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 #if defined(__linux__) && \
  30     (defined(__x86_64__) || defined(__arm__) || defined(__aarch64__))
  31    /* Use 2 MiB alignment so transparent hugepages can be used by KVM.
  32       Valgrind does not support alignments larger than 1 MiB,
  33       therefore we need special code which handles running on Valgrind. */
  34 #  define QEMU_VMALLOC_ALIGN (512 * 4096)
  35 #elif defined(__linux__) && defined(__s390x__)
  36    /* Use 1 MiB (segment size) alignment so gmap can be used by KVM. */
  37 #  define QEMU_VMALLOC_ALIGN (256 * 4096)
  38 #else
  39 #  define QEMU_VMALLOC_ALIGN getpagesize()
  40 #endif
  41
  42 #include "qemu/osdep.h"
  43 #include <termios.h>
  44 #include <termios.h>
  45
  46 #include <glib/gprintf.h>
  47
  48 #include "sysemu/sysemu.h"
  49 #include "trace.h"
  50 #include "qapi/error.h"
  51 #include "qemu/sockets.h"
  52 #include <sys/mman.h>
  53 #include <libgen.h>
  54 #include <sys/signal.h>
  55 #include "qemu/cutils.h"
  56
  57 #ifdef CONFIG_LINUX
  58 #include <sys/syscall.h>
  59 #endif
  60
  61 #ifdef __FreeBSD__
  62 #include <sys/sysctl.h>
  63 #endif
  64
  65 #include <qemu/mmap-alloc.h>
  66
  67 int qemu_get_thread_id(void)
  68 {
  69 #if defined(__linux__)
  70     return syscall(SYS_gettid);
  71 #else
  72     return getpid();
  73 #endif
  74 }
  75
  76 int qemu_daemon(int nochdir, int noclose)
  77 {
  78     return daemon(nochdir, noclose);
  79 }
  80
  81 void *qemu_oom_check(void *ptr)
  82 {
  83     if (ptr == NULL) {
  84         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
  85         abort();
  86     }
  87     return ptr;
  88 }
  89
  90 void *qemu_try_memalign(size_t alignment, size_t size)
  91 {
  92     void *ptr;
  93
  94     if (alignment < sizeof(void*)) {
  95         alignment = sizeof(void*);
  96     }
  97
  98 #if defined(_POSIX_C_SOURCE) && !defined(__sun__)
  99     int ret;
 100     ret = posix_memalign(&ptr, alignment, size);
 101     if (ret != 0) {
 102         errno = ret;
 103         ptr = NULL;
 104     }
 105 #elif defined(CONFIG_BSD)
 106     ptr = valloc(size);
 107 #else
 108     ptr = memalign(alignment, size);
 109 #endif
 110     trace_qemu_memalign(alignment, size, ptr);
 111     return ptr;
 112 }
 113
 114 void *qemu_memalign(size_t alignment, size_t size)
 115 {
 116     return qemu_oom_check(qemu_try_memalign(alignment, size));
 117 }
 118
 119 /* alloc shared memory pages */
 120 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment)
 121 {
 122     size_t align = QEMU_VMALLOC_ALIGN;
 123     void *ptr = qemu_ram_mmap(-1, size, align, false);
 124
 125     if (ptr == MAP_FAILED) {
 126         return NULL;
 127     }
 128
 129     if (alignment) {
 130         *alignment = align;
 131     }
 132
 133     trace_qemu_anon_ram_alloc(size, ptr);
 134     return ptr;
 135 }
 136
 137 void qemu_vfree(void *ptr)
 138 {
 139     trace_qemu_vfree(ptr);
 140     free(ptr);
 141 }
 142
 143 void qemu_anon_ram_free(void *ptr, size_t size)
 144 {
 145     trace_qemu_anon_ram_free(ptr, size);
 146     qemu_ram_munmap(ptr, size);
 147 }
 148
 149 void qemu_set_block(int fd)
 150 {
 151     int f;
 152     f = fcntl(fd, F_GETFL);
 153     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 154 }
 155
 156 void qemu_set_nonblock(int fd)
 157 {
 158     int f;
 159     f = fcntl(fd, F_GETFL);
 160     fcntl(fd, F_SETFL, f | O_NONBLOCK);
 161 }
 162
 163 int socket_set_fast_reuse(int fd)
 164 {
 165     int val = 1, ret;
 166
 167     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 168                      (const char *)&val, sizeof(val));
 169
 170     assert(ret == 0);
 171
 172     return ret;
 173 }
 174
 175 void qemu_set_cloexec(int fd)
 176 {
 177     int f;
 178     f = fcntl(fd, F_GETFD);
 179     fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 180 }
 181
 182 /*
 183  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 184  */
 185 int qemu_pipe(int pipefd[2])
 186 {
 187     int ret;
 188
 189 #ifdef CONFIG_PIPE2
 190     ret = pipe2(pipefd, O_CLOEXEC);
 191     if (ret != -1 || errno != ENOSYS) {
 192         return ret;
 193     }
 194 #endif
 195     ret = pipe(pipefd);
 196     if (ret == 0) {
 197         qemu_set_cloexec(pipefd[0]);
 198         qemu_set_cloexec(pipefd[1]);
 199     }
 200
 201     return ret;
 202 }
 203
 204 int qemu_utimens(const char *path, const struct timespec *times)
 205 {
 206     struct timeval tv[2], tv_now;
 207     struct stat st;
 208     int i;
 209 #ifdef CONFIG_UTIMENSAT
 210     int ret;
 211
