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 #define HUGETLBFS_MAGIC       0x958458f6
  50
  51 #include <termios.h>
  52 #include <unistd.h>
  53 #include <termios.h>
  54
  55 #include <glib/gprintf.h>
  56
  57 #include "config-host.h"
  58 #include "sysemu/sysemu.h"
  59 #include "trace.h"
  60 #include "qemu/sockets.h"
  61 #include <sys/mman.h>
  62 #include <libgen.h>
  63 #include <setjmp.h>
  64 #include <sys/signal.h>
  65
  66 #ifdef CONFIG_LINUX
  67 #include <sys/syscall.h>
  68 #include <sys/vfs.h>
  69 #endif
  70
  71 #ifdef __FreeBSD__
  72 #include <sys/sysctl.h>
  73 #endif
  74
  75 #include <qemu/mmap-alloc.h>
  76
  77 int qemu_get_thread_id(void)
  78 {
  79 #if defined(__linux__)
  80     return syscall(SYS_gettid);
  81 #else
  82     return getpid();
  83 #endif
  84 }
  85
  86 int qemu_daemon(int nochdir, int noclose)
  87 {
  88     return daemon(nochdir, noclose);
  89 }
  90
  91 void *qemu_oom_check(void *ptr)
  92 {
  93     if (ptr == NULL) {
  94         fprintf(stderr, "Failed to allocate memory: %s\n", strerror(errno));
  95         abort();
  96     }
  97     return ptr;
  98 }
  99
 100 void *qemu_try_memalign(size_t alignment, size_t size)
 101 {
 102     void *ptr;
 103
 104     if (alignment < sizeof(void*)) {
 105         alignment = sizeof(void*);
 106     }
 107
 108 #if defined(_POSIX_C_SOURCE) && !defined(__sun__)
 109     int ret;
 110     ret = posix_memalign(&ptr, alignment, size);
 111     if (ret != 0) {
 112         errno = ret;
 113         ptr = NULL;
 114     }
 115 #elif defined(CONFIG_BSD)
 116     ptr = valloc(size);
 117 #else
 118     ptr = memalign(alignment, size);
 119 #endif
 120     trace_qemu_memalign(alignment, size, ptr);
 121     return ptr;
 122 }
 123
 124 void *qemu_memalign(size_t alignment, size_t size)
 125 {
 126     return qemu_oom_check(qemu_try_memalign(alignment, size));
 127 }
 128
 129 /* alloc shared memory pages */
 130 void *qemu_anon_ram_alloc(size_t size, uint64_t *alignment)
 131 {
 132     size_t align = QEMU_VMALLOC_ALIGN;
 133     void *ptr = qemu_ram_mmap(-1, size, align, false);
 134
 135     if (ptr == MAP_FAILED) {
 136         return NULL;
 137     }
 138
 139     if (alignment) {
 140         *alignment = align;
 141     }
 142
 143     trace_qemu_anon_ram_alloc(size, ptr);
 144     return ptr;
 145 }
 146
 147 void qemu_vfree(void *ptr)
 148 {
 149     trace_qemu_vfree(ptr);
 150     free(ptr);
 151 }
 152
 153 void qemu_anon_ram_free(void *ptr, size_t size)
 154 {
 155     trace_qemu_anon_ram_free(ptr, size);
 156     qemu_ram_munmap(ptr, size);
 157 }
 158
 159 void qemu_set_block(int fd)
 160 {
 161     int f;
 162     f = fcntl(fd, F_GETFL);
 163     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 164 }
 165
 166 void qemu_set_nonblock(int fd)
 167 {
 168     int f;
 169     f = fcntl(fd, F_GETFL);
 170     fcntl(fd, F_SETFL, f | O_NONBLOCK);
 171 }
 172
 173 int socket_set_fast_reuse(int fd)
 174 {
 175     int val = 1, ret;
 176
 177     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 178                      (const char *)&val, sizeof(val));
 179
 180     assert(ret == 0);
 181
 182     return ret;
 183 }
 184
 185 void qemu_set_cloexec(int fd)
 186 {
 187     int f;
 188     f = fcntl(fd, F_GETFD);
 189     fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 190 }
 191
 192 /*
 193  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 194  */
 195 int qemu_pipe(int pipefd[2])
 196 {
 197     int ret;
 198
 199 #ifdef CONFIG_PIPE2
 200     ret = pipe2(pipefd, O_CLOEXEC);
 201     if (ret != -1 || errno != ENOSYS) {
 202         return ret;
 203     }
 204 #endif
 205     ret = pipe(pipefd);
 206     if (ret == 0) {
 207         qemu_set_cloexec(pipefd[0]);
 208         qemu_set_cloexec(pipefd[1]);
 209     }
 210
 211     return ret;
 212 }
 213
 214 int qemu_utimens(const char *path, const struct timespec *times)
 215 {
 216     struct timeval tv[2], tv_now;
 217     struct stat st;
 218     int i;
 219 #ifdef CONFIG_UTIMENSAT
 220     int ret;
 221
 222     ret = utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW);
 223     if (ret != -1 || errno != ENOSYS) {
 224         return ret;
 225     }
 226 #endif
 227     /* Fallback: use utimes() instead of utimensat() */
 228
 229     /* happy if special cases */
 230     if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) {
 231         return 0;
 232     }
 233     if (times[0].tv_nsec == UTIME_NOW && times[1].tv_nsec == UTIME_NOW) {
 234         return utimes(path, NULL);
 235     }
 236
