xref: /netbsd-src/sys/compat/linux/common/linux_misc.c (revision 93f9db1b75d415b78f73ed629beeb86235153473)
1 /*	$NetBSD: linux_misc.c,v 1.51 1998/10/04 10:17:54 fvdl Exp $	*/
2 
3 /*-
4  * Copyright (c) 1995, 1998 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Frank van der Linden and Eric Haszlakiewicz.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Linux compatibility module. Try to deal with various Linux system calls.
41  */
42 
43 /*
44  * These functions have been moved to multiarch to allow
45  * selection of which machines include them to be
46  * determined by the individual files.linux_<arch> files.
47  *
48  * Function in multiarch:
49  *	linux_sys_break			: linux_break.c
50  *	linux_sys_alarm			: linux_misc_notalpha.c
51  *	linux_sys_nice			: linux_misc_notalpha.c
52  *	linux_sys_readdir		: linux_misc_notalpha.c
53  *	linux_sys_time			: linux_misc_notalpha.c
54  *	linux_sys_utime			: linux_misc_notalpha.c
55  *	linux_sys_waitpid		: linux_misc_notalpha.c
56  *	linux_sys_old_mmap		: linux_oldmmap.c
57  *	linux_sys_oldolduname		: linux_oldolduname.c
58  *	linux_sys_oldselect		: linux_oldselect.c
59  *	linux_sys_olduname		: linux_olduname.c
60  *	linux_sys_pipe			: linux_pipe.c
61  */
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/namei.h>
66 #include <sys/proc.h>
67 #include <sys/dirent.h>
68 #include <sys/file.h>
69 #include <sys/stat.h>
70 #include <sys/filedesc.h>
71 #include <sys/ioctl.h>
72 #include <sys/kernel.h>
73 #include <sys/malloc.h>
74 #include <sys/mbuf.h>
75 #include <sys/mman.h>
76 #include <sys/mount.h>
77 #include <sys/ptrace.h>
78 #include <sys/resource.h>
79 #include <sys/resourcevar.h>
80 #include <sys/signal.h>
81 #include <sys/signalvar.h>
82 #include <sys/socket.h>
83 #include <sys/time.h>
84 #include <sys/times.h>
85 #include <sys/vnode.h>
86 #include <sys/uio.h>
87 #include <sys/wait.h>
88 #include <sys/utsname.h>
89 #include <sys/unistd.h>
90 
91 #include <sys/syscallargs.h>
92 
93 #include <vm/vm.h>
94 #include <vm/vm_param.h>
95 
96 #include <compat/linux/common/linux_types.h>
97 #include <compat/linux/common/linux_signal.h>
98 
99 #include <compat/linux/linux_syscallargs.h>
100 
101 #include <compat/linux/common/linux_fcntl.h>
102 #include <compat/linux/common/linux_mmap.h>
103 #include <compat/linux/common/linux_dirent.h>
104 #include <compat/linux/common/linux_util.h>
105 #include <compat/linux/common/linux_misc.h>
106 
107 
108 /* Local linux_misc.c functions: */
109 static void bsd_to_linux_statfs __P((struct statfs *, struct linux_statfs *));
110 
111 /*
112  * The information on a terminated (or stopped) process needs
113  * to be converted in order for Linux binaries to get a valid signal
114  * number out of it.
115  */
116 void
117 bsd_to_linux_wstat(status)
118 	int *status;
119 {
120 
121 	if (WIFSIGNALED(*status))
122 		*status = (*status & ~0177) |
123 		    native_to_linux_sig[WTERMSIG(*status)];
124 	else if (WIFSTOPPED(*status))
125 		*status = (*status & ~0xff00) |
126 		    (native_to_linux_sig[WSTOPSIG(*status)] << 8);
127 }
128 
129 /*
130  * This is very much the same as waitpid()
131  */
132 int
133 linux_sys_wait4(p, v, retval)
134 	struct proc *p;
135 	void *v;
136 	register_t *retval;
137 {
138 	struct linux_sys_wait4_args /* {
139 		syscallarg(int) pid;
140 		syscallarg(int *) status;
141 		syscallarg(int) options;
142 		syscallarg(struct rusage *) rusage;
143 	} */ *uap = v;
144 	struct sys_wait4_args w4a;
145 	int error, *status, tstat;
146 	caddr_t sg;
147 
148 	if (SCARG(uap, status) != NULL) {
149 		sg = stackgap_init(p->p_emul);
150 		status = (int *) stackgap_alloc(&sg, sizeof status);
151 	} else
152 		status = NULL;
153 
154 	SCARG(&w4a, pid) = SCARG(uap, pid);
155 	SCARG(&w4a, status) = status;
156 	SCARG(&w4a, options) = SCARG(uap, options);
157 	SCARG(&w4a, rusage) = SCARG(uap, rusage);
158 
159 	if ((error = sys_wait4(p, &w4a, retval)))
160 		return error;
161 
162 	sigdelset(&p->p_siglist, SIGCHLD);
163 
164 	if (status != NULL) {
165 		if ((error = copyin(status, &tstat, sizeof tstat)))
166 			return error;
167 
168 		bsd_to_linux_wstat(&tstat);
169 		return copyout(&tstat, SCARG(uap, status), sizeof tstat);
170 	}
171 
172 	return 0;
173 }
174 
175 /*
176  * Linux brk(2). The check if the new address is >= the old one is
177  * done in the kernel in Linux. NetBSD does it in the library.
