xref: /netbsd-src/lib/librumphijack/hijack.c (revision daf6c4152fcddc27c445489775ed1f66ab4ea9a9)
1 /*      $NetBSD: hijack.c,v 1.48 2011/02/17 17:18:08 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 #include <sys/cdefs.h>
29 __RCSID("$NetBSD: hijack.c,v 1.48 2011/02/17 17:18:08 pooka Exp $");
30 
31 #define __ssp_weak_name(fun) _hijack_ ## fun
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/event.h>
36 #include <sys/ioctl.h>
37 #include <sys/mount.h>
38 #include <sys/poll.h>
39 #include <sys/socket.h>
40 #include <sys/statvfs.h>
41 
42 #include <rump/rumpclient.h>
43 #include <rump/rump_syscalls.h>
44 
45 #include <assert.h>
46 #include <dlfcn.h>
47 #include <err.h>
48 #include <errno.h>
49 #include <fcntl.h>
50 #include <poll.h>
51 #include <pthread.h>
52 #include <signal.h>
53 #include <stdarg.h>
54 #include <stdbool.h>
55 #include <stdio.h>
56 #include <stdlib.h>
57 #include <string.h>
58 #include <time.h>
59 #include <unistd.h>
60 
61 enum dualcall {
62 	DUALCALL_WRITE, DUALCALL_WRITEV,
63 	DUALCALL_IOCTL, DUALCALL_FCNTL,
64 	DUALCALL_SOCKET, DUALCALL_ACCEPT, DUALCALL_BIND, DUALCALL_CONNECT,
65 	DUALCALL_GETPEERNAME, DUALCALL_GETSOCKNAME, DUALCALL_LISTEN,
66 	DUALCALL_RECVFROM, DUALCALL_RECVMSG,
67 	DUALCALL_SENDTO, DUALCALL_SENDMSG,
68 	DUALCALL_GETSOCKOPT, DUALCALL_SETSOCKOPT,
69 	DUALCALL_SHUTDOWN,
70 	DUALCALL_READ, DUALCALL_READV,
71 	DUALCALL_DUP2,
72 	DUALCALL_CLOSE,
73 	DUALCALL_POLLTS,
74 	DUALCALL_KEVENT,
75 	DUALCALL_STAT, DUALCALL_LSTAT, DUALCALL_FSTAT,
76 	DUALCALL_CHMOD, DUALCALL_LCHMOD, DUALCALL_FCHMOD,
77 	DUALCALL_CHOWN, DUALCALL_LCHOWN, DUALCALL_FCHOWN,
78 	DUALCALL_OPEN,
79 	DUALCALL_STATVFS1, DUALCALL_FSTATVFS1,
80 	DUALCALL_CHDIR, DUALCALL_FCHDIR,
81 	DUALCALL_LSEEK,
82 	DUALCALL_GETDENTS,
83 	DUALCALL_UNLINK, DUALCALL_SYMLINK, DUALCALL_READLINK,
84 	DUALCALL_RENAME,
85 	DUALCALL_MKDIR, DUALCALL_RMDIR,
86 	DUALCALL_UTIMES, DUALCALL_LUTIMES, DUALCALL_FUTIMES,
87 	DUALCALL_TRUNCATE, DUALCALL_FTRUNCATE,
88 	DUALCALL_FSYNC, DUALCALL_FSYNC_RANGE,
89 	DUALCALL_MOUNT, DUALCALL_UNMOUNT,
90 	DUALCALL__NUM
91 };
92 
93 #define RSYS_STRING(a) __STRING(a)
94 #define RSYS_NAME(a) RSYS_STRING(__CONCAT(RUMP_SYS_RENAME_,a))
95 
96 /*
97  * Would be nice to get this automatically in sync with libc.
98  * Also, this does not work for compat-using binaries!
99  */
100 #if !__NetBSD_Prereq__(5,99,7)
101 #define REALSELECT select
102 #define REALPOLLTS pollts
103 #define REALKEVENT kevent
104 #define REALSTAT __stat30
105 #define REALLSTAT __lstat30
106 #define REALFSTAT __fstat30
107 #define REALUTIMES utimes
108 #define REALLUTIMES lutimes
109 #define REALFUTIMES futimes
110 #else
111 #define REALSELECT _sys___select50
112 #define REALPOLLTS _sys___pollts50
113 #define REALKEVENT _sys___kevent50
114 #define REALSTAT __stat50
115 #define REALLSTAT __lstat50
116 #define REALFSTAT __fstat50
117 #define REALUTIMES __utimes50
118 #define REALLUTIMES __lutimes50
119 #define REALFUTIMES __futimes50
120 #endif
121 #define REALREAD _sys_read
122 #define REALGETDENTS __getdents30
123 #define REALMOUNT __mount50
124 
125 int REALSELECT(int, fd_set *, fd_set *, fd_set *, struct timeval *);
126 int REALPOLLTS(struct pollfd *, nfds_t,
127 	       const struct timespec *, const sigset_t *);
128 int REALKEVENT(int, const struct kevent *, size_t, struct kevent *, size_t,
129 	       const struct timespec *);
130 ssize_t REALREAD(int, void *, size_t);
131 int REALSTAT(const char *, struct stat *);
132 int REALLSTAT(const char *, struct stat *);
133 int REALFSTAT(int, struct stat *);
134 int REALGETDENTS(int, char *, size_t);
135 int REALUTIMES(const char *, const struct timeval [2]);
136 int REALLUTIMES(const char *, const struct timeval [2]);
137 int REALFUTIMES(int, const struct timeval [2]);
138 int REALMOUNT(const char *, const char *, int, void *, size_t);
139 
140 #define S(a) __STRING(a)
141 struct sysnames {
142 	enum dualcall scm_callnum;
143 	const char *scm_hostname;
144 	const char *scm_rumpname;
145 } syscnames[] = {
146 	{ DUALCALL_SOCKET,	"__socket30",	RSYS_NAME(SOCKET)	},
147 	{ DUALCALL_ACCEPT,	"accept",	RSYS_NAME(ACCEPT)	},
148 	{ DUALCALL_BIND,	"bind",		RSYS_NAME(BIND)		},
149 	{ DUALCALL_CONNECT,	"connect",	RSYS_NAME(CONNECT)	},
150 	{ DUALCALL_GETPEERNAME,	"getpeername",	RSYS_NAME(GETPEERNAME)	},
151 	{ DUALCALL_GETSOCKNAME,	"getsockname",	RSYS_NAME(GETSOCKNAME)	},
152 	{ DUALCALL_LISTEN,	"listen",	RSYS_NAME(LISTEN)	},
153 	{ DUALCALL_RECVFROM,	"recvfrom",	RSYS_NAME(RECVFROM)	},
154 	{ DUALCALL_RECVMSG,	"recvmsg",	RSYS_NAME(RECVMSG)	},
155 	{ DUALCALL_SENDTO,	"sendto",	RSYS_NAME(SENDTO)	},
156 	{ DUALCALL_SENDMSG,	"sendmsg",	RSYS_NAME(SENDMSG)	},
157 	{ DUALCALL_GETSOCKOPT,	"getsockopt",	RSYS_NAME(GETSOCKOPT)	},
158 	{ DUALCALL_SETSOCKOPT,	"setsockopt",	RSYS_NAME(SETSOCKOPT)	},
