xref: /freebsd/sys/amd64/linux32/linux32_machdep.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 2004 Tim J. Robbins
3  * Copyright (c) 2002 Doug Rabson
4  * Copyright (c) 2000 Marcel Moolenaar
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/kernel.h>
36 #include <sys/systm.h>
37 #include <sys/file.h>
38 #include <sys/fcntl.h>
39 #include <sys/clock.h>
40 #include <sys/imgact.h>
41 #include <sys/limits.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mman.h>
45 #include <sys/mutex.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/resource.h>
49 #include <sys/resourcevar.h>
50 #include <sys/sched.h>
51 #include <sys/syscallsubr.h>
52 #include <sys/sysproto.h>
53 #include <sys/unistd.h>
54 #include <sys/wait.h>
55 
56 #include <machine/frame.h>
57 #include <machine/pcb.h>
58 #include <machine/psl.h>
59 #include <machine/segments.h>
60 #include <machine/specialreg.h>
61 
62 #include <vm/vm.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_map.h>
65 
66 #include <compat/freebsd32/freebsd32_util.h>
67 #include <amd64/linux32/linux.h>
68 #include <amd64/linux32/linux32_proto.h>
69 #include <compat/linux/linux_ipc.h>
70 #include <compat/linux/linux_misc.h>
71 #include <compat/linux/linux_signal.h>
72 #include <compat/linux/linux_util.h>
73 #include <compat/linux/linux_emul.h>
74 
75 struct l_old_select_argv {
76 	l_int		nfds;
77 	l_uintptr_t	readfds;
78 	l_uintptr_t	writefds;
79 	l_uintptr_t	exceptfds;
80 	l_uintptr_t	timeout;
81 } __packed;
82 
83 int
84 linux_to_bsd_sigaltstack(int lsa)
85 {
86 	int bsa = 0;
87 
88 	if (lsa & LINUX_SS_DISABLE)
89 		bsa |= SS_DISABLE;
90 	if (lsa & LINUX_SS_ONSTACK)
91 		bsa |= SS_ONSTACK;
92 	return (bsa);
93 }
94 
95 static int	linux_mmap_common(struct thread *td, l_uintptr_t addr,
96 		    l_size_t len, l_int prot, l_int flags, l_int fd,
97 		    l_loff_t pos);
98 
99 int
100 bsd_to_linux_sigaltstack(int bsa)
101 {
102 	int lsa = 0;
103 
104 	if (bsa & SS_DISABLE)
105 		lsa |= LINUX_SS_DISABLE;
106 	if (bsa & SS_ONSTACK)
107 		lsa |= LINUX_SS_ONSTACK;
108 	return (lsa);
109 }
110 
111 static void
112 bsd_to_linux_rusage(struct rusage *ru, struct l_rusage *lru)
113 {
114 
115 	lru->ru_utime.tv_sec = ru->ru_utime.tv_sec;
116 	lru->ru_utime.tv_usec = ru->ru_utime.tv_usec;
117 	lru->ru_stime.tv_sec = ru->ru_stime.tv_sec;
118 	lru->ru_stime.tv_usec = ru->ru_stime.tv_usec;
119 	lru->ru_maxrss = ru->ru_maxrss;
120 	lru->ru_ixrss = ru->ru_ixrss;
121 	lru->ru_idrss = ru->ru_idrss;
122 	lru->ru_isrss = ru->ru_isrss;
123 	lru->ru_minflt = ru->ru_minflt;
124 	lru->ru_majflt = ru->ru_majflt;
125 	lru->ru_nswap = ru->ru_nswap;
126 	lru->ru_inblock = ru->ru_inblock;
127 	lru->ru_oublock = ru->ru_oublock;
128 	lru->ru_msgsnd = ru->ru_msgsnd;
129 	lru->ru_msgrcv = ru->ru_msgrcv;
130 	lru->ru_nsignals = ru->ru_nsignals;
131 	lru->ru_nvcsw = ru->ru_nvcsw;
132 	lru->ru_nivcsw = ru->ru_nivcsw;
133 }
134 
135 int
136 linux_execve(struct thread *td, struct linux_execve_args *args)
137 {
138 	struct image_args eargs;
139 	char *path;
140 	int error;
141 
142 	LCONVPATHEXIST(td, args->path, &path);
143 
144 #ifdef DEBUG
145 	if (ldebug(execve))
146 		printf(ARGS(execve, "%s"), path);
147 #endif
148 
149 	error = freebsd32_exec_copyin_args(&eargs, path, UIO_SYSSPACE,
150 	    args->argp, args->envp);
151 	free(path, M_TEMP);
152 	if (error == 0)
153 		error = kern_execve(td, &eargs, NULL);
154 	if (error == 0)
155 		/* Linux process can execute FreeBSD one, do not attempt
156 		 * to create emuldata for such process using
157 		 * linux_proc_init, this leads to a panic on KASSERT
158 		 * because such process has p->p_emuldata == NULL.
