1 /* $NetBSD: hijack.c,v 1.129 2020/02/10 09:10:58 kamil Exp $ */ 2 3 /*- 4 * Copyright (c) 2011 Antti Kantee. All Rights Reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS 16 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 17 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 18 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 21 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 /* 29 * XXX: rumphijack sort of works on glibc Linux. But it's not 30 * the same quality working as on NetBSD. 31 * autoconf HAVE_FOO vs. __NetBSD__ / __linux__ could be further 32 * improved. 33 */ 34 #include <rump/rumpuser_port.h> 35 36 #if !defined(lint) 37 __RCSID("$NetBSD: hijack.c,v 1.129 2020/02/10 09:10:58 kamil Exp $"); 38 #endif 39 40 #include <sys/param.h> 41 #include <sys/types.h> 42 #include <sys/ioctl.h> 43 #include <sys/mman.h> 44 #include <sys/mount.h> 45 #include <sys/socket.h> 46 #include <sys/stat.h> 47 #include <sys/time.h> 48 #include <sys/uio.h> 49 50 #ifdef __NetBSD__ 51 #include <sys/statvfs.h> 52 #endif 53 54 #ifdef HAVE_KQUEUE 55 #include <sys/event.h> 56 #endif 57 58 #ifdef __NetBSD__ 59 #include <sys/quotactl.h> 60 #endif 61 62 #include <assert.h> 63 #include <dlfcn.h> 64 #include <err.h> 65 #include <errno.h> 66 #include <fcntl.h> 67 #include <poll.h> 68 #include <pthread.h> 69 #include <signal.h> 70 #include <stdarg.h> 71 #include <stdbool.h> 72 #include <stdint.h> 73 #include <stdio.h> 74 #include <stdlib.h> 75 #include <string.h> 76 #include <time.h> 77 #include <unistd.h> 78 79 #include <rump/rumpclient.h> 80 #include <rump/rump_syscalls.h> 81 82 #include "hijack.h" 83 84 /* 85 * XXX: Consider autogenerating this, syscnames[] and syscalls[] with 86 * a DSL where the tool also checks the symbols exported by this library 87 * to make sure all relevant calls are accounted for. 88 */ 89 enum dualcall { 90 DUALCALL_WRITE, DUALCALL_WRITEV, DUALCALL_PWRITE, DUALCALL_PWRITEV, 91 DUALCALL_IOCTL, DUALCALL_FCNTL, 92 DUALCALL_SOCKET, DUALCALL_ACCEPT, 93 #ifndef __linux__ 94 DUALCALL_PACCEPT, 95 #endif 96 DUALCALL_BIND, DUALCALL_CONNECT, 97 DUALCALL_GETPEERNAME, DUALCALL_GETSOCKNAME, DUALCALL_LISTEN, 98 DUALCALL_RECVFROM, DUALCALL_RECVMSG, 99 DUALCALL_SENDTO, DUALCALL_SENDMSG, 100 DUALCALL_GETSOCKOPT, DUALCALL_SETSOCKOPT, 101 DUALCALL_SHUTDOWN, 102 DUALCALL_READ, DUALCALL_READV, DUALCALL_PREAD, DUALCALL_PREADV, 103 DUALCALL_DUP2, 104 DUALCALL_CLOSE, 105 DUALCALL_POLLTS, 106 107 #ifndef __linux__ 108 DUALCALL_STAT, DUALCALL_LSTAT, DUALCALL_FSTAT, 109 #endif 110 111 DUALCALL_CHMOD, DUALCALL_LCHMOD, DUALCALL_FCHMOD, 112 DUALCALL_CHOWN, DUALCALL_LCHOWN, DUALCALL_FCHOWN, 113 DUALCALL_OPEN, 114 DUALCALL_CHDIR, DUALCALL_FCHDIR, 115 DUALCALL_LSEEK, 116 DUALCALL_UNLINK, DUALCALL_SYMLINK, DUALCALL_READLINK, 117 DUALCALL_LINK, DUALCALL_RENAME, 118 DUALCALL_MKDIR, DUALCALL_RMDIR, 119 DUALCALL_UTIMES, DUALCALL_LUTIMES, DUALCALL_FUTIMES, 120 DUALCALL_UTIMENSAT, DUALCALL_FUTIMENS, 121 DUALCALL_TRUNCATE, DUALCALL_FTRUNCATE, 122 DUALCALL_FSYNC, 123 DUALCALL_ACCESS, 124 125 #ifndef __linux__ 126 DUALCALL___GETCWD, 127 DUALCALL_GETDENTS, 128 #endif 129 130 #ifndef __linux__ 131 DUALCALL_MKNOD, 132 #endif 133 134 #ifdef __NetBSD__ 135 DUALCALL_GETFH, DUALCALL_FHOPEN, DUALCALL_FHSTAT, DUALCALL_FHSTATVFS1, 136 #endif 137 138 #ifdef HAVE_KQUEUE 139 DUALCALL_KEVENT, 140 #endif 141 142 #ifdef __NetBSD__ 143 DUALCALL___SYSCTL, 144 DUALCALL_MODCTL, 145 #endif 146 147 #ifdef __NetBSD__ 148 DUALCALL_NFSSVC, 149 #endif 150 151 #ifdef __NetBSD__ 152 DUALCALL_STATVFS1, DUALCALL_FSTATVFS1, DUALCALL_GETVFSSTAT, 153 #endif 154 155 #ifdef __NetBSD__ 156 DUALCALL_MOUNT, DUALCALL_UNMOUNT, 157 #endif 158 159 #ifdef HAVE_FSYNC_RANGE 160 DUALCALL_FSYNC_RANGE, 161 #endif 162 163 #ifdef HAVE_CHFLAGS 164 DUALCALL_CHFLAGS, DUALCALL_LCHFLAGS, DUALCALL_FCHFLAGS, 165 #endif 166 167 #ifdef HAVE___QUOTACTL 168 DUALCALL_QUOTACTL, 169 #endif 170 #ifdef __NetBSD__ 171 DUALCALL_LINKAT, 172 #endif 173 DUALCALL__NUM 174 }; 175 176 #define RSYS_STRING(a) __STRING(a) 177 #define RSYS_NAME(a) RSYS_STRING(__CONCAT(RUMP_SYS_RENAME_,a)) 178 179 /* 180 * Would be nice to get this automatically in sync with libc. 181 * Also, this does not work for compat-using binaries (we should 182 * provide all previous interfaces, not just the current ones) 183 */ 184 #if defined(__NetBSD__) 185 186 #if !__NetBSD_Prereq__(5,99,7) 187 #define REALSELECT select 188 #define REALPOLLTS pollts 189 #define REALKEVENT kevent 190 #define REALSTAT __stat30 191 #define REALLSTAT __lstat30 192 #define REALFSTAT __fstat30 193 #define REALUTIMES utimes 194 #define REALLUTIMES lutimes 195 #define REALFUTIMES futimes 196 #define REALMKNOD mknod 197 #define REALFHSTAT __fhstat40 198 #else /* >= 5.99.7 */ 199 #define REALSELECT _sys___select50 200 #define REALPOLLTS _sys___pollts50 201 #define REALKEVENT _sys___kevent50 202 #define REALSTAT __stat50 203 #define REALLSTAT __lstat50 204 #define REALFSTAT __fstat50 205 #define REALUTIMES __utimes50 206 #define REALLUTIMES __lutimes50 207 #define REALFUTIMES __futimes50 208 #define REALMKNOD __mknod50 209 #define REALFHSTAT __fhstat50 210 #endif /* < 5.99.7 */ 211 212 #define REALREAD _sys_read 213 #define REALPREAD _sys_pread 214 #define REALPWRITE _sys_pwrite 215 #define REALGETDENTS __getdents30 216 #define REALMOUNT __mount50 217 #define REALGETFH __getfh30 218 #define REALFHOPEN __fhopen40 219 #if !__NetBSD_Prereq__(9,99,13) 220 #define REALSTATVFS1 statvfs1 221 #define REALFSTATVFS1 fstatvfs1 222 #define REALGETVFSSTAT getvfsstat 223 #define REALFHSTATVFS1 __fhstatvfs140 224 #else 225 #define REALSTATVFS1 __statvfs190 226 #define REALFSTATVFS1 __fstatvfs190 227 #define REALGETVFSSTAT __getvfsstat90 228 #define REALFHSTATVFS1 __fhstatvfs190 229 #endif 230 #define REALSOCKET __socket30 231 232 #define LSEEK_ALIAS _lseek 233 #define VFORK __vfork14 234 235 int REALSTAT(const char *, struct stat *); 236 int REALLSTAT(const char *, struct stat *); 237 int REALFSTAT(int, struct stat *); 238 int REALMKNOD(const char *, mode_t, dev_t); 239 int REALGETDENTS(int, char *, size_t); 240 241 int __getcwd(char *, size_t); 242 243 #elif defined(__linux__) /* glibc, really */ 244 245 #define REALREAD read 246 #define REALPREAD pread 247 #define REALPWRITE pwrite 248 #define REALSELECT select 249 #define REALPOLLTS ppoll 250 #define REALUTIMES utimes 251 #define REALLUTIMES lutimes 252 #define REALFUTIMES futimes 253 #define REALFHSTAT fhstat 254 #define REALSOCKET socket 255 256 #else /* !NetBSD && !linux */ 257 258 #error platform not supported 259 260 #endif /* platform */ 261 262 int REALSELECT(int, fd_set *, fd_set *, fd_set *, struct timeval *); 263 int REALPOLLTS(struct pollfd *, nfds_t, 264 const struct timespec *, const sigset_t *); 265 int REALKEVENT(int, const struct kevent *, size_t, struct kevent *, size_t, 266 const struct timespec *); 267 ssize_t REALREAD(int, void *, size_t); 268 ssize_t REALPREAD(int, void *, size_t, off_t); 269 ssize_t REALPWRITE(int, const void *, size_t, off_t); 270 int REALUTIMES(const char *, const struct timeval [2]); 271 int REALLUTIMES(const char *, const struct timeval [2]); 272 int REALFUTIMES(int, const struct timeval [2]); 273 int REALMOUNT(const char *, const char *, int, void *, size_t); 274 int REALGETFH(const char *, void *, size_t *); 275 int REALFHOPEN(const void *, size_t, int); 276 int REALFHSTAT(const void *, size_t, struct stat *); 277 int REALSTATVFS1(const char *, struct statvfs *, int); 278 int REALFSTATVFS1(int, struct statvfs *, int); 279 int REALFHSTATVFS1(const void *, size_t, struct statvfs *, int); 280 int REALGETVFSSTAT(struct statvfs *, size_t, int); 281 int REALSOCKET(int, int, int); 282 283 #define S(a) __STRING(a) 284 struct sysnames { 285 enum dualcall scm_callnum; 286 const char *scm_hostname; 287 const char *scm_rumpname; 288 } syscnames[] = { 289 { DUALCALL_SOCKET, S(REALSOCKET), RSYS_NAME(SOCKET) }, 290 { DUALCALL_ACCEPT, "accept", RSYS_NAME(ACCEPT) }, 291 #ifndef __linux__ 292 { DUALCALL_PACCEPT, "paccept", RSYS_NAME(PACCEPT) }, 293 #endif 294 { DUALCALL_BIND, "bind", RSYS_NAME(BIND) }, 295 { DUALCALL_CONNECT, "connect", RSYS_NAME(CONNECT) }, 296 { DUALCALL_GETPEERNAME, "getpeername", RSYS_NAME(GETPEERNAME) }, 297 { DUALCALL_GETSOCKNAME, "getsockname", RSYS_NAME(GETSOCKNAME) }, 298 { DUALCALL_LISTEN, "listen", RSYS_NAME(LISTEN) }, 299 { DUALCALL_RECVFROM, "recvfrom", RSYS_NAME(RECVFROM) }, 300 { DUALCALL_RECVMSG, "recvmsg", RSYS_NAME(RECVMSG) }, 301 { DUALCALL_SENDTO, "sendto", RSYS_NAME(SENDTO) }, 302 { DUALCALL_SENDMSG, "sendmsg", RSYS_NAME(SENDMSG) }, 303 { DUALCALL_GETSOCKOPT, "getsockopt", RSYS_NAME(GETSOCKOPT) }, 304 { DUALCALL_SETSOCKOPT, "setsockopt", RSYS_NAME(SETSOCKOPT) }, 305 { DUALCALL_SHUTDOWN, "shutdown", RSYS_NAME(SHUTDOWN) }, 306 { DUALCALL_READ, S(REALREAD), RSYS_NAME(READ) }, 307 { DUALCALL_READV, "readv", RSYS_NAME(READV) }, 308 { DUALCALL_PREAD, S(REALPREAD), RSYS_NAME(PREAD) }, 309 { DUALCALL_PREADV, "preadv", RSYS_NAME(PREADV) }, 310 { DUALCALL_WRITE, "write", RSYS_NAME(WRITE) }, 311 { DUALCALL_WRITEV, "writev", RSYS_NAME(WRITEV) }, 312 { DUALCALL_PWRITE, S(REALPWRITE), RSYS_NAME(PWRITE) }, 313 { DUALCALL_PWRITEV, "pwritev", RSYS_NAME(PWRITEV) }, 314 { DUALCALL_IOCTL, "ioctl", RSYS_NAME(IOCTL) }, 315 { DUALCALL_FCNTL, "fcntl", RSYS_NAME(FCNTL) }, 316 { DUALCALL_DUP2, "dup2", RSYS_NAME(DUP2) }, 317 { DUALCALL_CLOSE, "close", RSYS_NAME(CLOSE) }, 318 { DUALCALL_POLLTS, S(REALPOLLTS), RSYS_NAME(POLLTS) }, 319 #ifndef __linux__ 320 { DUALCALL_STAT, S(REALSTAT), RSYS_NAME(STAT) }, 321 { DUALCALL_LSTAT, S(REALLSTAT), RSYS_NAME(LSTAT) }, 322 { DUALCALL_FSTAT, S(REALFSTAT), RSYS_NAME(FSTAT) }, 323 #endif 324 { DUALCALL_CHOWN, "chown", RSYS_NAME(CHOWN) }, 325 { DUALCALL_LCHOWN, "lchown", RSYS_NAME(LCHOWN) }, 326 { DUALCALL_FCHOWN, "fchown", RSYS_NAME(FCHOWN) }, 327 { DUALCALL_CHMOD, "chmod", RSYS_NAME(CHMOD) }, 328 { DUALCALL_LCHMOD, "lchmod", RSYS_NAME(LCHMOD) }, 329 { DUALCALL_FCHMOD, "fchmod", RSYS_NAME(FCHMOD) }, 330 { DUALCALL_UTIMES, S(REALUTIMES), RSYS_NAME(UTIMES) }, 331 { DUALCALL_LUTIMES, S(REALLUTIMES), RSYS_NAME(LUTIMES) }, 332 { DUALCALL_FUTIMES, S(REALFUTIMES), RSYS_NAME(FUTIMES) }, 333 { DUALCALL_UTIMENSAT, "utimensat", RSYS_NAME(UTIMENSAT) }, 334 { DUALCALL_FUTIMENS, "futimens", RSYS_NAME(FUTIMENS) }, 335 { DUALCALL_OPEN, "open", RSYS_NAME(OPEN) }, 336 { DUALCALL_CHDIR, "chdir", RSYS_NAME(CHDIR) }, 337 { DUALCALL_FCHDIR, "fchdir", RSYS_NAME(FCHDIR) }, 338 { DUALCALL_LSEEK, "lseek", RSYS_NAME(LSEEK) }, 339 { DUALCALL_UNLINK, "unlink", RSYS_NAME(UNLINK) }, 340 { DUALCALL_SYMLINK, "symlink", RSYS_NAME(SYMLINK) }, 341 { DUALCALL_READLINK, "readlink", RSYS_NAME(READLINK) }, 342 { DUALCALL_LINK, "link", RSYS_NAME(LINK) }, 343 { DUALCALL_RENAME, "rename", RSYS_NAME(RENAME) }, 344 { DUALCALL_MKDIR, "mkdir", RSYS_NAME(MKDIR) }, 345 { DUALCALL_RMDIR, "rmdir", RSYS_NAME(RMDIR) }, 346 { DUALCALL_TRUNCATE, "truncate", RSYS_NAME(TRUNCATE) }, 347 { DUALCALL_FTRUNCATE, "ftruncate", RSYS_NAME(FTRUNCATE) }, 348 { DUALCALL_FSYNC, "fsync", RSYS_NAME(FSYNC) }, 349 { DUALCALL_ACCESS, "access", RSYS_NAME(ACCESS) }, 350 351 #ifndef __linux__ 352 { DUALCALL___GETCWD, "__getcwd", RSYS_NAME(__GETCWD) }, 353 { DUALCALL_GETDENTS, S(REALGETDENTS),RSYS_NAME(GETDENTS) }, 354 #endif 355 356 #ifndef __linux__ 357 { DUALCALL_MKNOD, S(REALMKNOD), RSYS_NAME(MKNOD) }, 358 #endif 359 360 #ifdef __NetBSD__ 361 { DUALCALL_GETFH, S(REALGETFH), RSYS_NAME(GETFH) }, 362 { DUALCALL_FHOPEN, S(REALFHOPEN), RSYS_NAME(FHOPEN) }, 363 { DUALCALL_FHSTAT, S(REALFHSTAT), RSYS_NAME(FHSTAT) }, 364 { DUALCALL_FHSTATVFS1, S(REALFHSTATVFS1),RSYS_NAME(FHSTATVFS1) }, 365 #endif 366 367 #ifdef HAVE_KQUEUE 368 { DUALCALL_KEVENT, S(REALKEVENT), RSYS_NAME(KEVENT) }, 369 #endif 370 371 #ifdef __NetBSD__ 372 { DUALCALL___SYSCTL, "__sysctl", RSYS_NAME(__SYSCTL) }, 373 { DUALCALL_MODCTL, "modctl", RSYS_NAME(MODCTL) }, 374 #endif 375 376 #ifdef __NetBSD__ 377 { DUALCALL_NFSSVC, "nfssvc", RSYS_NAME(NFSSVC) }, 378 #endif 379 380 #ifdef __NetBSD__ 381 { DUALCALL_STATVFS1, S(REALSTATVFS1),RSYS_NAME(STATVFS1) }, 382 { DUALCALL_FSTATVFS1, S(REALFSTATVFS1),RSYS_NAME(FSTATVFS1) }, 383 { DUALCALL_GETVFSSTAT, S(REALGETVFSSTAT),RSYS_NAME(GETVFSSTAT) }, 384 #endif 385 386 #ifdef __NetBSD__ 387 { DUALCALL_MOUNT, S(REALMOUNT), RSYS_NAME(MOUNT) }, 388 { DUALCALL_UNMOUNT, "unmount", RSYS_NAME(UNMOUNT) }, 389 #endif 390 391 #ifdef HAVE_FSYNC_RANGE 392 { DUALCALL_FSYNC_RANGE, "fsync_range", RSYS_NAME(FSYNC_RANGE) }, 393 #endif 394 395 #ifdef HAVE_CHFLAGS 396 { DUALCALL_CHFLAGS, "chflags", RSYS_NAME(CHFLAGS) }, 397 { DUALCALL_LCHFLAGS, "lchflags", RSYS_NAME(LCHFLAGS) }, 398 { DUALCALL_FCHFLAGS, "fchflags", RSYS_NAME(FCHFLAGS) }, 399 #endif /* HAVE_CHFLAGS */ 400 401 #ifdef HAVE___QUOTACTL 402 { DUALCALL_QUOTACTL, "__quotactl", RSYS_NAME(__QUOTACTL) }, 403 #endif /* HAVE___QUOTACTL */ 404 405 #ifdef __NetBSD__ 406 { DUALCALL_LINKAT, "linkat", RSYS_NAME(LINKAT) }, 407 #endif 408 }; 409 #undef S 410 411 struct bothsys { 412 void *bs_host; 413 void *bs_rump; 414 } syscalls[DUALCALL__NUM]; 415 #define GETSYSCALL(which, name) syscalls[DUALCALL_##name].bs_##which 416 417 static pid_t (*host_fork)(void); 418 static int (*host_daemon)(int, int); 419 static void * (*host_mmap)(void *, size_t, int, int, int, off_t); 420 421 /* 422 * This tracks if our process is in a subdirectory of /rump. 423 * It's preserved over exec. 424 */ 425 static bool pwdinrump; 426 427 enum pathtype { PATH_HOST, PATH_RUMP, PATH_RUMPBLANKET }; 428 429 static bool fd_isrump(int); 430 static enum pathtype path_isrump(const char *); 431 432 /* default FD_SETSIZE is 256 ==> default fdoff is 128 */ 433 static int hijack_fdoff = FD_SETSIZE/2; 434 435 /* 436 * Maintain a mapping table for the usual dup2 suspects. 437 * Could use atomic ops to operate on dup2vec, but an application 438 * racing there is not well-defined, so don't bother. 439 */ 440 /* note: you cannot change this without editing the env-passing code */ 441 #define DUP2HIGH 2 442 static uint32_t dup2vec[DUP2HIGH+1]; 443 #define DUP2BIT (1U<<31) 444 #define DUP2ALIAS (1<<30) 445 #define DUP2FDMASK ((1<<30)-1) 446 447 static bool 448 isdup2d(int fd) 449 { 450 451 return fd <= DUP2HIGH && fd >= 0 && dup2vec[fd] & DUP2BIT; 452 } 453 454 static int 455 mapdup2(int hostfd) 456 { 457 458 _DIAGASSERT(isdup2d(hostfd)); 459 return dup2vec[hostfd] & DUP2FDMASK; 460 } 461 462 static int 463 unmapdup2(int rumpfd) 464 { 465 int i; 466 467 for (i = 0; i <= DUP2HIGH; i++) { 468 if (dup2vec[i] & DUP2BIT && 469 (dup2vec[i] & DUP2FDMASK) == (unsigned)rumpfd) 470 return i; 471 } 472 return -1; 473 } 474 475 static void 476 setdup2(int hostfd, int rumpfd) 477 { 478 479 if (hostfd > DUP2HIGH) { 480 _DIAGASSERT(0); 481 return; 482 } 483 484 dup2vec[hostfd] = DUP2BIT | DUP2ALIAS | rumpfd; 485 } 486 487 static void 488 clrdup2(int hostfd) 489 { 490 491 if (hostfd > DUP2HIGH) { 492 _DIAGASSERT(0); 493 return; 494 } 495 496 dup2vec[hostfd] = 0; 497 } 498 499 static bool 500 killdup2alias(int rumpfd) 501 { 502 int hostfd; 503 504 if ((hostfd = unmapdup2(rumpfd)) == -1) 505 return false; 506 507 if (dup2vec[hostfd] & DUP2ALIAS) { 508 dup2vec[hostfd] &= ~DUP2ALIAS; 509 return true; 510 } 511 return false; 512 } 513 514 //#define DEBUGJACK 515 #ifdef DEBUGJACK 516 #define DPRINTF(x) mydprintf x 517 static void 518 mydprintf(const char *fmt, ...) 519 { 520 va_list ap; 521 522 if (isdup2d(STDERR_FILENO)) 523 return; 524 525 va_start(ap, fmt); 526 vfprintf(stderr, fmt, ap); 527 va_end(ap); 528 } 529 530 static const char * 531 whichfd(int fd) 532 { 533 534 if (fd == -1) 535 return "-1"; 536 else if (fd_isrump(fd)) 537 return "rump"; 538 else 539 return "host"; 540 } 541 542 static const char * 543 whichpath(const char *path) 544 { 545 546 if (path_isrump(path)) 547 return "rump"; 548 else 549 return "host"; 550 } 551 552 #else 553 #define DPRINTF(x) 554 #endif 555 556 #define ATCALL(type, name, rcname, args, proto, vars) \ 557 type name args \ 558 { \ 559 type (*fun) proto; \ 560 int isrump = -1; \ 561 \ 562 if (fd == AT_FDCWD || *path == '/') { \ 563 isrump = path_isrump(path); \ 564 } else { \ 565 isrump = fd_isrump(fd); \ 566 } \ 567 \ 568 DPRINTF(("%s -> %d:%s (%s)\n", __STRING(name), \ 569 fd, path, isrump ? "rump" : "host")); \ 570 \ 571 assert(isrump != -1); \ 572 if (isrump) { \ 573 fun = syscalls[rcname].bs_rump; \ 574 if (fd != AT_FDCWD) \ 575 fd = fd_host2rump(fd); \ 576 path = path_host2rump(path); \ 577 } else { \ 578 fun = syscalls[rcname].bs_host; \ 579 } \ 580 return fun vars; \ 581 } 582 583 #define FDCALL(type, name, rcname, args, proto, vars) \ 584 type name args \ 585 { \ 586 type (*fun) proto; \ 587 \ 588 DPRINTF(("%s -> %d (%s)\n", __STRING(name), fd, whichfd(fd))); \ 589 if (fd_isrump(fd)) { \ 590 fun = syscalls[rcname].bs_rump; \ 591 fd = fd_host2rump(fd); \ 592 } else { \ 593 fun = syscalls[rcname].bs_host; \ 594 } \ 595 \ 596 return fun vars; \ 597 } 598 599 #define PATHCALL(type, name, rcname, args, proto, vars) \ 600 type name args \ 601 { \ 602 type (*fun) proto; \ 603 enum pathtype pt; \ 604 \ 605 DPRINTF(("%s -> %s (%s)\n", __STRING(name), path, \ 606 whichpath(path))); \ 607 if ((pt = path_isrump(path)) != PATH_HOST) { \ 608 fun = syscalls[rcname].bs_rump; \ 609 if (pt == PATH_RUMP) \ 610 path = path_host2rump(path); \ 611 } else { \ 612 fun = syscalls[rcname].bs_host; \ 613 } \ 614 \ 615 return fun vars; \ 616 } 617 618 #define VFSCALL(bit, type, name, rcname, args, proto, vars) \ 619 type name args \ 620 { \ 621 type (*fun) proto; \ 622 \ 623 DPRINTF(("%s (0x%x, 0x%x)\n", __STRING(name), bit, vfsbits)); \ 624 if (vfsbits & bit) { \ 625 fun = syscalls[rcname].bs_rump; \ 626 } else { \ 627 fun = syscalls[rcname].bs_host; \ 628 } \ 629 \ 630 return fun vars; \ 631 } 632 633 /* 634 * These variables are set from the RUMPHIJACK string and control 635 * which operations can product rump kernel file descriptors. 636 * This should be easily extendable for future needs. 