1 /* $NetBSD: hijack.c,v 1.115 2015/03/04 23:42:26 pooka 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 1.115 2015/03/04 23:42:26 pooka 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, DUALCALL_BIND, DUALCALL_CONNECT, 93 DUALCALL_GETPEERNAME, DUALCALL_GETSOCKNAME, DUALCALL_LISTEN, 94 DUALCALL_RECVFROM, DUALCALL_RECVMSG, 95 DUALCALL_SENDTO, DUALCALL_SENDMSG, 96 DUALCALL_GETSOCKOPT, DUALCALL_SETSOCKOPT, 97 DUALCALL_SHUTDOWN, 98 DUALCALL_READ, DUALCALL_READV, DUALCALL_PREAD, DUALCALL_PREADV, 99 DUALCALL_DUP2, 100 DUALCALL_CLOSE, 101 DUALCALL_POLLTS, 102 103 #ifndef __linux__ 104 DUALCALL_STAT, DUALCALL_LSTAT, DUALCALL_FSTAT, 105 #endif 106 107 DUALCALL_CHMOD, DUALCALL_LCHMOD, DUALCALL_FCHMOD, 108 DUALCALL_CHOWN, DUALCALL_LCHOWN, DUALCALL_FCHOWN, 109 DUALCALL_OPEN, 110 DUALCALL_CHDIR, DUALCALL_FCHDIR, 111 DUALCALL_LSEEK, 112 DUALCALL_UNLINK, DUALCALL_SYMLINK, DUALCALL_READLINK, 113 DUALCALL_LINK, DUALCALL_RENAME, 114 DUALCALL_MKDIR, DUALCALL_RMDIR, 115 DUALCALL_UTIMES, DUALCALL_LUTIMES, DUALCALL_FUTIMES, 116 DUALCALL_UTIMENSAT, DUALCALL_FUTIMENS, 117 DUALCALL_TRUNCATE, DUALCALL_FTRUNCATE, 118 DUALCALL_FSYNC, 119 DUALCALL_ACCESS, 120 121 #ifndef __linux__ 122 DUALCALL___GETCWD, 123 DUALCALL_GETDENTS, 124 #endif 125 126 #ifndef __linux__ 127 DUALCALL_MKNOD, 128 #endif 129 130 #ifdef __NetBSD__ 131 DUALCALL_GETFH, DUALCALL_FHOPEN, DUALCALL_FHSTAT, DUALCALL_FHSTATVFS1, 132 #endif 133 134 #ifdef HAVE_KQUEUE 135 DUALCALL_KEVENT, 136 #endif 137 138 #ifdef __NetBSD__ 139 DUALCALL___SYSCTL, 140 #endif 141 142 #ifdef __NetBSD__ 143 DUALCALL_NFSSVC, 144 #endif 145 146 #ifdef __NetBSD__ 147 DUALCALL_STATVFS1, DUALCALL_FSTATVFS1, DUALCALL_GETVFSSTAT, 148 #endif 149 150 #ifdef __NetBSD__ 151 DUALCALL_MOUNT, DUALCALL_UNMOUNT, 152 #endif 153 154 #ifdef HAVE_FSYNC_RANGE 155 DUALCALL_FSYNC_RANGE, 156 #endif 157 158 #ifdef HAVE_CHFLAGS 159 DUALCALL_CHFLAGS, DUALCALL_LCHFLAGS, DUALCALL_FCHFLAGS, 160 #endif 161 162 #ifdef HAVE___QUOTACTL 163 DUALCALL_QUOTACTL, 164 #endif 165 DUALCALL__NUM 166 }; 167 168 #define RSYS_STRING(a) __STRING(a) 169 #define RSYS_NAME(a) RSYS_STRING(__CONCAT(RUMP_SYS_RENAME_,a)) 170 171 /* 172 * Would be nice to get this automatically in sync with libc. 173 * Also, this does not work for compat-using binaries (we should 174 * provide all previous interfaces, not just the current ones) 175 */ 176 #if defined(__NetBSD__) 177 178 #if !__NetBSD_Prereq__(5,99,7) 179 #define REALSELECT select 180 #define REALPOLLTS pollts 181 #define REALKEVENT kevent 182 #define REALSTAT __stat30 183 #define REALLSTAT __lstat30 184 #define REALFSTAT __fstat30 185 #define REALUTIMES utimes 186 #define REALLUTIMES lutimes 187 #define REALFUTIMES futimes 188 #define REALMKNOD mknod 189 #define REALFHSTAT __fhstat40 190 #else /* >= 5.99.7 */ 191 #define REALSELECT _sys___select50 192 #define REALPOLLTS _sys___pollts50 193 #define REALKEVENT _sys___kevent50 194 #define REALSTAT __stat50 195 #define REALLSTAT __lstat50 196 #define REALFSTAT __fstat50 197 #define REALUTIMES __utimes50 198 #define REALLUTIMES __lutimes50 199 #define REALFUTIMES __futimes50 200 #define REALMKNOD __mknod50 201 #define REALFHSTAT __fhstat50 202 #endif /* < 5.99.7 */ 203 204 #define REALREAD _sys_read 205 #define REALPREAD _sys_pread 206 #define REALPWRITE _sys_pwrite 207 #define REALGETDENTS __getdents30 208 #define REALMOUNT __mount50 209 #define REALGETFH __getfh30 210 #define REALFHOPEN __fhopen40 211 #define REALFHSTATVFS1 __fhstatvfs140 212 #define REALSOCKET __socket30 213 214 #define LSEEK_ALIAS _lseek 215 #define VFORK __vfork14 216 217 int REALSTAT(const char *, struct stat *); 218 int REALLSTAT(const char *, struct stat *); 219 int REALFSTAT(int, struct stat *); 220 int REALMKNOD(const char *, mode_t, dev_t); 221 int REALGETDENTS(int, char *, size_t); 222 223 int __getcwd(char *, size_t); 224 225 #elif defined(__linux__) /* glibc, really */ 226 227 #define REALREAD read 228 #define REALPREAD pread 229 #define REALPWRITE pwrite 230 #define REALSELECT select 231 #define REALPOLLTS ppoll 232 #define REALUTIMES utimes 233 #define REALLUTIMES lutimes 234 #define REALFUTIMES futimes 235 #define REALFHSTAT fhstat 236 #define REALSOCKET socket 237 238 #else /* !NetBSD && !linux */ 239 240 #error platform not supported 241 242 #endif /* platform */ 243 244 int REALSELECT(int, fd_set *, fd_set *, fd_set *, struct timeval *); 245 int REALPOLLTS(struct pollfd *, nfds_t, 246 const struct timespec *, const sigset_t *); 247 int REALKEVENT(int, const struct kevent *, size_t, struct kevent *, size_t, 248 const struct timespec *); 249 ssize_t REALREAD(int, void *, size_t); 250 ssize_t REALPREAD(int, void *, size_t, off_t); 251 ssize_t REALPWRITE(int, const void *, size_t, off_t); 252 int REALUTIMES(const char *, const struct timeval [2]); 253 int REALLUTIMES(const char *, const struct timeval [2]); 254 int REALFUTIMES(int, const struct timeval [2]); 255 int REALMOUNT(const char *, const char *, int, void *, size_t); 256 int REALGETFH(const char *, void *, size_t *); 257 int REALFHOPEN(const void *, size_t, int); 258 int REALFHSTAT(const void *, size_t, struct stat *); 259 int REALFHSTATVFS1(const void *, size_t, struct statvfs *, int); 260 int REALSOCKET(int, int, int); 261 262 #define S(a) __STRING(a) 263 struct sysnames { 264 enum dualcall scm_callnum; 265 const char *scm_hostname; 266 const char *scm_rumpname; 267 } syscnames[] = { 268 { DUALCALL_SOCKET, S(REALSOCKET), RSYS_NAME(SOCKET) }, 269 { DUALCALL_ACCEPT, "accept", RSYS_NAME(ACCEPT) }, 270 { DUALCALL_BIND, "bind", RSYS_NAME(BIND) }, 271 { DUALCALL_CONNECT, "connect", RSYS_NAME(CONNECT) }, 272 { DUALCALL_GETPEERNAME, "getpeername", RSYS_NAME(GETPEERNAME) }, 273 { DUALCALL_GETSOCKNAME, "getsockname", RSYS_NAME(GETSOCKNAME) }, 274 { DUALCALL_LISTEN, "listen", RSYS_NAME(LISTEN) }, 275 { DUALCALL_RECVFROM, "recvfrom", RSYS_NAME(RECVFROM) }, 276 { DUALCALL_RECVMSG, "recvmsg", RSYS_NAME(RECVMSG) }, 277 { DUALCALL_SENDTO, "sendto", RSYS_NAME(SENDTO) }, 278 { DUALCALL_SENDMSG, "sendmsg", RSYS_NAME(SENDMSG) }, 279 { DUALCALL_GETSOCKOPT, "getsockopt", RSYS_NAME(GETSOCKOPT) }, 280 { DUALCALL_SETSOCKOPT, "setsockopt", RSYS_NAME(SETSOCKOPT) }, 281 { DUALCALL_SHUTDOWN, "shutdown", RSYS_NAME(SHUTDOWN) }, 282 { DUALCALL_READ, S(REALREAD), RSYS_NAME(READ) }, 283 { DUALCALL_READV, "readv", RSYS_NAME(READV) }, 284 { DUALCALL_PREAD, S(REALPREAD), RSYS_NAME(PREAD) }, 285 { DUALCALL_PREADV, "preadv", RSYS_NAME(PREADV) }, 286 { DUALCALL_WRITE, "write", RSYS_NAME(WRITE) }, 287 { DUALCALL_WRITEV, "writev", RSYS_NAME(WRITEV) }, 288 { DUALCALL_PWRITE, S(REALPWRITE), RSYS_NAME(PWRITE) }, 289 { DUALCALL_PWRITEV, "pwritev", RSYS_NAME(PWRITEV) }, 290 { DUALCALL_IOCTL, "ioctl", RSYS_NAME(IOCTL) }, 291 { DUALCALL_FCNTL, "fcntl", RSYS_NAME(FCNTL) }, 292 { DUALCALL_DUP2, "dup2", RSYS_NAME(DUP2) }, 293 { DUALCALL_CLOSE, "close", RSYS_NAME(CLOSE) }, 294 { DUALCALL_POLLTS, S(REALPOLLTS), RSYS_NAME(POLLTS) }, 295 #ifndef __linux__ 296 { DUALCALL_STAT, S(REALSTAT), RSYS_NAME(STAT) }, 297 { DUALCALL_LSTAT, S(REALLSTAT), RSYS_NAME(LSTAT) }, 298 { DUALCALL_FSTAT, S(REALFSTAT), RSYS_NAME(FSTAT) }, 299 #endif 300 { DUALCALL_CHOWN, "chown", RSYS_NAME(CHOWN) }, 301 { DUALCALL_LCHOWN, "lchown", RSYS_NAME(LCHOWN) }, 302 { DUALCALL_FCHOWN, "fchown", RSYS_NAME(FCHOWN) }, 303 { DUALCALL_CHMOD, "chmod", RSYS_NAME(CHMOD) }, 304 { DUALCALL_LCHMOD, "lchmod", RSYS_NAME(LCHMOD) }, 305 { DUALCALL_FCHMOD, "fchmod", RSYS_NAME(FCHMOD) }, 306 { DUALCALL_UTIMES, S(REALUTIMES), RSYS_NAME(UTIMES) }, 307 { DUALCALL_LUTIMES, S(REALLUTIMES), RSYS_NAME(LUTIMES) }, 308 { DUALCALL_FUTIMES, S(REALFUTIMES), RSYS_NAME(FUTIMES) }, 309 { DUALCALL_UTIMENSAT, "utimensat", RSYS_NAME(UTIMENSAT) }, 310 { DUALCALL_FUTIMENS, "futimens", RSYS_NAME(FUTIMENS) }, 311 { DUALCALL_OPEN, "open", RSYS_NAME(OPEN) }, 312 { DUALCALL_CHDIR, "chdir", RSYS_NAME(CHDIR) }, 313 { DUALCALL_FCHDIR, "fchdir", RSYS_NAME(FCHDIR) }, 314 { DUALCALL_LSEEK, "lseek", RSYS_NAME(LSEEK) }, 315 { DUALCALL_UNLINK, "unlink", RSYS_NAME(UNLINK) }, 316 { DUALCALL_SYMLINK, "symlink", RSYS_NAME(SYMLINK) }, 317 { DUALCALL_READLINK, "readlink", RSYS_NAME(READLINK) }, 318 { DUALCALL_LINK, "link", RSYS_NAME(LINK) }, 319 { DUALCALL_RENAME, "rename", RSYS_NAME(RENAME) }, 320 { DUALCALL_MKDIR, "mkdir", RSYS_NAME(MKDIR) }, 321 { DUALCALL_RMDIR, "rmdir", RSYS_NAME(RMDIR) }, 322 { DUALCALL_TRUNCATE, "truncate", RSYS_NAME(TRUNCATE) }, 323 { DUALCALL_FTRUNCATE, "ftruncate", RSYS_NAME(FTRUNCATE) }, 324 { DUALCALL_FSYNC, "fsync", RSYS_NAME(FSYNC) }, 325 { DUALCALL_ACCESS, "access", RSYS_NAME(ACCESS) }, 326 327 #ifndef __linux__ 328 { DUALCALL___GETCWD, "__getcwd", RSYS_NAME(__GETCWD) }, 329 { DUALCALL_GETDENTS, S(REALGETDENTS),RSYS_NAME(GETDENTS) }, 330 #endif 331 332 #ifndef __linux__ 333 { DUALCALL_MKNOD, S(REALMKNOD), RSYS_NAME(MKNOD) }, 334 #endif 335 336 #ifdef __NetBSD__ 337 { DUALCALL_GETFH, S(REALGETFH), RSYS_NAME(GETFH) }, 338 { DUALCALL_FHOPEN, S(REALFHOPEN), RSYS_NAME(FHOPEN) }, 339 { DUALCALL_FHSTAT, S(REALFHSTAT), RSYS_NAME(FHSTAT) }, 340 { DUALCALL_FHSTATVFS1, S(REALFHSTATVFS1),RSYS_NAME(FHSTATVFS1) }, 341 #endif 342 343 #ifdef HAVE_KQUEUE 344 { DUALCALL_KEVENT, S(REALKEVENT), RSYS_NAME(KEVENT) }, 345 #endif 346 347 #ifdef __NetBSD__ 348 { DUALCALL___SYSCTL, "__sysctl", RSYS_NAME(__SYSCTL) }, 349 #endif 350 351 #ifdef __NetBSD__ 352 { DUALCALL_NFSSVC, "nfssvc", RSYS_NAME(NFSSVC) }, 353 #endif 354 355 #ifdef __NetBSD__ 356 { DUALCALL_STATVFS1, "statvfs1", RSYS_NAME(STATVFS1) }, 357 { DUALCALL_FSTATVFS1, "fstatvfs1", RSYS_NAME(FSTATVFS1) }, 358 { DUALCALL_GETVFSSTAT, "getvfsstat", RSYS_NAME(GETVFSSTAT) }, 359 #endif 360 361 #ifdef __NetBSD__ 362 { DUALCALL_MOUNT, S(REALMOUNT), RSYS_NAME(MOUNT) }, 363 { DUALCALL_UNMOUNT, "unmount", RSYS_NAME(UNMOUNT) }, 364 #endif 365 366 #ifdef HAVE_FSYNC_RANGE 367 { DUALCALL_FSYNC_RANGE, "fsync_range", RSYS_NAME(FSYNC_RANGE) }, 368 #endif 369 370 #ifdef HAVE_CHFLAGS 371 { DUALCALL_CHFLAGS, "chflags", RSYS_NAME(CHFLAGS) }, 372 { DUALCALL_LCHFLAGS, "lchflags", RSYS_NAME(LCHFLAGS) }, 373 { DUALCALL_FCHFLAGS, "fchflags", RSYS_NAME(FCHFLAGS) }, 374 #endif /* HAVE_CHFLAGS */ 375 376 #ifdef HAVE___QUOTACTL 377 { DUALCALL_QUOTACTL, "__quotactl", RSYS_NAME(__QUOTACTL) }, 378 #endif /* HAVE___QUOTACTL */ 379 380 }; 381 #undef S 382 383 struct bothsys { 384 void *bs_host; 385 void *bs_rump; 386 } syscalls[DUALCALL__NUM]; 387 #define GETSYSCALL(which, name) syscalls[DUALCALL_##name].bs_##which 388 389 static pid_t (*host_fork)(void); 390 static int (*host_daemon)(int, int); 391 static void * (*host_mmap)(void *, size_t, int, int, int, off_t); 392 393 /* 394 * This tracks if our process is in a subdirectory of /rump. 395 * It's preserved over exec. 396 */ 397 static bool pwdinrump; 398 399 enum pathtype { PATH_HOST, PATH_RUMP, PATH_RUMPBLANKET }; 400 401 static bool fd_isrump(int); 402 static enum pathtype path_isrump(const char *); 403 404 /* default FD_SETSIZE is 256 ==> default fdoff is 128 */ 405 static int hijack_fdoff = FD_SETSIZE/2; 406 407 /* 408 * Maintain a mapping table for the usual dup2 suspects. 409 * Could use atomic ops to operate on dup2vec, but an application 410 * racing there is not well-defined, so don't bother. 411 */ 412 /* note: you cannot change this without editing the env-passing code */ 413 #define DUP2HIGH 2 414 static uint32_t dup2vec[DUP2HIGH+1]; 415 #define DUP2BIT (1<<31) 416 #define DUP2ALIAS (1<<30) 417 #define DUP2FDMASK ((1<<30)-1) 418 419 static bool 420 isdup2d(int fd) 421 { 422 423 return fd <= DUP2HIGH && fd >= 0 && dup2vec[fd] & DUP2BIT; 424 } 425 426 static int 427 mapdup2(int hostfd) 428 { 429 430 _DIAGASSERT(isdup2d(hostfd)); 431 return dup2vec[hostfd] & DUP2FDMASK; 432 } 433 434 static int 435 unmapdup2(int rumpfd) 436 { 437 int i; 438 439 for (i = 0; i <= DUP2HIGH; i++) { 440 if (dup2vec[i] & DUP2BIT && 441 (dup2vec[i] & DUP2FDMASK) == (unsigned)rumpfd) 442 return i; 443 } 444 return -1; 445 } 446 447 static void 448 setdup2(int hostfd, int rumpfd) 449 { 450 451 if (hostfd > DUP2HIGH) { 452 _DIAGASSERT(0); 453 return; 454 } 455 456 dup2vec[hostfd] = DUP2BIT | DUP2ALIAS | rumpfd; 457 } 458 459 static void 460 clrdup2(int hostfd) 461 { 462 463 if (hostfd > DUP2HIGH) { 464 _DIAGASSERT(0); 465 return; 466 } 467 468 dup2vec[hostfd] = 0; 469 } 470 471 static bool 472 killdup2alias(int rumpfd) 473 { 474 int hostfd; 475 476 if ((hostfd = unmapdup2(rumpfd)) == -1) 477 return false; 478 479 if (dup2vec[hostfd] & DUP2ALIAS) { 480 dup2vec[hostfd] &= ~DUP2ALIAS; 481 return true; 482 } 483 return false; 484 } 485 486 //#define DEBUGJACK 487 #ifdef DEBUGJACK 488 #define DPRINTF(x) mydprintf x 489 static void 490 mydprintf(const char *fmt, ...) 491 { 492 va_list ap; 493 494 if (isdup2d(STDERR_FILENO)) 495 return; 496 497 va_start(ap, fmt); 498 vfprintf(stderr, fmt, ap); 499 va_end(ap); 500 } 501 502 static const char * 503 whichfd(int fd) 504 { 505 506 if (fd == -1) 507 return "-1"; 508 else if (fd_isrump(fd)) 509 return "rump"; 510 else 511 return "host"; 512 } 513 514 static const char * 515 whichpath(const char *path) 516 { 517 518 if (path_isrump(path)) 519 return "rump"; 520 else 521 return "host"; 522 } 523 524 #else 525 #define DPRINTF(x) 526 #endif 527 528 #define ATCALL(type, name, rcname, args, proto, vars) \ 529 type name args \ 530 { \ 531 type (*fun) proto; \ 532 int isrump = -1; \ 533 \ 534 if (fd == AT_FDCWD || *path == '/') { \ 535 isrump = path_isrump(path); \ 536 } else { \ 537 isrump = fd_isrump(fd); \ 538 } \ 539 \ 540 DPRINTF(("%s -> %d:%s (%s)\n", __STRING(name), \ 541 fd, path, isrump ? "rump" : "host")); \ 542 \ 543 assert(isrump != -1); \ 544 if (isrump) { \ 545 fun = syscalls[rcname].bs_rump; \ 546 if (fd != AT_FDCWD) \ 547 fd = fd_host2rump(fd); \ 548 path = path_host2rump(path); \ 549 } else { \ 550 fun = syscalls[rcname].bs_host; \ 551 } \ 552 return fun vars; \ 553 } 554 555 #define FDCALL(type, name, rcname, args, proto, vars) \ 556 type name args \ 557 { \ 558 type (*fun) proto; \ 559 \ 560 DPRINTF(("%s -> %d (%s)\n", __STRING(name), fd, whichfd(fd))); \ 561 if (fd_isrump(fd)) { \ 562 fun = syscalls[rcname].bs_rump; \ 563 fd = fd_host2rump(fd); \ 564 } else { \ 565 fun = syscalls[rcname].bs_host; \ 566 } \ 567 \ 568 return fun vars; \ 569 } 570 571 #define PATHCALL(type, name, rcname, args, proto, vars) \ 572 type name args \ 573 { \ 574 type (*fun) proto; \ 575 enum pathtype pt; \ 576 \ 577 DPRINTF(("%s -> %s (%s)\n", __STRING(name), path, \ 578 whichpath(path))); \ 579 if ((pt = path_isrump(path)) != PATH_HOST) { \ 580 fun = syscalls[rcname].bs_rump; \ 581 if (pt == PATH_RUMP) \ 582 path = path_host2rump(path); \ 583 } else { \ 584 fun = syscalls[rcname].bs_host; \ 585 } \ 586 \ 587 return fun vars; \ 588 } 589 590 #define VFSCALL(bit, type, name, rcname, args, proto, vars) \ 591 type name args \ 592 { \ 593 type (*fun) proto; \ 594 \ 595 DPRINTF(("%s (0x%x, 0x%x)\n", __STRING(name), bit, vfsbits)); \ 596 if (vfsbits & bit) { \ 597 fun = syscalls[rcname].bs_rump; \ 598 } else { \ 599 fun = syscalls[rcname].bs_host; \ 600 } \ 601 \ 602 return fun vars; \ 603 } 604 605 /* 606 * These variables are set from the RUMPHIJACK string and control 607 * which operations can product rump kernel file descriptors. 608 * This should be easily extendable for future needs. 609 */ 610 #define RUMPHIJACK_DEFAULT "path=/rump,socket=all:nolocal" 611 static bool rumpsockets[PF_MAX]; 612 static const char *rumpprefix; 613 static size_t rumpprefixlen; 614 615 static struct { 616 int pf; 617 const char *name; 618 } socketmap[] = { 619 { PF_LOCAL, "local" }, 620 { PF_INET, "inet" }, 621 #ifdef PF_LINK 622 { PF_LINK, "link" }, 623 #endif 624 #ifdef PF_OROUTE 625 { PF_OROUTE, "oroute" }, 626 #endif 627 { PF_ROUTE, "route" }, 628 { PF_INET6, "inet6" }, 629 #ifdef PF_MPLS 630 { PF_MPLS, "mpls" }, 631 #endif 632 { -1, NULL } 633 }; 634 635 static void 636 sockparser(char *buf) 637 { 638 char *p, *l = NULL; 639 bool value; 640 int i; 641 642 /* if "all" is present, it must be specified first */ 643 if (strncmp(buf, "all", strlen("all")) == 0) { 644 for (i = 0; i < (int)__arraycount(rumpsockets); i++) { 645 rumpsockets[i] = true; 646 } 647 buf += strlen("all"); 648 if (*buf == ':') 649 buf++; 650 } 651 652 for (p = strtok_r(buf, ":", &l); p; p = strtok_r(NULL, ":", &l)) { 653 value = true; 654 if (strncmp(p, "no", strlen("no")) == 0) { 655 value = false; 656 p += strlen("no"); 657 } 658 659 for (i = 0; socketmap[i].name; i++) { 660 if (strcmp(p, socketmap[i].name) == 0) { 661 rumpsockets[socketmap[i].pf] = value; 662 break; 663 } 664 } 665 if (socketmap[i].name == NULL) { 666 errx(1, "invalid socket specifier %s", p); 667 } 668 } 669 } 670 671 static void 672 pathparser(char *buf) 673 { 674 