159 		 */
160 		if (SV_PROC_ABI(td->td_proc) == SV_ABI_LINUX)
161 			error = linux_proc_init(td, 0, 0);
162 	return (error);
163 }
164 
165 CTASSERT(sizeof(struct l_iovec32) == 8);
166 
167 static int
168 linux32_copyinuio(struct l_iovec32 *iovp, l_ulong iovcnt, struct uio **uiop)
169 {
170 	struct l_iovec32 iov32;
171 	struct iovec *iov;
172 	struct uio *uio;
173 	uint32_t iovlen;
174 	int error, i;
175 
176 	*uiop = NULL;
177 	if (iovcnt > UIO_MAXIOV)
178 		return (EINVAL);
179 	iovlen = iovcnt * sizeof(struct iovec);
180 	uio = malloc(iovlen + sizeof(*uio), M_IOV, M_WAITOK);
181 	iov = (struct iovec *)(uio + 1);
182 	for (i = 0; i < iovcnt; i++) {
183 		error = copyin(&iovp[i], &iov32, sizeof(struct l_iovec32));
184 		if (error) {
185 			free(uio, M_IOV);
186 			return (error);
187 		}
188 		iov[i].iov_base = PTRIN(iov32.iov_base);
189 		iov[i].iov_len = iov32.iov_len;
190 	}
191 	uio->uio_iov = iov;
192 	uio->uio_iovcnt = iovcnt;
193 	uio->uio_segflg = UIO_USERSPACE;
194 	uio->uio_offset = -1;
195 	uio->uio_resid = 0;
196 	for (i = 0; i < iovcnt; i++) {
197 		if (iov->iov_len > INT_MAX - uio->uio_resid) {
198 			free(uio, M_IOV);
199 			return (EINVAL);
200 		}
201 		uio->uio_resid += iov->iov_len;
202 		iov++;
203 	}
204 	*uiop = uio;
205 	return (0);
206 }
207 
208 int
209 linux32_copyiniov(struct l_iovec32 *iovp32, l_ulong iovcnt, struct iovec **iovp,
210     int error)
211 {
212 	struct l_iovec32 iov32;
213 	struct iovec *iov;
214 	uint32_t iovlen;
215 	int i;
216 
217 	*iovp = NULL;
218 	if (iovcnt > UIO_MAXIOV)
219 		return (error);
220 	iovlen = iovcnt * sizeof(struct iovec);
221 	iov = malloc(iovlen, M_IOV, M_WAITOK);
222 	for (i = 0; i < iovcnt; i++) {
223 		error = copyin(&iovp32[i], &iov32, sizeof(struct l_iovec32));
224 		if (error) {
225 			free(iov, M_IOV);
226 			return (error);
227 		}
228 		iov[i].iov_base = PTRIN(iov32.iov_base);
229 		iov[i].iov_len = iov32.iov_len;
230 	}
231 	*iovp = iov;
232 	return(0);
233 
234 }
235 
236 int
237 linux_readv(struct thread *td, struct linux_readv_args *uap)
238 {
239 	struct uio *auio;
240 	int error;
241 
242 	error = linux32_copyinuio(uap->iovp, uap->iovcnt, &auio);
243 	if (error)
244 		return (error);
245 	error = kern_readv(td, uap->fd, auio);
246 	free(auio, M_IOV);
247 	return (error);
248 }
249 
250 int
251 linux_writev(struct thread *td, struct linux_writev_args *uap)
252 {
253 	struct uio *auio;
254 	int error;
255 
256 	error = linux32_copyinuio(uap->iovp, uap->iovcnt, &auio);
257 	if (error)
258 		return (error);
259 	error = kern_writev(td, uap->fd, auio);
260 	free(auio, M_IOV);
261 	return (error);
262 }
263 
264 struct l_ipc_kludge {
265 	l_uintptr_t msgp;
266 	l_long msgtyp;
267 } __packed;
268 
269 int
270 linux_ipc(struct thread *td, struct linux_ipc_args *args)
271 {
272 
273 	switch (args->what & 0xFFFF) {
274 	case LINUX_SEMOP: {
275 		struct linux_semop_args a;
276 
277 		a.semid = args->arg1;
278 		a.tsops = args->ptr;
279 		a.nsops = args->arg2;
280 		return (linux_semop(td, &a));
281 	}
282 	case LINUX_SEMGET: {
283 		struct linux_semget_args a;
284 
285 		a.key = args->arg1;
286 		a.nsems = args->arg2;
287 		a.semflg = args->arg3;
288 		return (linux_semget(td, &a));
289 	}
290 	case LINUX_SEMCTL: {
291 		struct linux_semctl_args a;
292 		int error;
293 
294 		a.semid = args->arg1;
295 		a.semnum = args->arg2;
296 		a.cmd = args->arg3;
297 		error = copyin(args->ptr, &a.arg, sizeof(a.arg));
298 		if (error)
299 			return (error);
300 		return (linux_semctl(td, &a));
301 	}
302 	case LINUX_MSGSND: {
303 		struct linux_msgsnd_args a;
304 
305 		a.msqid = args->arg1;
