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 <termios.h>
  51 #include <unistd.h>
  52 #include <termios.h>
  53
  54 #include <glib/gprintf.h>
  55
  56 #include "config-host.h"
  57 #include "sysemu/sysemu.h"
  58 #include "trace.h"
  59 #include "qemu/sockets.h"
  60 #include <sys/mman.h>
  61 #include <libgen.h>
  62 #include <setjmp.h>
  63 #include <sys/signal.h>
  64
  65 #ifdef CONFIG_LINUX
  66 #include <sys/syscall.h>
  67 #endif
  68
  69 #ifdef __FreeBSD__
  70 #include <sys/sysctl.h>
  71 #endif
  72
  73 #include <qemu/mmap-alloc.h>
  74
  75 int qemu_get_thread_id(void)
  76 {
  77 #if defined(__linux__)
  78     return syscall(SYS_gettid);
  79 #else
  80     return getpid();
  81 #endif
  82 }
  83
  84 int qemu_daemon(int nochdir, int noclose)
  85 {
  86     return daemon(nochdir, noclose);
  87 }
  88
  89 void *qemu_oom_check(void *ptr)
  90 {
  91     if (ptr == NULL) {
  92         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
  93         abort();
  94     }
  95     return ptr;
  96 }
  97
  98 void *qemu_try_memalign(size_t alignment, size_t size)
  99 {
 100     void *ptr;
 101
 102     if (alignment < sizeof(void*)) {
 103         alignment = sizeof(void*);
 104     }
 105
 106 #if defined(_POSIX_C_SOURCE) && !defined(__sun__)
 107     int ret;
 108     ret = posix_memalign(&ptr, alignment, size);
 109     if (ret != 0) {
 110         errno = ret;
 111         ptr = NULL;
 112     }
 113 #elif defined(CONFIG_BSD)
 114     ptr = valloc(size);
 115 #else
 116     ptr = memalign(alignment, size);
 117 #endif
 118     trace_qemu_memalign(alignment, size, ptr);
 119     return ptr;
 120 }
 121
 122 void *qemu_memalign(size_t alignment, size_t size)
 123 {
 124     return qemu_oom_check(qemu_try_memalign(alignment, size));
 125 }
 126
 127 /* alloc shared memory pages */
 128 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment)
 129 {
 130     size_t align = QEMU_VMALLOC_ALIGN;
 131     void *ptr = qemu_ram_mmap(-1, size, align, false);
 132
 133     if (ptr == MAP_FAILED) {
 134         return NULL;
 135     }
 136
 137     if (alignment) {
 138         *alignment = align;
 139     }
 140
 141     trace_qemu_anon_ram_alloc(size, ptr);
 142     return ptr;
 143 }
 144
 145 void qemu_vfree(void *ptr)
 146 {
 147     trace_qemu_vfree(ptr);
 148     free(ptr);
 149 }
 150
 151 void qemu_anon_ram_free(void *ptr, size_t size)
 152 {
 153     trace_qemu_anon_ram_free(ptr, size);
 154     qemu_ram_munmap(ptr, size);
 155 }
 156
 157 void qemu_set_block(int fd)
 158 {
 159     int f;
 160     f = fcntl(fd, F_GETFL);
 161     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 162 }
 163
 164 void qemu_set_nonblock(int fd)
 165 {
 166     int f;
 167     f = fcntl(fd, F_GETFL);
 168     fcntl(fd, F_SETFL, f | O_NONBLOCK);
 169 }
 170
 171 int socket_set_fast_reuse(int fd)
 172 {
 173     int val = 1, ret;
 174
 175     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 176                      (const char *)&val, sizeof(val));
 177
 178     assert(ret == 0);
 179
 180     return ret;
 181 }
 182
 183 void qemu_set_cloexec(int fd)
 184 {
 185     int f;
 186     f = fcntl(fd, F_GETFD);
 187     fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 188 }
 189
 190 /*
 191  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 192  */
 193 int qemu_pipe(int pipefd[2])
 194 {
 195     int ret;
 196
 197 #ifdef CONFIG_PIPE2
 198     ret = pipe2(pipefd, O_CLOEXEC);
 199     if (ret != -1 || errno != ENOSYS) {
 200         return ret;
 201     }
 202 #endif
 203     ret = pipe(pipefd);
 204     if (ret == 0) {
 205         qemu_set_cloexec(pipefd[0]);
 206         qemu_set_cloexec(pipefd[1]);
 207     }
 208
 209     return ret;
 210 }
 211
 212 int qemu_utimens(const char *path, const struct timespec *times)
 213 {
 214     struct timeval tv[2], tv_now;
 215     struct stat st;
 216     int i;
 217 #ifdef CONFIG_UTIMENSAT
 218     int ret;
 219
 220     ret = utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW);
 221     if (ret != -1 || errno != ENOSYS) {
 222         return ret;
 223     }
 224 #endif
 225     /* Fallback: use utimes() instead of utimensat() */
 226
 227     /* happy if special cases */
 228     if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) {
 229         return 0;
 230     }
 231     if (times[0].tv_nsec == UTIME_NOW && times[1].tv_nsec == UTIME_NOW) {
