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 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     size_t total = size + align;
 132     void *ptr = mmap(0, total, PROT_NONE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
 133     size_t offset = QEMU_ALIGN_UP((uintptr_t)ptr, align) - (uintptr_t)ptr;
 134     void *ptr1;
 135
 136     if (ptr == MAP_FAILED) {
 137         return NULL;
 138     }
 139
 140     if (alignment) {
 141         *alignment = align;
 142     }
 143
 144     ptr1 = mmap(ptr + offset, size, PROT_READ | PROT_WRITE,
 145                 MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
 146     if (ptr1 == MAP_FAILED) {
 147         munmap(ptr, total);
 148         return NULL;
 149     }
 150
 151     ptr += offset;
 152     total -= offset;
 153
 154     if (offset > 0) {
 155         munmap(ptr - offset, offset);
 156     }
 157     if (total > size + getpagesize()) {
 158         munmap(ptr + size + getpagesize(), total - size - getpagesize());
 159     }
 160
 161     trace_qemu_anon_ram_alloc(size, ptr);
 162     return ptr;
 163 }
 164
 165 void qemu_vfree(void *ptr)
 166 {
 167     trace_qemu_vfree(ptr);
 168     free(ptr);
 169 }
 170
 171 void qemu_anon_ram_free(void *ptr, size_t size)
 172 {
 173     trace_qemu_anon_ram_free(ptr, size);
 174     if (ptr) {
 175         munmap(ptr, size + getpagesize());
 176     }
 177 }
 178
 179 void qemu_set_block(int fd)
 180 {
 181     int f;
 182     f = fcntl(fd, F_GETFL);
 183     fcntl(fd, F_SETFL, f & ~O_NONBLOCK);
 184 }
 185
 186 void qemu_set_nonblock(int fd)
 187 {
 188     int f;
 189     f = fcntl(fd, F_GETFL);
 190     fcntl(fd, F_SETFL, f | O_NONBLOCK);
 191 }
 192
 193 int socket_set_fast_reuse(int fd)
 194 {
 195     int val = 1, ret;
 196
 197     ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
 198                      (const char *)&val, sizeof(val));
 199
 200     assert(ret == 0);
 201
 202     return ret;
 203 }
 204
 205 void qemu_set_cloexec(int fd)
 206 {
 207     int f;
 208     f = fcntl(fd, F_GETFD);
 209     fcntl(fd, F_SETFD, f | FD_CLOEXEC);
 210 }
 211
 212 /*
 213  * Creates a pipe with FD_CLOEXEC set on both file descriptors
 214  */
 215 int qemu_pipe(int pipefd[2])
 216 {
 217     int ret;
 218
 219 #ifdef CONFIG_PIPE2
 220     ret = pipe2(pipefd, O_CLOEXEC);
 221     if (ret != -1 || errno != ENOSYS) {
 222         return ret;
 223     }
 224 #endif
 225     ret = pipe(pipefd);
 226     if (ret == 0) {
 227         qemu_set_cloexec(pipefd[0]);
 228         qemu_set_cloexec(pipefd[1]);
 229     }
 230
 231     return ret;
 232 }
 233
 234 int qemu_utimens(const char *path, const struct timespec *times)
 235 {
 236     struct timeval tv[2], tv_now;
 237     struct stat st;
 238     int i;
 239 #ifdef CONFIG_UTIMENSAT
 240     int ret;
 241
 242     ret = utimensat(AT_FDCWD, path, times, AT_SYMLINK_NOFOLLOW);
 243     if (ret != -1 || errno != ENOSYS) {
 244         return ret;
 245     }
 246 #endif
 247     /* Fallback: use utimes() instead of utimensat() */
 248
 249     /* happy if special cases */
 250     if (times[0].tv_nsec == UTIME_OMIT && times[1].tv_nsec == UTIME_OMIT) {
 251         return 0;
 252     }
 253     if (times[0].tv_nsec == UTIME_NOW && times[1].tv_nsec == UTIME_NOW) {
 254         return utimes(path, NULL);
 255     }
 256
 257     /* prepare for hard cases */
 258     if (times[0].tv_nsec == UTIME_NOW || times[1].tv_nsec == UTIME_NOW) {
 259         gettimeofday(&tv_now, NULL);
 260     }
 261     if (times[0].tv_nsec == UTIME_OMIT || times[1].tv_nsec == UTIME_OMIT) {
 262         stat(path, &st);
 263     }
 264
 265     for (i = 0; i < 2; i++) {
 266         if (times[i].tv_nsec == UTIME_NOW) {
 267             tv[i].tv_sec = tv_now.tv_sec;
 268             tv[i].tv_usec = tv_now.tv_usec;
 269         } else if (times[i].tv_nsec == UTIME_OMIT) {
 270             tv[i].tv_sec = (i == 0) ? st.st_atime : st.st_mtime;
 271             tv[i].tv_usec = 0;
 272         } else {
 273             tv[i].tv_sec = times[i].tv_sec;
 274             tv[i].tv_usec = times[i].tv_nsec / 1000;
 275         }
 276     }
 277