306 		a.msgp = args->ptr;
307 		a.msgsz = args->arg2;
308 		a.msgflg = args->arg3;
309 		return (linux_msgsnd(td, &a));
310 	}
311 	case LINUX_MSGRCV: {
312 		struct linux_msgrcv_args a;
313 
314 		a.msqid = args->arg1;
315 		a.msgsz = args->arg2;
316 		a.msgflg = args->arg3;
317 		if ((args->what >> 16) == 0) {
318 			struct l_ipc_kludge tmp;
319 			int error;
320 
321 			if (args->ptr == 0)
322 				return (EINVAL);
323 			error = copyin(args->ptr, &tmp, sizeof(tmp));
324 			if (error)
325 				return (error);
326 			a.msgp = PTRIN(tmp.msgp);
327 			a.msgtyp = tmp.msgtyp;
328 		} else {
329 			a.msgp = args->ptr;
330 			a.msgtyp = args->arg5;
331 		}
332 		return (linux_msgrcv(td, &a));
333 	}
334 	case LINUX_MSGGET: {
335 		struct linux_msgget_args a;
336 
337 		a.key = args->arg1;
338 		a.msgflg = args->arg2;
339 		return (linux_msgget(td, &a));
340 	}
341 	case LINUX_MSGCTL: {
342 		struct linux_msgctl_args a;
343 
344 		a.msqid = args->arg1;
345 		a.cmd = args->arg2;
346 		a.buf = args->ptr;
347 		return (linux_msgctl(td, &a));
348 	}
349 	case LINUX_SHMAT: {
350 		struct linux_shmat_args a;
351 
352 		a.shmid = args->arg1;
353 		a.shmaddr = args->ptr;
354 		a.shmflg = args->arg2;
355 		a.raddr = PTRIN((l_uint)args->arg3);
356 		return (linux_shmat(td, &a));
357 	}
358 	case LINUX_SHMDT: {
359 		struct linux_shmdt_args a;
360 
361 		a.shmaddr = args->ptr;
362 		return (linux_shmdt(td, &a));
363 	}
364 	case LINUX_SHMGET: {
365 		struct linux_shmget_args a;
366 
367 		a.key = args->arg1;
368 		a.size = args->arg2;
369 		a.shmflg = args->arg3;
370 		return (linux_shmget(td, &a));
371 	}
372 	case LINUX_SHMCTL: {
373 		struct linux_shmctl_args a;
374 
375 		a.shmid = args->arg1;
376 		a.cmd = args->arg2;
377 		a.buf = args->ptr;
378 		return (linux_shmctl(td, &a));
379 	}
380 	default:
381 		break;
382 	}
383 
384 	return (EINVAL);
385 }
386 
387 int
388 linux_old_select(struct thread *td, struct linux_old_select_args *args)
389 {
390 	struct l_old_select_argv linux_args;
391 	struct linux_select_args newsel;
392 	int error;
393 
394 #ifdef DEBUG
395 	if (ldebug(old_select))
396 		printf(ARGS(old_select, "%p"), args->ptr);
397 #endif
398 
399 	error = copyin(args->ptr, &linux_args, sizeof(linux_args));
400 	if (error)
401 		return (error);
402 
403 	newsel.nfds = linux_args.nfds;
404 	newsel.readfds = PTRIN(linux_args.readfds);
405 	newsel.writefds = PTRIN(linux_args.writefds);
406 	newsel.exceptfds = PTRIN(linux_args.exceptfds);
407 	newsel.timeout = PTRIN(linux_args.timeout);
408 	return (linux_select(td, &newsel));
409 }
410 
411 int
412 linux_set_cloned_tls(struct thread *td, void *desc)
413 {
414 	struct user_segment_descriptor sd;
415 	struct l_user_desc info;
416 	struct pcb *pcb;
417 	int error;
418 	int a[2];
419 
420 	error = copyin(desc, &info, sizeof(struct l_user_desc));
421 	if (error) {
422 		printf(LMSG("copyin failed!"));
423 	} else {
424 		/* We might copy out the entry_number as GUGS32_SEL. */
425 		info.entry_number = GUGS32_SEL;
426 		error = copyout(&info, desc, sizeof(struct l_user_desc));
427 		if (error)
428 			printf(LMSG("copyout failed!"));
429 
430 		a[0] = LINUX_LDT_entry_a(&info);
431 		a[1] = LINUX_LDT_entry_b(&info);
432 
433 		memcpy(&sd, &a, sizeof(a));
434 #ifdef DEBUG
435 		if (ldebug(clone))
436 			printf("Segment created in clone with "
437 			    "CLONE_SETTLS: lobase: %x, hibase: %x, "
438 			    "lolimit: %x, hilimit: %x, type: %i, "
439 			    "dpl: %i, p: %i, xx: %i, long: %i, "
440 			    "def32: %i, gran: %i\n", sd.sd_lobase,
441 			    sd.sd_hibase, sd.sd_lolimit, sd.sd_hilimit,
442 			    sd.sd_type, sd.sd_dpl, sd.sd_p, sd.sd_xx,