 232         return utimes(path, NULL);
 233     }
 234
 235     /* prepare for hard cases */
 236     if (times[0].tv_nsec == UTIME_NOW || times[1].tv_nsec == UTIME_NOW) {
 237         gettimeofday(&tv_now, NULL);
 238     }
 239     if (times[0].tv_nsec == UTIME_OMIT || times[1].tv_nsec == UTIME_OMIT) {
 240         stat(path, &st);
 241     }
 242
 243     for (i = 0; i < 2; i++) {
 244         if (times[i].tv_nsec == UTIME_NOW) {
 245             tv[i].tv_sec = tv_now.tv_sec;
 246             tv[i].tv_usec = tv_now.tv_usec;
 247         } else if (times[i].tv_nsec == UTIME_OMIT) {
 248             tv[i].tv_sec = (i == 0) ? st.st_atime : st.st_mtime;
 249             tv[i].tv_usec = 0;
 250         } else {
 251             tv[i].tv_sec = times[i].tv_sec;
 252             tv[i].tv_usec = times[i].tv_nsec / 1000;
 253         }
 254     }
 255
 256     return utimes(path, &tv[0]);
 257 }
 258
 259 char *
 260 qemu_get_local_state_pathname(const char *relative_pathname)
 261 {
 262     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 263                            relative_pathname);
 264 }
 265
 266 void qemu_set_tty_echo(int fd, bool echo)
 267 {
 268     struct termios tty;
 269
 270     tcgetattr(fd, &tty);
 271
 272     if (echo) {
 273         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 274     } else {
 275         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 276     }
 277
 278     tcsetattr(fd, TCSANOW, &tty);
 279 }
 280
 281 static char exec_dir[PATH_MAX];
 282
 283 void qemu_init_exec_dir(const char *argv0)
 284 {
 285     char *dir;
 286     char *p = NULL;
 287     char buf[PATH_MAX];
 288
 289     assert(!exec_dir[0]);
 290
 291 #if defined(__linux__)
 292     {
 293         int len;
 294         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 295         if (len > 0) {
 296             buf[len] = 0;
 297             p = buf;
 298         }
 299     }
 300 #elif defined(__FreeBSD__)
 301     {
 302         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 303         size_t len = sizeof(buf) - 1;
 304
 305         *buf = '\0';
 306         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 307             *buf) {
 308             buf[sizeof(buf) - 1] = '\0';
 309             p = buf;
 310         }
 311     }
 312 #endif
 313     /* If we don't have any way of figuring out the actual executable
 314        location then try argv[0].  */
 315     if (!p) {
 316         if (!argv0) {
 317             return;
 318         }
 319         p = realpath(argv0, buf);
 320         if (!p) {
 321             return;
 322         }
 323     }
 324     dir = dirname(p);
 325
 326     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 327 }
 328
 329 char *qemu_get_exec_dir(void)
 330 {
 331     return g_strdup(exec_dir);
 332 }
 333
 334 static sigjmp_buf sigjump;
 335
 336 static void sigbus_handler(int signal)
 337 {
 338     siglongjmp(sigjump, 1);
 339 }
 340
 341 void os_mem_prealloc(int fd, char *area, size_t memory)
 342 {
 343     int ret;
 344     struct sigaction act, oldact;
 345     sigset_t set, oldset;
 346
 347     memset(&act, 0, sizeof(act));
 348     act.sa_handler = &sigbus_handler;
 349     act.sa_flags = 0;
 350
 351     ret = sigaction(SIGBUS, &act, &oldact);
 352     if (ret) {
 353         perror("os_mem_prealloc: failed to install signal handler");
 354         exit(1);
 355     }
 356
 357     /* unblock SIGBUS */
 358     sigemptyset(&set);
 359     sigaddset(&set, SIGBUS);
 360     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 361
 362     if (sigsetjmp(sigjump, 1)) {
 363         fprintf(stderr, "os_mem_prealloc: Insufficient free host memory "
 364                         "pages available to allocate guest RAM\n");
 365         exit(1);
 366     } else {
 367         int i;
 368         size_t hpagesize = qemu_fd_getpagesize(fd);
 369         size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 370
 371         /* MAP_POPULATE silently ignores failures */
 372         for (i = 0; i < numpages; i++) {
 373             memset(area + (hpagesize * i), 0, 1);
 374         }
 375
 376         ret = sigaction(SIGBUS, &oldact, NULL);
 377         if (ret) {
 378             perror("os_mem_prealloc: failed to reinstall signal handler");
 379             exit(1);
 380         }
 381