675 /* sanity-check */ 676 if (*buf != '/') 677 errx(1, "hijack path specifier must begin with ``/''"); 678 rumpprefixlen = strlen(buf); 679 if (rumpprefixlen < 2) 680 errx(1, "invalid hijack prefix: %s", buf); 681 if (buf[rumpprefixlen-1] == '/' && strspn(buf, "/") != rumpprefixlen) 682 errx(1, "hijack prefix may end in slash only if pure " 683 "slash, gave %s", buf); 684 685 if ((rumpprefix = strdup(buf)) == NULL) 686 err(1, "strdup"); 687 rumpprefixlen = strlen(rumpprefix); 688 } 689 690 static struct blanket { 691 const char *pfx; 692 size_t len; 693 } *blanket; 694 static int nblanket; 695 696 static void 697 blanketparser(char *buf) 698 { 699 char *p, *l = NULL; 700 int i; 701 702 for (nblanket = 0, p = buf; p; p = strchr(p+1, ':'), nblanket++) 703 continue; 704 705 blanket = malloc(nblanket * sizeof(*blanket)); 706 if (blanket == NULL) 707 err(1, "alloc blanket %d", nblanket); 708 709 for (p = strtok_r(buf, ":", &l), i = 0; p; 710 p = strtok_r(NULL, ":", &l), i++) { 711 blanket[i].pfx = strdup(p); 712 if (blanket[i].pfx == NULL) 713 err(1, "strdup blanket"); 714 blanket[i].len = strlen(p); 715 716 if (blanket[i].len == 0 || *blanket[i].pfx != '/') 717 errx(1, "invalid blanket specifier %s", p); 718 if (*(blanket[i].pfx + blanket[i].len-1) == '/') 719 errx(1, "invalid blanket specifier %s", p); 720 } 721 } 722 723 #define VFSBIT_NFSSVC 0x01 724 #define VFSBIT_GETVFSSTAT 0x02 725 #define VFSBIT_FHCALLS 0x04 726 static unsigned vfsbits; 727 728 static struct { 729 int bit; 730 const char *name; 731 } vfscalls[] = { 732 { VFSBIT_NFSSVC, "nfssvc" }, 733 { VFSBIT_GETVFSSTAT, "getvfsstat" }, 734 { VFSBIT_FHCALLS, "fhcalls" }, 735 { -1, NULL } 736 }; 737 738 static void 739 vfsparser(char *buf) 740 { 741 char *p, *l = NULL; 742 bool turnon; 743 unsigned int fullmask; 744 int i; 745 746 /* build the full mask and sanity-check while we're at it */ 747 fullmask = 0; 748 for (i = 0; vfscalls[i].name != NULL; i++) { 749 if (fullmask & vfscalls[i].bit) 750 errx(1, "problem exists between vi and chair"); 751 fullmask |= vfscalls[i].bit; 752 } 753 754 755 /* if "all" is present, it must be specified first */ 756 if (strncmp(buf, "all", strlen("all")) == 0) { 757 vfsbits = fullmask; 758 buf += strlen("all"); 759 if (*buf == ':') 760 buf++; 761 } 762 763 for (p = strtok_r(buf, ":", &l); p; p = strtok_r(NULL, ":", &l)) { 764 turnon = true; 765 if (strncmp(p, "no", strlen("no")) == 0) { 766 turnon = false; 767 p += strlen("no"); 768 } 769 770 for (i = 0; vfscalls[i].name; i++) { 771 if (strcmp(p, vfscalls[i].name) == 0) { 772 if (turnon) 773 vfsbits |= vfscalls[i].bit; 774 else 775 vfsbits &= ~vfscalls[i].bit; 776 break; 777 } 778 } 779 if (vfscalls[i].name == NULL) { 780 errx(1, "invalid vfscall specifier %s", p); 781 } 782 } 783 } 784 785 static bool rumpsysctl = false; 786 787 static void 788 sysctlparser(char *buf) 789 { 790 791 if (buf == NULL) { 792 rumpsysctl = true; 793 return; 794 } 795 796 if (strcasecmp(buf, "y") == 0 || strcasecmp(buf, "yes") == 0 || 797 strcasecmp(buf, "yep") == 0 || strcasecmp(buf, "tottakai") == 0) { 798 rumpsysctl = true; 799 return; 800 } 801 if (strcasecmp(buf, "n") == 0 || strcasecmp(buf, "no") == 0) { 802 rumpsysctl = false; 803 return; 804 } 805 806 errx(1, "sysctl value should be y(es)/n(o), gave: %s", buf); 807 } 808 809 static void 810 fdoffparser(char *buf) 811 { 812 unsigned long fdoff; 813 char *ep; 814 815 if (*buf == '-') { 816 errx(1, "fdoff must not be negative"); 817 } 818 fdoff = strtoul(buf, &ep, 10); 819 if (*ep != '\0') 820 errx(1, "invalid fdoff specifier \"%s\"", buf); 821 if (fdoff >= INT_MAX/2 || fdoff < 3) 822 errx(1, "fdoff out of range"); 823 hijack_fdoff = fdoff; 824 } 825 826 static struct { 827 void (*parsefn)(char *); 828 const char *name; 829 bool needvalues; 830 } hijackparse[] = { 831 { sockparser, "socket", true }, 832 { pathparser, "path", true }, 833 { blanketparser, "blanket", true }, 834 { vfsparser, "vfs", true }, 835 { sysctlparser, "sysctl", false }, 836 { fdoffparser, "fdoff", true }, 837 { NULL, NULL, false }, 838 }; 839 840 static void 841 parsehijack(char *hijack) 842 { 843 char *p, *p2, *l; 844 const char *hijackcopy; 845 bool nop2; 846 int i; 847 848 if ((hijackcopy = strdup(hijack)) == NULL) 849 err(1, "strdup"); 850 851 /* disable everything explicitly */ 852 for (i = 0; i < PF_MAX; i++) 853 rumpsockets[i] = false; 854 855 for (p = strtok_r(hijack, ",", &l); p; p = strtok_r(NULL, ",", &l)) { 856 nop2 = false; 857 p2 = strchr(p, '='); 858 if (!p2) { 859 nop2 = true; 860 p2 = p + strlen(p); 861 } 862 863 for (i = 0; hijackparse[i].parsefn; i++) { 864 if (strncmp(hijackparse[i].name, p, 865 (size_t)(p2-p)) == 0) { 866 if (nop2 && hijackparse[i].needvalues) 867 errx(1, "invalid hijack specifier: %s", 868 hijackcopy); 869 hijackparse[i].parsefn(nop2 ? NULL : p2+1); 870 break; 871 } 872 } 873 874 if (hijackparse[i].parsefn == NULL) 875 errx(1, "invalid hijack specifier name in %s", p); 876 } 877 878 } 879 880 static void __attribute__((constructor)) 881 rcinit(void) 882 { 883 char buf[1024]; 884 unsigned i, j; 885 886 host_fork = dlsym(RTLD_NEXT, "fork"); 887 host_daemon = dlsym(RTLD_NEXT, "daemon"); 888 host_mmap = dlsym(RTLD_NEXT, "mmap"); 889 890 /* 891 * In theory cannot print anything during lookups because 892 * we might not have the call vector set up. so, the errx() 893 * is a bit of a strech, but it might work. 894 */ 895 896 for (i = 0; i < DUALCALL__NUM; i++) { 897 /* build runtime O(1) access */ 898 for (j = 0; j < __arraycount(syscnames); j++) { 899 if (syscnames[j].scm_callnum == i) 900 break; 901 } 902 903 if (j == __arraycount(syscnames)) 904 errx(1, "rumphijack error: syscall pos %d missing", i); 905 906 syscalls[i].bs_host = dlsym(RTLD_NEXT, 907 syscnames[j].scm_hostname); 908 if (syscalls[i].bs_host == NULL) 909 errx(1, "hostcall %s not found!", 910 syscnames[j].scm_hostname); 911 912 syscalls[i].bs_rump = dlsym(RTLD_NEXT, 913 syscnames[j].scm_rumpname); 914 if (syscalls[i].bs_rump == NULL) 915 errx(1, "rumpcall %s not found!", 916 syscnames[j].scm_rumpname); 917 } 918 919 if (rumpclient_init() == -1) 920 err(1, "rumpclient init"); 921 922 /* check which syscalls we're supposed to hijack */ 923 if (getenv_r("RUMPHIJACK", buf, sizeof(buf)) == -1) { 924 strcpy(buf, RUMPHIJACK_DEFAULT); 925 } 926 parsehijack(buf); 927 928 /* set client persistence level */ 929 if (getenv_r("RUMPHIJACK_RETRYCONNECT", buf, sizeof(buf)) != -1) { 930 if (strcmp(buf, "die") == 0) 931 rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_DIE); 932 else if (strcmp(buf, "inftime") == 0) 933 rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_INFTIME); 934 else if (strcmp(buf, "once") == 0) 935 rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_ONCE); 936 else { 937 time_t timeout; 938 char *ep; 939 940 timeout = (time_t)strtoll(buf, &ep, 10); 941 if (timeout <= 0 || ep != buf + strlen(buf)) 942 errx(1, "RUMPHIJACK_RETRYCONNECT must be " 943 "keyword or integer, got: %s", buf); 944 945 rumpclient_setconnretry(timeout); 946 } 947 } 948 949 if (getenv_r("RUMPHIJACK__DUP2INFO", buf, sizeof(buf)) == 0) { 950 if (sscanf(buf, "%u,%u,%u", 951 &dup2vec[0], &dup2vec[1], &dup2vec[2]) != 3) { 952 warnx("invalid dup2mask: %s", buf); 953 memset(dup2vec, 0, sizeof(dup2vec)); 954 } 955 unsetenv("RUMPHIJACK__DUP2INFO"); 956 } 957 if (getenv_r("RUMPHIJACK__PWDINRUMP", buf, sizeof(buf)) == 0) { 958 pwdinrump = true; 959 unsetenv("RUMPHIJACK__PWDINRUMP"); 960 } 961 } 962 963 static int 964 fd_rump2host(int fd) 965 { 966 967 if (fd == -1) 968 return fd; 969 return fd + hijack_fdoff; 970 } 971 972 static int 973 fd_rump2host_withdup(int fd) 974 { 975 int hfd; 976 977 _DIAGASSERT(fd != -1); 978 hfd = unmapdup2(fd); 979 if (hfd != -1) { 980 _DIAGASSERT(hfd <= DUP2HIGH); 981 return hfd; 982 } 983 return fd_rump2host(fd); 984 } 985 986 static int 987 fd_host2rump(int fd) 988 { 989 990 if (!isdup2d(fd)) 991 return fd - hijack_fdoff; 992 else 993 return mapdup2(fd); 994 } 995 996 static bool 997 fd_isrump(int fd) 998 { 999 1000 return isdup2d(fd) || fd >= hijack_fdoff; 1001 } 1002 1003 #define assertfd(_fd_) assert(ISDUP2D(_fd_) || (_fd_) >= hijack_fdoff) 1004 1005 static enum pathtype 1006 path_isrump(const char *path) 1007 { 1008 size_t plen; 1009 int i; 1010 1011 if (rumpprefix == NULL && nblanket == 0) 1012 return PATH_HOST; 1013 1014 