 212     ret = utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW);
 213     if (ret != -1 || errno != ENOSYS) {
 214         return ret;
 215     }
 216 #endif
 217     /* Fallback: use utimes() instead of utimensat() */
 218
 219     /* happy if special cases */
 220     if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) {
 221         return 0;
 222     }
 223     if (times[0].tv_nsec == UTIME_NOW && times[1].tv_nsec == UTIME_NOW) {
 224         return utimes(path, NULL);
 225     }
 226
 227     /* prepare for hard cases */
 228     if (times[0].tv_nsec == UTIME_NOW || times[1].tv_nsec == UTIME_NOW) {
 229         gettimeofday(&tv_now, NULL);
 230     }
 231     if (times[0].tv_nsec == UTIME_OMIT || times[1].tv_nsec == UTIME_OMIT) {
 232         stat(path, &st);
 233     }
 234
 235     for (i = 0; i < 2; i++) {
 236         if (times[i].tv_nsec == UTIME_NOW) {
 237             tv[i].tv_sec = tv_now.tv_sec;
 238             tv[i].tv_usec = tv_now.tv_usec;
 239         } else if (times[i].tv_nsec == UTIME_OMIT) {
 240             tv[i].tv_sec = (i == 0) ? st.st_atime : st.st_mtime;
 241             tv[i].tv_usec = 0;
 242         } else {
 243             tv[i].tv_sec = times[i].tv_sec;
 244             tv[i].tv_usec = times[i].tv_nsec / 1000;
 245         }
 246     }
 247
 248     return utimes(path, &tv[0]);
 249 }
 250
 251 char *
 252 qemu_get_local_state_pathname(const char *relative_pathname)
 253 {
 254     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 255                            relative_pathname);
 256 }
 257
 258 void qemu_set_tty_echo(int fd, bool echo)
 259 {
 260     struct termios tty;
 261
 262     tcgetattr(fd, &tty);
 263
 264     if (echo) {
 265         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 266     } else {
 267         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 268     }
 269
 270     tcsetattr(fd, TCSANOW, &tty);
 271 }
 272
 273 static char exec_dir[PATH_MAX];
 274
 275 void qemu_init_exec_dir(const char *argv0)
 276 {
 277     char *dir;
 278     char *p = NULL;
 279     char buf[PATH_MAX];
 280
 281     assert(!exec_dir[0]);
 282
 283 #if defined(__linux__)
 284     {
 285         int len;
 286         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 287         if (len > 0) {
 288             buf[len] = 0;
 289             p = buf;
 290         }
 291     }
 292 #elif defined(__FreeBSD__)
 293     {
 294         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 295         size_t len = sizeof(buf) - 1;
 296
 297         *buf = '\0';
 298         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 299             *buf) {
 300             buf[sizeof(buf) - 1] = '\0';
 301             p = buf;
 302         }
 303     }
 304 #endif
 305     /* If we don't have any way of figuring out the actual executable
 306        location then try argv[0].  */
 307     if (!p) {
 308         if (!argv0) {
 309             return;
 310         }
 311         p = realpath(argv0, buf);
 312         if (!p) {
 313             return;
 314         }
 315     }
 316     dir = dirname(p);
 317
 318     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 319 }
 320
 321 char *qemu_get_exec_dir(void)
 322 {
 323     return g_strdup(exec_dir);
 324 }
 325
 326 static sigjmp_buf sigjump;
 327
 328 static void sigbus_handler(int signal)
 329 {
 330     siglongjmp(sigjump, 1);
 331 }
 332
 333 void os_mem_prealloc(int fd, char *area, size_t memory)
 334 {
 335     int ret;
 336     struct sigaction act, oldact;
 337     sigset_t set, oldset;
 338
 339     memset(&act, 0, sizeof(act));
 340     act.sa_handler = &sigbus_handler;
 341     act.sa_flags = 0;
 342
 343     ret = sigaction(SIGBUS, &act, &oldact);
 344     if (ret) {
 345         perror("os_mem_prealloc: failed to install signal handler");
 346         exit(1);
 347     }
 348
 349     /* unblock SIGBUS */
 350     sigemptyset(&set);
 351     sigaddset(&set, SIGBUS);
 352     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 353
 354     if (sigsetjmp(sigjump, 1)) {
 355         fprintf(stderr, "os_mem_prealloc: Insufficient free host memory "
 356                         "pages available to allocate guest RAM\n");
 357         exit(1);
 358     } else {
 359         int i;
 360         size_t hpagesize = qemu_fd_getpagesize(fd);
 361         size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 362
 363         /* MAP_POPULATE silently ignores failures */
 364         for (i = 0; i < numpages; i++) {
 365             memset(area + (hpagesize * i), 0, 1);
 366         }
 367