637 */ 638 #define RUMPHIJACK_DEFAULT "path=/rump,socket=all:nolocal" 639 static bool rumpsockets[PF_MAX]; 640 static const char *rumpprefix; 641 static size_t rumpprefixlen; 642 643 static struct { 644 int pf; 645 const char *name; 646 } socketmap[] = { 647 { PF_LOCAL, "local" }, 648 { PF_INET, "inet" }, 649 #ifdef PF_LINK 650 { PF_LINK, "link" }, 651 #endif 652 #ifdef PF_OROUTE 653 { PF_OROUTE, "oroute" }, 654 #endif 655 { PF_ROUTE, "route" }, 656 { PF_INET6, "inet6" }, 657 #ifdef PF_MPLS 658 { PF_MPLS, "mpls" }, 659 #endif 660 { -1, NULL } 661 }; 662 663 static void 664 sockparser(char *buf) 665 { 666 char *p, *l = NULL; 667 bool value; 668 int i; 669 670 /* if "all" is present, it must be specified first */ 671 if (strncmp(buf, "all", strlen("all")) == 0) { 672 for (i = 0; i < (int)__arraycount(rumpsockets); i++) { 673 rumpsockets[i] = true; 674 } 675 buf += strlen("all"); 676 if (*buf == ':') 677 buf++; 678 } 679 680 for (p = strtok_r(buf, ":", &l); p; p = strtok_r(NULL, ":", &l)) { 681 value = true; 682 if (strncmp(p, "no", strlen("no")) == 0) { 683 value = false; 684 p += strlen("no"); 685 } 686 687 for (i = 0; socketmap[i].name; i++) { 688 if (strcmp(p, socketmap[i].name) == 0) { 689 rumpsockets[socketmap[i].pf] = value; 690 break; 691 } 692 } 693 if (socketmap[i].name == NULL) { 694 errx(1, "invalid socket specifier %s", p); 695 } 696 } 697 } 698 699 static void 700 pathparser(char *buf) 701 { 702 703 /* sanity-check */ 704 if (*buf != '/') 705 errx(1, "hijack path specifier must begin with ``/''"); 706 rumpprefixlen = strlen(buf); 707 if (rumpprefixlen < 2) 708 errx(1, "invalid hijack prefix: %s", buf); 709 if (buf[rumpprefixlen-1] == '/' && strspn(buf, "/") != rumpprefixlen) 710 errx(1, "hijack prefix may end in slash only if pure " 711 "slash, gave %s", buf); 712 713 if ((rumpprefix = strdup(buf)) == NULL) 714 err(1, "strdup"); 715 rumpprefixlen = strlen(rumpprefix); 716 } 717 718 static struct blanket { 719 const char *pfx; 720 size_t len; 721 } *blanket; 722 static int nblanket; 723 724 static void 725 blanketparser(char *buf) 726 { 727 char *p, *l = NULL; 728 int i; 729 730 for (nblanket = 0, p = buf; p; p = strchr(p+1, ':'), nblanket++) 731 continue; 732 733 blanket = malloc(nblanket * sizeof(*blanket)); 734 if (blanket == NULL) 735 err(1, "alloc blanket %d", nblanket); 736 737 for (p = strtok_r(buf, ":", &l), i = 0; p; 738 p = strtok_r(NULL, ":", &l), i++) { 739 blanket[i].pfx = strdup(p); 740 if (blanket[i].pfx == NULL) 741 err(1, "strdup blanket"); 742 blanket[i].len = strlen(p); 743 744 if (blanket[i].len == 0 || *blanket[i].pfx != '/') 745 errx(1, "invalid blanket specifier %s", p); 746 if (*(blanket[i].pfx + blanket[i].len-1) == '/') 747 errx(1, "invalid blanket specifier %s", p); 748 } 749 } 750 751 #define VFSBIT_NFSSVC 0x01 752 #define VFSBIT_GETVFSSTAT 0x02 753 #define VFSBIT_FHCALLS 0x04 754 static unsigned vfsbits; 755 756 static struct { 757 int bit; 758 const char *name; 759 } vfscalls[] = { 760 { VFSBIT_NFSSVC, "nfssvc" }, 761 { VFSBIT_GETVFSSTAT, "getvfsstat" }, 762 { VFSBIT_FHCALLS, "fhcalls" }, 763 { -1, NULL } 764 }; 765 766 static void 767 vfsparser(char *buf) 768 { 769 char *p, *l = NULL; 770 bool turnon; 771 unsigned int fullmask; 772 int i; 773 774 /* build the full mask and sanity-check while we're at it */ 775 fullmask = 0; 776 for (i = 0; vfscalls[i].name != NULL; i++) { 777 if (fullmask & vfscalls[i].bit) 778 errx(1, "problem exists between vi and chair"); 779 fullmask |= vfscalls[i].bit; 780 } 781 782 783 /* if "all" is present, it must be specified first */ 784 if (strncmp(buf, "all", strlen("all")) == 0) { 785 vfsbits = fullmask; 786 buf += strlen("all"); 787 if (*buf == ':') 788 buf++; 789 } 790 791 for (p = strtok_r(buf, ":", &l); p; p = strtok_r(NULL, ":", &l)) { 792 turnon = true; 793 if (strncmp(p, "no", strlen("no")) == 0) { 794 turnon = false; 795 p += strlen("no"); 796 } 797 798 for (i = 0; vfscalls[i].name; i++) { 799 if (strcmp(p, vfscalls[i].name) == 0) { 800 if (turnon) 801 vfsbits |= vfscalls[i].bit; 802 else 803 vfsbits &= ~vfscalls[i].bit; 804 break; 805 } 806 } 807 if (vfscalls[i].name == NULL) { 808 errx(1, "invalid vfscall specifier %s", p); 809 } 810 } 811 } 812 813 static bool rumpsysctl = false; 814 815 static void 816 sysctlparser(char *buf) 817 { 818 819 if (buf == NULL) { 820 rumpsysctl = true; 821 return; 822 } 823 824 if (strcasecmp(buf, "y") == 0 || strcasecmp(buf, "yes") == 0 || 825 strcasecmp(buf, "yep") == 0 || strcasecmp(buf, "tottakai") == 0) { 826 rumpsysctl = true; 827 return; 828 } 829 if (strcasecmp(buf, "n") == 0 || strcasecmp(buf, "no") == 0) { 830 rumpsysctl = false; 831 return; 832 } 833 834 errx(1, "sysctl value should be y(es)/n(o), gave: %s", buf); 835 } 836 837 static bool rumpmodctl = false; 838 839 static void 840 modctlparser(char *buf) 841 { 842 843 if (buf == NULL) { 844 rumpmodctl = true; 845 return; 846 } 847 848 if (strcasecmp(buf, "y") == 0 || strcasecmp(buf, "yes") == 0 || 849 strcasecmp(buf, "yep") == 0 || strcasecmp(buf, "tottakai") == 0) { 850 rumpmodctl = true; 851 return; 852 } 853 if (strcasecmp(buf, "n") == 0 || strcasecmp(buf, "no") == 0) { 854 rumpmodctl = false; 855 return; 856 } 857 858 errx(1, "modctl value should be y(es)/n(o), gave: %s", buf); 859 } 860 861 static void 862 fdoffparser(char *buf) 863 { 864 unsigned long fdoff; 865 char *ep; 866 867 if (*buf == '-') { 868 errx(1, "fdoff must not be negative"); 869 } 870 fdoff = strtoul(buf, &ep, 10); 871 if (*ep != '\0') 872 errx(1, "invalid fdoff specifier \"%s\"", buf); 873 if (fdoff >= INT_MAX/2 || fdoff < 3) 874 errx(1, "fdoff out of range"); 875 hijack_fdoff = fdoff; 876 } 877 878 static struct { 879 void (*parsefn)(char *); 880 const char *name; 881 bool needvalues; 882 } hijackparse[] = { 883 { sockparser, "socket", true }, 884 { pathparser, "path", true }, 885 { blanketparser, "blanket", true }, 886 { vfsparser, "vfs", true }, 887 { sysctlparser, "sysctl", false }, 888 { modctlparser, "modctl", false }, 889 { fdoffparser, "fdoff", true }, 890 { NULL, NULL, false }, 891 }; 892 893 static void 894 parsehijack(char *hijack) 895 { 896 char *p, *p2, *l; 897 const char *hijackcopy; 898 bool nop2; 899 int i; 900 901 if ((hijackcopy = strdup(hijack)) == NULL) 902 err(1, "strdup"); 903 904 /* disable everything explicitly */ 905 for (i = 0; i < PF_MAX; i++) 906 rumpsockets[i] = false; 907 908 for (p = strtok_r(hijack, ",", &l); p; p = strtok_r(NULL, ",", &l)) { 909 nop2 = false; 910 p2 = strchr(p, '='); 911 if (!p2) { 912 nop2 = true; 913 p2 = p + strlen(p); 914 } 915 916 for (i = 0; hijackparse[i].parsefn; i++) { 917 if (strncmp(hijackparse[i].name, p, 918 (size_t)(p2-p)) == 0) { 919 if (nop2 && hijackparse[i].needvalues) 920 errx(1, "invalid hijack specifier: %s", 921 hijackcopy); 922 hijackparse[i].parsefn(nop2 ? NULL : p2+1); 923 break; 924 } 925 } 926 927 if (hijackparse[i].parsefn == NULL) 928 errx(1, "invalid hijack specifier name in %s", p); 929 } 930 931 } 932 933 static void __attribute__((constructor)) 934 rcinit(void) 935 { 936 char buf[1024]; 937 unsigned i, j; 938 939 host_fork = dlsym(RTLD_NEXT, "fork"); 940 host_daemon = dlsym(RTLD_NEXT, "daemon"); 941 if (host_mmap == NULL) 942 host_mmap = dlsym(RTLD_NEXT, "mmap"); 943 944 /* 945 * In theory cannot print anything during lookups because 946 * we might not have the call vector set up. so, the errx() 947 * is a bit of a strech, but it might work. 948 */ 949 950 for (i = 0; i < DUALCALL__NUM; i++) { 951 /* build runtime O(1) access */ 952 for (j = 0; j < __arraycount(syscnames); j++) { 953 if (syscnames[j].scm_callnum == i) 954 break; 955 } 956 957 if (j == __arraycount(syscnames)) 958 errx(1, "rumphijack error: syscall pos %d missing", i); 959 960 syscalls[i].bs_host = dlsym(RTLD_NEXT, 961 syscnames[j].scm_hostname); 962 if (syscalls[i].bs_host == NULL) 963 errx(1, "hostcall %s not found!", 964 syscnames[j].scm_hostname); 965 966 syscalls[i].bs_rump = dlsym(RTLD_NEXT, 967 syscnames[j].scm_rumpname); 968 if (syscalls[i].bs_rump == NULL) 969 errx(1, "rumpcall %s not found!", 970 syscnames[j].scm_rumpname); 971 } 972 973 if (rumpclient_init() == -1) 974 err(1, "rumpclient init"); 975 976 /* check which syscalls we're supposed to hijack */ 977 if (getenv_r("RUMPHIJACK", buf, sizeof(buf)) == -1) { 978 strcpy(buf, RUMPHIJACK_DEFAULT); 979 } 980 parsehijack(buf); 981 982 /* set client persistence level */ 983 if (getenv_r("RUMPHIJACK_RETRYCONNECT", buf, sizeof(buf)) != -1) { 984 if (strcmp(buf, "die") == 0) 985 rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_DIE); 986 else if (strcmp(buf, "inftime") == 0) 987 rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_INFTIME); 988 else if (strcmp(buf, "once") == 0) 989 rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_ONCE); 990 else { 991 time_t timeout; 992 char *ep; 993 994 timeout = (time_t)strtoll(buf, &ep, 10); 995 if (timeout <= 0 || ep != buf + strlen(buf)) 996 errx(1, "RUMPHIJACK_RETRYCONNECT must be " 997 "keyword or integer, got: %s", buf); 998 999 rumpclient_setconnretry(timeout); 1000 } 1001 } 1002 1003 if (getenv_r("RUMPHIJACK__DUP2INFO", buf, sizeof(buf)) == 0) { 1004 if (sscanf(buf, "%u,%u,%u", 1005 &dup2vec[0], &dup2vec[1], &dup2vec[2]) != 3) { 1006 warnx("invalid dup2mask: %s", buf); 1007 memset(dup2vec, 0, sizeof(dup2vec)); 1008 } 1009 unsetenv("RUMPHIJACK__DUP2INFO"); 1010 } 1011 if (getenv_r("RUMPHIJACK__PWDINRUMP", buf, sizeof(buf)) == 0) { 1012 pwdinrump = true; 1013 unsetenv("RUMPHIJACK__PWDINRUMP"); 1014 } 1015 } 1016 1017 static int 1018 fd_rump2host(int fd) 1019 { 1020 1021 if (fd == -1) 1022 return fd; 1023 return fd + hijack_fdoff; 1024 } 1025 1026 static int 1027 fd_rump2host_withdup(int fd) 1028 { 1029 int hfd; 1030 1031 _DIAGASSERT(fd != -1); 1032 hfd = unmapdup2(fd); 1033 if (hfd != -1) { 1034 _DIAGASSERT(hfd <= DUP2HIGH); 1035 return hfd; 1036 } 1037 return fd_rump2host(fd); 1038 } 1039 1040 static int 1041 fd_host2rump(int fd) 1042 { 1043 if (!isdup2d(fd)) 1044 return fd - hijack_fdoff; 1045 else 1046 return mapdup2(fd); 1047 } 1048 1049 static bool 1050 fd_isrump(int fd) 1051 { 1052 1053 return isdup2d(fd) || fd >= hijack_fdoff; 1054 } 1055 1056 #define assertfd(_fd_) assert(ISDUP2D(_fd_) || (_fd_) >= hijack_fdoff) 1057 1058 static enum pathtype 1059 path_isrump(const char *path) 1060 { 1061 size_t plen; 1062 int i; 1063 1064 if (rumpprefix == NULL && nblanket == 0) 1065 return PATH_HOST; 1066 1067 if (*path == '/') { 1068 plen = strlen(path); 1069 if (rumpprefix && plen >= rumpprefixlen) { 1070 if (strncmp(path, rumpprefix, rumpprefixlen) == 0 1071 && (plen == rumpprefixlen 1072 || *(path + rumpprefixlen) == '/')) { 1073 return PATH_RUMP; 1074 } 1075 } 1076 for (i = 0; i < nblanket; i++) { 1077 if (strncmp(path, blanket[i].pfx, blanket[i].len) == 0) 1078 return PATH_RUMPBLANKET; 1079 } 1080 1081 return PATH_HOST; 1082 } else { 1083 return pwdinrump ? PATH_RUMP : PATH_HOST; 1084 } 1085 } 1086 1087 static const char *rootpath = "/"; 1088 static const char * 1089 path_host2rump(const char *path) 1090 { 1091 const char *rv; 1092 1093 if (*path == '/') { 1094 rv = path + rumpprefixlen; 1095 if (*rv == '\0') 1096 rv = rootpath; 1097 } else { 1098 rv = path; 1099 } 1100 1101 return rv; 1102 } 1103 1104 static int 1105 dodup(int oldd, int minfd) 1106 { 1107 int (*op_fcntl)(int, int, ...); 1108 int newd; 1109 int isrump; 1110 1111 DPRINTF(("dup -> %d (minfd %d)\n", oldd, minfd)); 1112 if (fd_isrump(oldd)) { 1113 op_fcntl = GETSYSCALL(rump, FCNTL); 1114 oldd = fd_host2rump(oldd); 1115 if (minfd >= hijack_fdoff) 1116 minfd -= hijack_fdoff; 1117 isrump = 1; 1118 } else { 1119 if (minfd >= hijack_fdoff) { 1120 errno = EINVAL; 1121 return -1; 1122 } 1123 op_fcntl = GETSYSCALL(host, FCNTL); 1124 isrump = 0; 1125 } 1126 1127 newd = op_fcntl(oldd, F_DUPFD, minfd); 1128 1129 if (isrump) 1130 newd = fd_rump2host(newd); 1131 DPRINTF(("dup <- %d\n", newd)); 1132 1133 return newd; 1134 } 1135 1136 /* 1137 * Check that host fd value does not exceed fdoffset and if necessary 1138 * dup the file descriptor so that it doesn't collide with the dup2mask. 1139 */ 1140 static int 1141 fd_host2host(int fd) 1142 { 1143 int (*op_fcntl)(int, int, ...) = GETSYSCALL(host, FCNTL); 1144 int (*op_close)(int) = GETSYSCALL(host, CLOSE); 1145 int ofd, i; 1146 1147 if (fd >= hijack_fdoff) { 1148 op_close(fd); 1149 errno = ENFILE; 1150 return -1; 1151 } 1152 1153 for (i = 1; isdup2d(fd); i++) { 1154 ofd = fd; 1155 fd = op_fcntl(ofd, F_DUPFD, i); 1156 op_close(ofd); 1157 } 1158 1159 return fd; 1160 } 1161 1162 int 1163 open(const char *path, int flags, ...) 1164 { 1165 int (*op_open)(const char *, int, ...); 1166 bool isrump; 1167 va_list ap; 1168 enum pathtype pt; 1169 int fd; 1170 1171 DPRINTF(("open -> %s (%s)\n", path, whichpath(path))); 1172 1173 if ((pt = path_isrump(path)) != PATH_HOST) { 1174 if (pt == PATH_RUMP) 1175 path = path_host2rump(path); 1176 op_open = GETSYSCALL(rump, OPEN); 1177 isrump = true; 1178 } else { 1179 op_open = GETSYSCALL(host, OPEN); 1180 isrump = false; 1181 } 1182 1183 va_start(ap, flags); 1184 fd = op_open(path, flags, va_arg(ap, mode_t)); 1185 va_end(ap); 1186 1187 if (isrump) 1188 fd = fd_rump2host(fd); 1189 else 1190 fd = fd_host2host(fd); 1191 1192 DPRINTF(("open <- %d (%s)\n", fd, whichfd(fd))); 1193 return fd; 1194 } 1195 1196 int 1197 chdir(const char *path) 1198 { 1199 int (*op_chdir)(const char *); 1200 enum pathtype pt; 1201 int rv; 1202 1203 if ((pt = path_isrump(path)) != PATH_HOST) { 1204 op_chdir = GETSYSCALL(rump, CHDIR); 1205 if (pt == PATH_RUMP) 1206 path = path_host2rump(path); 1207 } else { 1208 op_chdir = GETSYSCALL(host, CHDIR); 1209 } 1210 1211 rv = op_chdir(path); 1212 if (rv == 0) 1213 pwdinrump = pt != PATH_HOST; 1214 1215 return rv; 1216 } 1217 1218 int 1219 fchdir(int fd) 1220 { 1221 int (*op_fchdir)(int); 1222 bool isrump; 1223 int rv; 1224 1225 if (fd_isrump(fd)) { 1226 op_fchdir = GETSYSCALL(rump, FCHDIR); 1227 isrump = true; 1228 fd = fd_host2rump(fd); 1229 } else { 1230 op_fchdir = GETSYSCALL(host, FCHDIR); 1231 isrump = false; 1232 } 1233 1234 rv = op_fchdir(fd); 1235 if (rv == 0) { 1236 pwdinrump = isrump; 1237 } 1238 1239 return rv; 1240 } 1241 1242 #ifndef __linux__ 1243 int 1244 __getcwd(char *bufp, size_t len) 1245 { 1246 int (*op___getcwd)(char *, size_t); 1247 size_t prefixgap; 1248 bool iamslash; 1249 int rv; 1250 1251 if (pwdinrump && rumpprefix) { 1252 if (rumpprefix[rumpprefixlen-1] == '/') 1253 iamslash = true; 1254 else 1255 iamslash = false; 1256 1257 if (iamslash) 1258 prefixgap = rumpprefixlen - 1; /* ``//+path'' */ 1259 else 1260 prefixgap = rumpprefixlen; /* ``/pfx+/path'' */ 1261 if (len <= prefixgap) { 1262 errno = ERANGE; 1263 return -1; 1264 } 1265 1266 op___getcwd = GETSYSCALL(rump, __GETCWD); 1267 rv = op___getcwd(bufp + prefixgap, len - prefixgap); 1268 if (rv == -1) 1269 return rv; 1270 1271 /* augment the "/" part only for a non-root path */ 1272 memcpy(bufp, rumpprefix, rumpprefixlen); 1273 1274 /* append / only to non-root cwd */ 1275 if (rv != 2) 1276 bufp[prefixgap] = '/'; 1277 1278 /* don't append extra slash in the purely-slash case */ 1279 if (rv == 2 && !iamslash) 1280 bufp[rumpprefixlen] = '\0'; 1281 } else if (pwdinrump) { 1282 /* assume blanket. we can't provide a prefix here */ 1283 op___getcwd = GETSYSCALL(rump, __GETCWD); 1284 rv = op___getcwd(bufp, len); 1285 } else { 1286 op___getcwd = GETSYSCALL(host, __GETCWD); 1287 rv = op___getcwd(bufp, len); 1288 } 1289 1290 return rv; 1291 } 1292 #endif 1293 1294 static int 1295 moveish(const char *from, const char *to, 1296 int (*rump_op)(const char *, const char *), 1297 int (*host_op)(const char *, const char *)) 1298 { 1299 int (*op)(const char *, const char *); 1300 enum pathtype ptf, ptt; 1301 1302 if ((ptf = path_isrump(from)) != PATH_HOST) { 1303 if ((ptt = path_isrump(to)) == PATH_HOST) { 1304 errno = EXDEV; 1305 return -1; 1306 } 1307 1308 if (ptf == PATH_RUMP) 1309 from = path_host2rump(from); 1310 if (ptt == PATH_RUMP) 1311 to = path_host2rump(to); 1312 op = rump_op; 1313 } else { 1314 if (path_isrump(to) != PATH_HOST) { 1315 errno = EXDEV; 1316 return -1; 1317 } 1318 1319 op = host_op; 1320 } 1321 1322 return op(from, to); 1323 } 1324 1325 #ifdef __NetBSD__ 1326 int 1327 linkat(int fromfd, const char *from, int tofd, const char *to, int flags) 1328 { 1329 if (fromfd != AT_FDCWD || tofd != AT_FDCWD 1330 || flags != AT_SYMLINK_FOLLOW) 1331 return ENOSYS; 1332 1333 return moveish(from, to, 1334 GETSYSCALL(rump, LINK), GETSYSCALL(host, LINK)); 1335 } 1336 #endif 1337 1338 int 1339 link(const char *from, const char *to) 1340 { 1341 return moveish(from, to, 1342 GETSYSCALL(rump, LINK), GETSYSCALL(host, LINK)); 1343 } 1344 1345 int 1346 rename(const char *from, const char *to) 1347 { 1348 return moveish(from, to, 1349 GETSYSCALL(rump, RENAME), GETSYSCALL(host, RENAME)); 1350 } 1351 1352 int 1353 REALSOCKET(int domain, int type, int protocol) 1354 { 1355 int (*op_socket)(int, int, int); 1356 int fd; 1357 bool isrump; 1358 1359 isrump = domain < PF_MAX && rumpsockets[domain]; 1360 1361 if (isrump) 1362 op_socket = GETSYSCALL(rump, SOCKET); 1363 else 1364 op_socket = GETSYSCALL(host, SOCKET); 1365 fd = op_socket(domain, type, protocol); 1366 1367 if (isrump) 1368 fd = fd_rump2host(fd); 1369 else 1370 fd = fd_host2host(fd); 1371 DPRINTF(("socket <- %d\n", fd)); 1372 1373 return fd; 1374 } 1375 1376 int 1377 accept(int s, struct sockaddr *addr, socklen_t *addrlen) 1378 { 1379 int (*op_accept)(int, struct sockaddr *, socklen_t *); 1380 int fd; 1381 bool isrump; 1382 1383 isrump = fd_isrump(s); 1384 1385 DPRINTF(("accept -> %d", s)); 1386 if (isrump) { 1387 op_accept = GETSYSCALL(rump, ACCEPT); 1388 s = fd_host2rump(s); 1389 } else { 1390 op_accept = GETSYSCALL(host, ACCEPT); 1391 } 1392 fd = op_accept(s, addr, addrlen); 1393 if (fd != -1 && isrump) 1394 fd = fd_rump2host(fd); 1395 else 1396 fd = fd_host2host(fd); 1397 1398 DPRINTF((" <- %d\n", fd)); 1399 1400 return fd; 1401 } 1402 1403 #ifndef __linux__ 1404 int 1405 paccept(int s, struct sockaddr *addr, socklen_t *addrlen, 1406 const sigset_t * restrict sigmask, int flags) 1407 { 1408 int (*op_paccept)(int, struct sockaddr *, socklen_t *, 1409 const sigset_t * restrict, int); 1410 int fd; 1411 bool isrump; 1412 1413 isrump = fd_isrump(s); 1414 1415 DPRINTF(("paccept -> %d", s)); 1416 if (isrump) { 1417 op_paccept = GETSYSCALL(rump, PACCEPT); 1418 s = fd_host2rump(s); 1419 } else { 1420 op_paccept = GETSYSCALL(host, PACCEPT); 1421 } 1422 fd = op_paccept(s, addr, addrlen, sigmask, flags); 1423 if (fd != -1 && isrump) 1424 fd = fd_rump2host(fd); 1425 else 1426 fd = fd_host2host(fd); 1427 1428 DPRINTF((" <- %d\n", fd)); 1429 1430 return fd; 1431 } 1432 #endif 1433 1434 /* 1435 * ioctl() and fcntl() are varargs calls and need special treatment. 