if (*path == '/') { 1015 plen = strlen(path); 1016 if (rumpprefix && plen >= rumpprefixlen) { 1017 if (strncmp(path, rumpprefix, rumpprefixlen) == 0 1018 && (plen == rumpprefixlen 1019 || *(path + rumpprefixlen) == '/')) { 1020 return PATH_RUMP; 1021 } 1022 } 1023 for (i = 0; i < nblanket; i++) { 1024 if (strncmp(path, blanket[i].pfx, blanket[i].len) == 0) 1025 return PATH_RUMPBLANKET; 1026 } 1027 1028 return PATH_HOST; 1029 } else { 1030 return pwdinrump ? PATH_RUMP : PATH_HOST; 1031 } 1032 } 1033 1034 static const char *rootpath = "/"; 1035 static const char * 1036 path_host2rump(const char *path) 1037 { 1038 const char *rv; 1039 1040 if (*path == '/') { 1041 rv = path + rumpprefixlen; 1042 if (*rv == '\0') 1043 rv = rootpath; 1044 } else { 1045 rv = path; 1046 } 1047 1048 return rv; 1049 } 1050 1051 static int 1052 dodup(int oldd, int minfd) 1053 { 1054 int (*op_fcntl)(int, int, ...); 1055 int newd; 1056 int isrump; 1057 1058 DPRINTF(("dup -> %d (minfd %d)\n", oldd, minfd)); 1059 if (fd_isrump(oldd)) { 1060 op_fcntl = GETSYSCALL(rump, FCNTL); 1061 oldd = fd_host2rump(oldd); 1062 if (minfd >= hijack_fdoff) 1063 minfd -= hijack_fdoff; 1064 isrump = 1; 1065 } else { 1066 op_fcntl = GETSYSCALL(host, FCNTL); 1067 isrump = 0; 1068 } 1069 1070 newd = op_fcntl(oldd, F_DUPFD, minfd); 1071 1072 if (isrump) 1073 newd = fd_rump2host(newd); 1074 DPRINTF(("dup <- %d\n", newd)); 1075 1076 return newd; 1077 } 1078 1079 /* 1080 * Check that host fd value does not exceed fdoffset and if necessary 1081 * dup the file descriptor so that it doesn't collide with the dup2mask. 1082 */ 1083 static int 1084 fd_host2host(int fd) 1085 { 1086 int (*op_fcntl)(int, int, ...) = GETSYSCALL(host, FCNTL); 1087 int (*op_close)(int) = GETSYSCALL(host, CLOSE); 1088 int ofd, i; 1089 1090 if (fd >= hijack_fdoff) { 1091 op_close(fd); 1092 errno = ENFILE; 1093 return -1; 1094 } 1095 1096 for (i = 1; isdup2d(fd); i++) { 1097 ofd = fd; 1098 fd = op_fcntl(ofd, F_DUPFD, i); 1099 op_close(ofd); 1100 } 1101 1102 return fd; 1103 } 1104 1105 int 1106 open(const char *path, int flags, ...) 1107 { 1108 int (*op_open)(const char *, int, ...); 1109 bool isrump; 1110 va_list ap; 1111 enum pathtype pt; 1112 int fd; 1113 1114 DPRINTF(("open -> %s (%s)\n", path, whichpath(path))); 1115 1116 if ((pt = path_isrump(path)) != PATH_HOST) { 1117 if (pt == PATH_RUMP) 1118 path = path_host2rump(path); 1119 op_open = GETSYSCALL(rump, OPEN); 1120 isrump = true; 1121 } else { 1122 op_open = GETSYSCALL(host, OPEN); 1123 isrump = false; 1124 } 1125 1126 va_start(ap, flags); 1127 fd = op_open(path, flags, va_arg(ap, mode_t)); 1128 va_end(ap); 1129 1130 if (isrump) 1131 fd = fd_rump2host(fd); 1132 else 1133 fd = fd_host2host(fd); 1134 1135 DPRINTF(("open <- %d (%s)\n", fd, whichfd(fd))); 1136 return fd; 1137 } 1138 1139 int 1140 chdir(const char *path) 1141 { 1142 int (*op_chdir)(const char *); 1143 enum pathtype pt; 1144 int rv; 1145 1146 if ((pt = path_isrump(path)) != PATH_HOST) { 1147 op_chdir = GETSYSCALL(rump, CHDIR); 1148 if (pt == PATH_RUMP) 1149 path = path_host2rump(path); 1150 } else { 1151 op_chdir = GETSYSCALL(host, CHDIR); 1152 } 1153 1154 rv = op_chdir(path); 1155 if (rv == 0) 1156 pwdinrump = pt != PATH_HOST; 1157 1158 return rv; 1159 } 1160 1161 int 1162 fchdir(int fd) 1163 { 1164 int (*op_fchdir)(int); 1165 bool isrump; 1166 int rv; 1167 1168 if (fd_isrump(fd)) { 1169 op_fchdir = GETSYSCALL(rump, FCHDIR); 1170 isrump = true; 1171 fd = fd_host2rump(fd); 1172 } else { 1173 op_fchdir = GETSYSCALL(host, FCHDIR); 1174 isrump = false; 1175 } 1176 1177 rv = op_fchdir(fd); 1178 if (rv == 0) { 1179 pwdinrump = isrump; 1180 } 1181 1182 return rv; 1183 } 1184 1185 #ifndef __linux__ 1186 int 1187 __getcwd(char *bufp, size_t len) 1188 { 1189 int (*op___getcwd)(char *, size_t); 1190 size_t prefixgap; 1191 bool iamslash; 1192 int rv; 1193 1194 if (pwdinrump && rumpprefix) { 1195 if (rumpprefix[rumpprefixlen-1] == '/') 1196 iamslash = true; 1197 else 1198 iamslash = false; 1199 1200 if (iamslash) 1201 prefixgap = rumpprefixlen - 1; /* ``//+path'' */ 1202 else 1203 prefixgap = rumpprefixlen; /* ``/pfx+/path'' */ 1204 if (len <= prefixgap) { 1205 errno = ERANGE; 1206 return -1; 1207 } 1208 1209 op___getcwd = GETSYSCALL(rump, __GETCWD); 1210 rv = op___getcwd(bufp + prefixgap, len - prefixgap); 1211 if (rv == -1) 1212 return rv; 1213 1214 /* augment the "/" part only for a non-root path */ 1215 memcpy(bufp, rumpprefix, rumpprefixlen); 1216 1217 /* append / only to non-root cwd */ 1218 if (rv != 2) 1219 bufp[prefixgap] = '/'; 1220 1221 /* don't append extra slash in the purely-slash case */ 1222 if (rv == 2 && !iamslash) 1223 bufp[rumpprefixlen] = '\0'; 1224 } else if (pwdinrump) { 1225 /* assume blanket. we can't provide a prefix here */ 1226 op___getcwd = GETSYSCALL(rump, __GETCWD); 1227 rv = op___getcwd(bufp, len); 1228 } else { 1229 op___getcwd = GETSYSCALL(host, __GETCWD); 1230 rv = op___getcwd(bufp, len); 1231 } 1232 1233 return rv; 1234 } 1235 #endif 1236 1237 static int 1238 moveish(const char *from, const char *to, 1239 int (*rump_op)(const char *, const char *), 1240 int (*host_op)(const char *, const char *)) 1241 { 1242 int (*op)(const char *, const char *); 1243 enum pathtype ptf, ptt; 1244 1245 if ((ptf = path_isrump(from)) != PATH_HOST) { 1246 if ((ptt = path_isrump(to)) == PATH_HOST) { 1247 errno = EXDEV; 1248 return -1; 1249 } 1250 1251 if (ptf == PATH_RUMP) 1252 from = path_host2rump(from); 1253 if (ptt == PATH_RUMP) 1254 to = path_host2rump(to); 1255 op = rump_op; 1256 } else { 1257 if (path_isrump(to) != PATH_HOST) { 1258 errno = EXDEV; 1259 return -1; 1260 } 1261 1262 op = host_op; 1263 } 1264 1265 return op(from, to); 1266 } 1267 1268 int 1269 link(const char *from, const char *to) 1270 { 1271 return moveish(from, to, 1272 GETSYSCALL(rump, LINK), GETSYSCALL(host, LINK)); 1273 } 1274 1275 int 1276 rename(const char *from, const char *to) 1277 { 1278 return moveish(from, to, 1279 GETSYSCALL(rump, RENAME), GETSYSCALL(host, RENAME)); 1280 } 1281 1282 int 1283 REALSOCKET(int domain, int type, int protocol) 1284 { 1285 int (*op_socket)(int, int, int); 1286 int fd; 1287 bool isrump; 1288 1289 isrump = domain < PF_MAX && rumpsockets[domain]; 1290 1291 if (isrump) 1292 op_socket = GETSYSCALL(rump, SOCKET); 1293 else 1294 op_socket = GETSYSCALL(host, SOCKET); 1295 fd = op_socket(domain, type, protocol); 1296 1297 if (isrump) 1298 fd = fd_rump2host(fd); 1299 else 1300 fd = fd_host2host(fd); 1301 DPRINTF(("socket <- %d\n", fd)); 1302 1303 return fd; 1304 } 1305 1306 int 1307 accept(int s, struct sockaddr *addr, socklen_t *addrlen) 1308 { 1309 int (*op_accept)(int, struct sockaddr *, socklen_t *); 1310 int fd; 1311 bool isrump; 1312 1313 isrump = fd_isrump(s); 1314 1315 DPRINTF(("accept -> %d", s)); 1316 if (isrump) { 1317 op_accept = GETSYSCALL(rump, ACCEPT); 1318 s = fd_host2rump(s); 1319 } else { 1320 op_accept = GETSYSCALL(host, ACCEPT); 1321 } 1322 fd = op_accept(s, addr, addrlen); 1323 if (fd != -1 && isrump) 1324 fd = fd_rump2host(fd); 1325 else 1326 fd = fd_host2host(fd); 1327 1328 DPRINTF((" <- %d\n", fd)); 1329 1330 return fd; 1331 } 1332 1333 /* 1334 * ioctl() and fcntl() are varargs calls and need special treatment. 1335 */ 1336 1337 /* 1338 * Various [Linux] libc's have various signatures for ioctl so we 1339 * need to handle the discrepancies. On NetBSD, we use the 1340 * one with unsigned long cmd. 1341 */ 1342 int 1343 #ifdef HAVE_IOCTL_CMD_INT 1344 ioctl(int fd, int cmd, ...) 1345 { 1346 int (*op_ioctl)(int, int cmd, ...); 1347 #else 1348 ioctl(int fd, unsigned long cmd, ...) 1349 { 1350 int (*op_ioctl)(int, unsigned long cmd, ...); 1351 #endif 1352 va_list ap; 1353 int rv; 1354 1355 DPRINTF(("ioctl -> %d\n", fd)); 1356 if (fd_isrump(fd)) { 1357 fd = fd_host2rump(fd); 1358 op_ioctl = GETSYSCALL(rump, IOCTL); 1359 } else { 1360 op_ioctl = GETSYSCALL(host, IOCTL); 1361 } 1362 1363 va_start(ap, cmd); 1364 rv = op_ioctl(fd, cmd, va_arg(ap, void *)); 1365 va_end(ap); 1366 return rv; 1367 } 1368 1369 int 1370 fcntl(int fd, int cmd, ...) 1371 { 1372 int (*op_fcntl)(int, int, ...); 1373 va_list ap; 1374 int rv, minfd; 1375 1376 DPRINTF(("fcntl -> %d (cmd %d)\n", fd, cmd)); 1377 1378 switch (cmd) { 1379 case F_DUPFD: 1380 va_start(ap, cmd); 1381 minfd = va_arg(ap, int); 1382 va_end(ap); 1383 return dodup(fd, minfd); 1384 1385 #ifdef F_CLOSEM 1386 case F_CLOSEM: { 1387 int maxdup2, i; 1388 1389 /* 1390 * So, if fd < HIJACKOFF, we want to do a host closem. 