util/osdep.c

   1 /*
   2  * QEMU low level functions
   3  *
   4  * Copyright (c) 2003 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu/osdep.h"
  25 #include "qapi/error.h"
  26 #include "qemu/cutils.h"
  27 #include "qemu/sockets.h"
  28 #include "qemu/error-report.h"
  29 #include "qemu/madvise.h"
  30 #include "qemu/mprotect.h"
  31 #include "qemu/hw-version.h"
  32 #include "monitor/monitor.h"
  33
  34 static const char *hw_version = QEMU_HW_VERSION;
  35
  36 int socket_set_cork(int fd, int v)
  37 {
  38 #if defined(SOL_TCP) && defined(TCP_CORK)
  39     return setsockopt(fd, SOL_TCP, TCP_CORK, &v, sizeof(v));
  40 #else
  41     return 0;
  42 #endif
  43 }
  44
  45 int socket_set_nodelay(int fd)
  46 {
  47     int v = 1;
  48     return setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &v, sizeof(v));
  49 }
  50
  51 int qemu_madvise(void *addr, size_t len, int advice)
  52 {
  53     if (advice == QEMU_MADV_INVALID) {
  54         errno = EINVAL;
  55         return -1;
  56     }
  57 #if defined(CONFIG_MADVISE)
  58     return madvise(addr, len, advice);
  59 #elif defined(CONFIG_POSIX_MADVISE)
  60     return posix_madvise(addr, len, advice);
  61 #else
  62     errno = EINVAL;
  63     return -1;
  64 #endif
  65 }
  66
  67 static int qemu_mprotect__osdep(void *addr, size_t size, int prot)
  68 {
  69     g_assert(!((uintptr_t)addr & ~qemu_real_host_page_mask()));
  70     g_assert(!(size & ~qemu_real_host_page_mask()));
  71
  72 #ifdef _WIN32
  73     DWORD old_protect;
  74
  75     if (!VirtualProtect(addr, size, prot, &old_protect)) {
  76         g_autofree gchar *emsg = g_win32_error_message(GetLastError());
  77         error_report("%s: VirtualProtect failed: %s", __func__, emsg);
  78         return -1;
  79     }
  80     return 0;
  81 #else
  82     if (mprotect(addr, size, prot)) {
  83         error_report("%s: mprotect failed: %s", __func__, strerror(errno));
  84         return -1;
  85     }
  86     return 0;
  87 #endif
  88 }
  89
  90 int qemu_mprotect_rw(void *addr, size_t size)
  91 {
  92 #ifdef _WIN32
  93     return qemu_mprotect__osdep(addr, size, PAGE_READWRITE);
  94 #else
  95     return qemu_mprotect__osdep(addr, size, PROT_READ | PROT_WRITE);
  96 #endif
  97 }
  98
  99 int qemu_mprotect_rwx(void *addr, size_t size)
 100 {
 101 #ifdef _WIN32
 102     return qemu_mprotect__osdep(addr, size, PAGE_EXECUTE_READWRITE);
 103 #else
 104     return qemu_mprotect__osdep(addr, size, PROT_READ | PROT_WRITE | PROT_EXEC);
 105 #endif
 106 }
 107
 108 int qemu_mprotect_none(void *addr, size_t size)
 109 {
 110 #ifdef _WIN32
 111     return qemu_mprotect__osdep(addr, size, PAGE_NOACCESS);
 112 #else
 113     return qemu_mprotect__osdep(addr, size, PROT_NONE);
 114 #endif
 115 }
 116
 117 #ifndef _WIN32
 118
 119 static int fcntl_op_setlk = -1;
 120 static int fcntl_op_getlk = -1;
 121
 122 /*
 123  * Dups an fd and sets the flags
 124  */
 125 int qemu_dup_flags(int fd, int flags)
 126 {
 127     int ret;
 128     int serrno;
 129     int dup_flags;
 130
 131     ret = qemu_dup(fd);
 132     if (ret == -1) {
 133         goto fail;
 134     }
 135
 136     dup_flags = fcntl(ret, F_GETFL);
 137     if (dup_flags == -1) {
 138         goto fail;
 139     }
 140
 141     if ((flags & O_SYNC) != (dup_flags & O_SYNC)) {
 142         errno = EINVAL;
 143         goto fail;
 144     }
 145
 146     /* Set/unset flags that we can with fcntl */
 147     if (fcntl(ret, F_SETFL, flags) == -1) {
 148         goto fail;
 149     }
 150
 151     /* Truncate the file in the cases that open() would truncate it */
 152     if (flags & O_TRUNC ||
 153             ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))) {
 154         if (ftruncate(ret, 0) == -1) {
 155             goto fail;