1436 */ 1437 1438 /* 1439 * Various [Linux] libc's have various signatures for ioctl so we 1440 * need to handle the discrepancies. On NetBSD, we use the 1441 * one with unsigned long cmd. 1442 */ 1443 int 1444 #ifdef HAVE_IOCTL_CMD_INT 1445 ioctl(int fd, int cmd, ...) 1446 { 1447 int (*op_ioctl)(int, int cmd, ...); 1448 #else 1449 ioctl(int fd, unsigned long cmd, ...) 1450 { 1451 int (*op_ioctl)(int, unsigned long cmd, ...); 1452 #endif 1453 va_list ap; 1454 int rv; 1455 1456 DPRINTF(("ioctl -> %d (%s)\n", fd, whichfd(fd))); 1457 if (fd_isrump(fd)) { 1458 fd = fd_host2rump(fd); 1459 op_ioctl = GETSYSCALL(rump, IOCTL); 1460 } else { 1461 op_ioctl = GETSYSCALL(host, IOCTL); 1462 } 1463 1464 va_start(ap, cmd); 1465 rv = op_ioctl(fd, cmd, va_arg(ap, void *)); 1466 va_end(ap); 1467 DPRINTF(("ioctl <- %d\n", rv)); 1468 return rv; 1469 } 1470 1471 int 1472 fcntl(int fd, int cmd, ...) 1473 { 1474 int (*op_fcntl)(int, int, ...); 1475 va_list ap; 1476 int rv, minfd; 1477 1478 DPRINTF(("fcntl -> %d (cmd %d)\n", fd, cmd)); 1479 1480 switch (cmd) { 1481 case F_DUPFD_CLOEXEC: /* Ignore CLOEXEC bit for now */ 1482 case F_DUPFD: 1483 va_start(ap, cmd); 1484 minfd = va_arg(ap, int); 1485 va_end(ap); 1486 return dodup(fd, minfd); 1487 1488 #ifdef F_CLOSEM 1489 case F_CLOSEM: { 1490 int maxdup2, i; 1491 1492 /* 1493 * So, if fd < HIJACKOFF, we want to do a host closem. 1494 */ 1495 1496 if (fd < hijack_fdoff) { 1497 int closemfd = fd; 1498 1499 if (rumpclient__closenotify(&closemfd, 1500 RUMPCLIENT_CLOSE_FCLOSEM) == -1) 1501 return -1; 1502 op_fcntl = GETSYSCALL(host, FCNTL); 1503 rv = op_fcntl(closemfd, cmd); 1504 if (rv) 1505 return rv; 1506 } 1507 1508 /* 1509 * Additionally, we want to do a rump closem, but only 1510 * for the file descriptors not dup2'd. 1511 */ 1512 1513 for (i = 0, maxdup2 = -1; i <= DUP2HIGH; i++) { 1514 if (dup2vec[i] & DUP2BIT) { 1515 int val; 1516 1517 val = dup2vec[i] & DUP2FDMASK; 1518 maxdup2 = MAX(val, maxdup2); 1519 } 1520 } 1521 1522 if (fd >= hijack_fdoff) 1523 fd -= hijack_fdoff; 1524 else 1525 fd = 0; 1526 fd = MAX(maxdup2+1, fd); 1527 1528 /* hmm, maybe we should close rump fd's not within dup2mask? */ 1529 return rump_sys_fcntl(fd, F_CLOSEM); 1530 } 1531 #endif /* F_CLOSEM */ 1532 1533 #ifdef F_MAXFD 1534 case F_MAXFD: 1535 /* 1536 * For maxfd, if there's a rump kernel fd, return 1537 * it hostified. Otherwise, return host's MAXFD 1538 * return value. 1539 */ 1540 if ((rv = rump_sys_fcntl(fd, F_MAXFD)) != -1) { 1541 /* 1542 * This might go a little wrong in case 1543 * of dup2 to [012], but I'm not sure if 1544 * there's a justification for tracking 1545 * that info. Consider e.g. 1546 * dup2(rumpfd, 2) followed by rump_sys_open() 1547 * returning 1. We should return 1+HIJACKOFF, 1548 * not 2+HIJACKOFF. However, if [01] is not 1549 * open, the correct return value is 2. 1550 */ 1551 return fd_rump2host(fd); 1552 } else { 1553 op_fcntl = GETSYSCALL(host, FCNTL); 1554 return op_fcntl(fd, F_MAXFD); 1555 } 1556 /*NOTREACHED*/ 1557 #endif /* F_MAXFD */ 1558 1559 default: 1560 if (fd_isrump(fd)) { 1561 fd = fd_host2rump(fd); 1562 op_fcntl = GETSYSCALL(rump, FCNTL); 1563 } else { 1564 op_fcntl = GETSYSCALL(host, FCNTL); 1565 } 1566 1567 va_start(ap, cmd); 1568 rv = op_fcntl(fd, cmd, va_arg(ap, void *)); 1569 va_end(ap); 1570 return rv; 1571 } 1572 /*NOTREACHED*/ 1573 } 1574 1575 int 1576 close(int fd) 1577 { 1578 int (*op_close)(int); 1579 int rv; 1580 1581 DPRINTF(("close -> %d\n", fd)); 1582 if (fd_isrump(fd)) { 1583 bool undup2 = false; 1584 int ofd; 1585 1586 if (isdup2d(ofd = fd)) { 1587 undup2 = true; 1588 } 1589 1590 fd = fd_host2rump(fd); 1591 if (!undup2 && killdup2alias(fd)) { 1592 return 0; 1593 } 1594 1595 op_close = GETSYSCALL(rump, CLOSE); 1596 rv = op_close(fd); 1597 if (rv == 0 && undup2) { 1598 clrdup2(ofd); 1599 } 1600 } else { 1601 if (rumpclient__closenotify(&fd, RUMPCLIENT_CLOSE_CLOSE) == -1) 1602 return -1; 1603 op_close = GETSYSCALL(host, CLOSE); 1604 rv = op_close(fd); 1605 } 1606 1607 return rv; 1608 } 1609 1610 /* 1611 * write cannot issue a standard debug printf due to recursion 1612 */ 1613 ssize_t 1614 write(int fd, const void *buf, size_t blen) 1615 { 1616 ssize_t (*op_write)(int, const void *, size_t); 1617 1618 if (fd_isrump(fd)) { 1619 fd = fd_host2rump(fd); 1620 op_write = GETSYSCALL(rump, WRITE); 1621 } else { 1622 op_write = GETSYSCALL(host, WRITE); 1623 } 1624 1625 return op_write(fd, buf, blen); 1626 } 1627 1628 /* 1629 * file descriptor passing 1630 * 1631 * we intercept sendmsg and recvmsg to convert file descriptors in 1632 * control messages. an attempt to send a descriptor from a different kernel 1633 * is rejected. (ENOTSUP) 1634 */ 1635 1636 static int 1637 _msg_convert_fds(struct msghdr *msg, int (*func)(int), bool dryrun) 1638 { 1639 struct cmsghdr *cmsg; 1640 1641 for (cmsg = CMSG_FIRSTHDR(msg); cmsg != NULL; 1642 cmsg = CMSG_NXTHDR(msg, cmsg)) { 1643 if (cmsg->cmsg_level == SOL_SOCKET && 1644 cmsg->cmsg_type == SCM_RIGHTS) { 1645 int *fdp = (void *)CMSG_DATA(cmsg); 1646 const size_t size = 1647 cmsg->cmsg_len - __CMSG_ALIGN(sizeof(*cmsg)); 1648 const int nfds = (int)(size / sizeof(int)); 1649 const int * const efdp = fdp + nfds; 1650 1651 while (fdp < efdp) { 1652 const int newval = func(*fdp); 1653 1654 if (newval < 0) { 1655 return ENOTSUP; 1656 } 1657 if (!dryrun) 1658 *fdp = newval; 1659 fdp++; 1660 } 1661 } 1662 } 1663 return 0; 1664 } 1665 1666 static int 1667 msg_convert_fds(struct msghdr *msg, int (*func)(int)) 1668 { 1669 1670 return _msg_convert_fds(msg, func, false); 1671 } 1672 1673 static int 1674 msg_check_fds(struct msghdr *msg, int (*func)(int)) 1675 { 1676 1677 return _msg_convert_fds(msg, func, true); 1678 } 1679 1680 ssize_t 1681 recvmsg(int fd, struct msghdr *msg, int flags) 1682 { 1683 ssize_t (*op_recvmsg)(int, struct msghdr *, int); 1684 ssize_t ret; 1685 const bool isrump = fd_isrump(fd); 1686 1687 if (isrump) { 1688 fd = fd_host2rump(fd); 1689 op_recvmsg = GETSYSCALL(rump, RECVMSG); 1690 } else { 1691 op_recvmsg = GETSYSCALL(host, RECVMSG); 1692 } 1693 ret = op_recvmsg(fd, msg, flags); 1694 if (ret == -1) { 1695 return ret; 1696 } 1697 /* 1698 * convert descriptors in the message. 1699 */ 1700 if (isrump) { 1701 msg_convert_fds(msg, fd_rump2host); 1702 } else { 1703 msg_convert_fds(msg, fd_host2host); 1704 } 1705 return ret; 1706 } 1707 1708 ssize_t 1709 recv(int fd, void *buf, size_t len, int flags) 1710 { 1711 1712 return recvfrom(fd, buf, len, flags, NULL, NULL); 1713 } 1714 1715 ssize_t 1716 send(int fd, const void *buf, size_t len, int flags) 1717 { 1718 1719 return sendto(fd, buf, len, flags, NULL, 0); 1720 } 1721 1722 static int 1723 fd_check_rump(int fd) 1724 { 1725 1726 return fd_isrump(fd) ? 0 : -1; 1727 } 1728 1729 static int 1730 fd_check_host(int fd) 1731 { 1732 1733 return !fd_isrump(fd) ? 0 : -1; 1734 } 1735 1736 ssize_t 1737 sendmsg(int fd, const struct msghdr *msg, int flags) 1738 { 1739 ssize_t (*op_sendmsg)(int, const struct msghdr *, int); 1740 const bool isrump = fd_isrump(fd); 1741 int error; 1742 1743 /* 1744 * reject descriptors from a different kernel. 1745 */ 1746 error = msg_check_fds(__UNCONST(msg), 1747 isrump ? fd_check_rump: fd_check_host); 1748 if (error != 0) { 1749 errno = error; 1750 return -1; 1751 } 1752 /* 1753 * convert descriptors in the message to raw values. 1754 */ 1755 if (isrump) { 1756 fd = fd_host2rump(fd); 1757 /* 1758 * XXX we directly modify the given message assuming: 1759 * - cmsg is writable (typically on caller's stack) 1760 * - caller don't care cmsg's contents after calling sendmsg. 1761 * (thus no need to restore values) 1762 * 1763 * it's safer to copy and modify instead. 1764 */ 1765 msg_convert_fds(__UNCONST(msg), fd_host2rump); 1766 op_sendmsg = GETSYSCALL(rump, SENDMSG); 1767 } else { 1768 op_sendmsg = GETSYSCALL(host, SENDMSG); 1769 } 1770 return op_sendmsg(fd, msg, flags); 1771 } 1772 1773 /* 1774 * dup2 is special. we allow dup2 of a rump kernel fd to 0-2 since 1775 * many programs do that. dup2 of a rump kernel fd to another value 1776 * not >= fdoff is an error. 1777 * 1778 * Note: cannot rump2host newd, because it is often hardcoded. 1779 */ 1780 int 1781 dup2(int oldd, int newd) 1782 { 1783 int (*host_dup2)(int, int); 1784 int rv; 1785 1786 DPRINTF(("dup2 -> %d (o) -> %d (n)\n", oldd, newd)); 1787 1788 if (fd_isrump(oldd)) { 1789 int (*op_close)(int) = GETSYSCALL(host, CLOSE); 1790 1791 /* only allow fd 0-2 for cross-kernel dup */ 1792 if (!