443 			    sd.sd_long, sd.sd_def32, sd.sd_gran);
444 #endif
445 		pcb = td->td_pcb;
446 		pcb->pcb_gsbase = (register_t)info.base_addr;
447 /* XXXKIB	pcb->pcb_gs32sd = sd; */
448 		td->td_frame->tf_gs = GSEL(GUGS32_SEL, SEL_UPL);
449 		set_pcb_flags(pcb, PCB_GS32BIT | PCB_32BIT);
450 	}
451 
452 	return (error);
453 }
454 
455 int
456 linux_set_upcall_kse(struct thread *td, register_t stack)
457 {
458 
459 	td->td_frame->tf_rsp = stack;
460 
461 	return (0);
462 }
463 
464 #define STACK_SIZE  (2 * 1024 * 1024)
465 #define GUARD_SIZE  (4 * PAGE_SIZE)
466 
467 int
468 linux_mmap2(struct thread *td, struct linux_mmap2_args *args)
469 {
470 
471 #ifdef DEBUG
472 	if (ldebug(mmap2))
473 		printf(ARGS(mmap2, "0x%08x, %d, %d, 0x%08x, %d, %d"),
474 		    args->addr, args->len, args->prot,
475 		    args->flags, args->fd, args->pgoff);
476 #endif
477 
478 	return (linux_mmap_common(td, PTROUT(args->addr), args->len, args->prot,
479 		args->flags, args->fd, (uint64_t)(uint32_t)args->pgoff *
480 		PAGE_SIZE));
481 }
482 
483 int
484 linux_mmap(struct thread *td, struct linux_mmap_args *args)
485 {
486 	int error;
487 	struct l_mmap_argv linux_args;
488 
489 	error = copyin(args->ptr, &linux_args, sizeof(linux_args));
490 	if (error)
491 		return (error);
492 
493 #ifdef DEBUG
494 	if (ldebug(mmap))
495 		printf(ARGS(mmap, "0x%08x, %d, %d, 0x%08x, %d, %d"),
496 		    linux_args.addr, linux_args.len, linux_args.prot,
497 		    linux_args.flags, linux_args.fd, linux_args.pgoff);
498 #endif
499 
500 	return (linux_mmap_common(td, linux_args.addr, linux_args.len,
501 	    linux_args.prot, linux_args.flags, linux_args.fd,
502 	    (uint32_t)linux_args.pgoff));
503 }
504 
505 static int
506 linux_mmap_common(struct thread *td, l_uintptr_t addr, l_size_t len, l_int prot,
507     l_int flags, l_int fd, l_loff_t pos)
508 {
509 	struct proc *p = td->td_proc;
510 	struct mmap_args /* {
511 		caddr_t addr;
512 		size_t len;
513 		int prot;
514 		int flags;
515 		int fd;
516 		long pad;
517 		off_t pos;
518 	} */ bsd_args;
519 	int error;
520 	struct file *fp;
521 
522 	error = 0;
523 	bsd_args.flags = 0;
524 	fp = NULL;
525 
526 	/*
527 	 * Linux mmap(2):
528 	 * You must specify exactly one of MAP_SHARED and MAP_PRIVATE
529 	 */
530 	if (!((flags & LINUX_MAP_SHARED) ^ (flags & LINUX_MAP_PRIVATE)))
531 		return (EINVAL);
532 
533 	if (flags & LINUX_MAP_SHARED)
534 		bsd_args.flags |= MAP_SHARED;
535 	if (flags & LINUX_MAP_PRIVATE)
536 		bsd_args.flags |= MAP_PRIVATE;
537 	if (flags & LINUX_MAP_FIXED)
538 		bsd_args.flags |= MAP_FIXED;
539 	if (flags & LINUX_MAP_ANON) {
540 		/* Enforce pos to be on page boundary, then ignore. */
541 		if ((pos & PAGE_MASK) != 0)
542 			return (EINVAL);
543 		pos = 0;
544 		bsd_args.flags |= MAP_ANON;
545 	} else
546 		bsd_args.flags |= MAP_NOSYNC;
547 	if (flags & LINUX_MAP_GROWSDOWN)
548 		bsd_args.flags |= MAP_STACK;
549 
550 	/*
551 	 * PROT_READ, PROT_WRITE, or PROT_EXEC implies PROT_READ and PROT_EXEC
552 	 * on Linux/i386. We do this to ensure maximum compatibility.
553 	 * Linux/ia64 does the same in i386 emulation mode.
554 	 */
555 	bsd_args.prot = prot;
556 	if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC))
557 		bsd_args.prot |= PROT_READ | PROT_EXEC;
558 
559 	/* Linux does not check file descriptor when MAP_ANONYMOUS is set. */
560 	bsd_args.fd = (bsd_args.flags & MAP_ANON) ? -1 : fd;
561 	if (bsd_args.fd != -1) {
562 		/*
563 		 * Linux follows Solaris mmap(2) description:
564 		 * The file descriptor fildes is opened with
565 		 * read permission, regardless of the
566 		 * protection options specified.