159 	{ DUALCALL_SHUTDOWN,	"shutdown",	RSYS_NAME(SHUTDOWN)	},
160 	{ DUALCALL_READ,	S(REALREAD),	RSYS_NAME(READ)		},
161 	{ DUALCALL_READV,	"readv",	RSYS_NAME(READV)	},
162 	{ DUALCALL_WRITE,	"write",	RSYS_NAME(WRITE)	},
163 	{ DUALCALL_WRITEV,	"writev",	RSYS_NAME(WRITEV)	},
164 	{ DUALCALL_IOCTL,	"ioctl",	RSYS_NAME(IOCTL)	},
165 	{ DUALCALL_FCNTL,	"fcntl",	RSYS_NAME(FCNTL)	},
166 	{ DUALCALL_DUP2,	"dup2",		RSYS_NAME(DUP2)		},
167 	{ DUALCALL_CLOSE,	"close",	RSYS_NAME(CLOSE)	},
168 	{ DUALCALL_POLLTS,	S(REALPOLLTS),	RSYS_NAME(POLLTS)	},
169 	{ DUALCALL_KEVENT,	S(REALKEVENT),	RSYS_NAME(KEVENT)	},
170 	{ DUALCALL_STAT,	S(REALSTAT),	RSYS_NAME(STAT)		},
171 	{ DUALCALL_LSTAT,	S(REALLSTAT),	RSYS_NAME(LSTAT)	},
172 	{ DUALCALL_FSTAT,	S(REALFSTAT),	RSYS_NAME(FSTAT)	},
173 	{ DUALCALL_CHOWN,	"chown",	RSYS_NAME(CHOWN)	},
174 	{ DUALCALL_LCHOWN,	"lchown",	RSYS_NAME(LCHOWN)	},
175 	{ DUALCALL_FCHOWN,	"fchown",	RSYS_NAME(FCHOWN)	},
176 	{ DUALCALL_CHMOD,	"chmod",	RSYS_NAME(CHMOD)	},
177 	{ DUALCALL_LCHMOD,	"lchmod",	RSYS_NAME(LCHMOD)	},
178 	{ DUALCALL_FCHMOD,	"fchmod",	RSYS_NAME(FCHMOD)	},
179 	{ DUALCALL_UTIMES,	S(REALUTIMES),	RSYS_NAME(UTIMES)	},
180 	{ DUALCALL_LUTIMES,	S(REALLUTIMES),	RSYS_NAME(LUTIMES)	},
181 	{ DUALCALL_FUTIMES,	S(REALFUTIMES),	RSYS_NAME(FUTIMES)	},
182 	{ DUALCALL_OPEN,	"open",		RSYS_NAME(OPEN)		},
183 	{ DUALCALL_STATVFS1,	"statvfs1",	RSYS_NAME(STATVFS1)	},
184 	{ DUALCALL_FSTATVFS1,	"fstatvfs1",	RSYS_NAME(FSTATVFS1)	},
185 	{ DUALCALL_CHDIR,	"chdir",	RSYS_NAME(CHDIR)	},
186 	{ DUALCALL_FCHDIR,	"fchdir",	RSYS_NAME(FCHDIR)	},
187 	{ DUALCALL_LSEEK,	"lseek",	RSYS_NAME(LSEEK)	},
188 	{ DUALCALL_GETDENTS,	"__getdents30",	RSYS_NAME(GETDENTS)	},
189 	{ DUALCALL_UNLINK,	"unlink",	RSYS_NAME(UNLINK)	},
190 	{ DUALCALL_SYMLINK,	"symlink",	RSYS_NAME(SYMLINK)	},
191 	{ DUALCALL_READLINK,	"readlink",	RSYS_NAME(READLINK)	},
192 	{ DUALCALL_RENAME,	"rename",	RSYS_NAME(RENAME)	},
193 	{ DUALCALL_MKDIR,	"mkdir",	RSYS_NAME(MKDIR)	},
194 	{ DUALCALL_RMDIR,	"rmdir",	RSYS_NAME(RMDIR)	},
195 	{ DUALCALL_TRUNCATE,	"truncate",	RSYS_NAME(TRUNCATE)	},
196 	{ DUALCALL_FTRUNCATE,	"ftruncate",	RSYS_NAME(FTRUNCATE)	},
197 	{ DUALCALL_FSYNC,	"fsync",	RSYS_NAME(FSYNC)	},
198 	{ DUALCALL_FSYNC_RANGE,	"fsync_range",	RSYS_NAME(FSYNC_RANGE)	},
199 	{ DUALCALL_MOUNT,	S(REALMOUNT),	RSYS_NAME(MOUNT)	},
200 	{ DUALCALL_UNMOUNT,	"unmount",	RSYS_NAME(UNMOUNT)	},
201 };
202 #undef S
203 
204 struct bothsys {
205 	void *bs_host;
206 	void *bs_rump;
207 } syscalls[DUALCALL__NUM];
208 #define GETSYSCALL(which, name) syscalls[DUALCALL_##name].bs_##which
209 
210 pid_t	(*host_fork)(void);
211 int	(*host_daemon)(int, int);
212 int	(*host_execve)(const char *, char *const[], char *const[]);
213 
214 /* ok, we need *two* bits per dup2'd fd to track fd+HIJACKOFF aliases */
215 static uint32_t dup2mask;
216 #define ISDUP2D(fd) (((fd) < 16) && (1<<(fd) & dup2mask))
217 #define SETDUP2(fd) \
218     do { if ((fd) < 16) dup2mask |= (1<<(fd)); } while (/*CONSTCOND*/0)
219 #define CLRDUP2(fd) \
220     do { if ((fd) < 16) dup2mask &= ~(1<<(fd)); } while (/*CONSTCOND*/0)
221 #define ISDUP2ALIAS(fd) (((fd) < 16) && (1<<((fd)+16) & dup2mask))
222 #define SETDUP2ALIAS(fd) \
223     do { if ((fd) < 16) dup2mask |= (1<<((fd)+16)); } while (/*CONSTCOND*/0)
224 #define CLRDUP2ALIAS(fd) \
225     do { if ((fd) < 16) dup2mask &= ~(1<<((fd)+16)); } while (/*CONSTCOND*/0)
226 
227 //#define DEBUGJACK
228 #ifdef DEBUGJACK
229 #define DPRINTF(x) mydprintf x
230 static void
231 mydprintf(const char *fmt, ...)
232 {
233 	va_list ap;
234 
235 	if (ISDUP2D(STDERR_FILENO))
236 		return;
237 
238 	va_start(ap, fmt);
239 	vfprintf(stderr, fmt, ap);
240 	va_end(ap);
241 }
242 
243 #else
244 #define DPRINTF(x)
245 #endif
246 
247 #define FDCALL(type, name, rcname, args, proto, vars)			\
248 type name args								\
249 {									\
250 	type (*fun) proto;						\
251 									\
252 	DPRINTF(("%s -> %d\n", __STRING(name), fd));			\
253 	if (fd_isrump(fd)) {						\
254 		fun = syscalls[rcname].bs_rump;				\
255 		fd = fd_host2rump(fd);					\
256 	} else {							\
257 		fun = syscalls[rcname].bs_host;				\
258 	}								\
259 									\
260 	return fun vars;						\
261 }
262 
263 #define PATHCALL(type, name, rcname, args, proto, vars)			\
264 type name args								\
265 {									\
266 	type (*fun) proto;						\
267 									\
268 	DPRINTF(("%s -> %s\n", __STRING(name), path));			\
269 	if (path_isrump(path)) {					\
270 		fun = syscalls[rcname].bs_rump;				\
271 		path = path_host2rump(path);				\
272 	} else {							\
273 		fun = syscalls[rcname].bs_host;				\
274 	}								\
275 									\
276 	return fun vars;						\
277 }
278 
279 /*
280  * This is called from librumpclient in case of LD_PRELOAD.
281  * It ensures correct RTLD_NEXT.