 156         }
 157     }
 158
 159     return ret;
 160
 161 fail:
 162     serrno = errno;
 163     if (ret != -1) {
 164         close(ret);
 165     }
 166     errno = serrno;
 167     return -1;
 168 }
 169
 170 int qemu_dup(int fd)
 171 {
 172     int ret;
 173 #ifdef F_DUPFD_CLOEXEC
 174     ret = fcntl(fd, F_DUPFD_CLOEXEC, 0);
 175 #else
 176     ret = dup(fd);
 177     if (ret != -1) {
 178         qemu_set_cloexec(ret);
 179     }
 180 #endif
 181     return ret;
 182 }
 183
 184 static int qemu_parse_fdset(const char *param)
 185 {
 186     return qemu_parse_fd(param);
 187 }
 188
 189 static void qemu_probe_lock_ops(void)
 190 {
 191     if (fcntl_op_setlk == -1) {
 192 #ifdef F_OFD_SETLK
 193         int fd;
 194         int ret;
 195         struct flock fl = {
 196             .l_whence = SEEK_SET,
 197             .l_start  = 0,
 198             .l_len    = 0,
 199             .l_type   = F_WRLCK,
 200         };
 201
 202         fd = open("/dev/null", O_RDWR);
 203         if (fd < 0) {
 204             fprintf(stderr,
 205                     "Failed to open /dev/null for OFD lock probing: %s\n",
 206                     strerror(errno));
 207             fcntl_op_setlk = F_SETLK;
 208             fcntl_op_getlk = F_GETLK;
 209             return;
 210         }
 211         ret = fcntl(fd, F_OFD_GETLK, &fl);
 212         close(fd);
 213         if (!ret) {
 214             fcntl_op_setlk = F_OFD_SETLK;
 215             fcntl_op_getlk = F_OFD_GETLK;
 216         } else {
 217             fcntl_op_setlk = F_SETLK;
 218             fcntl_op_getlk = F_GETLK;
 219         }
 220 #else
 221         fcntl_op_setlk = F_SETLK;
 222         fcntl_op_getlk = F_GETLK;
 223 #endif
 224     }
 225 }
 226
 227 bool qemu_has_ofd_lock(void)
 228 {
 229     qemu_probe_lock_ops();
 230 #ifdef F_OFD_SETLK
 231     return fcntl_op_setlk == F_OFD_SETLK;
 232 #else
 233     return false;
 234 #endif
 235 }
 236
 237 static int qemu_lock_fcntl(int fd, int64_t start, int64_t len, int fl_type)
 238 {
 239     int ret;
 240     struct flock fl = {
 241         .l_whence = SEEK_SET,
 242         .l_start  = start,
 243         .l_len    = len,
 244         .l_type   = fl_type,
 245     };
 246     qemu_probe_lock_ops();
 247     ret = RETRY_ON_EINTR(fcntl(fd, fcntl_op_setlk, &fl));
 248     return ret == -1 ? -errno : 0;
 249 }
 250
 251 int qemu_lock_fd(int fd, int64_t start, int64_t len, bool exclusive)
 252 {
 253     return qemu_lock_fcntl(fd, start, len, exclusive ? F_WRLCK : F_RDLCK);
 254 }
 255
 256 int qemu_unlock_fd(int fd, int64_t start, int64_t len)
 257 {
 258     return qemu_lock_fcntl(fd, start, len, F_UNLCK);
 259 }
 260
 261 int qemu_lock_fd_test(int fd, int64_t start, int64_t len, bool exclusive)
 262 {
 263     int ret;
 264     struct flock fl = {
 265         .l_whence = SEEK_SET,
 266         .l_start  = start,
 267         .l_len    = len,
 268         .l_type   = exclusive ? F_WRLCK : F_RDLCK,
 269     };
 270     qemu_probe_lock_ops();
 271     ret = fcntl(fd, fcntl_op_getlk, &fl);
 272     if (ret == -1) {
 273         return -errno;
 274     } else {
 275         return fl.l_type == F_UNLCK ? 0 : -EAGAIN;
 276     }
 277 }
 278 #endif
 279
 280 static int qemu_open_cloexec(const char *name, int flags, mode_t mode)
 281 {
 282     int ret;
 283 #ifdef O_CLOEXEC
 284     ret = open(name, flags | O_CLOEXEC, mode);
 285 #else
 286     ret = open(name, flags, mode);
 287     if (ret >= 0) {
 288         qemu_set_cloexec(ret);
 289     }
 290 #endif
 291     return ret;
 292 }
 293
 294 /*
 295  * Opens a file with FD_CLOEXEC set
 296  */
 297 static int
 298 qemu_open_internal(const char *name, int flags, mode_t mode, Error **errp)
 299 {
 300     int ret;
 301
 302 #ifndef _WIN32
 303     const char *fdset_id_str;