 382         pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 383     }
 384 }
 385
 386
 387 static struct termios oldtty;
 388
 389 static void term_exit(void)
 390 {
 391     tcsetattr(0, TCSANOW, &oldtty);
 392 }
 393
 394 static void term_init(void)
 395 {
 396     struct termios tty;
 397
 398     tcgetattr(0, &tty);
 399     oldtty = tty;
 400
 401     tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
 402                           |INLCR|IGNCR|ICRNL|IXON);
 403     tty.c_oflag |= OPOST;
 404     tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
 405     tty.c_cflag &= ~(CSIZE|PARENB);
 406     tty.c_cflag |= CS8;
 407     tty.c_cc[VMIN] = 1;
 408     tty.c_cc[VTIME] = 0;
 409
 410     tcsetattr(0, TCSANOW, &tty);
 411
 412     atexit(term_exit);
 413 }
 414
 415 int qemu_read_password(char *buf, int buf_size)
 416 {
 417     uint8_t ch;
 418     int i, ret;
 419
 420     printf("password: ");
 421     fflush(stdout);
 422     term_init();
 423     i = 0;
 424     for (;;) {
 425         ret = read(0, &ch, 1);
 426         if (ret == -1) {
 427             if (errno == EAGAIN || errno == EINTR) {
 428                 continue;
 429             } else {
 430                 break;
 431             }
 432         } else if (ret == 0) {
 433             ret = -1;
 434             break;
 435         } else {
 436             if (ch == '\r' ||
 437                 ch == '\n') {
 438                 ret = 0;
 439                 break;
 440             }
 441             if (i < (buf_size - 1)) {
 442                 buf[i++] = ch;
 443             }
 444         }
 445     }
 446     term_exit();
 447     buf[i] = '\0';
 448     printf("\n");
 449     return ret;
 450 }
 451
 452
 453 pid_t qemu_fork(Error **errp)
 454 {
 455     sigset_t oldmask, newmask;
 456     struct sigaction sig_action;
 457     int saved_errno;
 458     pid_t pid;
 459
 460     /*
 461      * Need to block signals now, so that child process can safely
 462      * kill off caller's signal handlers without a race.
 463      */
 464     sigfillset(&newmask);
 465     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 466         error_setg_errno(errp, errno,
 467                          "cannot block signals");
 468         return -1;
 469     }
 470
 471     pid = fork();
 472     saved_errno = errno;
 473
 474     if (pid < 0) {
 475         /* attempt to restore signal mask, but ignore failure, to
 476          * avoid obscuring the fork failure */
 477         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 478         error_setg_errno(errp, saved_errno,
 479                          "cannot fork child process");
 480         errno = saved_errno;
 481         return -1;
 482     } else if (pid) {
 483         /* parent process */
 484
 485         /* Restore our original signal mask now that the child is
 486          * safely running. Only documented failures are EFAULT (not
 487          * possible, since we are using just-grabbed mask) or EINVAL
 488          * (not possible, since we are using correct arguments).  */
 489         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 490     } else {
 491         /* child process */
 492         size_t i;
 493
 494         /* Clear out all signal handlers from parent so nothing
 495          * unexpected can happen in our child once we unblock
 496          * signals */
 497         sig_action.sa_handler = SIG_DFL;
 498         sig_action.sa_flags = 0;
 499         sigemptyset(&sig_action.sa_mask);
 500
 501         for (i = 1; i < NSIG; i++) {
 502             /* Only possible errors are EFAULT or EINVAL The former
 503              * won't happen, the latter we expect, so no need to check
 504              * return value */
 505             (void)sigaction(i, &sig_action, NULL);
 506         }
 507
 508         /* Unmask all signals in child, since we've no idea what the
 509          * caller's done with their signal mask and don't want to
 510          * propagate that to children */
 511         sigemptyset(&newmask);
 512         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 513             Error *local_err = NULL;
 514             error_setg_errno(&local_err, errno,
 515                              "cannot unblock signals");
 516             error_report_err(local_err);
 517             _exit(1);
 518         }
 519     }
 520     return pid;
 521 }