(newd >= 0 && newd <= 2 && !fd_isrump(newd))) { 1793 errno = EBADF; 1794 return -1; 1795 } 1796 1797 /* regular dup2? */ 1798 if (fd_isrump(newd)) { 1799 newd = fd_host2rump(newd); 1800 rv = rump_sys_dup2(oldd, newd); 1801 return fd_rump2host(rv); 1802 } 1803 1804 /* 1805 * dup2 rump => host? just establish an 1806 * entry in the mapping table. 1807 */ 1808 op_close(newd); 1809 setdup2(newd, fd_host2rump(oldd)); 1810 rv = 0; 1811 } else { 1812 host_dup2 = syscalls[DUALCALL_DUP2].bs_host; 1813 if (rumpclient__closenotify(&newd, RUMPCLIENT_CLOSE_DUP2) == -1) 1814 return -1; 1815 rv = host_dup2(oldd, newd); 1816 } 1817 1818 return rv; 1819 } 1820 1821 int 1822 dup(int oldd) 1823 { 1824 1825 return dodup(oldd, 0); 1826 } 1827 1828 pid_t 1829 fork(void) 1830 { 1831 pid_t rv; 1832 1833 DPRINTF(("fork\n")); 1834 1835 rv = rumpclient__dofork(host_fork); 1836 1837 DPRINTF(("fork returns %d\n", rv)); 1838 return rv; 1839 } 1840 #ifdef VFORK 1841 /* we do not have the luxury of not requiring a stackframe */ 1842 #define __strong_alias_macro(m, f) __strong_alias(m, f) 1843 __strong_alias_macro(VFORK,fork); 1844 #endif 1845 1846 int 1847 daemon(int nochdir, int noclose) 1848 { 1849 struct rumpclient_fork *rf; 1850 1851 if ((rf = rumpclient_prefork()) == NULL) 1852 return -1; 1853 1854 if (host_daemon(nochdir, noclose) == -1) 1855 return -1; 1856 1857 if (rumpclient_fork_init(rf) == -1) 1858 return -1; 1859 1860 return 0; 1861 } 1862 1863 int 1864 execve(const char *path, char *const argv[], char *const envp[]) 1865 { 1866 char buf[128]; 1867 char *dup2str; 1868 const char *pwdinrumpstr; 1869 char **newenv; 1870 size_t nelem; 1871 int rv, sverrno; 1872 int bonus = 2, i = 0; 1873 1874 snprintf(buf, sizeof(buf), "RUMPHIJACK__DUP2INFO=%u,%u,%u", 1875 dup2vec[0], dup2vec[1], dup2vec[2]); 1876 dup2str = strdup(buf); 1877 if (dup2str == NULL) { 1878 errno = ENOMEM; 1879 return -1; 1880 } 1881 1882 if (pwdinrump) { 1883 pwdinrumpstr = "RUMPHIJACK__PWDINRUMP=true"; 1884 bonus++; 1885 } else { 1886 pwdinrumpstr = NULL; 1887 } 1888 1889 for (nelem = 0; envp && envp[nelem]; nelem++) 1890 continue; 1891 newenv = malloc(sizeof(*newenv) * (nelem+bonus)); 1892 if (newenv == NULL) { 1893 free(dup2str); 1894 errno = ENOMEM; 1895 return -1; 1896 } 1897 memcpy(newenv, envp, nelem*sizeof(*newenv)); 1898 newenv[nelem+i] = dup2str; 1899 i++; 1900 1901 if (pwdinrumpstr) { 1902 newenv[nelem+i] = __UNCONST(pwdinrumpstr); 1903 i++; 1904 } 1905 newenv[nelem+i] = NULL; 1906 _DIAGASSERT(i < bonus); 1907 1908 rv = rumpclient_exec(path, argv, newenv); 1909 1910 _DIAGASSERT(rv != 0); 1911 sverrno = errno; 1912 free(newenv); 1913 free(dup2str); 1914 errno = sverrno; 1915 return rv; 1916 } 1917 1918 /* 1919 * select is done by calling poll. 1920 */ 1921 int 1922 REALSELECT(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, 1923 struct timeval *timeout) 1924 { 1925 struct pollfd *pfds; 1926 struct timespec ts, *tsp = NULL; 1927 nfds_t realnfds; 1928 int i, j; 1929 int rv, incr; 1930 1931 DPRINTF(("select %d %p %p %p %p\n", nfds, 1932 readfds, writefds, exceptfds, timeout)); 1933 1934 /* 1935 * Well, first we must scan the fds to figure out how many 1936 * fds there really are. This is because up to and including 1937 * nb5 poll() silently refuses nfds > process_maxopen_fds. 1938 * Seems to be fixed in current, thank the maker. 1939 * god damn cluster...bomb. 1940 */ 1941 1942 for (i = 0, realnfds = 0; i < nfds; i++) { 1943 if (readfds && FD_ISSET(i, readfds)) { 1944 realnfds++; 1945 continue; 1946 } 1947 if (writefds && FD_ISSET(i, writefds)) { 1948 realnfds++; 1949 continue; 1950 } 1951 if (exceptfds && FD_ISSET(i, exceptfds)) { 1952 realnfds++; 1953 continue; 1954 } 1955 } 1956 1957 if (realnfds) { 1958 pfds = calloc(realnfds, sizeof(*pfds)); 1959 if (!pfds) 1960 return -1; 1961 } else { 1962 pfds = NULL; 1963 } 1964 1965 for (i = 0, j = 0; i < nfds; i++) { 1966 incr = 0; 1967 if (readfds && FD_ISSET(i, readfds)) { 1968 pfds[j].fd = i; 1969 pfds[j].events |= POLLIN; 1970 incr=1; 1971 } 1972 if (writefds && FD_ISSET(i, writefds)) { 1973 pfds[j].fd = i; 1974 pfds[j].events |= POLLOUT; 1975 incr=1; 1976 } 1977 if (exceptfds && FD_ISSET(i, exceptfds)) { 1978 pfds[j].fd = i; 1979 pfds[j].events |= POLLHUP|POLLERR; 1980 incr=1; 1981 } 1982 if (incr) 1983 j++; 1984 } 1985 assert(j == (int)realnfds); 1986 1987 if (timeout) { 1988 TIMEVAL_TO_TIMESPEC(timeout, &ts); 1989 tsp = &ts; 1990 } 1991 rv = REALPOLLTS(pfds, realnfds, tsp, NULL); 1992 /* 1993 * "If select() returns with an error the descriptor sets 1994 * will be unmodified" 1995 */ 1996 if (rv < 0) 1997 goto out; 1998 1999 /* 2000 * zero out results (can't use FD_ZERO for the 2001 * obvious select-me-not reason). whee. 2002 * 2003 * We do this here since some software ignores the return 2004 * value of select, and hence if the timeout expires, it may 2005 * assume all input descriptors have activity. 2006 */ 2007 for (i = 0; i < nfds; i++) { 2008 if (readfds) 2009 FD_CLR(i, readfds); 2010 if (writefds) 2011 FD_CLR(i, writefds); 2012 if (exceptfds) 2013 FD_CLR(i, exceptfds); 2014 } 2015 if (rv == 0) 2016 goto out; 2017 2018 /* 2019 * We have >0 fds with activity. Harvest the results. 2020 */ 2021 for (i = 0; i < (int)realnfds; i++) { 2022 if (readfds) { 2023 if (pfds[i].revents & POLLIN) { 2024 FD_SET(pfds[i].fd, readfds); 2025 } 2026 } 2027 if (writefds) { 2028 if (pfds[i].revents & POLLOUT) { 2029 FD_SET(pfds[i].fd, writefds); 2030 } 2031 } 2032 if (exceptfds) { 2033 if (pfds[i].revents & (POLLHUP|POLLERR)) { 2034 FD_SET(pfds[i].fd, exceptfds); 2035 } 2036 } 2037 } 2038 2039 out: 2040 free(pfds); 2041 return rv; 2042 } 2043 2044 static void 2045 checkpoll(struct pollfd *fds, nfds_t nfds, int *hostcall, int *rumpcall) 2046 { 2047 nfds_t i; 2048 2049 for (i = 0; i < nfds; i++) { 2050 if (fds[i].fd == -1) 2051 continue; 2052 2053 if (fd_isrump(fds[i].fd)) 2054 (*rumpcall)++; 2055 else 2056 (*hostcall)++; 2057 } 2058 } 2059 2060 static void 2061 adjustpoll(struct pollfd *fds, nfds_t nfds, int (*fdadj)(int)) 2062 { 2063 nfds_t i; 2064 2065 for (i = 0; i < nfds; i++) { 2066 fds[i].fd = fdadj(fds[i].fd); 2067 } 2068 } 2069 2070 /* 2071 * poll is easy as long as the call comes in the fds only in one 2072 * kernel. otherwise its quite tricky... 2073 */ 2074 struct pollarg { 2075 struct pollfd *pfds; 2076 nfds_t nfds; 2077 const struct timespec *ts; 2078 const sigset_t *sigmask; 2079 int pipefd; 2080 int errnum; 2081 }; 2082 2083 static void * 2084 hostpoll(void *arg) 2085 { 2086 int (*op_pollts)(struct pollfd *, nfds_t, const struct timespec *, 2087 const sigset_t *); 2088 struct pollarg *parg = arg; 2089 intptr_t rv; 2090 2091 op_pollts = GETSYSCALL(host, POLLTS); 2092 rv = op_pollts(parg->pfds, parg->nfds, parg->ts, parg->sigmask); 2093 if (rv == -1) 2094 parg->errnum = errno; 2095 rump_sys_write(parg->pipefd, &rv, sizeof(rv)); 2096 2097 return (void *)rv; 2098 } 2099 2100 int 2101 REALPOLLTS(struct pollfd *fds, nfds_t nfds, const struct timespec *ts, 2102 const sigset_t *sigmask) 2103 { 2104 int (*op_pollts)(struct pollfd *, nfds_t, const struct timespec *, 2105 const sigset_t *); 2106 int (*host_close)(int); 2107 int hostcall = 0, rumpcall = 0; 2108 pthread_t pt; 2109 nfds_t i; 2110 int rv; 2111 2112 DPRINTF(("poll %p %d %p %p\n", fds, (int)nfds, ts, sigmask)); 2113 checkpoll(fds, nfds, &hostcall, &rumpcall); 2114 2115 if (hostcall && rumpcall) { 2116 struct pollfd *pfd_host = NULL, *pfd_rump = NULL; 2117 int rpipe[2] = {-1,-1}, hpipe[2] = {-1,-1}; 2118 struct pollarg parg; 2119 void *trv_val; 2120 int sverrno = 0, rv_rump, rv_host, errno_rump, errno_host; 2121 2122 /* 2123 * ok, this is where it gets tricky. We must support 2124 * this since it's a very common operation in certain 2125 * types of software (telnet, netcat, etc). We allocate 2126 * two vectors and run two poll commands in separate 2127 * threads. Whichever returns first "wins" and the 2128 * other kernel's fds won't show activity. 2129 */ 2130 rv = -1; 2131 2132 /* allocate full vector for O(n) joining after call */ 2133 pfd_host = malloc(sizeof(*pfd_host)*(nfds+1)); 2134 if (!pfd_host) 2135 goto out; 2136 pfd_rump = malloc(sizeof(*pfd_rump)*(nfds+1)); 2137 if (!pfd_rump) { 2138 goto out; 2139 } 2140 2141 /* 2142 * then, open two pipes, one for notifications 2143 * to each kernel. 2144 * 2145 * At least the rump pipe should probably be 2146 * cached, along with the helper threads. This 2147 * should give a microbenchmark improvement (haven't 2148 * experienced a macro-level problem yet, though). 