1391 */ 1392 1393 if (fd < hijack_fdoff) { 1394 int closemfd = fd; 1395 1396 if (rumpclient__closenotify(&closemfd, 1397 RUMPCLIENT_CLOSE_FCLOSEM) == -1) 1398 return -1; 1399 op_fcntl = GETSYSCALL(host, FCNTL); 1400 rv = op_fcntl(closemfd, cmd); 1401 if (rv) 1402 return rv; 1403 } 1404 1405 /* 1406 * Additionally, we want to do a rump closem, but only 1407 * for the file descriptors not dup2'd. 1408 */ 1409 1410 for (i = 0, maxdup2 = -1; i <= DUP2HIGH; i++) { 1411 if (dup2vec[i] & DUP2BIT) { 1412 int val; 1413 1414 val = dup2vec[i] & DUP2FDMASK; 1415 maxdup2 = MAX(val, maxdup2); 1416 } 1417 } 1418 1419 if (fd >= hijack_fdoff) 1420 fd -= hijack_fdoff; 1421 else 1422 fd = 0; 1423 fd = MAX(maxdup2+1, fd); 1424 1425 /* hmm, maybe we should close rump fd's not within dup2mask? */ 1426 return rump_sys_fcntl(fd, F_CLOSEM); 1427 } 1428 #endif /* F_CLOSEM */ 1429 1430 #ifdef F_MAXFD 1431 case F_MAXFD: 1432 /* 1433 * For maxfd, if there's a rump kernel fd, return 1434 * it hostified. Otherwise, return host's MAXFD 1435 * return value. 1436 */ 1437 if ((rv = rump_sys_fcntl(fd, F_MAXFD)) != -1) { 1438 /* 1439 * This might go a little wrong in case 1440 * of dup2 to [012], but I'm not sure if 1441 * there's a justification for tracking 1442 * that info. Consider e.g. 1443 * dup2(rumpfd, 2) followed by rump_sys_open() 1444 * returning 1. We should return 1+HIJACKOFF, 1445 * not 2+HIJACKOFF. However, if [01] is not 1446 * open, the correct return value is 2. 1447 */ 1448 return fd_rump2host(fd); 1449 } else { 1450 op_fcntl = GETSYSCALL(host, FCNTL); 1451 return op_fcntl(fd, F_MAXFD); 1452 } 1453 /*NOTREACHED*/ 1454 #endif /* F_MAXFD */ 1455 1456 default: 1457 if (fd_isrump(fd)) { 1458 fd = fd_host2rump(fd); 1459 op_fcntl = GETSYSCALL(rump, FCNTL); 1460 } else { 1461 op_fcntl = GETSYSCALL(host, FCNTL); 1462 } 1463 1464 va_start(ap, cmd); 1465 rv = op_fcntl(fd, cmd, va_arg(ap, void *)); 1466 va_end(ap); 1467 return rv; 1468 } 1469 /*NOTREACHED*/ 1470 } 1471 1472 int 1473 close(int fd) 1474 { 1475 int (*op_close)(int); 1476 int rv; 1477 1478 DPRINTF(("close -> %d\n", fd)); 1479 if (fd_isrump(fd)) { 1480 bool undup2 = false; 1481 int ofd; 1482 1483 if (isdup2d(ofd = fd)) { 1484 undup2 = true; 1485 } 1486 1487 fd = fd_host2rump(fd); 1488 if (!undup2 && killdup2alias(fd)) { 1489 return 0; 1490 } 1491 1492 op_close = GETSYSCALL(rump, CLOSE); 1493 rv = op_close(fd); 1494 if (rv == 0 && undup2) { 1495 clrdup2(ofd); 1496 } 1497 } else { 1498 if (rumpclient__closenotify(&fd, RUMPCLIENT_CLOSE_CLOSE) == -1) 1499 return -1; 1500 op_close = GETSYSCALL(host, CLOSE); 1501 rv = op_close(fd); 1502 } 1503 1504 return rv; 1505 } 1506 1507 /* 1508 * write cannot issue a standard debug printf due to recursion 1509 */ 1510 ssize_t 1511 write(int fd, const void *buf, size_t blen) 1512 { 1513 ssize_t (*op_write)(int, const void *, size_t); 1514 1515 if (fd_isrump(fd)) { 1516 fd = fd_host2rump(fd); 1517 op_write = GETSYSCALL(rump, WRITE); 1518 } else { 1519 op_write = GETSYSCALL(host, WRITE); 1520 } 1521 1522 return op_write(fd, buf, blen); 1523 } 1524 1525 /* 1526 * file descriptor passing 1527 * 1528 * we intercept sendmsg and recvmsg to convert file descriptors in 1529 * control messages. an attempt to send a descriptor from a different kernel 1530 * is rejected. (ENOTSUP) 1531 */ 1532 1533 static int 1534 msg_convert(struct msghdr *msg, int (*func)(int)) 1535 { 1536 struct cmsghdr *cmsg; 1537 1538 for (cmsg = CMSG_FIRSTHDR(msg); cmsg != NULL; 1539 cmsg = CMSG_NXTHDR(msg, cmsg)) { 1540 if (cmsg->cmsg_level == SOL_SOCKET && 1541 cmsg->cmsg_type == SCM_RIGHTS) { 1542 int *fdp = (void *)CMSG_DATA(cmsg); 1543 const size_t size = 1544 cmsg->cmsg_len - __CMSG_ALIGN(sizeof(*cmsg)); 1545 const int nfds = (int)(size / sizeof(int)); 1546 const int * const efdp = fdp + nfds; 1547 1548 while (fdp < efdp) { 1549 const int newval = func(*fdp); 1550 1551 if (newval < 0) { 1552 return ENOTSUP; 1553 } 1554 *fdp = newval; 1555 fdp++; 1556 } 1557 } 1558 } 1559 return 0; 1560 } 1561 1562 ssize_t 1563 recvmsg(int fd, struct msghdr *msg, int flags) 1564 { 1565 ssize_t (*op_recvmsg)(int, struct msghdr *, int); 1566 ssize_t ret; 1567 const bool isrump = fd_isrump(fd); 1568 1569 if (isrump) { 1570 fd = fd_host2rump(fd); 1571 op_recvmsg = GETSYSCALL(rump, RECVMSG); 1572 } else { 1573 op_recvmsg = GETSYSCALL(host, RECVMSG); 1574 } 1575 ret = op_recvmsg(fd, msg, flags); 1576 if (ret == -1) { 1577 return ret; 1578 } 1579 /* 1580 * convert descriptors in the message. 1581 */ 1582 if (isrump) { 1583 msg_convert(msg, fd_rump2host); 1584 } else { 1585 msg_convert(msg, fd_host2host); 1586 } 1587 return ret; 1588 } 1589 1590 ssize_t 1591 recv(int fd, void *buf, size_t len, int flags) 1592 { 1593 1594 return recvfrom(fd, buf, len, flags, NULL, NULL); 1595 } 1596 1597 ssize_t 1598 send(int fd, const void *buf, size_t len, int flags) 1599 { 1600 1601 return sendto(fd, buf, len, flags, NULL, 0); 1602 } 1603 1604 static int 1605 fd_check_rump(int fd) 1606 { 1607 1608 return fd_isrump(fd) ? 0 : -1; 1609 } 1610 1611 static int 1612 fd_check_host(int fd) 1613 { 1614 1615 return !fd_isrump(fd) ? 0 : -1; 1616 } 1617 1618 ssize_t 1619 sendmsg(int fd, const struct msghdr *msg, int flags) 1620 { 1621 ssize_t (*op_sendmsg)(int, const struct msghdr *, int); 1622 const bool isrump = fd_isrump(fd); 1623 int error; 1624 1625 /* 1626 * reject descriptors from a different kernel. 1627 */ 1628 error = msg_convert(__UNCONST(msg), 1629 isrump ? fd_check_rump: fd_check_host); 1630 if (error != 0) { 1631 errno = error; 1632 return -1; 1633 } 1634 /* 1635 * convert descriptors in the message to raw values. 1636 */ 1637 if (isrump) { 1638 fd = fd_host2rump(fd); 1639 /* 1640 * XXX we directly modify the given message assuming: 1641 * - cmsg is writable (typically on caller's stack) 1642 * - caller don't care cmsg's contents after calling sendmsg. 1643 * (thus no need to restore values) 1644 * 1645 * it's safer to copy and modify instead. 1646 */ 1647 msg_convert(__UNCONST(msg), fd_host2rump); 1648 op_sendmsg = GETSYSCALL(rump, SENDMSG); 1649 } else { 1650 op_sendmsg = GETSYSCALL(host, SENDMSG); 1651 } 1652 return op_sendmsg(fd, msg, flags); 1653 } 1654 1655 /* 1656 * dup2 is special. we allow dup2 of a rump kernel fd to 0-2 since 1657 * many programs do that. dup2 of a rump kernel fd to another value 1658 * not >= fdoff is an error. 1659 * 1660 * Note: cannot rump2host newd, because it is often hardcoded. 1661 */ 1662 int 1663 dup2(int oldd, int newd) 1664 { 1665 int (*host_dup2)(int, int); 1666 int rv; 1667 1668 DPRINTF(("dup2 -> %d (o) -> %d (n)\n", oldd, newd)); 1669 1670 if (fd_isrump(oldd)) { 1671 int (*op_close)(int) = GETSYSCALL(host, CLOSE); 1672 1673 /* only allow fd 0-2 for cross-kernel dup */ 1674 if (!(newd >= 0 && newd <= 2 && !fd_isrump(newd))) { 1675 errno = EBADF; 1676 return -1; 1677 } 1678 1679 /* regular dup2? */ 1680 if (fd_isrump(newd)) { 1681 newd = fd_host2rump(newd); 1682 rv = rump_sys_dup2(oldd, newd); 1683 return fd_rump2host(rv); 1684 } 1685 1686 /* 1687 * dup2 rump => host? just establish an 1688 * entry in the mapping table. 