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