567 		 */
568 
569 		if ((error = fget(td, bsd_args.fd, &fp)) != 0)
570 			return (error);
571 		if (fp->f_type != DTYPE_VNODE) {
572 			fdrop(fp, td);
573 			return (EINVAL);
574 		}
575 
576 		/* Linux mmap() just fails for O_WRONLY files */
577 		if (!(fp->f_flag & FREAD)) {
578 			fdrop(fp, td);
579 			return (EACCES);
580 		}
581 
582 		fdrop(fp, td);
583 	}
584 
585 	if (flags & LINUX_MAP_GROWSDOWN) {
586 		/*
587 		 * The Linux MAP_GROWSDOWN option does not limit auto
588 		 * growth of the region.  Linux mmap with this option
589 		 * takes as addr the inital BOS, and as len, the initial
590 		 * region size.  It can then grow down from addr without
591 		 * limit.  However, Linux threads has an implicit internal
592 		 * limit to stack size of STACK_SIZE.  Its just not
593 		 * enforced explicitly in Linux.  But, here we impose
594 		 * a limit of (STACK_SIZE - GUARD_SIZE) on the stack
595 		 * region, since we can do this with our mmap.
596 		 *
597 		 * Our mmap with MAP_STACK takes addr as the maximum
598 		 * downsize limit on BOS, and as len the max size of
599 		 * the region.  It then maps the top SGROWSIZ bytes,
600 		 * and auto grows the region down, up to the limit
601 		 * in addr.
602 		 *
603 		 * If we don't use the MAP_STACK option, the effect
604 		 * of this code is to allocate a stack region of a
605 		 * fixed size of (STACK_SIZE - GUARD_SIZE).
606 		 */
607 
608 		if ((caddr_t)PTRIN(addr) + len > p->p_vmspace->vm_maxsaddr) {
609 			/*
610 			 * Some Linux apps will attempt to mmap
611 			 * thread stacks near the top of their
612 			 * address space.  If their TOS is greater
613 			 * than vm_maxsaddr, vm_map_growstack()
614 			 * will confuse the thread stack with the
615 			 * process stack and deliver a SEGV if they
616 			 * attempt to grow the thread stack past their
617 			 * current stacksize rlimit.  To avoid this,
618 			 * adjust vm_maxsaddr upwards to reflect
619 			 * the current stacksize rlimit rather
620 			 * than the maximum possible stacksize.
621 			 * It would be better to adjust the
622 			 * mmap'ed region, but some apps do not check
623 			 * mmap's return value.
624 			 */
625 			PROC_LOCK(p);
626 			p->p_vmspace->vm_maxsaddr = (char *)LINUX32_USRSTACK -
627 			    lim_cur(p, RLIMIT_STACK);
628 			PROC_UNLOCK(p);
629 		}
630 
631 		/*
632 		 * This gives us our maximum stack size and a new BOS.
633 		 * If we're using VM_STACK, then mmap will just map
634 		 * the top SGROWSIZ bytes, and let the stack grow down
635 		 * to the limit at BOS.  If we're not using VM_STACK
636 		 * we map the full stack, since we don't have a way
637 		 * to autogrow it.
638 		 */
639 		if (len > STACK_SIZE - GUARD_SIZE) {
640 			bsd_args.addr = (caddr_t)PTRIN(addr);
641 			bsd_args.len = len;
642 		} else {
643 			bsd_args.addr = (caddr_t)PTRIN(addr) -
644 			    (STACK_SIZE - GUARD_SIZE - len);
645 			bsd_args.len = STACK_SIZE - GUARD_SIZE;
646 		}
647 	} else {
648 		bsd_args.addr = (caddr_t)PTRIN(addr);
649 		bsd_args.len  = len;
650 	}
651 	bsd_args.pos = pos;
652 
653 #ifdef DEBUG
654 	if (ldebug(mmap))
655 		printf("-> %s(%p, %d, %d, 0x%08x, %d, 0x%x)\n",
656 		    __func__,
657 		    (void *)bsd_args.addr, (int)bsd_args.len, bsd_args.prot,
658 		    bsd_args.flags, bsd_args.fd, (int)bsd_args.pos);
659 #endif
660 	error = mmap(td, &bsd_args);
661 #ifdef DEBUG
662 	if (ldebug(mmap))
663 		printf("-> %s() return: 0x%x (0x%08x)\n",
664 			__func__, error, (u_int)td->td_retval[0]);
665 #endif
666 	return (error);
667 }
668 
669 int
670 linux_mprotect(struct thread *td, struct linux_mprotect_args *uap)
671 {
672 	struct mprotect_args bsd_args;
673 
674 	bsd_args.addr = uap->addr;
675 	bsd_args.len = uap->len;
676 	bsd_args.prot = uap->prot;
677 	if (bsd_args.prot & (PROT_READ | PROT_WRITE | PROT_EXEC))
678 		bsd_args.prot |= PROT_READ | PROT_EXEC;
679 	return (mprotect(td, &bsd_args));
680 }