282  *
283  * ... except, it's apparently extremely difficult to force
284  * at least gcc to generate an actual stack frame here.  So
285  * sprinkle some volatile foobar and baz to throw the optimizer
286  * off the scent and generate a variable assignment with the
287  * return value.  The posterboy for this meltdown is amd64
288  * with -O2.  At least with gcc 4.1.3 i386 works regardless of
289  * optimization.
290  */
291 volatile int rumphijack_unrope; /* there, unhang yourself */
292 static void *
293 hijackdlsym(void *handle, const char *symbol)
294 {
295 	void *rv;
296 
297 	rv = dlsym(handle, symbol);
298 	rumphijack_unrope = *(volatile int *)rv;
299 
300 	return (void *)rv;
301 }
302 
303 static int pwdinrump = 0;
304 
305 /* low calorie sockets? */
306 static bool hostlocalsockets = true;
307 
308 static void __attribute__((constructor))
309 rcinit(void)
310 {
311 	char buf[64];
312 	extern void *(*rumpclient_dlsym)(void *, const char *);
313 	unsigned i, j;
314 
315 	rumpclient_dlsym = hijackdlsym;
316 	host_fork = dlsym(RTLD_NEXT, "fork");
317 	host_daemon = dlsym(RTLD_NEXT, "daemon");
318 	host_execve = dlsym(RTLD_NEXT, "execve");
319 
320 	/*
321 	 * In theory cannot print anything during lookups because
322 	 * we might not have the call vector set up.  so, the errx()
323 	 * is a bit of a strech, but it might work.
324 	 */
325 
326 	for (i = 0; i < DUALCALL__NUM; i++) {
327 		/* build runtime O(1) access */
328 		for (j = 0; j < __arraycount(syscnames); j++) {
329 			if (syscnames[j].scm_callnum == i)
330 				break;
331 		}
332 
333 		if (j == __arraycount(syscnames))
334 			errx(1, "rumphijack error: syscall pos %d missing", i);
335 
336 		syscalls[i].bs_host = dlsym(RTLD_NEXT,
337 		    syscnames[j].scm_hostname);
338 		if (syscalls[i].bs_host == NULL)
339 			errx(1, "hostcall %s not found missing",
340 			    syscnames[j].scm_hostname);
341 
342 		syscalls[i].bs_rump = dlsym(RTLD_NEXT,
343 		    syscnames[j].scm_rumpname);
344 		if (syscalls[i].bs_rump == NULL)
345 			errx(1, "rumpcall %s not found missing",
346 			    syscnames[j].scm_rumpname);
347 	}
348 
349 	if (rumpclient_init() == -1)
350 		err(1, "rumpclient init");
351 
352 	/* set client persistence level */
353 	if (getenv_r("RUMPHIJACK_RETRYCONNECT", buf, sizeof(buf)) != -1) {
354 		if (strcmp(buf, "die") == 0)
355 			rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_DIE);
356 		else if (strcmp(buf, "inftime") == 0)
357 			rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_INFTIME);
358 		else if (strcmp(buf, "once") == 0)
359 			rumpclient_setconnretry(RUMPCLIENT_RETRYCONN_ONCE);
360 		else {
361 			time_t timeout;
362 			char *ep;
363 
364 			timeout = (time_t)strtoll(buf, &ep, 10);
365 			if (timeout <= 0 || ep != buf + strlen(buf))
366 				errx(1, "RUMPHIJACK_RETRYCONNECT must be "
367 				    "keyword or integer, got: %s", buf);
368 
369 			rumpclient_setconnretry(timeout);
370 		}
371 	}
372 
373 	if (getenv_r("RUMPHIJACK__DUP2MASK", buf, sizeof(buf)) == 0) {
374 		dup2mask = strtoul(buf, NULL, 10);
375 		unsetenv("RUMPHIJACK__DUP2MASK");
376 	}
377 	if (getenv_r("RUMPHIJACK__PWDINRUMP", buf, sizeof(buf)) == 0) {
378 		pwdinrump = strtoul(buf, NULL, 10);
379 		unsetenv("RUMPHIJACK__PWDINRUMP");
380 	}
381 }
382 
383 /* XXX: need runtime selection.  low for now due to FD_SETSIZE */
384 #define HIJACK_FDOFF 128
385 static int
386 fd_rump2host(int fd)
387 {
388 
389 	if (fd == -1)
390 		return fd;
391 
392 	if (!ISDUP2D(fd))
393 		fd += HIJACK_FDOFF;
394 
395 	return fd;
396 }
397 
398 static int
399 fd_host2rump(int fd)
400 {
401 
402 	if (!ISDUP2D(fd))
403 		fd -= HIJACK_FDOFF;
404 	return fd;
405 }
406 
407 static bool
408 fd_isrump(int fd)
409 {
410 
411 	return ISDUP2D(fd) || fd >= HIJACK_FDOFF;
412 }
413 
414 #define assertfd(_fd_) assert(ISDUP2D(_fd_) || (_fd_) >= HIJACK_FDOFF)
415 
416 #define RUMPPREFIX "/rump"
417 static int
418 path_isrump(const char *path)
419 {
420 
421 	if (*path == '/') {
422 		if (strncmp(path, RUMPPREFIX, sizeof(RUMPPREFIX)-1) == 0)
423 			return 1;
424 		return 0;
425 	} else {
426 		return pwdinrump;
427 	}
428 }
429 
430 static const char *rootpath = "/";
431 static const char *
432 path_host2rump(const char *path)
433 {
434 	const char *rv;
435 
436 	if (*path == '/') {
437 		rv = path + (sizeof(RUMPPREFIX)-1);
438 		if (*rv == '\0')
439 			rv = rootpath;
440 	} else {
441 		rv = path;
442 	}
443 
444 	return rv;
445 }
446 
447 static int
448 dodup(int oldd, int minfd)
449 {
450 	int (*op_fcntl)(int, int, ...);
451 	int newd;
452 	int isrump;
453 
454 	DPRINTF(("dup -> %d (minfd %d)\n", oldd, minfd));
455 	if (fd_isrump(oldd)) {
456 		op_fcntl = GETSYSCALL(rump, FCNTL);
457 		oldd = fd_host2rump(oldd);
458 		isrump = 1;
459 	} else {
460 		op_fcntl = GETSYSCALL(host, FCNTL);
461 		isrump = 0;
462 	}
463 
464 	newd = op_fcntl(oldd, F_DUPFD, minfd);
465 
466 	if (isrump)
467 		newd = fd_rump2host(newd);
468 	DPRINTF(("dup <- %d\n", newd));
469 
470 	return newd;
471 }
472 
473 /*
474  * dup a host file descriptor so that it doesn't collide with the dup2mask
475  */
476 static int
477 fd_dupgood(int fd)
478 {
479 	int (*op_fcntl)(int, int, ...) = GETSYSCALL(host, FCNTL);
480 	int (*op_close)(int) = GETSYSCALL(host, CLOSE);
481 	int ofd, i;
482 
483 	for (i = 1; ISDUP2D(fd); i++) {
484 		ofd = fd;
485 		fd = op_fcntl(ofd, F_DUPFD, i);
486 		op_close(ofd);
487 	}
488 
489 	return fd;
490 }
491 
492 int
493 open(const char *path, int flags, ...)
