved_errno;
 480         return -1;
 481     } else if (pid) {
 482         /* parent process */
 483
 484         /* Restore our original signal mask now that the child is
 485          * safely running. Only documented failures are EFAULT (not
 486          * possible, since we are using just-grabbed mask) or EINVAL
 487          * (not possible, since we are using correct arguments).  */
 488         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 489     } else {
 490         /* child process */
 491         size_t i;
 492
 493         /* Clear out all signal handlers from parent so nothing
 494          * unexpected can happen in our child once we unblock
 495          * signals */
 496         sig_action.sa_handler = SIG_DFL;
 497         sig_action.sa_flags = 0;
 498         sigemptyset(&sig_action.sa_mask);
 499
 500         for (i = 1; i < NSIG; i++) {
 501             /* Only possible errors are EFAULT or EINVAL The former
 502              * won't happen, the latter we expect, so no need to check
 503              * return value */
 504             (void)sigaction(i, &sig_action, NULL);
 505         }
 506
 507         /* Unmask all signals in child, since we've no idea what the
 508          * caller's done with their signal mask and don't want to
 509          * propagate that to children */
 510         sigemptyset(&newmask);
 511         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 512             Error *local_err = NULL;
 513             error_setg_errno(&local_err, errno,
 514                              "cannot unblock signals");
 515             error_report_err(local_err);
 516             _exit(1);
 517         }
 518     }
 519     return pid;
 520 }