2149 */ 2150 if ((rv = rump_sys_pipe(rpipe)) == -1) { 2151 sverrno = errno; 2152 } 2153 if (rv == 0 && (rv = pipe(hpipe)) == -1) { 2154 sverrno = errno; 2155 } 2156 2157 /* split vectors (or signal errors) */ 2158 for (i = 0; i < nfds; i++) { 2159 int fd; 2160 2161 fds[i].revents = 0; 2162 if (fds[i].fd == -1) { 2163 pfd_host[i].fd = -1; 2164 pfd_rump[i].fd = -1; 2165 } else if (fd_isrump(fds[i].fd)) { 2166 pfd_host[i].fd = -1; 2167 fd = fd_host2rump(fds[i].fd); 2168 if (fd == rpipe[0] || fd == rpipe[1]) { 2169 fds[i].revents = POLLNVAL; 2170 if (rv != -1) 2171 rv++; 2172 } 2173 pfd_rump[i].fd = fd; 2174 pfd_rump[i].events = fds[i].events; 2175 } else { 2176 pfd_rump[i].fd = -1; 2177 fd = fds[i].fd; 2178 if (fd == hpipe[0] || fd == hpipe[1]) { 2179 fds[i].revents = POLLNVAL; 2180 if (rv != -1) 2181 rv++; 2182 } 2183 pfd_host[i].fd = fd; 2184 pfd_host[i].events = fds[i].events; 2185 } 2186 pfd_rump[i].revents = pfd_host[i].revents = 0; 2187 } 2188 if (rv) { 2189 goto out; 2190 } 2191 2192 pfd_host[nfds].fd = hpipe[0]; 2193 pfd_host[nfds].events = POLLIN; 2194 pfd_rump[nfds].fd = rpipe[0]; 2195 pfd_rump[nfds].events = POLLIN; 2196 2197 /* 2198 * then, create a thread to do host part and meanwhile 2199 * do rump kernel part right here 2200 */ 2201 2202 parg.pfds = pfd_host; 2203 parg.nfds = nfds+1; 2204 parg.ts = ts; 2205 parg.sigmask = sigmask; 2206 parg.pipefd = rpipe[1]; 2207 pthread_create(&pt, NULL, hostpoll, &parg); 2208 2209 op_pollts = GETSYSCALL(rump, POLLTS); 2210 rv_rump = op_pollts(pfd_rump, nfds+1, ts, NULL); 2211 errno_rump = errno; 2212 write(hpipe[1], &rv, sizeof(rv)); 2213 pthread_join(pt, &trv_val); 2214 rv_host = (int)(intptr_t)trv_val; 2215 errno_host = parg.errnum; 2216 2217 /* strip cross-thread notification from real results */ 2218 if (rv_host > 0 && pfd_host[nfds].revents & POLLIN) { 2219 rv_host--; 2220 } 2221 if (rv_rump > 0 && pfd_rump[nfds].revents & POLLIN) { 2222 rv_rump--; 2223 } 2224 2225 /* then merge the results into what's reported to the caller */ 2226 if (rv_rump > 0 || rv_host > 0) { 2227 /* SUCCESS */ 2228 2229 rv = 0; 2230 if (rv_rump > 0) { 2231 for (i = 0; i < nfds; i++) { 2232 if (pfd_rump[i].fd != -1) 2233 fds[i].revents 2234 = pfd_rump[i].revents; 2235 } 2236 rv += rv_rump; 2237 } 2238 if (rv_host > 0) { 2239 for (i = 0; i < nfds; i++) { 2240 if (pfd_host[i].fd != -1) 2241 fds[i].revents 2242 = pfd_host[i].revents; 2243 } 2244 rv += rv_host; 2245 } 2246 assert(rv > 0); 2247 sverrno = 0; 2248 } else if (rv_rump == -1 || rv_host == -1) { 2249 /* ERROR */ 2250 2251 /* just pick one kernel at "random" */ 2252 rv = -1; 2253 if (rv_host == -1) { 2254 sverrno = errno_host; 2255 } else if (rv_rump == -1) { 2256 sverrno = errno_rump; 2257 } 2258 } else { 2259 /* TIMEOUT */ 2260 2261 rv = 0; 2262 assert(rv_rump == 0 && rv_host == 0); 2263 } 2264 2265 out: 2266 host_close = GETSYSCALL(host, CLOSE); 2267 if (rpipe[0] != -1) 2268 rump_sys_close(rpipe[0]); 2269 if (rpipe[1] != -1) 2270 rump_sys_close(rpipe[1]); 2271 if (hpipe[0] != -1) 2272 host_close(hpipe[0]); 2273 if (hpipe[1] != -1) 2274 host_close(hpipe[1]); 2275 free(pfd_host); 2276 free(pfd_rump); 2277 errno = sverrno; 2278 } else { 2279 if (hostcall) { 2280 op_pollts = GETSYSCALL(host, POLLTS); 2281 } else { 2282 op_pollts = GETSYSCALL(rump, POLLTS); 2283 adjustpoll(fds, nfds, fd_host2rump); 2284 } 2285 2286 rv = op_pollts(fds, nfds, ts, sigmask); 2287 if (rumpcall) 2288 adjustpoll(fds, nfds, fd_rump2host_withdup); 2289 } 2290 2291 return rv; 2292 } 2293 2294 int 2295 poll(struct pollfd *fds, nfds_t nfds, int timeout) 2296 { 2297 struct timespec ts; 2298 struct timespec *tsp = NULL; 2299 2300 if (timeout != INFTIM) { 2301 ts.tv_sec = timeout / 1000; 2302 ts.tv_nsec = (timeout % 1000) * 1000*1000; 2303 2304 tsp = &ts; 2305 } 2306 2307 return REALPOLLTS(fds, nfds, tsp, NULL); 2308 } 2309 2310 #ifdef HAVE_KQUEUE 2311 int 2312 REALKEVENT(int kq, const struct kevent *changelist, size_t nchanges, 2313 struct kevent *eventlist, size_t nevents, 2314 const struct timespec *timeout) 2315 { 2316 int (*op_kevent)(int, const struct kevent *, size_t, 2317 struct kevent *, size_t, const struct timespec *); 2318 const struct kevent *ev; 2319 size_t i; 2320 2321 /* 2322 * Check that we don't attempt to kevent rump kernel fd's. 2323 * That needs similar treatment to select/poll, but is slightly 2324 * trickier since we need to manage to different kq descriptors. 2325 * (TODO, in case you're wondering). 2326 */ 2327 for (i = 0; i < nchanges; i++) { 2328 ev = &changelist[i]; 2329 if (ev->filter == EVFILT_READ || ev->filter == EVFILT_WRITE || 2330 ev->filter == EVFILT_VNODE) { 2331 if (fd_isrump((int)ev->ident)) { 2332 errno = ENOTSUP; 2333 return -1; 2334 } 2335 } 2336 } 2337 2338 op_kevent = GETSYSCALL(host, KEVENT); 2339 return op_kevent(kq, changelist, nchanges, eventlist, nevents, timeout); 2340 } 2341 #endif /* HAVE_KQUEUE */ 2342 2343 /* 2344 * mmapping from a rump kernel is not supported, so disallow it. 2345 */ 2346 void * 2347 mmap(void *addr, size_t len, int prot, int flags, int fd, off_t offset) 2348 { 2349 2350 if (flags & MAP_FILE && fd_isrump(fd)) { 2351 errno = ENOSYS; 2352 return MAP_FAILED; 2353 } 2354 if (__predict_false(host_mmap == NULL)) { 2355 host_mmap = rumphijack_dlsym(RTLD_NEXT, "mmap"); 2356 } 2357 return host_mmap(addr, len, prot, flags, fd, offset); 2358 } 2359 2360 #ifdef __NetBSD__ 2361 /* 2362 * these go to one or the other on a per-process configuration 2363 */ 2364 int __sysctl(const int *, unsigned int, void *, size_t *, const void *, size_t); 2365 int 2366 __sysctl(const int *name, unsigned int namelen, void *old, size_t *oldlenp, 2367 const void *new, size_t newlen) 2368 { 2369 int (*op___sysctl)(const int *, unsigned int, void *, size_t *, 2370 const void *, size_t); 2371 2372 if (rumpsysctl) { 2373 op___sysctl = GETSYSCALL(rump, __SYSCTL); 2374 } else { 2375 op___sysctl = GETSYSCALL(host, __SYSCTL); 2376 /* we haven't inited yet */ 2377 if (__predict_false(op___sysctl == NULL)) { 2378 op___sysctl = rumphijack_dlsym(RTLD_NEXT, "__sysctl"); 2379 } 2380 } 2381 2382 return op___sysctl(name, namelen, old, oldlenp, new, newlen); 2383 } 2384 int modctl(int, void *); 2385 int 2386 modctl(int operation, void *argp) 2387 { 2388 int (*op_modctl)(int operation, void *argp); 2389 2390 if (rumpmodctl) { 2391 op_modctl = GETSYSCALL(rump, MODCTL); 2392 } else { 2393 op_modctl = GETSYSCALL(host, MODCTL); 2394 } 2395 2396 return op_modctl(operation, argp); 2397 } 2398 #endif 2399 2400 /* 2401 * Rest are std type calls. 2402 */ 2403 2404 #ifdef HAVE_UTIMENSAT 2405 ATCALL(int, utimensat, DUALCALL_UTIMENSAT, \ 2406 (int fd, const char *path, const struct timespec t[2], int f), \ 2407 (int, const char *, const struct timespec [2], int), 2408 (fd, path, t, f)) 2409 #endif 2410 2411 FDCALL(int, bind, DUALCALL_BIND, \ 2412 (int fd, const struct sockaddr *name, socklen_t namelen), \ 2413 (int, const struct sockaddr *, socklen_t), \ 2414 (fd, name, namelen)) 2415 2416 FDCALL(int, connect, DUALCALL_CONNECT, \ 2417 (int fd, const struct sockaddr *name, socklen_t namelen), \ 2418 (int, const struct sockaddr *, socklen_t), \ 2419 (fd, name, namelen)) 2420 2421 FDCALL(int, getpeername, DUALCALL_GETPEERNAME, \ 2422 (int fd, struct sockaddr *name, socklen_t *namelen), \ 2423 (int, struct sockaddr *, socklen_t *), \ 2424 (fd, name, namelen)) 2425 2426 FDCALL(int, getsockname, DUALCALL_GETSOCKNAME, \ 2427 (int fd, struct sockaddr *name, socklen_t *namelen), \ 2428 (int, struct sockaddr *, socklen_t *), \ 2429 (fd, name, namelen)) 2430 2431 FDCALL(int, listen, DUALCALL_LISTEN, \ 2432 (int fd, int backlog), \ 2433 (int, int), \ 2434 (fd, backlog)) 2435 2436 FDCALL(ssize_t, recvfrom, DUALCALL_RECVFROM, \ 2437 (int fd, void *buf, size_t len, int flags, \ 2438 struct sockaddr *from, socklen_t *fromlen), \ 2439 (int, void *, size_t, int, struct sockaddr *, socklen_t *), \ 2440 (fd, buf, len, flags, from, fromlen)) 2441 2442 FDCALL(ssize_t, sendto, DUALCALL_SENDTO, \ 2443 (int fd, const void *buf, size_t len, int flags, \ 2444 const struct sockaddr *to, socklen_t tolen), \ 2445 (int, const void *, size_t, int, \ 2446 const struct sockaddr *, socklen_t), \ 2447 (fd, buf, len, flags, to, tolen)) 2448 2449 FDCALL(int, getsockopt, DUALCALL_GETSOCKOPT, \ 2450 (int fd, int level, int optn, void *optval, socklen_t *optlen), \ 2451 (int, int, int, void *, socklen_t *), \ 2452 (fd, level, optn, optval, optlen)) 2453 2454 FDCALL(int, setsockopt, DUALCALL_SETSOCKOPT, \ 2455 (int fd, int level, int optn, \ 2456 const void *optval, socklen_t optlen), \ 2457 (int, int, int, const void *, socklen_t), \ 2458 (fd, level, optn, optval, optlen)) 2459 2460 FDCALL(int, shutdown, DUALCALL_SHUTDOWN, \ 2461 (int fd, int how), \ 2462 (int, int), \ 2463 (fd, how)) 2464 2465 FDCALL(ssize_t, REALREAD, DUALCALL_READ, \ 2466 (int fd, void *buf, size_t buflen), \ 2467 (int, void *, size_t), \ 2468 (fd, buf, buflen)) 2469 2470 #ifdef __linux__ 2471 ssize_t __read_chk(int, void *, size_t) 2472 __attribute__((alias("read"))); 2473 #endif 2474 2475 FDCALL(ssize_t, readv, DUALCALL_READV, \ 2476 (int fd, const struct iovec *iov, int iovcnt), \ 2477 (int, const struct iovec *, int), \ 2478 (fd, iov, iovcnt)) 2479 2480 FDCALL(ssize_t, REALPREAD, DUALCALL_PREAD, \ 2481 (int fd, void *buf, size_t nbytes, off_t offset), \ 2482 (int, void *, size_t, off_t), \ 2483 (fd, buf, nbytes, offset)) 2484 2485 FDCALL(ssize_t, preadv, DUALCALL_PREADV, \ 2486 (int