494 {
495 	int (*op_open)(const char *, int, ...);
496 	bool isrump;
497 	va_list ap;
498 	int fd;
499 
500 	if (path_isrump(path)) {
501 		path = path_host2rump(path);
502 		op_open = GETSYSCALL(rump, OPEN);
503 		isrump = true;
504 	} else {
505 		op_open = GETSYSCALL(host, OPEN);
506 		isrump = false;
507 	}
508 
509 	va_start(ap, flags);
510 	fd = op_open(path, flags, va_arg(ap, mode_t));
511 	va_end(ap);
512 
513 	if (isrump)
514 		fd = fd_rump2host(fd);
515 	else
516 		fd = fd_dupgood(fd);
517 	return fd;
518 }
519 
520 int
521 chdir(const char *path)
522 {
523 	int (*op_chdir)(const char *);
524 	bool isrump;
525 	int rv;
526 
527 	if (path_isrump(path)) {
528 		op_chdir = GETSYSCALL(rump, CHDIR);
529 		isrump = true;
530 		path = path_host2rump(path);
531 	} else {
532 		op_chdir = GETSYSCALL(host, CHDIR);
533 		isrump = false;
534 	}
535 
536 	rv = op_chdir(path);
537 	if (rv == 0) {
538 		if (isrump)
539 			pwdinrump = true;
540 		else
541 			pwdinrump = false;
542 	}
543 
544 	return rv;
545 }
546 
547 int
548 fchdir(int fd)
549 {
550 	int (*op_fchdir)(int);
551 	bool isrump;
552 	int rv;
553 
554 	if (fd_isrump(fd)) {
555 		op_fchdir = GETSYSCALL(rump, FCHDIR);
556 		isrump = true;
557 		fd = fd_host2rump(fd);
558 	} else {
559 		op_fchdir = GETSYSCALL(host, FCHDIR);
560 		isrump = false;
561 	}
562 
563 	rv = op_fchdir(fd);
564 	if (rv == 0) {
565 		if (isrump)
566 			pwdinrump = true;
567 		else
568 			pwdinrump = false;
569 	}
570 
571 	return rv;
572 }
573 
574 int __socket30(int, int, int);
575 int
576 __socket30(int domain, int type, int protocol)
577 {
578 	int (*op_socket)(int, int, int);
579 	int fd;
580 	bool dohost;
581 
582 	dohost = hostlocalsockets && (domain == AF_LOCAL);
583 
584 	if (dohost)
585 		op_socket = GETSYSCALL(host, SOCKET);
586 	else
587 		op_socket = GETSYSCALL(rump, SOCKET);
588 	fd = op_socket(domain, type, protocol);
589 
590 	if (!dohost)
591 		fd = fd_rump2host(fd);
592 	else
593 		fd = fd_dupgood(fd);
594 	DPRINTF(("socket <- %d\n", fd));
595 
596 	return fd;
597 }
598 
599 int
600 accept(int s, struct sockaddr *addr, socklen_t *addrlen)
601 {
602 	int (*op_accept)(int, struct sockaddr *, socklen_t *);
603 	int fd;
604 	bool isrump;
605 
606 	isrump = fd_isrump(s);
607 
608 	DPRINTF(("accept -> %d", s));
609 	if (isrump) {
610 		op_accept = GETSYSCALL(rump, ACCEPT);
611 		s = fd_host2rump(s);
612 	} else {
613 		op_accept = GETSYSCALL(host, ACCEPT);
614 	}
615 	fd = op_accept(s, addr, addrlen);
616 	if (fd != -1 && isrump)
617 		fd = fd_rump2host(fd);
618 	else
619 		fd = fd_dupgood(fd);
620 
621 	DPRINTF((" <- %d\n", fd));
622 
623 	return fd;
624 }
625 
626 /*
627  * ioctl and fcntl are varargs calls and need special treatment
628  */
629 int
630 ioctl(int fd, unsigned long cmd, ...)
631 {
632 	int (*op_ioctl)(int, unsigned long cmd, ...);
633 	va_list ap;
634 	int rv;
635 
636 	DPRINTF(("ioctl -> %d\n", fd));
637 	if (fd_isrump(fd)) {
638 		fd = fd_host2rump(fd);
639 		op_ioctl = GETSYSCALL(rump, IOCTL);
640 	} else {
641 		op_ioctl = GETSYSCALL(host, IOCTL);
642 	}
643 
644 	va_start(ap, cmd);
645 	rv = op_ioctl(fd, cmd, va_arg(ap, void *));
646 	va_end(ap);
647 	return rv;
648 }
649 
650 #include <syslog.h>
651 int
652 fcntl(int fd, int cmd, ...)
653 {
654 	int (*op_fcntl)(int, int, ...);
655 	va_list ap;
656 	int rv, minfd, i;
657 
658 	DPRINTF(("fcntl -> %d (cmd %d)\n", fd, cmd));
659 
660 	switch (cmd) {
661 	case F_DUPFD:
662 		va_start(ap, cmd);
663 		minfd = va_arg(ap, int);
664 		va_end(ap);
665 		return dodup(fd, minfd);
666 
667 	case F_CLOSEM:
668 		/*
669 		 * So, if fd < HIJACKOFF, we want to do a host closem.
670 		 */
671 
672 		if (fd < HIJACK_FDOFF) {
673 			int closemfd = fd;
674 
675 			if (rumpclient__closenotify(&closemfd,
676 			    RUMPCLIENT_CLOSE_FCLOSEM) == -1)
677 				return -1;
678 			op_fcntl = GETSYSCALL(host, FCNTL);
679 			rv = op_fcntl(closemfd, cmd);
680 			if (rv)
681 				return rv;
682 		}
683 
684 		/*
685 		 * Additionally, we want to do a rump closem, but only
686 		 * for the file descriptors not within the dup2mask.
687 		 */
688 
689 		/* why don't we offer fls()? */
690 		for (i = 15; i >= 0; i--) {
691 			if (ISDUP2D(i))
692 				break;
693 		}
694 
695 		if (fd >= HIJACK_FDOFF)
696 			fd -= HIJACK_FDOFF;
697 		else
698 			fd = 0;
699 		fd = MAX(i+1, fd);
700 
701 		/* hmm, maybe we should close rump fd's not within dup2mask? */
702 
703 		return rump_sys_fcntl(fd, F_CLOSEM);
704 
705 	case F_MAXFD:
706 		/*
707 		 * For maxfd, if there's a rump kernel fd, return
708 		 * it hostified.  Otherwise, return host's MAXFD
709 		 * return value.
710 		 */
711 		if ((rv = rump_sys_fcntl(fd, F_MAXFD)) != -1) {
712 			/*
713 			 * This might go a little wrong in case
714 			 * of dup2 to [012], but I'm not sure if
715 			 * there's a justification for tracking
716 			 * that info.  Consider e.g.
717 			 * dup2(rumpfd, 2) followed by rump_sys_open()
718 			 * returning 1.  We should return 1+HIJACKOFF,
719 			 * not 2+HIJACKOFF.  However, if [01] is not
720 			 * open, the correct return value is 2.