 237     /* prepare for hard cases */
 238     if (times[0].tv_nsec == UTIME_NOW || times[1].tv_nsec == UTIME_NOW) {
 239         gettimeofday(&tv_now, NULL);
 240     }
 241     if (times[0].tv_nsec == UTIME_OMIT || times[1].tv_nsec == UTIME_OMIT) {
 242         stat(path, &st);
 243     }
 244
 245     for (i = 0; i < 2; i++) {
 246         if (times[i].tv_nsec == UTIME_NOW) {
 247             tv[i].tv_sec = tv_now.tv_sec;
 248             tv[i].tv_usec = tv_now.tv_usec;
 249         } else if (times[i].tv_nsec == UTIME_OMIT) {
 250             tv[i].tv_sec = (i == 0) ? st.st_atime : st.st_mtime;
 251             tv[i].tv_usec = 0;
 252         } else {
 253             tv[i].tv_sec = times[i].tv_sec;
 254             tv[i].tv_usec = times[i].tv_nsec / 1000;
 255         }
 256     }
 257
 258     return utimes(path, &tv[0]);
 259 }
 260
 261 char *
 262 qemu_get_local_state_pathname(const char *relative_pathname)
 263 {
 264     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 265                            relative_pathname);
 266 }
 267
 268 void qemu_set_tty_echo(int fd, bool echo)
 269 {
 270     struct termios tty;
 271
 272     tcgetattr(fd, &tty);
 273
 274     if (echo) {
 275         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 276     } else {
 277         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 278     }
 279
 280     tcsetattr(fd, TCSANOW, &tty);
 281 }
 282
 283 static char exec_dir[PATH_MAX];
 284
 285 void qemu_init_exec_dir(const char *argv0)
 286 {
 287     char *dir;
 288     char *p = NULL;
 289     char buf[PATH_MAX];
 290
 291     assert(!exec_dir[0]);
 292
 293 #if defined(__linux__)
 294     {
 295         int len;
 296         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 297         if (len > 0) {
 298             buf[len] = 0;
 299             p = buf;
 300         }
 301     }
 302 #elif defined(__FreeBSD__)
 303     {
 304         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 305         size_t len = sizeof(buf) - 1;
 306
 307         *buf = '\0';
 308         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 309             *buf) {
 310             buf[sizeof(buf) - 1] = '\0';
 311             p = buf;
 312         }
 313     }
 314 #endif
 315     /* If we don't have any way of figuring out the actual executable
 316        location then try argv[0].  */
 317     if (!p) {
 318         if (!argv0) {
 319             return;
 320         }
 321         p = realpath(argv0, buf);
 322         if (!p) {
 323             return;
 324         }
 325     }
 326     dir = dirname(p);
 327
 328     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 329 }
 330
 331 char *qemu_get_exec_dir(void)
 332 {
 333     return g_strdup(exec_dir);
 334 }
 335
 336 static sigjmp_buf sigjump;
 337
 338 static void sigbus_handler(int signal)
 339 {
 340     siglongjmp(sigjump, 1);
 341 }
 342
 343 static size_t fd_getpagesize(int fd)
 344 {
 345 #ifdef CONFIG_LINUX
 346     struct statfs fs;
 347     int ret;
 348
 349     if (fd != -1) {
 350         do {
 351             ret = fstatfs(fd, &fs);
 352         } while (ret != 0 && errno == EINTR);
 353
 354         if (ret == 0 && fs.f_type == HUGETLBFS_MAGIC) {
 355             return fs.f_bsize;
 356         }
 357     }
 358 #endif
 359
 360     return getpagesize();
 361 }
 362
 363 void os_mem_prealloc(int fd, char *area, size_t memory)
 364 {
 365     int ret;
 366     struct sigaction act, oldact;
 367     sigset_t set, oldset;
 368
 369     memset(&act, 0, sizeof(act));
 370     act.sa_handler = &sigbus_handler;
 371     act.sa_flags = 0;
 372
 373     ret = sigaction(SIGBUS, &act, &oldact);
 374     if (ret) {
 375         perror("os_mem_prealloc: failed to install signal handler");
 376         exit(1);
 377     }
 378
 379     /* unblock SIGBUS */
 380     sigemptyset(&set);
 381     sigaddset(&set, SIGBUS);
 382     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 383
 384     if (sigsetjmp(sigjump, 1)) {
 385         fprintf(stderr, "os_mem_prealloc: Insufficient free host memory "
 386                         "pages available to allocate guest RAM\n");
 387         exit(1);
 388     } else {
 389         int i;
 390         size_t hpagesize = fd_getpagesize(fd);
 391         size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 392
 393         /* MAP_POPULATE silently ignores failures */
 394         for (i = 0; i < numpages; i++) {
 395             memset(area + (hpagesize * i), 0, 1);
 396         }
 397
 398         ret = sigaction(SIGBUS, &oldact, NULL);