178  */
179 int
180 linux_sys_brk(p, v, retval)
181 	struct proc *p;
182 	void *v;
183 	register_t *retval;
184 {
185 	struct linux_sys_brk_args /* {
186 		syscallarg(char *) nsize;
187 	} */ *uap = v;
188 	char *nbrk = SCARG(uap, nsize);
189 	struct sys_obreak_args oba;
190 	struct vmspace *vm = p->p_vmspace;
191 	caddr_t oldbrk;
192 
193 	oldbrk = vm->vm_daddr + ctob(vm->vm_dsize);
194 	/*
195 	 * XXX inconsistent.. Linux always returns at least the old
196 	 * brk value, but it will be page-aligned if this fails,
197 	 * and possibly not page aligned if it succeeds (the user
198 	 * supplied pointer is returned).
199 	 */
200 	SCARG(&oba, nsize) = nbrk;
201 
202 	if ((caddr_t) nbrk > vm->vm_daddr && sys_obreak(p, &oba, retval) == 0)
203 		retval[0] = (register_t)nbrk;
204 	else
205 		retval[0] = (register_t)oldbrk;
206 
207 	return 0;
208 }
209 
210 /*
211  * Convert BSD statfs structure to Linux statfs structure.
212  * The Linux structure has less fields, and it also wants
213  * the length of a name in a dir entry in a field, which
214  * we fake (probably the wrong way).
215  */
216 static void
217 bsd_to_linux_statfs(bsp, lsp)
218 	struct statfs *bsp;
219 	struct linux_statfs *lsp;
220 {
221 
222 	lsp->l_ftype = bsp->f_type;
223 	lsp->l_fbsize = bsp->f_bsize;
224 	lsp->l_fblocks = bsp->f_blocks;
225 	lsp->l_fbfree = bsp->f_bfree;
226 	lsp->l_fbavail = bsp->f_bavail;
227 	lsp->l_ffiles = bsp->f_files;
228 	lsp->l_fffree = bsp->f_ffree;
229 	lsp->l_ffsid.val[0] = bsp->f_fsid.val[0];
230 	lsp->l_ffsid.val[1] = bsp->f_fsid.val[1];
231 	lsp->l_fnamelen = MAXNAMLEN;	/* XXX */
232 }
233 
234 /*
235  * Implement the fs stat functions. Straightforward.