721 			 */
722 			return fd_rump2host(fd);
723 		} else {
724 			op_fcntl = GETSYSCALL(host, FCNTL);
725 			return op_fcntl(fd, F_MAXFD);
726 		}
727 		/*NOTREACHED*/
728 
729 	default:
730 		if (fd_isrump(fd)) {
731 			fd = fd_host2rump(fd);
732 			op_fcntl = GETSYSCALL(rump, FCNTL);
733 		} else {
734 			op_fcntl = GETSYSCALL(host, FCNTL);
735 		}
736 
737 		va_start(ap, cmd);
738 		rv = op_fcntl(fd, cmd, va_arg(ap, void *));
739 		va_end(ap);
740 		return rv;
741 	}
742 	/*NOTREACHED*/
743 }
744 
745 int
746 close(int fd)
747 {
748 	int (*op_close)(int);
749 	int rv;
750 
751 	DPRINTF(("close -> %d\n", fd));
752 	if (fd_isrump(fd)) {
753 		int undup2 = 0;
754 
755 		fd = fd_host2rump(fd);
756 		if (ISDUP2ALIAS(fd)) {
757 			_DIAGASSERT(ISDUP2D(fd));
758 			CLRDUP2ALIAS(fd);
759 			return 0;
760 		}
761 
762 		if (ISDUP2D(fd))
763 			undup2 = 1;
764 		op_close = GETSYSCALL(rump, CLOSE);
765 		rv = op_close(fd);
766 		if (rv == 0 && undup2)
767 			CLRDUP2(fd);
768 	} else {
769 		if (rumpclient__closenotify(&fd, RUMPCLIENT_CLOSE_CLOSE) == -1)
770 			return -1;
771 		op_close = GETSYSCALL(host, CLOSE);
772 		rv = op_close(fd);
773 	}
774 
775 	return rv;
776 }
777 
778 /*
779  * write cannot issue a standard debug printf due to recursion
780  */
781 ssize_t
782 write(int fd, const void *buf, size_t blen)
783 {
784 	ssize_t (*op_write)(int, const void *, size_t);
785 
786 	if (fd_isrump(fd)) {
787 		fd = fd_host2rump(fd);
788 		op_write = GETSYSCALL(rump, WRITE);
789 	} else {
790 		op_write = GETSYSCALL(host, WRITE);
791 	}
792 
793 	return op_write(fd, buf, blen);
794 }
795 
796 /*
797  * dup2 is special.  we allow dup2 of a rump kernel fd to 0-2 since
798  * many programs do that.  dup2 of a rump kernel fd to another value
799  * not >= fdoff is an error.
800  *
801  * Note: cannot rump2host newd, because it is often hardcoded.
802  */
803 int
804 dup2(int oldd, int newd)
805 {
806 	int (*host_dup2)(int, int);
807 	int rv;
808 
809 	DPRINTF(("dup2 -> %d (o) -> %d (n)\n", oldd, newd));
810 
811 	if (fd_isrump(oldd)) {
812 		if (!(newd >= 0 && newd <= 2))
813 			return EBADF;
814 		oldd = fd_host2rump(oldd);
815 		if (oldd == newd) {
816 			SETDUP2(newd);
817 			SETDUP2ALIAS(newd);
818 			return newd;
819 		}
820 		rv = rump_sys_dup2(oldd, newd);
821 		if (rv != -1)
822 			SETDUP2(newd);
823 	} else {
824 		host_dup2 = syscalls[DUALCALL_DUP2].bs_host;
825 		if (rumpclient__closenotify(&newd, RUMPCLIENT_CLOSE_DUP2) == -1)
826 			return -1;
827 		rv = host_dup2(oldd, newd);
828 	}
829 
830 	return rv;
831 }
832 
833 int
834 dup(int oldd)
835 {
836 
837 	return dodup(oldd, 0);
838 }
839 
840 pid_t
841 fork()
842 {
843 	pid_t rv;
844 
845 	DPRINTF(("fork\n"));
846 
847 	rv = rumpclient__dofork(host_fork);
848 
849 	DPRINTF(("fork returns %d\n", rv));
850 	return rv;
851 }
852 /* we do not have the luxury of not requiring a stackframe */
853 __strong_alias(__vfork14,fork);
854 
855 int
856 daemon(int nochdir, int noclose)
857 {
858 	struct rumpclient_fork *rf;
859 
860 	if ((rf = rumpclient_prefork()) == NULL)
861 		return -1;
862 
863 	if (host_daemon(nochdir, noclose) == -1)
864 		return -1;
865 
866 	if (rumpclient_fork_init(rf) == -1)
867 		return -1;
868 
869 	return 0;
870 }
871 
872 int
873 execve(const char *path, char *const argv[], char *const envp[])
874 {
875 	char buf[128];
876 	char *dup2str;
877 	char *pwdinrumpstr;
878 	char **newenv;
879 	size_t nelem;
880 	int rv, sverrno;
881 	int bonus = 1, i = 0;
882 
883 	if (dup2mask) {
884 		snprintf(buf, sizeof(buf), "RUMPHIJACK__DUP2MASK=%u", dup2mask);
885 		dup2str = malloc(strlen(buf)+1);
886 		if (dup2str == NULL)
887 			return ENOMEM;
888 		strcpy(dup2str, buf);
889 		bonus++;
890 	} else {
891 		dup2str = NULL;
892 	}
893 
894 	if (pwdinrump) {
895 		snprintf(buf, sizeof(buf), "RUMPHIJACK__PWDINRUMP=%u",
896 		    pwdinrump);
897 		pwdinrumpstr = malloc(strlen(buf)+1);
898 		if (pwdinrumpstr == NULL) {
899 			free(dup2str);
900 			return ENOMEM;
901 		}
902 		strcpy(pwdinrumpstr, buf);
903 		bonus++;
904 	} else {
905 		pwdinrumpstr = NULL;
906 	}
907 
908 	for (nelem = 0; envp && envp[nelem]; nelem++)
909 		continue;
910 	newenv = malloc(sizeof(*newenv) * nelem+bonus);
911 	if (newenv == NULL) {
912 		free(dup2str);
913 		free(pwdinrumpstr);
914 		return ENOMEM;
915 	}
916 	memcpy(newenv, envp, nelem*sizeof(*newenv));
917 	if (dup2str) {
918 		newenv[nelem+i] = dup2str;
919 		i++;
920 	}
921 	if (pwdinrumpstr) {
922 		newenv[nelem+i] = pwdinrumpstr;
923 		i++;
924 	}
925 	newenv[nelem+i] = NULL;
926 	_DIAGASSERT(i < bonus);
927 
928 	rv = rumpclient_exec(path, argv, newenv);
929 
930 	_DIAGASSERT(rv != 0);
931 	sverrno = errno;
932 	free(newenv);
933 	free(dup2str);
934 	errno = sverrno;
935 	return rv;
936 }
937 
938 /*
939  * select is done by calling poll.
940  */
941 int
942 REALSELECT(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds,
943 	struct timeval *timeout)
944 {
945 	struct pollfd *pfds;
946 	struct timespec ts, *tsp = NULL;
947 	nfds_t realnfds;
948 	int i, j;
949 	int rv, incr;
950 
951 	DPRINTF(("select\n"));
952 
953 	/*
954 	 * Well, first we must scan the fds to figure out how many
955 	 * fds there really are.  This is because up to and including
956 	 * nb5 poll() silently refuses nfds > process_maxopen_fds.
957 	 * Seems to be fixed in current, thank the maker.
958 	 * god damn cluster...bomb.
959 	 */
960 
961 	for (i = 0, realnfds = 0; i < nfds; i++) {
962 		if (readfds && FD_ISSET(i, readfds)) {
963 			realnfds++;
964 			continue;
965 		}
966 		if (writefds && FD_ISSET(i, writefds)) {
967 			realnfds++;
968 			continue;
969 		}
970 		if (exceptfds && FD_ISSET(i, exceptfds)) {
971 			realnfds++;
972 			continue;
973 		}
974 	}
975 
976 	if (realnfds) {
977 		pfds = calloc(realnfds, sizeof(*pfds));
978 		if (!pfds)
979 			return -1;
980 	} else {
981 		pfds = NULL;
982 	}
983 
984 	for (i = 0, j = 0; i < nfds; i++) {
985 		incr = 0;
986 		if (readfds && FD_ISSET(i, readfds)) {
987 			pfds[j].fd = i;
988 			pfds[j].events |= POLLIN;
989 			incr=1;
990 		}
991 		if (writefds && FD_ISSET(i, writefds)) {
992 			pfds[j].fd = i;
993 			pfds[j].events |= POLLOUT;
994 			incr=1;
995 		}
996 		if (exceptfds && FD_ISSET(i, exceptfds)) {
997 			pfds[j].fd = i;
998 			pfds[j].events |= POLLHUP|POLLERR;
999 			incr=1;
1000 		}
1001 		if (incr)
1002 			j++;
1003 	}
1004 	assert(j == (int)realnfds);
1005 
1006 	if (timeout) {
1007 		TIMEVAL_TO_TIMESPEC(timeout, &ts);
1008 		tsp = &ts;
1009 	}
1010 	rv = REALPOLLTS(pfds, realnfds, tsp, NULL);
1011 	/*
1012 	 * "If select() returns with an error the descriptor sets
1013 	 * will be unmodified"
1014 	 */
1015 	if (rv < 0)
1016 		goto out;
1017 
1018 	/*
1019 	 * zero out results (can't use FD_ZERO for the
1020 	 * obvious select-me-not reason).  whee.