1689 */ 1690 op_close(newd); 1691 setdup2(newd, fd_host2rump(oldd)); 1692 rv = 0; 1693 } else { 1694 host_dup2 = syscalls[DUALCALL_DUP2].bs_host; 1695 if (rumpclient__closenotify(&newd, RUMPCLIENT_CLOSE_DUP2) == -1) 1696 return -1; 1697 rv = host_dup2(oldd, newd); 1698 } 1699 1700 return rv; 1701 } 1702 1703 int 1704 dup(int oldd) 1705 { 1706 1707 return dodup(oldd, 0); 1708 } 1709 1710 pid_t 1711 fork(void) 1712 { 1713 pid_t rv; 1714 1715 DPRINTF(("fork\n")); 1716 1717 rv = rumpclient__dofork(host_fork); 1718 1719 DPRINTF(("fork returns %d\n", rv)); 1720 return rv; 1721 } 1722 #ifdef VFORK 1723 /* we do not have the luxury of not requiring a stackframe */ 1724 __strong_alias(VFORK,fork); 1725 #endif 1726 1727 int 1728 daemon(int nochdir, int noclose) 1729 { 1730 struct rumpclient_fork *rf; 1731 1732 if ((rf = rumpclient_prefork()) == NULL) 1733 return -1; 1734 1735 if (host_daemon(nochdir, noclose) == -1) 1736 return -1; 1737 1738 if (rumpclient_fork_init(rf) == -1) 1739 return -1; 1740 1741 return 0; 1742 } 1743 1744 int 1745 execve(const char *path, char *const argv[], char *const envp[]) 1746 { 1747 char buf[128]; 1748 char *dup2str; 1749 const char *pwdinrumpstr; 1750 char **newenv; 1751 size_t nelem; 1752 int rv, sverrno; 1753 int bonus = 2, i = 0; 1754 1755 snprintf(buf, sizeof(buf), "RUMPHIJACK__DUP2INFO=%u,%u,%u", 1756 dup2vec[0], dup2vec[1], dup2vec[2]); 1757 dup2str = strdup(buf); 1758 if (dup2str == NULL) { 1759 errno = ENOMEM; 1760 return -1; 1761 } 1762 1763 if (pwdinrump) { 1764 pwdinrumpstr = "RUMPHIJACK__PWDINRUMP=true"; 1765 bonus++; 1766 } else { 1767 pwdinrumpstr = NULL; 1768 } 1769 1770 for (nelem = 0; envp && envp[nelem]; nelem++) 1771 continue; 1772 newenv = malloc(sizeof(*newenv) * (nelem+bonus)); 1773 if (newenv == NULL) { 1774 free(dup2str); 1775 errno = ENOMEM; 1776 return -1; 1777 } 1778 memcpy(newenv, envp, nelem*sizeof(*newenv)); 1779 newenv[nelem+i] = dup2str; 1780 i++; 1781 1782 if (pwdinrumpstr) { 1783 newenv[nelem+i] = __UNCONST(pwdinrumpstr); 1784 i++; 1785 } 1786 newenv[nelem+i] = NULL; 1787 _DIAGASSERT(i < bonus); 1788 1789 rv = rumpclient_exec(path, argv, newenv); 1790 1791 _DIAGASSERT(rv != 0); 1792 sverrno = errno; 1793 free(newenv); 1794 free(dup2str); 1795 errno = sverrno; 1796 return rv; 1797 } 1798 1799 /* 1800 * select is done by calling poll. 1801 */ 1802 int 1803 REALSELECT(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, 1804 struct timeval *timeout) 1805 { 1806 struct pollfd *pfds; 1807 struct timespec ts, *tsp = NULL; 1808 nfds_t realnfds; 1809 int i, j; 1810 int rv, incr; 1811 1812 DPRINTF(("select %d %p %p %p %p\n", nfds, 1813 readfds, writefds, exceptfds, timeout)); 1814 1815 /* 1816 * Well, first we must scan the fds to figure out how many 1817 * fds there really are. This is because up to and including 1818 * nb5 poll() silently refuses nfds > process_maxopen_fds. 1819 * Seems to be fixed in current, thank the maker. 1820 * god damn cluster...bomb. 1821 */ 1822 1823 for (i = 0, realnfds = 0; i < nfds; i++) { 1824 if (readfds && FD_ISSET(i, readfds)) { 1825 realnfds++; 1826 continue; 1827 } 1828 if (writefds && FD_ISSET(i, writefds)) { 1829 realnfds++; 1830 continue; 1831 } 1832 if (exceptfds && FD_ISSET(i, exceptfds)) { 1833 realnfds++; 1834 continue; 1835 } 1836 } 1837 1838 if (realnfds) { 1839 pfds = calloc(realnfds, sizeof(*pfds)); 1840 if (!pfds) 1841 return -1; 1842 } else { 1843 pfds = NULL; 1844 } 1845 1846 for (i = 0, j = 0; i < nfds; i++) { 1847 incr = 0; 1848 if (readfds && FD_ISSET(i, readfds)) { 1849 pfds[j].fd = i; 1850 pfds[j].events |= POLLIN; 1851 incr=1; 1852 } 1853 if (writefds && FD_ISSET(i, writefds)) { 1854 pfds[j].fd = i; 1855 pfds[j].events |= POLLOUT; 1856 incr=1; 1857 } 1858 if (exceptfds && FD_ISSET(i, exceptfds)) { 1859 pfds[j].fd = i; 1860 pfds[j].events |= POLLHUP|POLLERR; 1861 incr=1; 1862 } 1863 if (incr) 1864 j++; 1865 } 1866 assert(j == (int)realnfds); 1867 1868 if (timeout) { 1869 TIMEVAL_TO_TIMESPEC(timeout, &ts); 1870 tsp = &ts; 1871 } 1872 rv = REALPOLLTS(pfds, realnfds, tsp, NULL); 1873 /* 1874 * "If select() returns with an error the descriptor sets 1875 * will be unmodified" 1876 */ 1877 if (rv < 0) 1878 goto out; 1879 1880 /* 1881 * zero out results (can't use FD_ZERO for the 1882 * obvious select-me-not reason). whee. 1883 * 1884 * We do this here since some software ignores the return 1885 * value of select, and hence if the timeout expires, it may 1886 * assume all input descriptors have activity. 1887 */ 1888 for (i = 0; i < nfds; i++) { 1889 if (readfds) 1890 FD_CLR(i, readfds); 1891 if (writefds) 1892 FD_CLR(i, writefds); 1893 if (exceptfds) 1894 FD_CLR(i, exceptfds); 1895 } 1896 if (rv == 0) 1897 goto out; 1898 1899 /* 1900 * We have >0 fds with activity. Harvest the results. 1901 */ 1902 for (i = 0; i < (int)realnfds; i++) { 1903 if (readfds) { 1904 if (pfds[i].revents & POLLIN) { 1905 FD_SET(pfds[i].fd, readfds); 1906 } 1907 } 1908 if (writefds) { 1909 if (pfds[i].revents & POLLOUT) { 1910 FD_SET(pfds[i].fd, writefds); 1911 } 1912 } 1913 if (exceptfds) { 1914 if (pfds[i].revents & (POLLHUP|POLLERR)) { 1915 FD_SET(pfds[i].fd, exceptfds); 1916 } 1917 } 1918 } 1919 1920 out: 1921 free(pfds); 1922 return rv; 1923 } 1924 1925 static void 1926 checkpoll(struct pollfd *fds, nfds_t nfds, int *hostcall, int *rumpcall) 1927 { 1928 nfds_t i; 1929 1930 for (i = 0; i < nfds; i++) { 1931 if (fds[i].fd == -1) 1932 continue; 1933 1934 if (fd_isrump(fds[i].fd)) 1935 (*rumpcall)++; 1936 else 1937 (*hostcall)++; 1938 } 1939 } 1940 1941 static void 1942 adjustpoll(struct pollfd *fds, nfds_t nfds, int (*fdadj)(int)) 1943 { 1944 nfds_t i; 1945 1946 for (i = 0; i < nfds; i++) { 1947 fds[i].fd = fdadj(fds[i].fd); 1948 } 1949 } 1950 1951 /* 1952 * poll is easy as long as the call comes in the fds only in one 1953 * kernel. otherwise its quite tricky... 1954 */ 1955 struct pollarg { 1956 struct pollfd *pfds; 1957 nfds_t nfds; 1958 const struct timespec *ts; 1959 const sigset_t *sigmask; 1960 int pipefd; 1961 int errnum; 1962 }; 1963 1964 static void * 1965 hostpoll(void *arg) 1966 { 1967 int (*op_pollts)(struct pollfd *, nfds_t, const struct timespec *, 1968 const sigset_t *); 1969 struct pollarg *parg = arg; 1970 intptr_t rv; 1971 1972 op_pollts = GETSYSCALL(host, POLLTS); 1973 rv = op_pollts(parg->pfds, parg->nfds, parg->ts, parg->sigmask); 1974 if (rv == -1) 1975 parg->errnum = errno; 1976 rump_sys_write(parg->pipefd, &rv, sizeof(rv)); 1977 1978 return (void *)rv; 1979 } 1980 1981 int 1982 REALPOLLTS(struct pollfd *fds, nfds_t nfds, const struct timespec *ts, 1983 const sigset_t *sigmask) 1984 { 1985 int (*op_pollts)(struct pollfd *, nfds_t, const struct timespec *, 1986 const sigset_t *); 1987 int (*host_close)(int); 1988 int hostcall = 0, rumpcall = 0; 1989 pthread_t pt; 1990 nfds_t i; 1991 int rv; 1992 1993 DPRINTF(("poll %p %d %p %p\n", fds, (int)nfds, ts, sigmask)); 1994 checkpoll(fds, nfds, &hostcall, &rumpcall); 1995 1996 if (hostcall && rumpcall) { 1997 struct pollfd *pfd_host = NULL, *pfd_rump = NULL; 1998 int rpipe[2] = {-1,-1}, hpipe[2] = {-1,-1}; 1999 struct pollarg parg; 2000 void *trv_val; 2001 int sverrno = 0, rv_rump, rv_host, errno_rump, errno_host; 2002 2003 /* 2004 * ok, this is where it gets tricky. We must support 2005 * this since it's a very common operation in certain 2006 * types of software (telnet, netcat, etc). We allocate 2007 * two vectors and run two poll commands in separate 2008 * threads. Whichever returns first "wins" and the 2009 * other kernel's fds won't show activity. 2010 */ 2011 rv = -1; 2012 2013 /* allocate full vector for O(n) joining after call */ 2014 pfd_host = malloc(sizeof(*pfd_host)*(nfds+1)); 2015 if (!pfd_host) 2016 goto out; 2017 pfd_rump = malloc(sizeof(*pfd_rump)*(nfds+1)); 2018 if (!pfd_rump) { 2019 goto out; 2020 } 2021 2022 /* 2023 * then, open two pipes, one for notifications 2024 * to each kernel. 2025 * 2026 * At least the rump pipe should probably be 2027 * cached, along with the helper threads. This 2028 * should give a microbenchmark improvement (haven't 2029 * experienced a macro-level problem yet, though). 2030 */ 2031 if ((rv = rump_sys_pipe(rpipe)) == -1) { 2032 sverrno = errno; 2033 } 2034 if (rv == 0 && (rv = pipe(hpipe)) == -1) { 2035 sverrno = errno; 2036 } 2037 2038 /* split vectors (or signal errors) */ 2039 for (i = 0; i < nfds; i++) { 2040 int fd; 2041 2042 fds[i].revents = 0; 2043 if (fds[i].fd == -1) { 2044 pfd_host[i].fd = -1; 2045 pfd_rump[i].fd = -1; 2046 } else if (fd_isrump(fds[i].fd)) { 2047 pfd_host[i].fd = -1; 2048 fd = fd_host2rump(fds[i].fd); 2049 if (fd == rpipe[0] || fd == rpipe[1]) { 2050 fds[i].revents = POLLNVAL; 2051 if (rv != -1) 2052 rv++; 2053 } 2054 pfd_rump[i].fd = fd; 2055 pfd_rump[i].events = fds[i].events; 2056 } else { 2057 pfd_rump[i].fd = -1; 2058 fd = fds[i].fd; 2059 if (fd == hpipe[0] || fd == hpipe[1]) { 2060 fds[i].revents = POLLNVAL; 2061 if (rv != -1) 2062 rv++; 2063 } 2064 pfd_host[i].fd = fd; 2065 pfd_host[i].events = fds[i].events; 2066 } 2067 pfd_rump[i].revents = pfd_host[i].revents = 0; 2068 } 2069 if (rv) { 2070 goto out; 2071 } 2072 2073 pfd_host[nfds].fd = hpipe[0]; 2074 pfd_host[nfds].events = POLLIN; 2075 pfd_rump[nfds].fd = rpipe[0]; 2076 pfd_rump[nfds].events = POLLIN; 2077 2078 /* 2079 * then, create a thread to do host part and meanwhile 2080 * do rump kernel part right here 2081 */ 2082 2083 parg.pfds = pfd_host; 2084 parg.nfds = nfds+1; 2085 parg.ts = ts; 2086 parg.sigmask = sigmask; 2087 parg.pipefd = rpipe[1]; 2088 pthread_create(&pt, NULL, hostpoll, &parg); 2089 2090 op_pollts = GETSYSCALL(rump, POLLTS); 2091 rv_rump = op_pollts(pfd_rump, nfds+1, ts, NULL); 2092 errno_rump = errno; 2093 write(hpipe[1], &rv, sizeof(rv)); 2094 pthread_join(pt, &trv_val); 2095 rv_host = (int)(intptr_t)trv_val; 2096 errno_host = parg.errnum; 2097 2098 /* strip cross-thread notification from real results */ 2099 if (rv_host > 0 && pfd_host[nfds].revents & POLLIN) { 2100 rv_host--; 2101 } 2102 if (rv_rump > 0 && pfd_rump[nfds].revents & POLLIN) { 2103 rv_rump--; 2104 } 2105 2106 /* then merge the results into what's reported to the caller */ 2107 if (rv_rump > 0 || rv_host > 0) { 2108 /* SUCCESS */ 2109 2110 rv = 0; 2111 if (rv_rump > 0) { 2112 for (i = 0; i < nfds; i++) { 2113 if (pfd_rump[i].fd != -1) 2114 fds[i].revents 2115 = pfd_rump[i].revents; 2116 } 2117 rv += rv_rump; 2118 } 2119 if (rv_host > 0) { 2120 for (i = 0; i < nfds; i++) { 2121 if (pfd_host[i].fd != -1) 2122 fds[i].revents 2123 = pfd_host[i].revents; 2124 } 2125 rv += rv_host; 2126 } 2127 assert(rv > 0); 2128 sverrno = 0; 2129 } else if (rv_rump == -1 || rv_host == -1) { 2130 /* ERROR */ 2131 2132 /* just pick one kernel at "random" */ 2133 rv = -1; 2134 if (rv_host == -1) { 2135 sverrno = errno_host; 2136 } else if (rv_rump == -1) { 2137 sverrno = errno_rump; 2138 } 2139 } else { 2140 /* TIMEOUT */ 2141 2142 rv = 0; 2143 assert(rv_rump == 0 && rv_host == 0); 2144 } 2145 2146 out: 2147 host_close = GETSYSCALL(host, CLOSE); 2148 if (rpipe[0] != -1) 2149 rump_sys_close(rpipe[0]); 2150 if (rpipe[1] != -1) 2151 rump_sys_close(rpipe[1]); 2152 if (hpipe[0] != -1) 2153 host_close(hpipe[0]); 2154 if (hpipe[1] != -1) 2155 host_close(hpipe[1]); 2156 free(pfd_host); 2157 free(pfd_rump); 2158 errno = sverrno; 2159 } else { 2160 if (hostcall) { 2161 op_pollts = GETSYSCALL(host, POLLTS); 2162 } else { 2163 op_pollts = GETSYSCALL(rump, POLLTS); 2164 adjustpoll(fds, nfds, fd_host2rump); 2165 } 2166 2167 rv = op_pollts(fds, nfds, ts, sigmask); 2168 if (rumpcall) 2169 adjustpoll(fds, nfds, fd_rump2host_withdup); 2170 } 2171 2172 return rv; 2173 } 2174 2175 int 2176 poll(struct pollfd *fds, nfds_t nfds, int timeout) 2177 { 2178 struct timespec ts; 2179 struct timespec *tsp = NULL; 2180 2181 if (timeout != INFTIM) { 2182 ts.tv_sec = timeout / 1000; 2183 ts.tv_nsec = (timeout % 1000) * 1000*1000; 2184 2185 tsp = &ts; 2186 } 2187 2188 return REALPOLLTS(fds, nfds, tsp, NULL); 2189 } 2190 2191 #ifdef HAVE_KQUEUE 2192 int 2193 REALKEVENT(int kq, const struct kevent *changelist, size_t nchanges, 2194 struct kevent *eventlist, size_t nevents, 2195 const struct timespec *timeout) 2196 { 2197 int (*op_kevent)(int, const struct kevent *, size_t, 2198 struct kevent *, size_t, const struct timespec *); 2199 const struct kevent *ev; 2200 size_t i; 2201 2202 /* 2203 * Check that we don't attempt to kevent rump kernel fd's. 2204 * That needs similar treatment to select/poll, but is slightly 2205 * trickier since we need to manage to different kq descriptors. 2206 * (TODO, in case you're wondering). 2207 */ 2208 for (i = 0; i < nchanges; i++) { 2209 ev = &changelist[i]; 2210 if (ev->filter == EVFILT_READ || ev->filter == EVFILT_WRITE || 2211 ev->filter == EVFILT_VNODE) { 2212 if (fd_isrump((int)ev->ident)) { 2213 errno = ENOTSUP; 2214 return -1; 2215 } 2216 } 2217 } 2218 2219 op_kevent = GETSYSCALL(host, KEVENT); 2220 return op_kevent(kq, changelist, nchanges, eventlist, nevents, timeout); 2221 } 2222 #endif /* HAVE_KQUEUE */ 2223 2224 /* 2225 * mmapping from a rump kernel is not supported, so disallow it. 2226 */ 2227 void * 2228 mmap(void *addr, size_t len, int prot, int flags, int fd, off_t offset) 2229 { 2230 2231 if (flags & MAP_FILE && fd_isrump(fd)) { 2232 errno = ENOSYS; 2233 return MAP_FAILED; 2234 } 2235 return host_mmap(addr, len, prot, flags, fd, offset); 2236 } 2237 2238 #ifdef __NetBSD__ 2239 /* 2240 * these go to one or the other on a per-process configuration 2241 */ 2242 int __sysctl(const int *, unsigned int, void *, size_t *, const void *, size_t); 2243 int 2244 __sysctl(const int *name, unsigned int namelen, void *old, size_t *oldlenp, 2245 const void *new, size_t newlen) 2246 { 2247 int (*op___sysctl)(const int *, unsigned int, void *, size_t *, 2248 const void *, size_t); 2249 2250 if (rumpsysctl) { 2251 op___sysctl = GETSYSCALL(rump, __SYSCTL); 2252 } else { 2253 op___sysctl = GETSYSCALL(host, __SYSCTL); 2254 /* we haven't inited yet */ 2255 if (__predict_false(op___sysctl == NULL)) { 2256 op___sysctl = rumphijack_dlsym(RTLD_NEXT, "__sysctl"); 2257 } 2258 } 2259 2260 return op___sysctl(name, namelen, old, oldlenp, new, newlen); 2261 } 2262 #endif 2263 2264 /* 2265 * Rest are std type calls. 2266 */ 2267 2268 ATCALL(int, utimensat, DUALCALL_UTIMENSAT, \ 2269 (int fd, const char *path, const struct timespec t[2], int f), \ 2270 (int, const char *, const struct timespec [2], int), 2271 (fd, path, t, f)) 2272 2273 FDCALL(int, bind, DUALCALL_BIND, \ 2274 (int fd, const struct sockaddr *name, socklen_t namelen), \ 2275 (int, const struct sockaddr *, socklen_t), \ 2276 (fd, name, namelen)) 2277 2278 FDCALL(int, connect, DUALCALL_CONNECT, \ 2279 (int fd, const struct sockaddr *name, socklen_t namelen), \ 2280 (int, const struct sockaddr *, socklen_t), \ 2281 (fd, name, namelen)) 2282 2283 FDCALL(int, getpeername, DUALCALL_GETPEERNAME, \ 2284 (int fd, struct sockaddr *name, socklen_t *namelen), \ 2285 (int, struct sockaddr *, socklen_t *), \ 2286 (fd, name, namelen)) 2287 2288 FDCALL(int, getsockname, DUALCALL_GETSOCKNAME, \ 2289 (int fd, struct sockaddr *name, socklen_t *namelen), \ 2290 (int, struct sockaddr *, socklen_t *), \ 2291 (fd, name, namelen)) 2292 2293 FDCALL(int, listen, DUALCALL_LISTEN, \ 2294 (int fd, int backlog), \ 2295 (int, int), \ 2296 (fd, backlog)) 2297 2298 FDCALL(ssize_t, recvfrom, DUALCALL_RECVFROM, \ 2299 (int fd, void *buf, size_t len, int flags, \ 2300 struct sockaddr *from, socklen_t *fromlen), \ 2301 (int, void *, size_t, int, struct sockaddr *, socklen_t *), \ 2302 (fd, buf, len, flags, from, fromlen)) 2303 2304 FDCALL(ssize_t, sendto, DUALCALL_SENDTO, \ 2305 (int fd, const void *buf, size_t len, int flags, \ 2306 const struct sockaddr *to, socklen_t tolen), \ 2307 (int, const void *, size_t, int, \ 2308 const struct sockaddr *, socklen_t), \ 2309 (fd, buf, len, flags, to, tolen)) 2310 2311 FDCALL(int, getsockopt, DUALCALL_GETSOCKOPT, \ 2312 (int fd, int level, int optn, void *optval, socklen_t *optlen), \ 2313 (int, int, int, void *, socklen_t *), \ 2314 (fd, level, optn, optval, optlen)) 2315 2316 FDCALL(int, setsockopt, DUALCALL_SETSOCKOPT, \ 2317 (int fd, int level, int optn, \ 2318 const void *optval, socklen_t optlen), \ 2319 (int, int, int, const void *, socklen_t), \ 2320 (fd, level, optn, optval, optlen)) 2321 2322 FDCALL(int, shutdown, DUALCALL_SHUTDOWN, \ 2323 (int fd, int how), \ 2324 (int, int), \ 2325 (fd, how)) 2326 2327 FDCALL(ssize_t, REALREAD, DUALCALL_READ, \ 2328 (int fd, void *buf, size_t buflen), \ 2329 (int, void *, size_t), \ 2330 (fd, buf, buflen)) 2331 2332 #ifdef __linux__ 2333 ssize_t __read_chk(int, void *, size_t) 2334 __attribute__((alias("read"))); 2335 #endif 2336 2337 FDCALL(ssize_t, readv, DUALCALL_READV, \ 2338 (int fd, const struct iovec *iov, int iovcnt), \ 2339 (int, const struct iovec *, int), \ 2340 (fd, iov, iovcnt)) 2341 2342 FDCALL(ssize_t, REALPREAD, DUALCALL_PREAD, \ 2343 (int fd, void *buf, size_t nbytes, off_t offset), \ 2344 (int, void *, size_t, off_t), \ 2345 (fd, buf, nbytes, offset)) 2346 2347 FDCALL(ssize_t, preadv, DUALCALL_PREADV, \ 2348 (int fd, const struct iovec *iov, int iovcnt, off_t offset), \ 2349 (int, const struct iovec *, int, off_t), \ 2350 (fd, iov, iovcnt, offset)) 2351 2352 FDCALL(ssize_t, writev, DUALCALL_WRITEV, \ 2353 (int fd, const struct iovec *iov, int iovcnt), \ 2354 (int, const struct iovec *, int), \ 2355 (fd, iov, iovcnt)) 2356 2357 FDCALL(ssize_t, REALPWRITE, DUALCALL_PWRITE, \ 2358 (int fd, const void *buf, size_t nbytes, off_t offset), \ 2359 (int, const void *, size_t, off_t), \ 2360 (fd, buf, nbytes, offset)) 2361 2362 FDCALL(ssize_t, pwritev, DUALCALL_PWRITEV, \ 2363 (int fd, const struct iovec *iov, int iovcnt, off_t offset), \ 2364 (int, const struct iovec *, int, off_t), \ 