236  */
237 int
238 linux_sys_statfs(p, v, retval)
239 	struct proc *p;
240 	void *v;
241 	register_t *retval;
242 {
243 	struct linux_sys_statfs_args /* {
244 		syscallarg(char *) path;
245 		syscallarg(struct linux_statfs *) sp;
246 	} */ *uap = v;
247 	struct statfs btmp, *bsp;
248 	struct linux_statfs ltmp;
249 	struct sys_statfs_args bsa;
250 	caddr_t sg;
251 	int error;
252 
253 	sg = stackgap_init(p->p_emul);
254 	bsp = (struct statfs *) stackgap_alloc(&sg, sizeof (struct statfs));
255 
256 	LINUX_CHECK_ALT_EXIST(p, &sg, SCARG(uap, path));
257 
258 	SCARG(&bsa, path) = SCARG(uap, path);
259 	SCARG(&bsa, buf) = bsp;
260 
261 	if ((error = sys_statfs(p, &bsa, retval)))
262 		return error;
263 
264 	if ((error = copyin((caddr_t) bsp, (caddr_t) &btmp, sizeof btmp)))
265 		return error;
266 
267 	bsd_to_linux_statfs(&btmp, &ltmp);
268 
269 	return copyout((caddr_t) &ltmp, (caddr_t) SCARG(uap, sp), sizeof ltmp);
270 }
271 
272 int
273 linux_sys_fstatfs(p, v, retval)
274 	struct proc *p;
275 	void *v;
276 	register_t *retval;
277 {
278 	struct linux_sys_fstatfs_args /* {
279 		syscallarg(int) fd;
280 		syscallarg(struct linux_statfs *) sp;
281 	} */ *uap = v;
282 	struct statfs btmp, *bsp;
283 	struct linux_statfs ltmp;
284 	struct sys_fstatfs_args bsa;
285 	caddr_t sg;
286 	int error;
287 
288 	sg = stackgap_init(p->p_emul);
289 	bsp = (struct statfs *) stackgap_alloc(&sg, sizeof (struct statfs));
290 
291 	SCARG(&bsa, fd) = SCARG(uap, fd);
292 	SCARG(&bsa, buf) = bsp;
293 
294 	if ((error = sys_fstatfs(p, &bsa, retval)))
295 		return error;
296 
297 	if ((error = copyin((caddr_t) bsp, (caddr_t) &btmp, sizeof btmp)))
298 		return error;
299 
300 	bsd_to_linux_statfs(&btmp, &ltmp);
301 
302 	return copyout((caddr_t) &ltmp, (caddr_t) SCARG(uap, sp), sizeof ltmp);
303 }
304 
305 /*
306  * uname(). Just copy the info from the various strings stored in the
307  * kernel, and put it in the Linux utsname structure. That structure
308  * is almost the same as the NetBSD one, only it has fields 65 characters
309  * long, and an extra domainname field.
310  */
311 int
312 linux_sys_uname(p, v, retval)
313 	struct proc *p;
314 	void *v;
315 	register_t *retval;
316 {
317 	struct linux_sys_uname_args /* {
318 		syscallarg(struct linux_utsname *) up;
319 	} */ *uap = v;
320 	extern char ostype[], hostname[], osrelease[], version[], machine[],
321 	    domainname[];
322 	struct linux_utsname luts;
323 	int len;
324 	char *cp;
325 
326 	strncpy(luts.l_sysname, ostype, sizeof(luts.l_sysname));
327 	strncpy(luts.l_nodename, hostname, sizeof(luts.l_nodename));
328 	strncpy(luts.l_release, osrelease, sizeof(luts.l_release));
329 	strncpy(luts.l_version, version, sizeof(luts.l_version));
330 	strncpy(luts.l_machine, machine, sizeof(luts.l_machine));
331 	strncpy(luts.l_domainname, domainname, sizeof(luts.l_domainname));
332 
333 	/* This part taken from the the uname() in libc */
334 	len = sizeof(luts.l_version);
335 	for (cp = luts.l_version; len--; ++cp) {
336 		if (*cp == '\n' || *cp == '\t') {
337 			if (len > 1)
338 				*cp = ' ';
339 			else
340 				*cp = '\0';
341 		}
342 	}
343 
344 	return copyout(&luts, SCARG(uap, up), sizeof(luts));
345 }
346 
347 /* Used directly on: alpha, mips, ppc, sparc, sparc64 */
348 /* Used indirectly on: arm, i386, m68k */
349 
350 /*
351  * New type Linux mmap call.
352  * Only called directly on machines with >= 6 free regs.