 368         ret = sigaction(SIGBUS, &oldact, NULL);
 369         if (ret) {
 370             perror("os_mem_prealloc: failed to reinstall signal handler");
 371             exit(1);
 372         }
 373
 374         pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 375     }
 376 }
 377
 378
 379 static struct termios oldtty;
 380
 381 static void term_exit(void)
 382 {
 383     tcsetattr(0, TCSANOW, &oldtty);
 384 }
 385
 386 static void term_init(void)
 387 {
 388     struct termios tty;
 389
 390     tcgetattr(0, &tty);
 391     oldtty = tty;
 392
 393     tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
 394                           |INLCR|IGNCR|ICRNL|IXON);
 395     tty.c_oflag |= OPOST;
 396     tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
 397     tty.c_cflag &= ~(CSIZE|PARENB);
 398     tty.c_cflag |= CS8;
 399     tty.c_cc[VMIN] = 1;
 400     tty.c_cc[VTIME] = 0;
 401
 402     tcsetattr(0, TCSANOW, &tty);
 403
 404     atexit(term_exit);
 405 }
 406
 407 int qemu_read_password(char *buf, int buf_size)
 408 {
 409     uint8_t ch;
 410     int i, ret;
 411
 412     printf("password: ");
 413     fflush(stdout);
 414     term_init();
 415     i = 0;
 416     for (;;) {
 417         ret = read(0, &ch, 1);
 418         if (ret == -1) {
 419             if (errno == EAGAIN || errno == EINTR) {
 420                 continue;
 421             } else {
 422                 break;
 423             }
 424         } else if (ret == 0) {
 425             ret = -1;
 426             break;
 427         } else {
 428             if (ch == '\r' ||
 429                 ch == '\n') {
 430                 ret = 0;
 431                 break;
 432             }
 433             if (i < (buf_size - 1)) {
 434                 buf[i++] = ch;
 435             }
 436         }
 437     }
 438     term_exit();
 439     buf[i] = '\0';
 440     printf("\n");
 441     return ret;
 442 }
 443
 444
 445 pid_t qemu_fork(Error **errp)
 446 {
 447     sigset_t oldmask, newmask;
 448     struct sigaction sig_action;
 449     int saved_errno;
 450     pid_t pid;
 451
 452     /*
 453      * Need to block signals now, so that child process can safely
 454      * kill off caller's signal handlers without a race.
 455      */
 456     sigfillset(&newmask);
 457     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 458         error_setg_errno(errp, errno,
 459                          "cannot block signals");
 460         return -1;
 461     }
 462
 463     pid = fork();
 464     saved_errno = errno;
 465
 466     if (pid < 0) {
 467         /* attempt to restore signal mask, but ignore failure, to
 468          * avoid obscuring the fork failure */
 469         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 470         error_setg_errno(errp, saved_errno,
 471                          "cannot fork child process");
 472         errno = saved_errno;
 473         return -1;
 474     } else if (pid) {
 475         /* parent process */
 476
 477         /* Restore our original signal mask now that the child is
 478          * safely running. Only documented failures are EFAULT (not
 479          * possible, since we are using just-grabbed mask) or EINVAL
 480          * (not possible, since we are using correct arguments).  */
 481         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 482     } else {
 483         /* child process */
 484         size_t i;
 485
 486         /* Clear out all signal handlers from parent so nothing
 487          * unexpected can happen in our child once we unblock
 488          * signals */
 489         sig_action.sa_handler = SIG_DFL;
 490         sig_action.sa_flags = 0;
 491         sigemptyset(&sig_action.sa_mask);
 492
 493         for (i = 1; i < NSIG; i++) {
 494             /* Only possible errors are EFAULT or EINVAL The former
 495              * won't happen, the latter we expect, so no need to check
 496              * return value */
 497             (void)sigaction(i, &sig_action, NULL);
 498         }
 499
 500         /* Unmask all signals in child, since we've no idea what the
 501          * caller's done with their signal mask and don't want to
 502          * propagate that to children */
 503         sigemptyset(&newmask);
 504         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 505             Error *local_err = NULL;
 506             error_setg_errno(&local_err, errno,
 507                              "cannot unblock signals");
 508             error_report_err(local_err);
 509             _exit(1);
 510         }
 511     }
 512     return pid;
 513 }