2365 (fd, iov, iovcnt, offset)) 2366 2367 #ifndef __linux__ 2368 FDCALL(int, REALFSTAT, DUALCALL_FSTAT, \ 2369 (int fd, struct stat *sb), \ 2370 (int, struct stat *), \ 2371 (fd, sb)) 2372 #endif 2373 2374 #ifdef __NetBSD__ 2375 FDCALL(int, fstatvfs1, DUALCALL_FSTATVFS1, \ 2376 (int fd, struct statvfs *buf, int flags), \ 2377 (int, struct statvfs *, int), \ 2378 (fd, buf, flags)) 2379 #endif 2380 2381 FDCALL(off_t, lseek, DUALCALL_LSEEK, \ 2382 (int fd, off_t offset, int whence), \ 2383 (int, off_t, int), \ 2384 (fd, offset, whence)) 2385 #ifdef LSEEK_ALIAS 2386 __strong_alias(LSEEK_ALIAS,lseek); 2387 #endif 2388 2389 #ifndef __linux__ 2390 FDCALL(int, REALGETDENTS, DUALCALL_GETDENTS, \ 2391 (int fd, char *buf, size_t nbytes), \ 2392 (int, char *, size_t), \ 2393 (fd, buf, nbytes)) 2394 #endif 2395 2396 FDCALL(int, fchown, DUALCALL_FCHOWN, \ 2397 (int fd, uid_t owner, gid_t group), \ 2398 (int, uid_t, gid_t), \ 2399 (fd, owner, group)) 2400 2401 FDCALL(int, fchmod, DUALCALL_FCHMOD, \ 2402 (int fd, mode_t mode), \ 2403 (int, mode_t), \ 2404 (fd, mode)) 2405 2406 FDCALL(int, ftruncate, DUALCALL_FTRUNCATE, \ 2407 (int fd, off_t length), \ 2408 (int, off_t), \ 2409 (fd, length)) 2410 2411 FDCALL(int, fsync, DUALCALL_FSYNC, \ 2412 (int fd), \ 2413 (int), \ 2414 (fd)) 2415 2416 #ifdef HAVE_FSYNC_RANGE 2417 FDCALL(int, fsync_range, DUALCALL_FSYNC_RANGE, \ 2418 (int fd, int how, off_t start, off_t length), \ 2419 (int, int, off_t, off_t), \ 2420 (fd, how, start, length)) 2421 #endif 2422 2423 FDCALL(int, futimes, DUALCALL_FUTIMES, \ 2424 (int fd, const struct timeval *tv), \ 2425 (int, const struct timeval *), \ 2426 (fd, tv)) 2427 2428 FDCALL(int, futimens, DUALCALL_FUTIMENS, \ 2429 (int fd, const struct timespec *ts), \ 2430 (int, const struct timespec *), \ 2431 (fd, ts)) 2432 2433 #ifdef HAVE_CHFLAGS 2434 FDCALL(int, fchflags, DUALCALL_FCHFLAGS, \ 2435 (int fd, u_long flags), \ 2436 (int, u_long), \ 2437 (fd, flags)) 2438 #endif 2439 2440 /* 2441 * path-based selectors 2442 */ 2443 2444 #ifndef __linux__ 2445 PATHCALL(int, REALSTAT, DUALCALL_STAT, \ 2446 (const char *path, struct stat *sb), \ 2447 (const char *, struct stat *), \ 2448 (path, sb)) 2449 2450 PATHCALL(int, REALLSTAT, DUALCALL_LSTAT, \ 2451 (const char *path, struct stat *sb), \ 2452 (const char *, struct stat *), \ 2453 (path, sb)) 2454 #endif 2455 2456 PATHCALL(int, chown, DUALCALL_CHOWN, \ 2457 (const char *path, uid_t owner, gid_t group), \ 2458 (const char *, uid_t, gid_t), \ 2459 (path, owner, group)) 2460 2461 PATHCALL(int, lchown, DUALCALL_LCHOWN, \ 2462 (const char *path, uid_t owner, gid_t group), \ 2463 (const char *, uid_t, gid_t), \ 2464 (path, owner, group)) 2465 2466 PATHCALL(int, chmod, DUALCALL_CHMOD, \ 2467 (const char *path, mode_t mode), \ 2468 (const char *, mode_t), \ 2469 (path, mode)) 2470 2471 PATHCALL(int, lchmod, DUALCALL_LCHMOD, \ 2472 (const char *path, mode_t mode), \ 2473 (const char *, mode_t), \ 2474 (path, mode)) 2475 2476 #ifdef __NetBSD__ 2477 PATHCALL(int, statvfs1, DUALCALL_STATVFS1, \ 2478 (const char *path, struct statvfs *buf, int flags), \ 2479 (const char *, struct statvfs *, int), \ 2480 (path, buf, flags)) 2481 #endif 2482 2483 PATHCALL(int, unlink, DUALCALL_UNLINK, \ 2484 (const char *path), \ 2485 (const char *), \ 2486 (path)) 2487 2488 PATHCALL(int, symlink, DUALCALL_SYMLINK, \ 2489 (const char *target, const char *path), \ 2490 (const char *, const char *), \ 2491 (target, path)) 2492 2493 /* 2494 * readlink() can be called from malloc which can be called 2495 * from dlsym() during init 2496 */ 2497 ssize_t 2498 readlink(const char *path, char *buf, size_t bufsiz) 2499 { 2500 int (*op_readlink)(const char *, char *, size_t); 2501 enum pathtype pt; 2502 2503 if ((pt = path_isrump(path)) != PATH_HOST) { 2504 op_readlink = GETSYSCALL(rump, READLINK); 2505 if (pt == PATH_RUMP) 2506 path = path_host2rump(path); 2507 } else { 2508 op_readlink = GETSYSCALL(host, READLINK); 2509 } 2510 2511 if (__predict_false(op_readlink == NULL)) { 2512 errno = ENOENT; 2513 return -1; 2514 } 2515 2516 return op_readlink(path, buf, bufsiz); 2517 } 2518 2519 PATHCALL(int, mkdir, DUALCALL_MKDIR, \ 2520 (const char *path, mode_t mode), \ 2521 (const char *, mode_t), \ 2522 (path, mode)) 2523 2524 PATHCALL(int, rmdir, DUALCALL_RMDIR, \ 2525 (const char *path), \ 2526 (const char *), \ 2527 (path)) 2528 2529 PATHCALL(int, utimes, DUALCALL_UTIMES, \ 2530 (const char *path, const struct timeval *tv), \ 2531 (const char *, const struct timeval *), \ 2532 (path, tv)) 2533 2534 PATHCALL(int, lutimes, DUALCALL_LUTIMES, \ 2535 (const char *path, const struct timeval *tv), \ 2536 (const char *, const struct timeval *), \ 2537 (path, tv)) 2538 2539 #ifdef HAVE_CHFLAGS 2540 PATHCALL(int, chflags, DUALCALL_CHFLAGS, \ 2541 (const char *path, u_long flags), \ 2542 (const char *, u_long), \ 2543 (path, flags)) 2544 2545 PATHCALL(int, lchflags, DUALCALL_LCHFLAGS, \ 2546 (const char *path, u_long flags), \ 2547 (const char *, u_long), \ 2548 (path, flags)) 2549 #endif /* HAVE_CHFLAGS */ 2550 2551 PATHCALL(int, truncate, DUALCALL_TRUNCATE, \ 2552 (const char *path, off_t length), \ 2553 (const char *, off_t), \ 2554 (path, length)) 2555 2556 PATHCALL(int, access, DUALCALL_ACCESS, \ 2557 (const char *path, int mode), \ 2558 (const char *, int), \ 2559 (path, mode)) 2560 2561 #ifndef __linux__ 2562 PATHCALL(int, REALMKNOD, DUALCALL_MKNOD, \ 2563 (const char *path, mode_t mode, dev_t dev), \ 2564 (const char *, mode_t, dev_t), \ 2565 (path, mode, dev)) 2566 #endif 2567 2568 /* 2569 * Note: with mount the decisive parameter is the mount 2570 * destination directory. This is because we don't really know 2571 * about the "source" directory in a generic call (and besides, 2572 * it might not even exist, cf. nfs). 2573 */ 2574 #ifdef __NetBSD__ 2575 PATHCALL(int, REALMOUNT, DUALCALL_MOUNT, \ 2576 (const char *type, const char *path, int flags, \ 2577 void *data, size_t dlen), \ 2578 (const char *, const char *, int, void *, size_t), \ 2579 (type, path, flags, data, dlen)) 2580 2581 PATHCALL(int, unmount, DUALCALL_UNMOUNT, \ 2582 (const char *path, int flags), \ 2583 (const char *, int), \ 2584 (path, flags)) 2585 #endif /* __NetBSD__ */ 2586 2587 #ifdef HAVE___QUOTACTL 2588 PATHCALL(int, __quotactl, DUALCALL_QUOTACTL, \ 2589 (const char *path, struct quotactl_args *args), \ 2590 (const char *, struct quotactl_args *), \ 2591 (path, args)) 2592 #endif /* HAVE___QUOTACTL */ 2593 2594 #ifdef __NetBSD__ 2595 PATHCALL(int, REALGETFH, DUALCALL_GETFH, \ 2596 (const char *path, void *fhp, size_t *fh_size), \ 2597 (const char *, void *, size_t *), \ 2598 (path, fhp, fh_size)) 2599 #endif 2600 2601 /* 2602 * These act different on a per-process vfs configuration 2603 */ 2604 2605 #ifdef __NetBSD__ 2606 VFSCALL(VFSBIT_GETVFSSTAT, int, getvfsstat, DUALCALL_GETVFSSTAT, \ 2607 (struct statvfs *buf, size_t buflen, int flags), \ 2608 (struct statvfs *, size_t, int), \ 2609 (buf, buflen, flags)) 2610 #endif 2611 2612 #ifdef __NetBSD__ 2613 VFSCALL(VFSBIT_FHCALLS, int, REALFHOPEN, DUALCALL_FHOPEN, \ 2614 (const void *fhp, size_t fh_size, int flags), \ 2615 (const char *, size_t, int), \ 2616 (fhp, fh_size, flags)) 2617 2618 VFSCALL(VFSBIT_FHCALLS, int, REALFHSTAT, DUALCALL_FHSTAT, \ 2619 (const void *fhp, size_t fh_size, struct stat *sb), \ 2620 (const char *, size_t, struct stat *), \ 2621 (fhp, fh_size, sb)) 2622 2623 VFSCALL(VFSBIT_FHCALLS, int, REALFHSTATVFS1, DUALCALL_FHSTATVFS1, \ 2624 (const void *fhp, size_t fh_size, struct statvfs *sb, int flgs),\ 2625 (const char *, size_t, struct statvfs *, int), \ 2626 (fhp, fh_size, sb, flgs)) 2627 #endif 2628 2629 2630 #ifdef __NetBSD__ 2631 2632 /* finally, put nfssvc here. "keep the namespace clean" */ 2633 #include <nfs/rpcv2.h> 2634 #include <nfs/nfs.h> 2635 2636 int 2637 nfssvc(int flags, void *argstructp) 2638 { 2639 int (*op_nfssvc)(int, void *); 2640 2641 if (vfsbits & VFSBIT_NFSSVC){ 2642 struct nfsd_args *nfsdargs; 2643 2644 /* massage the socket descriptor if necessary */ 2645 if (flags == NFSSVC_ADDSOCK) { 2646 nfsdargs = argstructp; 2647 nfsdargs->sock = fd_host2rump(nfsdargs->sock); 2648 } 2649 op_nfssvc = GETSYSCALL(rump, NFSSVC); 2650 } else 2651 op_nfssvc = GETSYSCALL(host, NFSSVC); 2652 2653 return op_nfssvc(flags, argstructp); 2654 } 2655 #endif /* __NetBSD__ */ 2656