 278     return utimes(path, &tv[0]);
 279 }
 280
 281 char *
 282 qemu_get_local_state_pathname(const char *relative_pathname)
 283 {
 284     return g_strdup_printf("%s/%s", CONFIG_QEMU_LOCALSTATEDIR,
 285                            relative_pathname);
 286 }
 287
 288 void qemu_set_tty_echo(int fd, bool echo)
 289 {
 290     struct termios tty;
 291
 292     tcgetattr(fd, &tty);
 293
 294     if (echo) {
 295         tty.c_lflag |= ECHO | ECHONL | ICANON | IEXTEN;
 296     } else {
 297         tty.c_lflag &= ~(ECHO | ECHONL | ICANON | IEXTEN);
 298     }
 299
 300     tcsetattr(fd, TCSANOW, &tty);
 301 }
 302
 303 static char exec_dir[PATH_MAX];
 304
 305 void qemu_init_exec_dir(const char *argv0)
 306 {
 307     char *dir;
 308     char *p = NULL;
 309     char buf[PATH_MAX];
 310
 311     assert(!exec_dir[0]);
 312
 313 #if defined(__linux__)
 314     {
 315         int len;
 316         len = readlink("/proc/self/exe", buf, sizeof(buf) - 1);
 317         if (len > 0) {
 318             buf[len] = 0;
 319             p = buf;
 320         }
 321     }
 322 #elif defined(__FreeBSD__)
 323     {
 324         static int mib[4] = {CTL_KERN, KERN_PROC, KERN_PROC_PATHNAME, -1};
 325         size_t len = sizeof(buf) - 1;
 326
 327         *buf = '\0';
 328         if (!sysctl(mib, ARRAY_SIZE(mib), buf, &len, NULL, 0) &&
 329             *buf) {
 330             buf[sizeof(buf) - 1] = '\0';
 331             p = buf;
 332         }
 333     }
 334 #endif
 335     /* If we don't have any way of figuring out the actual executable
 336        location then try argv[0].  */
 337     if (!p) {
 338         if (!argv0) {
 339             return;
 340         }
 341         p = realpath(argv0, buf);
 342         if (!p) {
 343             return;
 344         }
 345     }
 346     dir = dirname(p);
 347
 348     pstrcpy(exec_dir, sizeof(exec_dir), dir);
 349 }
 350
 351 char *qemu_get_exec_dir(void)
 352 {
 353     return g_strdup(exec_dir);
 354 }
 355
 356 static sigjmp_buf sigjump;
 357
 358 static void sigbus_handler(int signal)
 359 {
 360     siglongjmp(sigjump, 1);
 361 }
 362
 363 static size_t fd_getpagesize(int fd)
 364 {
 365 #ifdef CONFIG_LINUX
 366     struct statfs fs;
 367     int ret;
 368
 369     if (fd != -1) {
 370         do {
 371             ret = fstatfs(fd, &fs);
 372         } while (ret != 0 && errno == EINTR);
 373
 374         if (ret == 0 && fs.f_type == HUGETLBFS_MAGIC) {
 375             return fs.f_bsize;
 376         }
 377     }
 378 #endif
 379
 380     return getpagesize();
 381 }
 382
 383 void os_mem_prealloc(int fd, char *area, size_t memory)
 384 {
 385     int ret;
 386     struct sigaction act, oldact;
 387     sigset_t set, oldset;
 388
 389     memset(&act, 0, sizeof(act));
 390     act.sa_handler = &sigbus_handler;
 391     act.sa_flags = 0;
 392
 393     ret = sigaction(SIGBUS, &act, &oldact);
 394     if (ret) {
 395         perror("os_mem_prealloc: failed to install signal handler");
 396         exit(1);
 397     }
 398
 399     /* unblock SIGBUS */
 400     sigemptyset(&set);
 401     sigaddset(&set, SIGBUS);
 402     pthread_sigmask(SIG_UNBLOCK, &set, &oldset);
 403
 404     if (sigsetjmp(sigjump, 1)) {
 405         fprintf(stderr, "os_mem_prealloc: Insufficient free host memory "
 406                         "pages available to allocate guest RAM\n");
 407         exit(1);
 408     } else {
 409         int i;
 410         size_t hpagesize = fd_getpagesize(fd);
 411         size_t numpages = DIV_ROUND_UP(memory, hpagesize);
 412
 413         /* MAP_POPULATE silently ignores failures */
 414         for (i = 0; i < numpages; i++) {
 415             memset(area + (hpagesize * i), 0, 1);
 416         }
 417
 418         ret = sigaction(SIGBUS, &oldact, NULL);
 419         if (ret) {
 420             perror("os_mem_prealloc: failed to reinstall signal handler");
 421             exit(1);
 422         }
 423
 424         pthread_sigmask(SIG_SETMASK, &oldset, NULL);
 425     }
 426 }
 427
 428
 429 static struct termios oldtty;
 430
 431 static void term_exit(void)
 432 {
 433     tcsetattr(0, TCSANOW, &oldtty);
 434 }
 435
 436 static void term_init(void)