1021 	 *
1022 	 * We do this here since some software ignores the return
1023 	 * value of select, and hence if the timeout expires, it may
1024 	 * assume all input descriptors have activity.
1025 	 */
1026 	for (i = 0; i < nfds; i++) {
1027 		if (readfds)
1028 			FD_CLR(i, readfds);
1029 		if (writefds)
1030 			FD_CLR(i, writefds);
1031 		if (exceptfds)
1032 			FD_CLR(i, exceptfds);
1033 	}
1034 	if (rv == 0)
1035 		goto out;
1036 
1037 	/*
1038 	 * We have >0 fds with activity.  Harvest the results.
1039 	 */
1040 	for (i = 0; i < (int)realnfds; i++) {
1041 		if (readfds) {
1042 			if (pfds[i].revents & POLLIN) {
1043 				FD_SET(pfds[i].fd, readfds);
1044 			}
1045 		}
1046 		if (writefds) {
1047 			if (pfds[i].revents & POLLOUT) {
1048 				FD_SET(pfds[i].fd, writefds);
1049 			}
1050 		}
1051 		if (exceptfds) {
1052 			if (pfds[i].revents & (POLLHUP|POLLERR)) {
1053 				FD_SET(pfds[i].fd, exceptfds);
1054 			}
1055 		}
1056 	}
1057 
1058  out:
1059 	free(pfds);
1060 	return rv;
1061 }
1062 
1063 static void
1064 checkpoll(struct pollfd *fds, nfds_t nfds, int *hostcall, int *rumpcall)
1065 {
1066 	nfds_t i;
1067 
1068 	for (i = 0; i < nfds; i++) {
1069 		if (fds[i].fd == -1)
1070 			continue;
1071 
1072 		if (fd_isrump(fds[i].fd))
1073 			(*rumpcall)++;
1074 		else
1075 			(*hostcall)++;
1076 	}
1077 }
1078 
1079 static void
1080 adjustpoll(struct pollfd *fds, nfds_t nfds, int (*fdadj)(int))
1081 {
1082 	nfds_t i;
1083 
1084 	for (i = 0; i < nfds; i++) {
1085 		fds[i].fd = fdadj(fds[i].fd);
1086 	}
1087 }
1088 
1089 /*
1090  * poll is easy as long as the call comes in the fds only in one
1091  * kernel.  otherwise its quite tricky...
1092  */
1093 struct pollarg {
1094 	struct pollfd *pfds;
1095 	nfds_t nfds;
1096 	const struct timespec *ts;
1097 	const sigset_t *sigmask;
1098 	int pipefd;
1099 	int errnum;
1100 };
1101 
1102 static void *
1103 hostpoll(void *arg)
1104 {
1105 	int (*op_pollts)(struct pollfd *, nfds_t, const struct timespec *,
1106 			 const sigset_t *);
1107 	struct pollarg *parg = arg;
1108 	intptr_t rv;
1109 
1110 	op_pollts = GETSYSCALL(host, POLLTS);
1111 	rv = op_pollts(parg->pfds, parg->nfds, parg->ts, parg->sigmask);
1112 	if (rv == -1)
1113 		parg->errnum = errno;
1114 	rump_sys_write(parg->pipefd, &rv, sizeof(rv));
1115 
1116 	return (void *)(intptr_t)rv;
1117 }
1118 
1119 int
1120 REALPOLLTS(struct pollfd *fds, nfds_t nfds, const struct timespec *ts,
1121 	const sigset_t *sigmask)
1122 {
1123 	int (*op_pollts)(struct pollfd *, nfds_t, const struct timespec *,
1124 			 const sigset_t *);
1125 	int (*host_close)(int);
1126 	int hostcall = 0, rumpcall = 0;
1127 	pthread_t pt;
1128 	nfds_t i;
1129 	int rv;
1130 
1131 	DPRINTF(("poll\n"));
1132 	checkpoll(fds, nfds, &hostcall, &rumpcall);
1133 
1134 	if (hostcall && rumpcall) {
1135 		struct pollfd *pfd_host = NULL, *pfd_rump = NULL;
1136 		int rpipe[2] = {-1,-1}, hpipe[2] = {-1,-1};
1137 		struct pollarg parg;
1138 		uintptr_t lrv;
1139 		int sverrno = 0, trv;
1140 
1141 		/*
1142 		 * ok, this is where it gets tricky.  We must support
1143 		 * this since it's a very common operation in certain
1144 		 * types of software (telnet, netcat, etc).  We allocate
1145 		 * two vectors and run two poll commands in separate
1146 		 * threads.  Whichever returns first "wins" and the
1147 		 * other kernel's fds won't show activity.
1148 		 */
1149 		rv = -1;
1150 
1151 		/* allocate full vector for O(n) joining after call */
1152 		pfd_host = malloc(sizeof(*pfd_host)*(nfds+1));
1153 		if (!pfd_host)
1154 			goto out;
1155 		pfd_rump = malloc(sizeof(*pfd_rump)*(nfds+1));
1156 		if (!pfd_rump) {
1157 			goto out;
1158 		}
1159 
1160 		/* split vectors */
1161 		for (i = 0; i < nfds; i++) {
1162 			if (fds[i].fd == -1) {
1163 				pfd_host[i].fd = -1;
1164 				pfd_rump[i].fd = -1;
1165 			} else if (fd_isrump(fds[i].fd)) {
1166 				pfd_host[i].fd = -1;
1167 				pfd_rump[i].fd = fd_host2rump(fds[i].fd);
1168 				pfd_rump[i].events = fds[i].events;
1169 			} else {
1170 				pfd_rump[i].fd = -1;
1171 				pfd_host[i].fd = fds[i].fd;
1172 				pfd_host[i].events = fds[i].events;
1173 			}
1174 			pfd_rump[i].revents = pfd_host[i].revents = 0;
1175 			fds[i].revents = 0;
1176 		}
1177 
1178 		/*
1179 		 * then, open two pipes, one for notifications
1180 		 * to each kernel.