681 
682 int
683 linux_iopl(struct thread *td, struct linux_iopl_args *args)
684 {
685 	int error;
686 
687 	if (args->level < 0 || args->level > 3)
688 		return (EINVAL);
689 	if ((error = priv_check(td, PRIV_IO)) != 0)
690 		return (error);
691 	if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
692 		return (error);
693 	td->td_frame->tf_rflags = (td->td_frame->tf_rflags & ~PSL_IOPL) |
694 	    (args->level * (PSL_IOPL / 3));
695 
696 	return (0);
697 }
698 
699 int
700 linux_pipe(struct thread *td, struct linux_pipe_args *args)
701 {
702 	int error;
703 	int fildes[2];
704 
705 #ifdef DEBUG
706 	if (ldebug(pipe))
707 		printf(ARGS(pipe, "*"));
708 #endif
709 
710 	error = kern_pipe(td, fildes);
711 	if (error)
712 		return (error);
713 
714 	/* XXX: Close descriptors on error. */
715 	return (copyout(fildes, args->pipefds, sizeof fildes));
716 }
717 
718 int
719 linux_sigaction(struct thread *td, struct linux_sigaction_args *args)
720 {
721 	l_osigaction_t osa;
722 	l_sigaction_t act, oact;
723 	int error;
724 
725 #ifdef DEBUG
726 	if (ldebug(sigaction))
727 		printf(ARGS(sigaction, "%d, %p, %p"),
728 		    args->sig, (void *)args->nsa, (void *)args->osa);
729 #endif
730 
731 	if (args->nsa != NULL) {
732 		error = copyin(args->nsa, &osa, sizeof(l_osigaction_t));
733 		if (error)
734 			return (error);
735 		act.lsa_handler = osa.lsa_handler;
736 		act.lsa_flags = osa.lsa_flags;
737 		act.lsa_restorer = osa.lsa_restorer;
738 		LINUX_SIGEMPTYSET(act.lsa_mask);
739 		act.lsa_mask.__bits[0] = osa.lsa_mask;
740 	}
741 
742 	error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL,
743 	    args->osa ? &oact : NULL);
744 
745 	if (args->osa != NULL && !error) {
746 		osa.lsa_handler = oact.lsa_handler;
747 		osa.lsa_flags = oact.lsa_flags;
748 		osa.lsa_restorer = oact.lsa_restorer;
749 		osa.lsa_mask = oact.lsa_mask.__bits[0];
750 		error = copyout(&osa, args->osa, sizeof(l_osigaction_t));
751 	}
752 
753 	return (error);
754 }
755 
756 /*
757  * Linux has two extra args, restart and oldmask.  We don't use these,
758  * but it seems that "restart" is actually a context pointer that
759  * enables the signal to happen with a different register set.
760  */
761 int
762 linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args)
763 {
764 	sigset_t sigmask;
765 	l_sigset_t mask;
766 
767 #ifdef DEBUG
768 	if (ldebug(sigsuspend))
769 		printf(ARGS(sigsuspend, "%08lx"), (unsigned long)args->mask);
770 #endif
771 
772 	LINUX_SIGEMPTYSET(mask);
773 	mask.__bits[0] = args->mask;
774 	linux_to_bsd_sigset(&mask, &sigmask);
775 	return (kern_sigsuspend(td, sigmask));
776 }
777 
778 int
779 linux_rt_sigsuspend(struct thread *td, struct linux_rt_sigsuspend_args *uap)
780 {
781 	l_sigset_t lmask;
782 	sigset_t sigmask;
783 	int error;
784 
785 #ifdef DEBUG
786 	if (ldebug(rt_sigsuspend))
787 		printf(ARGS(rt_sigsuspend, "%p, %d"),
788 		    (void *)uap->newset, uap->sigsetsize);
789 #endif
790 
791 	if (uap->sigsetsize != sizeof(l_sigset_t))
792 		return (EINVAL);
793 
794 	error = copyin(uap->newset, &lmask, sizeof(l_sigset_t));
795 	if (error)
796 		return (error);
797 
798 	linux_to_bsd_sigset(&lmask, &sigmask);
799 	return (kern_sigsuspend(td, sigmask));
800 }
801 
802 int
803 linux_pause(struct thread *td, struct linux_pause_args *args)
804 {
805 	struct proc *p = td->td_proc;
806 	sigset_t sigmask;
807 
808 #ifdef DEBUG
809 	if (ldebug(pause))
810 		printf(ARGS(pause, ""));
811 #endif
812 
813 	PROC_LOCK(p);
814 	sigmask = td->td_sigmask;
815 	PROC_UNLOCK(p);
816 	return (kern_sigsuspend(td, sigmask));
817 }
818 
819 int
820 linux_sigaltstack(struct thread *td, struct linux_sigaltstack_args *uap)
821 {
822 	stack_t ss, oss;
823 	l_stack_t lss;
824 	int error;
825 
826 #ifdef DEBUG
827 	if (ldebug(sigaltstack))
828 		printf(ARGS(sigaltstack, "%p, %p"), uap->uss, uap->uoss);
829 #endif
830 
831 	if (uap->uss != NULL) {