 437 {
 438     struct termios tty;
 439
 440     tcgetattr(0, &tty);
 441     oldtty = tty;
 442
 443     tty.c_iflag &= ~(IGNBRK|BRKINT|PARMRK|ISTRIP
 444                           |INLCR|IGNCR|ICRNL|IXON);
 445     tty.c_oflag |= OPOST;
 446     tty.c_lflag &= ~(ECHO|ECHONL|ICANON|IEXTEN);
 447     tty.c_cflag &= ~(CSIZE|PARENB);
 448     tty.c_cflag |= CS8;
 449     tty.c_cc[VMIN] = 1;
 450     tty.c_cc[VTIME] = 0;
 451
 452     tcsetattr(0, TCSANOW, &tty);
 453
 454     atexit(term_exit);
 455 }
 456
 457 int qemu_read_password(char *buf, int buf_size)
 458 {
 459     uint8_t ch;
 460     int i, ret;
 461
 462     printf("password: ");
 463     fflush(stdout);
 464     term_init();
 465     i = 0;
 466     for (;;) {
 467         ret = read(0, &ch, 1);
 468         if (ret == -1) {
 469             if (errno == EAGAIN || errno == EINTR) {
 470                 continue;
 471             } else {
 472                 break;
 473             }
 474         } else if (ret == 0) {
 475             ret = -1;
 476             break;
 477         } else {
 478             if (ch == '\r' ||
 479                 ch == '\n') {
 480                 ret = 0;
 481                 break;
 482             }
 483             if (i < (buf_size - 1)) {
 484                 buf[i++] = ch;
 485             }
 486         }
 487     }
 488     term_exit();
 489     buf[i] = '\0';
 490     printf("\n");
 491     return ret;
 492 }
 493
 494
 495 pid_t qemu_fork(Error **errp)
 496 {
 497     sigset_t oldmask, newmask;
 498     struct sigaction sig_action;
 499     int saved_errno;
 500     pid_t pid;
 501
 502     /*
 503      * Need to block signals now, so that child process can safely
 504      * kill off caller's signal handlers without a race.
 505      */
 506     sigfillset(&newmask);
 507     if (pthread_sigmask(SIG_SETMASK, &newmask, &oldmask) != 0) {
 508         error_setg_errno(errp, errno,
 509                          "cannot block signals");
 510         return -1;
 511     }
 512
 513     pid = fork();
 514     saved_errno = errno;
 515
 516     if (pid < 0) {
 517         /* attempt to restore signal mask, but ignore failure, to
 518          * avoid obscuring the fork failure */
 519         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 520         error_setg_errno(errp, saved_errno,
 521                          "cannot fork child process");
 522         errno = saved_errno;
 523         return -1;
 524     } else if (pid) {
 525         /* parent process */
 526
 527         /* Restore our original signal mask now that the child is
 528          * safely running. Only documented failures are EFAULT (not
 529          * possible, since we are using just-grabbed mask) or EINVAL
 530          * (not possible, since we are using correct arguments).  */
 531         (void)pthread_sigmask(SIG_SETMASK, &oldmask, NULL);
 532     } else {
 533         /* child process */
 534         size_t i;
 535
 536         /* Clear out all signal handlers from parent so nothing
 537          * unexpected can happen in our child once we unblock
 538          * signals */
 539         sig_action.sa_handler = SIG_DFL;
 540         sig_action.sa_flags = 0;
 541         sigemptyset(&sig_action.sa_mask);
 542
 543         for (i = 1; i < NSIG; i++) {
 544             /* Only possible errors are EFAULT or EINVAL The former
 545              * won't happen, the latter we expect, so no need to check
 546              * return value */
 547             (void)sigaction(i, &sig_action, NULL);
 548         }
 549
 550         /* Unmask all signals in child, since we've no idea what the
 551          * caller's done with their signal mask and don't want to
 552          * propagate that to children */
 553         sigemptyset(&newmask);
 554         if (pthread_sigmask(SIG_SETMASK, &newmask, NULL) != 0) {
 555             Error *local_err = NULL;
 556             error_setg_errno(&local_err, errno,
 557                              "cannot unblock signals");
 558             error_report_err(local_err);
 559             _exit(1);
 560         }
 561     }
 562     return pid;
 563 }