 304
 305     /* Attempt dup of fd from fd set */
 306     if (strstart(name, "/dev/fdset/", &fdset_id_str)) {
 307         int64_t fdset_id;
 308         int dupfd;
 309
 310         fdset_id = qemu_parse_fdset(fdset_id_str);
 311         if (fdset_id == -1) {
 312             error_setg(errp, "Could not parse fdset %s", name);
 313             errno = EINVAL;
 314             return -1;
 315         }
 316
 317         dupfd = monitor_fdset_dup_fd_add(fdset_id, flags);
 318         if (dupfd == -1) {
 319             error_setg_errno(errp, errno, "Could not dup FD for %s flags %x",
 320                              name, flags);
 321             return -1;
 322         }
 323
 324         return dupfd;
 325     }
 326 #endif
 327
 328     ret = qemu_open_cloexec(name, flags, mode);
 329
 330     if (ret == -1) {
 331         const char *action = flags & O_CREAT ? "create" : "open";
 332 #ifdef O_DIRECT
 333         /* Give more helpful error message for O_DIRECT */
 334         if (errno == EINVAL && (flags & O_DIRECT)) {
 335             ret = open(name, flags & ~O_DIRECT, mode);
 336             if (ret != -1) {
 337                 close(ret);
 338                 error_setg(errp, "Could not %s '%s': "
 339                            "filesystem does not support O_DIRECT",
 340                            action, name);
 341                 errno = EINVAL; /* restore first open()'s errno */
 342                 return -1;
 343             }
 344         }
 345 #endif /* O_DIRECT */
 346         error_setg_errno(errp, errno, "Could not %s '%s'",
 347                          action, name);
 348     }
 349
 350     return ret;
 351 }
 352
 353
 354 int qemu_open(const char *name, int flags, Error **errp)
 355 {
 356     assert(!(flags & O_CREAT));
 357
 358     return qemu_open_internal(name, flags, 0, errp);
 359 }
 360
 361
 362 int qemu_create(const char *name, int flags, mode_t mode, Error **errp)
 363 {
 364     assert(!(flags & O_CREAT));
 365
 366     return qemu_open_internal(name, flags | O_CREAT, mode, errp);
 367 }
 368
 369
 370 int qemu_open_old(const char *name, int flags, ...)
 371 {
 372     va_list ap;
 373     mode_t mode = 0;
 374     int ret;
 375
 376     va_start(ap, flags);
 377     if (flags & O_CREAT) {
 378         mode = va_arg(ap, int);
 379     }
 380     va_end(ap);
 381
 382     ret = qemu_open_internal(name, flags, mode, NULL);
 383
 384 #ifdef O_DIRECT
 385     if (ret == -1 && errno == EINVAL && (flags & O_DIRECT)) {
 386         error_report("file system may not support O_DIRECT");
 387         errno = EINVAL; /* in case it was clobbered */
 388     }
 389 #endif /* O_DIRECT */
 390
 391     return ret;
 392 }
 393
 394 int qemu_close(int fd)
 395 {
 396     int64_t fdset_id;
 397
 398     /* Close fd that was dup'd from an fdset */
 399     fdset_id = monitor_fdset_dup_fd_find(fd);
 400     if (fdset_id != -1) {
 401         int ret;
 402
 403         ret = close(fd);
 404         if (ret == 0) {
 405             monitor_fdset_dup_fd_remove(fd);
 406         }
 407
 408         return ret;
 409     }
 410
 411     return close(fd);
 412 }
 413
 414 /*
 415  * Delete a file from the filesystem, unless the filename is /dev/fdset/...
 416  *
 417  * Returns: On success, zero is returned.  On error, -1 is returned,
 418  * and errno is set appropriately.
 419  */
 420 int qemu_unlink(const char *name)
 421 {
 422     if (g_str_has_prefix(name, "/dev/fdset/")) {
 423         return 0;
 424     }
 425
 426     return unlink(name);
 427 }
 428
 429 /*
 430  * A variant of write(2) which handles partial write.
 431  *
 432  * Return the number of bytes transferred.
 433  * Set errno if fewer than `count' bytes are written.
 434  *
 435  * This function don't work with non-blocking fd's.