fd, const struct iovec *iov, int iovcnt, off_t offset), \ 2487 (int, const struct iovec *, int, off_t), \ 2488 (fd, iov, iovcnt, offset)) 2489 2490 FDCALL(ssize_t, writev, DUALCALL_WRITEV, \ 2491 (int fd, const struct iovec *iov, int iovcnt), \ 2492 (int, const struct iovec *, int), \ 2493 (fd, iov, iovcnt)) 2494 2495 FDCALL(ssize_t, REALPWRITE, DUALCALL_PWRITE, \ 2496 (int fd, const void *buf, size_t nbytes, off_t offset), \ 2497 (int, const void *, size_t, off_t), \ 2498 (fd, buf, nbytes, offset)) 2499 2500 FDCALL(ssize_t, pwritev, DUALCALL_PWRITEV, \ 2501 (int fd, const struct iovec *iov, int iovcnt, off_t offset), \ 2502 (int, const struct iovec *, int, off_t), \ 2503 (fd, iov, iovcnt, offset)) 2504 2505 #ifndef __linux__ 2506 FDCALL(int, REALFSTAT, DUALCALL_FSTAT, \ 2507 (int fd, struct stat *sb), \ 2508 (int, struct stat *), \ 2509 (fd, sb)) 2510 #endif 2511 2512 #ifdef __NetBSD__ 2513 FDCALL(int, REALFSTATVFS1, DUALCALL_FSTATVFS1, \ 2514 (int fd, struct statvfs *buf, int flags), \ 2515 (int, struct statvfs *, int), \ 2516 (fd, buf, flags)) 2517 #endif 2518 2519 FDCALL(off_t, lseek, DUALCALL_LSEEK, \ 2520 (int fd, off_t offset, int whence), \ 2521 (int, off_t, int), \ 2522 (fd, offset, whence)) 2523 #ifdef LSEEK_ALIAS 2524 __strong_alias(LSEEK_ALIAS,lseek); 2525 #endif 2526 2527 #ifndef __linux__ 2528 FDCALL(int, REALGETDENTS, DUALCALL_GETDENTS, \ 2529 (int fd, char *buf, size_t nbytes), \ 2530 (int, char *, size_t), \ 2531 (fd, buf, nbytes)) 2532 #endif 2533 2534 FDCALL(int, fchown, DUALCALL_FCHOWN, \ 2535 (int fd, uid_t owner, gid_t group), \ 2536 (int, uid_t, gid_t), \ 2537 (fd, owner, group)) 2538 2539 FDCALL(int, fchmod, DUALCALL_FCHMOD, \ 2540 (int fd, mode_t mode), \ 2541 (int, mode_t), \ 2542 (fd, mode)) 2543 2544 FDCALL(int, ftruncate, DUALCALL_FTRUNCATE, \ 2545 (int fd, off_t length), \ 2546 (int, off_t), \ 2547 (fd, length)) 2548 2549 FDCALL(int, fsync, DUALCALL_FSYNC, \ 2550 (int fd), \ 2551 (int), \ 2552 (fd)) 2553 2554 #ifdef HAVE_FSYNC_RANGE 2555 FDCALL(int, fsync_range, DUALCALL_FSYNC_RANGE, \ 2556 (int fd, int how, off_t start, off_t length), \ 2557 (int, int, off_t, off_t), \ 2558 (fd, how, start, length)) 2559 #endif 2560 2561 FDCALL(int, futimes, DUALCALL_FUTIMES, \ 2562 (int fd, const struct timeval *tv), \ 2563 (int, const struct timeval *), \ 2564 (fd, tv)) 2565 2566 FDCALL(int, futimens, DUALCALL_FUTIMENS, \ 2567 (int fd, const struct timespec *ts), \ 2568 (int, const struct timespec *), \ 2569 (fd, ts)) 2570 2571 #ifdef HAVE_CHFLAGS 2572 FDCALL(int, fchflags, DUALCALL_FCHFLAGS, \ 2573 (int fd, u_long flags), \ 2574 (int, u_long), \ 2575 (fd, flags)) 2576 #endif 2577 2578 /* 2579 * path-based selectors 2580 */ 2581 2582 #ifndef __linux__ 2583 PATHCALL(int, REALSTAT, DUALCALL_STAT, \ 2584 (const char *path, struct stat *sb), \ 2585 (const char *, struct stat *), \ 2586 (path, sb)) 2587 2588 PATHCALL(int, REALLSTAT, DUALCALL_LSTAT, \ 2589 (const char *path, struct stat *sb), \ 2590 (const char *, struct stat *), \ 2591 (path, sb)) 2592 #endif 2593 2594 PATHCALL(int, chown, DUALCALL_CHOWN, \ 2595 (const char *path, uid_t owner, gid_t group), \ 2596 (const char *, uid_t, gid_t), \ 2597 (path, owner, group)) 2598 2599 PATHCALL(int, lchown, DUALCALL_LCHOWN, \ 2600 (const char *path, uid_t owner, gid_t group), \ 2601 (const char *, uid_t, gid_t), \ 2602 (path, owner, group)) 2603 2604 PATHCALL(int, chmod, DUALCALL_CHMOD, \ 2605 (const char *path, mode_t mode), \ 2606 (const char *, mode_t), \ 2607 (path, mode)) 2608 2609 PATHCALL(int, lchmod, DUALCALL_LCHMOD, \ 2610 (const char *path, mode_t mode), \ 2611 (const char *, mode_t), \ 2612 (path, mode)) 2613 2614 #ifdef __NetBSD__ 2615 PATHCALL(int, REALSTATVFS1, DUALCALL_STATVFS1, \ 2616 (const char *path, struct statvfs *buf, int flags), \ 2617 (const char *, struct statvfs *, int), \ 2618 (path, buf, flags)) 2619 #endif 2620 2621 PATHCALL(int, unlink, DUALCALL_UNLINK, \ 2622 (const char *path), \ 2623 (const char *), \ 2624 (path)) 2625 2626 PATHCALL(int, symlink, DUALCALL_SYMLINK, \ 2627 (const char *target, const char *path), \ 2628 (const char *, const char *), \ 2629 (target, path)) 2630 2631 /* 2632 * readlink() can be called from malloc which can be called 2633 * from dlsym() during init 2634 */ 2635 ssize_t 2636 readlink(const char *path, char *buf, size_t bufsiz) 2637 { 2638 int (*op_readlink)(const char *, char *, size_t); 2639 enum pathtype pt; 2640 2641 if ((pt = path_isrump(path)) != PATH_HOST) { 2642 op_readlink = GETSYSCALL(rump, READLINK); 2643 if (pt == PATH_RUMP) 2644 path = path_host2rump(path); 2645 } else { 2646 op_readlink = GETSYSCALL(host, READLINK); 2647 } 2648 2649 if (__predict_false(op_readlink == NULL)) { 2650 errno = ENOENT; 2651 return -1; 2652 } 2653 2654 return op_readlink(path, buf, bufsiz); 2655 } 2656 2657 PATHCALL(int, mkdir, DUALCALL_MKDIR, \ 2658 (const char *path, mode_t mode), \ 2659 (const char *, mode_t), \ 2660 (path, mode)) 2661 2662 PATHCALL(int, rmdir, DUALCALL_RMDIR, \ 2663 (const char *path), \ 2664 (const char *), \ 2665 (path)) 2666 2667 PATHCALL(int, utimes, DUALCALL_UTIMES, \ 2668 (const char *path, const struct timeval *tv), \ 2669 (const char *, const struct timeval *), \ 2670 (path, tv)) 2671 2672 PATHCALL(int, lutimes, DUALCALL_LUTIMES, \ 2673 (const char *path, const struct timeval *tv), \ 2674 (const char *, const struct timeval *), \ 2675 (path, tv)) 2676 2677 #ifdef HAVE_CHFLAGS 2678 PATHCALL(int, chflags, DUALCALL_CHFLAGS, \ 2679 (const char *path, u_long flags), \ 2680 (const char *, u_long), \ 2681 (path, flags)) 2682 2683 PATHCALL(int, lchflags, DUALCALL_LCHFLAGS, \ 2684 (const char *path, u_long flags), \ 2685 (const char *, u_long), \ 2686 (path, flags)) 2687 #endif /* HAVE_CHFLAGS */ 2688 2689 PATHCALL(int, truncate, DUALCALL_TRUNCATE, \ 2690 (const char *path, off_t length), \ 2691 (const char *, off_t), \ 2692 (path, length)) 2693 2694 PATHCALL(int, access, DUALCALL_ACCESS, \ 2695 (const char *path, int mode), \ 2696 (const char *, int), \ 2697 (path, mode)) 2698 2699 #ifndef __linux__ 2700 PATHCALL(int, REALMKNOD, DUALCALL_MKNOD, \ 2701 (const char *path, mode_t mode, dev_t dev), \ 2702 (const char *, mode_t, dev_t), \ 2703 (path, mode, dev)) 2704 #endif 2705 2706 /* 2707 * Note: with mount the decisive parameter is the mount 2708 * destination directory. This is because we don't really know 2709 * about the "source" directory in a generic call (and besides, 2710 * it might not even exist, cf. nfs). 2711 */ 2712 #ifdef __NetBSD__ 2713 PATHCALL(int, REALMOUNT, DUALCALL_MOUNT, \ 2714 (const char *type, const char *path, int flags, \ 2715 void *data, size_t dlen), \ 2716 (const char *, const char *, int, void *, size_t), \ 2717 (type, path, flags, data, dlen)) 2718 2719 PATHCALL(int, unmount, DUALCALL_UNMOUNT, \ 2720 (const char *path, int flags), \ 2721 (const char *, int), \ 2722 (path, flags)) 2723 #endif /* __NetBSD__ */ 2724 2725 #ifdef HAVE___QUOTACTL 2726 PATHCALL(int, __quotactl, DUALCALL_QUOTACTL, \ 2727 (const char *path, struct quotactl_args *args), \ 2728 (const char *, struct quotactl_args *), \ 2729 (path, args)) 2730 #endif /* HAVE___QUOTACTL */ 2731 2732 #ifdef __NetBSD__ 2733 PATHCALL(int, REALGETFH, DUALCALL_GETFH, \ 2734 (const char *path, void *fhp, size_t *fh_size), \ 2735 (const char *, void *, size_t *), \ 2736 (path, fhp, fh_size)) 2737 #endif 2738 2739 /* 2740 * These act different on a per-process vfs configuration 2741 */ 2742 2743 #ifdef __NetBSD__ 2744 VFSCALL(VFSBIT_GETVFSSTAT, int, REALGETVFSSTAT, DUALCALL_GETVFSSTAT, \ 2745 (struct statvfs *buf, size_t buflen, int flags), \ 2746 (struct statvfs *, size_t, int), \ 2747 (buf, buflen, flags)) 2748 #endif 2749 2750 #ifdef __NetBSD__ 2751 VFSCALL(VFSBIT_FHCALLS, int, REALFHOPEN, DUALCALL_FHOPEN, \ 2752 (const void *fhp, size_t fh_size, int flags), \ 2753 (const char *, size_t, int), \ 2754 (fhp, fh_size, flags)) 2755 2756 VFSCALL(VFSBIT_FHCALLS, int, REALFHSTAT, DUALCALL_FHSTAT, \ 2757 (const void *fhp, size_t fh_size, struct stat *sb), \ 2758 (const char *, size_t, struct stat *), \ 2759 (fhp, fh_size, sb)) 2760 2761 VFSCALL(VFSBIT_FHCALLS, int, REALFHSTATVFS1, DUALCALL_FHSTATVFS1, \ 2762 (const void *fhp, size_t fh_size, struct statvfs *sb, int flgs),\ 2763 (const char *, size_t, struct statvfs *, int), \ 2764 (fhp, fh_size, sb, flgs)) 2765 #endif 2766 2767 2768 #ifdef __NetBSD__ 2769 2770 /* finally, put nfssvc here. "keep the namespace clean" */ 2771 #include <nfs/rpcv2.h> 2772 #include <nfs/nfs.h> 2773 2774 int 2775 nfssvc(int flags, void *argstructp) 2776 { 2777 int (*op_nfssvc)(int, void *); 2778 2779 if (vfsbits & VFSBIT_NFSSVC){ 2780 struct nfsd_args *nfsdargs; 2781 2782 /* massage the socket descriptor if necessary */ 2783 if (flags == NFSSVC_ADDSOCK) { 2784 nfsdargs = argstructp; 2785 nfsdargs->sock = fd_host2rump(nfsdargs->sock); 2786 } 2787 op_nfssvc = GETSYSCALL(rump, NFSSVC); 2788 } else 2789 op_nfssvc = GETSYSCALL(host, NFSSVC); 2790 2791 return op_nfssvc(flags, argstructp); 2792 } 2793 #endif /* __NetBSD__ */ 2794