 399         if (ret) {
 400             perror("os_mem_prealloc: failed to reinstall signal handler");
 401             exit(1);
 402         }
 403
 404         pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 405     }
 406 }
 407
 408
 409 static struct termios oldtty;
 410
 411 static void term_exit(void)
 412 {
 413     tcsetattr(0, TCSANOW, &oldtty);
 414 }
 415
 416 static void term_init(void)
 417 {
 418     struct termios tty;
 419
 420     tcgetattr(0, &tty);
 421     oldtty = tty;
 422
 423     tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
 424                           |INLCR|IGNCR|ICRNL|IXON);
 425     tty.c_oflag |= OPOST;
 426     tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
 427     tty.c_cflag &= ~(CSIZE|PARENB);
 428     tty.c_cflag |= CS8;
 429     tty.c_cc[VMIN] = 1;
 430     tty.c_cc[VTIME] = 0;
 431
 432     tcsetattr(0, TCSANOW, &tty);
 433
 434     atexit(term_exit);
 435 }
 436
 437 int qemu_read_password(char *buf, int buf_size)
 438 {
 439     uint8_t ch;
 440     int i, ret;
 441
 442     printf("password: ");
 443     fflush(stdout);
 444     term_init();
 445     i = 0;
 446     for (;;) {
 447         ret = read(0, &ch, 1);
 448         if (ret == -1) {
 449             if (errno == EAGAIN || errno == EINTR) {
 450                 continue;
 451             } else {
 452                 break;
 453             }
 454         } else if (ret == 0) {
 455             ret = -1;
 456             break;
 457         } else {
 458             if (ch == '\r' ||
 459                 ch == '\n') {
 460                 ret = 0;
 461                 break;
 462             }
 463             if (i < (buf_size - 1)) {
 464                 buf[i++] = ch;
 465             }
 466         }
 467     }
 468     term_exit();
 469     buf[i] = '\0';
 470     printf("\n");
 471     return ret;
 472 }
 473
 474
 475 pid_t qemu_fork(Error **errp)
 476 {
 477     sigset_t oldmask, newmask;
 478     struct sigaction sig_action;
 479     int saved_errno;
 480     pid_t pid;
 481
 482     /*
 483      * Need to block signals now, so that child process can safely
 484      * kill off caller's signal handlers without a race.
 485      */
 486     sigfillset(&newmask);
 487     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 488         error_setg_errno(errp, errno,
 489                          "cannot block signals");
 490         return -1;
 491     }
 492
 493     pid = fork();
 494     saved_errno = errno;
 495
 496     if (pid < 0) {
 497         /* attempt to restore signal mask, but ignore failure, to
 498          * avoid obscuring the fork failure */
 499         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 500         error_setg_errno(errp, saved_errno,
 501                          "cannot fork child process");
 502         errno = saved_errno;
 503         return -1;
 504     } else if (pid) {
 505         /* parent process */
 506
 507         /* Restore our original signal mask now that the child is
 508          * safely running. Only documented failures are EFAULT (not
 509          * possible, since we are using just-grabbed mask) or EINVAL
 510          * (not possible, since we are using correct arguments).  */
 511         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 512     } else {
 513         /* child process */
 514         size_t i;
 515
 516         /* Clear out all signal handlers from parent so nothing
 517          * unexpected can happen in our child once we unblock
 518          * signals */
 519         sig_action.sa_handler = SIG_DFL;
 520         sig_action.sa_flags = 0;
 521         sigemptyset(&sig_action.sa_mask);
 522
 523         for (i = 1; i < NSIG; i++) {
 524             /* Only possible errors are EFAULT or EINVAL The former
 525              * won't happen, the latter we expect, so no need to check
 526              * return value */
 527             (void)sigaction(i, &sig_action, NULL);
 528         }
 529
 530         /* Unmask all signals in child, since we've no idea what the
 531          * caller's done with their signal mask and don't want to
 532          * propagate that to children */
 533         sigemptyset(&newmask);
 534         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 535             Error *local_err = NULL;
 536             error_setg_errno(&local_err, errno,
 537                              "cannot unblock signals");
 538             error_report_err(local_err);
 539             _exit(1);
 540         }
 541     }
 542     return pid;
 543 }