 436  * Any of the possibilities with non-blocking fd's is bad:
 437  *   - return a short write (then name is wrong)
 438  *   - busy wait adding (errno == EAGAIN) to the loop
 439  */
 440 ssize_t qemu_write_full(int fd, const void *buf, size_t count)
 441 {
 442     ssize_t ret = 0;
 443     ssize_t total = 0;
 444
 445     while (count) {
 446         ret = write(fd, buf, count);
 447         if (ret < 0) {
 448             if (errno == EINTR)
 449                 continue;
 450             break;
 451         }
 452
 453         count -= ret;
 454         buf += ret;
 455         total += ret;
 456     }
 457
 458     return total;
 459 }
 460
 461 /*
 462  * Opens a socket with FD_CLOEXEC set
 463  */
 464 int qemu_socket(int domain, int type, int protocol)
 465 {
 466     int ret;
 467
 468 #ifdef SOCK_CLOEXEC
 469     ret = socket(domain, type | SOCK_CLOEXEC, protocol);
 470     if (ret != -1 || errno != EINVAL) {
 471         return ret;
 472     }
 473 #endif
 474     ret = socket(domain, type, protocol);
 475     if (ret >= 0) {
 476         qemu_set_cloexec(ret);
 477     }
 478
 479     return ret;
 480 }
 481
 482 /*
 483  * Accept a connection and set FD_CLOEXEC
 484  */
 485 int qemu_accept(int s, struct sockaddr *addr, socklen_t *addrlen)
 486 {
 487     int ret;
 488
 489 #ifdef CONFIG_ACCEPT4
 490     ret = accept4(s, addr, addrlen, SOCK_CLOEXEC);
 491     if (ret != -1 || errno != ENOSYS) {
 492         return ret;
 493     }
 494 #endif
 495     ret = accept(s, addr, addrlen);
 496     if (ret >= 0) {
 497         qemu_set_cloexec(ret);
 498     }
 499
 500     return ret;
 501 }
 502
 503 ssize_t qemu_send_full(int s, const void *buf, size_t count)
 504 {
 505     ssize_t ret = 0;
 506     ssize_t total = 0;
 507
 508     while (count) {
 509         ret = send(s, buf, count, 0);
 510         if (ret < 0) {
 511             if (errno == EINTR) {
 512                 continue;
 513             }
 514             break;
 515         }
 516
 517         count -= ret;
 518         buf += ret;
 519         total += ret;
 520     }
 521
 522     return total;
 523 }
 524
 525 void qemu_set_hw_version(const char *version)
 526 {
 527     hw_version = version;
 528 }
 529
 530 const char *qemu_hw_version(void)
 531 {
 532     return hw_version;
 533 }
 534
 535 #ifdef _WIN32
 536 static void socket_cleanup(void)
 537 {
 538     WSACleanup();
 539 }
 540 #endif
 541
 542 int socket_init(void)
 543 {
 544 #ifdef _WIN32
 545     WSADATA Data;
 546     int ret, err;
 547
 548     ret = WSAStartup(MAKEWORD(2, 2), &Data);
 549     if (ret != 0) {
 550         err = WSAGetLastError();
 551         fprintf(stderr, "WSAStartup: %d\n", err);
 552         return -1;
 553     }
 554     atexit(socket_cleanup);
 555 #endif
 556     return 0;
 557 }
 558
 559
 560 #ifndef CONFIG_IOVEC
 561 static ssize_t
 562 readv_writev(int fd, const struct iovec *iov, int iov_cnt, bool do_write)
 563 {
 564     unsigned i = 0;
 565     ssize_t ret = 0;
 566     ssize_t off = 0;
 567     while (i < iov_cnt) {
 568         ssize_t r = do_write
 569             ? write(fd, iov[i].iov_base + off, iov[i].iov_len - off)
 570             : read(fd, iov[i].iov_base + off, iov[i].iov_len - off);
 571         if (r > 0) {
 572             ret += r;
 573             off += r;
 574             if (off < iov[i].iov_len) {
 575                 continue;
 576             }
 577         } else if (!r) {
 578             break;
 579         } else if (errno == EINTR) {
 580             continue;
 581         } else {
 582             /* else it is some "other" error,
 583              * only return if there was no data processed. */
 584             if (ret == 0) {
 585                 ret = -1;
 586             }
 587             break;
 588         }
 589         off = 0;
 590         i++;
 591     }
 592     return ret;
 593 }
 594
 595 ssize_t
 596 readv(int fd, const struct iovec *iov, int iov_cnt)
 597 {
 598     return readv_writev(fd, iov, iov_cnt, false);
 599 }
 600
 601 ssize_t
 602 writev(int fd, const struct iovec *iov, int iov_cnt)
 603 {
 604     return readv_writev(fd, iov, iov_cnt, true);
 605 }
 606 #endif
 607
 608 /*
 609  * Make sure data goes on disk, but if possible do not bother to
 610  * write out the inode just for timestamp updates.
 611  *
 612  * Unfortunately even in 2009 many operating systems do not support
 613  * fdatasync and have to fall back to fsync.
 614  */
 615 int qemu_fdatasync(int fd)
 616 {
 617 #ifdef CONFIG_FDATASYNC
 618     return fdatasync(fd);
 619 #else
 620     return fsync(fd);
 621 #endif
 622 }