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