1181 		 */
1182 		if (rump_sys_pipe(rpipe) == -1)
1183 			goto out;
1184 		if (pipe(hpipe) == -1)
1185 			goto out;
1186 
1187 		pfd_host[nfds].fd = hpipe[0];
1188 		pfd_host[nfds].events = POLLIN;
1189 		pfd_rump[nfds].fd = rpipe[0];
1190 		pfd_rump[nfds].events = POLLIN;
1191 
1192 		/*
1193 		 * then, create a thread to do host part and meanwhile
1194 		 * do rump kernel part right here
1195 		 */
1196 
1197 		parg.pfds = pfd_host;
1198 		parg.nfds = nfds+1;
1199 		parg.ts = ts;
1200 		parg.sigmask = sigmask;
1201 		parg.pipefd = rpipe[1];
1202 		pthread_create(&pt, NULL, hostpoll, &parg);
1203 
1204 		op_pollts = GETSYSCALL(rump, POLLTS);
1205 		lrv = op_pollts(pfd_rump, nfds+1, ts, NULL);
1206 		sverrno = errno;
1207 		write(hpipe[1], &rv, sizeof(rv));
1208 		pthread_join(pt, (void *)&trv);
1209 
1210 		/* check who "won" and merge results */
1211 		if (lrv != 0 && pfd_host[nfds].revents & POLLIN) {
1212 			rv = trv;
1213 
1214 			for (i = 0; i < nfds; i++) {
1215 				if (pfd_rump[i].fd != -1)
1216 					fds[i].revents = pfd_rump[i].revents;
1217 			}
1218 			sverrno = parg.errnum;
1219 		} else if (trv != 0 && pfd_rump[nfds].revents & POLLIN) {
1220 			rv = trv;
1221 
1222 			for (i = 0; i < nfds; i++) {
1223 				if (pfd_host[i].fd != -1)
1224 					fds[i].revents = pfd_host[i].revents;
1225 			}
1226 		} else {
1227 			rv = 0;
1228 		}
1229 
1230  out:
1231 		host_close = GETSYSCALL(host, CLOSE);
1232 		if (rpipe[0] != -1)
1233 			rump_sys_close(rpipe[0]);
1234 		if (rpipe[1] != -1)
1235 			rump_sys_close(rpipe[1]);
1236 		if (hpipe[0] != -1)
1237 			host_close(hpipe[0]);
1238 		if (hpipe[1] != -1)
1239 			host_close(hpipe[1]);
1240 		free(pfd_host);
1241 		free(pfd_rump);
1242 		errno = sverrno;
1243 	} else {
1244 		if (hostcall) {
1245 			op_pollts = GETSYSCALL(host, POLLTS);
1246 		} else {
1247 			op_pollts = GETSYSCALL(rump, POLLTS);
1248 			adjustpoll(fds, nfds, fd_host2rump);
1249 		}
1250 
1251 		rv = op_pollts(fds, nfds, ts, sigmask);
1252 		if (rumpcall)
1253 			adjustpoll(fds, nfds, fd_rump2host);
1254 	}
1255 
1256 	return rv;
1257 }
1258 
1259 int
1260 poll(struct pollfd *fds, nfds_t nfds, int timeout)
1261 {
1262 	struct timespec ts;
1263 	struct timespec *tsp = NULL;
1264 
1265 	if (timeout != INFTIM) {
1266 		ts.tv_sec = timeout / 1000;
1267 		ts.tv_nsec = (timeout % 1000) * 1000*1000;
1268 
1269 		tsp = &ts;
1270 	}
1271 
1272 	return REALPOLLTS(fds, nfds, tsp, NULL);
1273 }
1274 
1275 int
1276 REALKEVENT(int kq, const struct kevent *changelist, size_t nchanges,
1277 	struct kevent *eventlist, size_t nevents,
1278 	const struct timespec *timeout)
1279 {
1280 	int (*op_kevent)(int, const struct kevent *, size_t,
1281 		struct kevent *, size_t, const struct timespec *);
1282 	const struct kevent *ev;
1283 	size_t i;
1284 
1285 	/*
1286 	 * Check that we don't attempt to kevent rump kernel fd's.
1287 	 * That needs similar treatment to select/poll, but is slightly
1288 	 * trickier since we need to manage to different kq descriptors.
1289 	 * (TODO, in case you're wondering).
1290 	 */
1291 	for (i = 0; i < nchanges; i++) {
1292 		ev = &changelist[i];
1293 		if (ev->filter == EVFILT_READ || ev->filter == EVFILT_WRITE ||
1294 		    ev->filter == EVFILT_VNODE) {
1295 			if (fd_isrump((int)ev->ident))
1296 				return ENOTSUP;
1297 		}
1298 	}
1299 
1300 	op_kevent = GETSYSCALL(host, KEVENT);
1301 	return op_kevent(kq, changelist, nchanges, eventlist, nevents, timeout);
1302 }
1303 
1304 /*
1305  * Rest are std type calls.
1306  */
1307 
1308 FDCALL(int, bind, DUALCALL_BIND,					\
1309 	(int fd, const struct sockaddr *name, socklen_t namelen),	\
1310 	(int, const struct sockaddr *, socklen_t),			\
1311 	(fd, name, namelen))
1312 
1313 FDCALL(int, connect, DUALCALL_CONNECT,					\
1314 	(int fd, const struct sockaddr *name, socklen_t namelen),	\
1315 	(int, const struct sockaddr *, socklen_t),			\
1316 	(fd, name, namelen))
1317 
1318 FDCALL(int, getpeername, DUALCALL_GETPEERNAME,				\
1319 	(int fd, struct sockaddr *name, socklen_t *namelen),		\
1320 	(int, struct sockaddr *, socklen_t *),				\
1321 	(fd, name, namelen))
1322 
1323 FDCALL(int, getsockname, DUALCALL_GETSOCKNAME, 				\
1324 	(int fd, struct sockaddr *name, socklen_t *namelen),		\
1325 	(int, struct sockaddr *, socklen_t *),				\
1326 	(fd, name, namelen))
1327 
1328 FDCALL(int, listen, DUALCALL_LISTEN,	 				\
1329 	(int fd, int backlog),						\
1330 	(int, int),							\
1331 	(fd, backlog))
1332 
1333 FDCALL(ssize_t, recvfrom, DUALCALL_RECVFROM, 				\
1334 	(int fd, void *buf, size_t len, int flags,			\
1335 	    struct sockaddr *from, socklen_t *fromlen),			\
1336 	(int, void *, size_t, int, struct sockaddr *, socklen_t *),	\
1337 	(fd, buf, len, flags, from, fromlen))
1338 
1339 FDCALL(ssize_t, sendto, DUALCALL_SENDTO, 				\
1340 	(int fd, const void *buf, size_t len, int flags,		\
1341 	    const struct sockaddr *to, socklen_t tolen),		\
1342 	(int, const void *, size_t, int,				\
1343 	    const struct sockaddr *, socklen_t),			\
1344 	(fd, buf, len, flags, to, tolen))
1345 
1346 FDCALL(ssize_t, recvmsg, DUALCALL_RECVMSG, 				\
1347 	(int fd, struct msghdr *msg, int flags),			\
1348 	(int, struct msghdr *, int),					\
1349 	(fd, msg, flags))
1350 
1351 FDCALL(ssize_t, sendmsg, DUALCALL_SENDMSG, 				\
1352 	(int fd, const struct msghdr *msg, int flags),			\
1353 	(int, const struct msghdr *, int),				\
1354 	(fd, msg, flags))
1355 
1356 FDCALL(int, getsockopt, DUALCALL_GETSOCKOPT, 				\
1357 	(int fd, int level, int optn, void *optval, socklen_t *optlen),	\
1358 	(int, int, int, void *, socklen_t *),				\
1359 	(fd, level, optn, optval, optlen))
1360 
1361 FDCALL(int, setsockopt, DUALCALL_SETSOCKOPT, 				\
1362 	(int fd, int level, int optn,					\
1363 	    const void *optval, socklen_t optlen),			\
1364 	(int, int, int, const void *, socklen_t),			\
1365 	(fd, level, optn, optval, optlen))
1366 
1367 FDCALL(int, shutdown, DUALCALL_SHUTDOWN, 				\
1368 	(int fd, int how),						\
1369 	(int, int),							\
1370 	(fd, how))
1371 
1372 #if _FORTIFY_SOURCE > 0