832 		error = copyin(uap->uss, &lss, sizeof(l_stack_t));
833 		if (error)
834 			return (error);
835 
836 		ss.ss_sp = PTRIN(lss.ss_sp);
837 		ss.ss_size = lss.ss_size;
838 		ss.ss_flags = linux_to_bsd_sigaltstack(lss.ss_flags);
839 	}
840 	error = kern_sigaltstack(td, (uap->uss != NULL) ? &ss : NULL,
841 	    (uap->uoss != NULL) ? &oss : NULL);
842 	if (!error && uap->uoss != NULL) {
843 		lss.ss_sp = PTROUT(oss.ss_sp);
844 		lss.ss_size = oss.ss_size;
845 		lss.ss_flags = bsd_to_linux_sigaltstack(oss.ss_flags);
846 		error = copyout(&lss, uap->uoss, sizeof(l_stack_t));
847 	}
848 
849 	return (error);
850 }
851 
852 int
853 linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args)
854 {
855 	struct ftruncate_args sa;
856 
857 #ifdef DEBUG
858 	if (ldebug(ftruncate64))
859 		printf(ARGS(ftruncate64, "%u, %jd"), args->fd,
860 		    (intmax_t)args->length);
861 #endif
862 
863 	sa.fd = args->fd;
864 	sa.length = args->length;
865 	return ftruncate(td, &sa);
866 }
867 
868 int
869 linux_gettimeofday(struct thread *td, struct linux_gettimeofday_args *uap)
870 {
871 	struct timeval atv;
872 	l_timeval atv32;
873 	struct timezone rtz;
874 	int error = 0;
875 
876 	if (uap->tp) {
877 		microtime(&atv);
878 		atv32.tv_sec = atv.tv_sec;
879 		atv32.tv_usec = atv.tv_usec;
880 		error = copyout(&atv32, uap->tp, sizeof(atv32));
881 	}
882 	if (error == 0 && uap->tzp != NULL) {
883 		rtz.tz_minuteswest = tz_minuteswest;
884 		rtz.tz_dsttime = tz_dsttime;
885 		error = copyout(&rtz, uap->tzp, sizeof(rtz));
886 	}
887 	return (error);
888 }
889 
890 int
891 linux_settimeofday(struct thread *td, struct linux_settimeofday_args *uap)
892 {
893 	l_timeval atv32;
894 	struct timeval atv, *tvp;
895 	struct timezone atz, *tzp;
896 	int error;
897 
898 	if (uap->tp) {
899 		error = copyin(uap->tp, &atv32, sizeof(atv32));
900 		if (error)
901 			return (error);
902 		atv.tv_sec = atv32.tv_sec;
903 		atv.tv_usec = atv32.tv_usec;
904 		tvp = &atv;
905 	} else
906 		tvp = NULL;
907 	if (uap->tzp) {
908 		error = copyin(uap->tzp, &atz, sizeof(atz));
909 		if (error)
910 			return (error);
911 		tzp = &atz;
912 	} else
913 		tzp = NULL;
914 	return (kern_settimeofday(td, tvp, tzp));
915 }
916 
917 int
918 linux_getrusage(struct thread *td, struct linux_getrusage_args *uap)
919 {
920 	struct l_rusage s32;
921 	struct rusage s;
922 	int error;
923 
924 	error = kern_getrusage(td, uap->who, &s);
925 	if (error != 0)
926 		return (error);
927 	if (uap->rusage != NULL) {
928 		bsd_to_linux_rusage(&s, &s32);
929 		error = copyout(&s32, uap->rusage, sizeof(s32));
930 	}
931 	return (error);
932 }
933 
934 int
935 linux_sched_rr_get_interval(struct thread *td,
936     struct linux_sched_rr_get_interval_args *uap)
937 {
938 	struct timespec ts;
939 	struct l_timespec ts32;
940 	int error;
941 
942 	error = kern_sched_rr_get_interval(td, uap->pid, &ts);
943 	if (error != 0)
944 		return (error);
945 	ts32.tv_sec = ts.tv_sec;
946 	ts32.tv_nsec = ts.tv_nsec;
947 	return (copyout(&ts32, uap->interval, sizeof(ts32)));
948 }
949 
950 int
951 linux_set_thread_area(struct thread *td,
952     struct linux_set_thread_area_args *args)
953 {
954 	struct l_user_desc info;
955 	struct user_segment_descriptor sd;
956 	struct pcb *pcb;
957 	int a[2];
958 	int error;
959 
960 	error = copyin(args->desc, &info, sizeof(struct l_user_desc));
961 	if (error)
962 		return (error);
963 
964 #ifdef DEBUG
965 	if (ldebug(set_thread_area))
966 		printf(ARGS(set_thread_area, "%i, %x, %x, %i, %i, %i, "
967 		    "%i, %i, %i"), info.entry_number, info.base_addr,
968 		    info.limit, info.seg_32bit, info.contents,
969 		    info.read_exec_only, info.limit_in_pages,
970 		    info.seg_not_present, info.useable);
971 #endif
972 
973 	/*
974 	 * Semantics of Linux version: every thread in the system has array
975 	 * of three TLS descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown.