1373 #define STUB(fun) __ssp_weak_name(fun)
1374 ssize_t _sys_readlink(const char * __restrict, char * __restrict, size_t);
1375 ssize_t
1376 STUB(readlink)(const char * __restrict path, char * __restrict buf,
1377     size_t bufsiz)
1378 {
1379 	return _sys_readlink(path, buf, bufsiz);
1380 }
1381 
1382 char *_sys_getcwd(char *, size_t);
1383 char *
1384 STUB(getcwd)(char *buf, size_t size)
1385 {
1386 	return _sys_getcwd(buf, size);
1387 }
1388 #else
1389 #define STUB(fun) fun
1390 #endif
1391 
1392 FDCALL(ssize_t, REALREAD, DUALCALL_READ,				\
1393 	(int fd, void *buf, size_t buflen),				\
1394 	(int, void *, size_t),						\
1395 	(fd, buf, buflen))
1396 
1397 FDCALL(ssize_t, readv, DUALCALL_READV, 					\
1398 	(int fd, const struct iovec *iov, int iovcnt),			\
1399 	(int, const struct iovec *, int),				\
1400 	(fd, iov, iovcnt))
1401 
1402 FDCALL(ssize_t, writev, DUALCALL_WRITEV, 				\
1403 	(int fd, const struct iovec *iov, int iovcnt),			\
1404 	(int, const struct iovec *, int),				\
1405 	(fd, iov, iovcnt))
1406 
1407 FDCALL(int, REALFSTAT, DUALCALL_FSTAT,					\
1408 	(int fd, struct stat *sb),					\
1409 	(int, struct stat *),						\
1410 	(fd, sb))
1411 
1412 FDCALL(int, fstatvfs1, DUALCALL_FSTATVFS1,				\
1413 	(int fd, struct statvfs *buf, int flags),			\
1414 	(int, struct statvfs *, int),					\
1415 	(fd, buf, flags))
1416 
1417 FDCALL(off_t, lseek, DUALCALL_LSEEK,					\
1418 	(int fd, off_t offset, int whence),				\
1419 	(int, off_t, int),						\
1420 	(fd, offset, whence))
1421 
1422 FDCALL(int, REALGETDENTS, DUALCALL_GETDENTS,				\
1423 	(int fd, char *buf, size_t nbytes),				\
1424 	(int, char *, size_t),						\
1425 	(fd, buf, nbytes))
1426 
1427 FDCALL(int, fchown, DUALCALL_FCHOWN,					\
1428 	(int fd, uid_t owner, gid_t group),				\
1429 	(int, uid_t, gid_t),						\
1430 	(fd, owner, group))
1431 
1432 FDCALL(int, fchmod, DUALCALL_FCHMOD,					\
1433 	(int fd, mode_t mode),						\
1434 	(int, mode_t),							\
1435 	(fd, mode))
1436 
1437 FDCALL(int, ftruncate, DUALCALL_FTRUNCATE,				\
1438 	(int fd, off_t length),						\
1439 	(int, off_t),							\
1440 	(fd, length))
1441 
1442 FDCALL(int, fsync, DUALCALL_FSYNC,					\
1443 	(int fd),							\
1444 	(int),								\
1445 	(fd))
1446 
1447 FDCALL(int, fsync_range, DUALCALL_FSYNC_RANGE,				\
1448 	(int fd, int how, off_t start, off_t length),			\
1449 	(int, int, off_t, off_t),					\
1450 	(fd, how, start, length))
1451 
1452 FDCALL(int, futimes, DUALCALL_FUTIMES,					\
1453 	(int fd, const struct timeval *tv),				\
1454 	(int, const struct timeval *),					\
1455 	(fd, tv))
1456 
1457 /*
1458  * path-based selectors
1459  */
1460 
1461 PATHCALL(int, REALSTAT, DUALCALL_STAT,					\
1462 	(const char *path, struct stat *sb),				\
1463 	(const char *, struct stat *),					\
1464 	(path, sb))
1465 
1466 PATHCALL(int, REALLSTAT, DUALCALL_LSTAT,				\
1467 	(const char *path, struct stat *sb),				\
1468 	(const char *, struct stat *),					\
1469 	(path, sb))
1470 
1471 PATHCALL(int, chown, DUALCALL_CHOWN,					\
1472 	(const char *path, uid_t owner, gid_t group),			\
1473 	(const char *, uid_t, gid_t),					\
1474 	(path, owner, group))
1475 
1476 PATHCALL(int, lchown, DUALCALL_LCHOWN,					\
1477 	(const char *path, uid_t owner, gid_t group),			\
1478 	(const char *, uid_t, gid_t),					\
1479 	(path, owner, group))
1480 
1481 PATHCALL(int, chmod, DUALCALL_CHMOD,					\
1482 	(const char *path, mode_t mode),				\
1483 	(const char *, mode_t),						\
1484 	(path, mode))
1485 
1486 PATHCALL(int, lchmod, DUALCALL_LCHMOD,					\
1487 	(const char *path, mode_t mode),				\
1488 	(const char *, mode_t),						\
1489 	(path, mode))
1490 
1491 PATHCALL(int, statvfs1, DUALCALL_STATVFS1,				\
1492 	(const char *path, struct statvfs *buf, int flags),		\
1493 	(const char *, struct statvfs *, int),				\
1494 	(path, buf, flags))
1495 
1496 PATHCALL(int, unlink, DUALCALL_UNLINK,					\
1497 	(const char *path),						\
1498 	(const char *),							\
1499 	(path))
1500 
1501 PATHCALL(int, symlink, DUALCALL_SYMLINK,				\
1502 	(const char *path, const char *target),				\
1503 	(const char *, const char *),					\
1504 	(path, target))
1505 
1506 PATHCALL(ssize_t, readlink, DUALCALL_READLINK,				\
1507 	(const char *path, char *buf, size_t bufsiz),			\
1508 	(const char *, char *, size_t),					\
1509 	(path, buf, bufsiz))
1510 
1511 /* XXX: cross-kernel renames need to be blocked */
1512 PATHCALL(int, rename, DUALCALL_RENAME,					\
1513 	(const char *path, const char *to),				\
1514 	(const char *, const char *),					\
1515 	(path, to))
1516 
1517 PATHCALL(int, mkdir, DUALCALL_MKDIR,					\
1518 	(const char *path, mode_t mode),				\
1519 	(const char *, mode_t),						\
1520 	(path, mode))
1521 
1522 PATHCALL(int, rmdir, DUALCALL_RMDIR,					\
1523 	(const char *path),						\
1524 	(const char *),							\
1525 	(path))
1526 
1527 PATHCALL(int, utimes, DUALCALL_UTIMES,					\
1528 	(const char *path, const struct timeval *tv),			\
1529 	(const char *, const struct timeval *),				\
1530 	(path, tv))
1531 
1532 PATHCALL(int, lutimes, DUALCALL_LUTIMES,				\
1533 	(const char *path, const struct timeval *tv),			\
1534 	(const char *, const struct timeval *),				\
1535 	(path, tv))
1536 
1537 PATHCALL(int, truncate, DUALCALL_TRUNCATE,				\
1538 	(const char *path, off_t length),				\
1539 	(const char *, off_t),						\
1540 	(path, length))
1541 
1542 /*
1543  * Note: with mount the decisive parameter is the mount
1544  * destination directory.  This is because we don't really know
1545  * about the "source" directory in a generic call (and besides,
1546  * it might not even exist, cf. nfs).
1547  */
1548 PATHCALL(int, REALMOUNT, DUALCALL_MOUNT,				\
1549 	(const char *type, const char *path, int flags,			\
1550 	    void *data, size_t dlen),					\
1551 	(const char *, const char *, int, void *, size_t),		\
1552 	(type, path, flags, data, dlen))
1553 
1554 PATHCALL(int, unmount, DUALCALL_UNMOUNT,				\
1555 	(const char *path, int flags),					\
1556 	(const char *, int),						\
1557 	(path, flags))
1558