976 	 * This syscall loads one of the selected TLS decriptors with a value
977 	 * and also loads GDT descriptors 6, 7 and 8 with the content of
978 	 * the per-thread descriptors.
979 	 *
980 	 * Semantics of FreeBSD version: I think we can ignore that Linux has
981 	 * three per-thread descriptors and use just the first one.
982 	 * The tls_array[] is used only in [gs]et_thread_area() syscalls and
983 	 * for loading the GDT descriptors. We use just one GDT descriptor
984 	 * for TLS, so we will load just one.
985 	 *
986 	 * XXX: This doesn't work when a user space process tries to use more
987 	 * than one TLS segment. Comment in the Linux source says wine might
988 	 * do this.
989 	 */
990 
991 	/*
992 	 * GLIBC reads current %gs and call set_thread_area() with it.
993 	 * We should let GUDATA_SEL and GUGS32_SEL proceed as well because
994 	 * we use these segments.
995 	 */
996 	switch (info.entry_number) {
997 	case GUGS32_SEL:
998 	case GUDATA_SEL:
999 	case 6:
1000 	case -1:
1001 		info.entry_number = GUGS32_SEL;
1002 		break;
1003 	default:
1004 		return (EINVAL);
1005 	}
1006 
1007 	/*
1008 	 * We have to copy out the GDT entry we use.
1009 	 *
1010 	 * XXX: What if a user space program does not check the return value
1011 	 * and tries to use 6, 7 or 8?
1012 	 */
1013 	error = copyout(&info, args->desc, sizeof(struct l_user_desc));
1014 	if (error)
1015 		return (error);
1016 
1017 	if (LINUX_LDT_empty(&info)) {
1018 		a[0] = 0;
1019 		a[1] = 0;
1020 	} else {
1021 		a[0] = LINUX_LDT_entry_a(&info);
1022 		a[1] = LINUX_LDT_entry_b(&info);
1023 	}
1024 
1025 	memcpy(&sd, &a, sizeof(a));
1026 #ifdef DEBUG
1027 	if (ldebug(set_thread_area))
1028 		printf("Segment created in set_thread_area: "
1029 		    "lobase: %x, hibase: %x, lolimit: %x, hilimit: %x, "
1030 		    "type: %i, dpl: %i, p: %i, xx: %i, long: %i, "
1031 		    "def32: %i, gran: %i\n",
1032 		    sd.sd_lobase,
1033 		    sd.sd_hibase,
1034 		    sd.sd_lolimit,
1035 		    sd.sd_hilimit,
1036 		    sd.sd_type,
1037 		    sd.sd_dpl,
1038 		    sd.sd_p,
1039 		    sd.sd_xx,
1040 		    sd.sd_long,
1041 		    sd.sd_def32,
1042 		    sd.sd_gran);
1043 #endif
1044 
1045 	pcb = td->td_pcb;
1046 	pcb->pcb_gsbase = (register_t)info.base_addr;
1047 	set_pcb_flags(pcb, PCB_32BIT | PCB_GS32BIT);
1048 	update_gdt_gsbase(td, info.base_addr);
1049 
1050 	return (0);
1051 }
1052 
1053 int
1054 linux_wait4(struct thread *td, struct linux_wait4_args *args)
1055 {
1056 	int error, options;
1057 	struct rusage ru, *rup;
1058 	struct l_rusage lru;
1059 
1060 #ifdef DEBUG
1061 	if (ldebug(wait4))
1062 		printf(ARGS(wait4, "%d, %p, %d, %p"),
1063 		    args->pid, (void *)args->status, args->options,
1064 		    (void *)args->rusage);
1065 #endif
1066 
1067 	options = (args->options & (WNOHANG | WUNTRACED));
1068 	/* WLINUXCLONE should be equal to __WCLONE, but we make sure */
1069 	if (args->options & __WCLONE)
1070 		options |= WLINUXCLONE;
1071 
1072 	if (args->rusage != NULL)
1073 		rup = &ru;
1074 	else
1075 		rup = NULL;
1076 	error = linux_common_wait(td, args->pid, args->status, options, rup);
1077 	if (error)
1078 		return (error);
1079 	if (args->rusage != NULL) {
1080 		bsd_to_linux_rusage(rup, &lru);
1081 		error = copyout(&lru, args->rusage, sizeof(lru));
1082 	}
1083 
1084 	return (error);
1085 }
1086