353  */
354 int
355 linux_sys_mmap(p, v, retval)
356 	struct proc *p;
357 	void *v;
358 	register_t *retval;
359 {
360 	struct linux_sys_mmap_args /* {
361 		syscallarg(unsigned long) addr;
362 		syscallarg(size_t) len;
363 		syscallarg(int) prot;
364 		syscallarg(int) flags;
365 		syscallarg(int) fd;
366 		syscallarg(off_t) offset;
367 	} */ *uap = v;
368 	struct sys_mmap_args cma;
369 	int flags;
370 
371 	flags = 0;
372 	flags |= cvtto_bsd_mask(SCARG(uap,flags), LINUX_MAP_SHARED, MAP_SHARED);
373 	flags |= cvtto_bsd_mask(SCARG(uap,flags), LINUX_MAP_PRIVATE, MAP_PRIVATE);
374 	flags |= cvtto_bsd_mask(SCARG(uap,flags), LINUX_MAP_FIXED, MAP_FIXED);
375 	flags |= cvtto_bsd_mask(SCARG(uap,flags), LINUX_MAP_ANON, MAP_ANON);
376 	/* XXX XAX ERH: Any other flags here?  There are more defined... */
377 
378 	SCARG(&cma,addr) = (void *)SCARG(uap, addr);
379 	SCARG(&cma,len) = SCARG(uap, len);
380 	SCARG(&cma,prot) = SCARG(uap, prot);
381 	if (SCARG(&cma,prot) & VM_PROT_WRITE) /* XXX */
382 		SCARG(&cma,prot) |= VM_PROT_READ;
383 	SCARG(&cma,flags) = flags;
384 	SCARG(&cma,fd) = SCARG(uap, fd);
385 	SCARG(&cma,pad) = 0;
386 	SCARG(&cma,pos) = SCARG(uap, offset);
387 
388 	return sys_mmap(p, &cma, retval);
389 }
390 
391 int
392 linux_sys_mremap(p, v, retval)
393 	struct proc *p;
394 	void *v;
395 	register_t *retval;
396 {
397 	struct linux_sys_mremap_args /* {
398 		syscallarg(void *) old_address;
399 		syscallarg(size_t) old_size;
400 		syscallarg(size_t) new_size;
401 		syscallarg(u_long) flags;
402 	} */ *uap = v;
403 	struct sys_munmap_args mua;
404 	size_t old_size, new_size;
405 	int error;
406 
407 	old_size = round_page(SCARG(uap, old_size));
408 	new_size = round_page(SCARG(uap, new_size));
409 
410 	/*
411 	 * Growing mapped region.
412 	 */
413 	if (new_size > old_size) {
414 		/*
415 		 * XXX Implement me.  What we probably want to do is
416 		 * XXX dig out the guts of the old mapping, mmap that
417 		 * XXX object again with the new size, then munmap
418 		 * XXX the old mapping.
419 		 */
420 		*retval = 0;
421 		return (ENOMEM);
422 	}
423 
424 	/*
425 	 * Shrinking mapped region.
426 	 */
427 	if (new_size < old_size) {
428 		SCARG(&mua, addr) = (caddr_t)SCARG(uap, old_address) +
429 		    new_size;
430 		SCARG(&mua, len) = old_size - new_size;
431 		error = sys_munmap(p, &mua, retval);
432 		*retval = error ? 0 : (register_t)SCARG(uap, old_address);
433 		return (error);
434 	}
435 
436 	/*
437 	 * No change.
438 	 */
439 	*retval = (register_t)SCARG(uap, old_address);
440 	return (0);
441 }
442 
443 int
444 linux_sys_msync(p, v, retval)
445 	struct proc *p;
446 	void *v;
447 	register_t *retval;
448 {
449 	struct linux_sys_msync_args /* {
450 		syscallarg(caddr_t) addr;
451 		syscallarg(int) len;
452 		syscallarg(int) fl;
453 	} */ *uap = v;
454 
455 	struct sys___msync13_args bma;
456 
457 	/* flags are ignored */
458 	SCARG(&bma, addr) = SCARG(uap, addr);
459 	SCARG(&bma, len) = SCARG(uap, len);
460 	SCARG(&bma, flags) = SCARG(uap, fl);
461 
462 	return sys___msync13(p, &bma, retval);
463 }
464 
465 /*
466  * This code is partly stolen from src/lib/libc/compat-43/times.c
467  * XXX - CLK_TCK isn't declared in /sys, just in <time.h>, done here
468  */
469 
470 #define CLK_TCK 100
471 #define	CONVTCK(r)	(r.tv_sec * CLK_TCK + r.tv_usec / (1000000 / CLK_TCK))
472 
473 int
474 linux_sys_times(p, v, retval)
475 	struct proc *p;
476 	void *v;
477 	register_t *retval;
478 {
479 	struct linux_sys_times_args /* {
480 		syscallarg(struct times *) tms;
481 	} */ *uap = v;
482 	struct timeval t;
483 	struct linux_tms ltms;
484 	struct rusage ru;
485 	int error, s;
486 
487 	calcru(p, &ru.ru_utime, &ru.ru_stime, NULL);
488 	ltms.ltms_utime = CONVTCK(ru.ru_utime);
489 	ltms.ltms_stime = CONVTCK(ru.ru_stime);
490 
491 	ltms.ltms_cutime = CONVTCK(p->p_stats->p_cru.ru_utime);
492 	ltms.ltms_cstime = CONVTCK(p->p_stats->p_cru.ru_stime);
493 
494 	if ((error = copyout(&ltms, SCARG(uap, tms), sizeof ltms)))
495 		return error;
496 
497 	s = splclock();
498 	timersub(&time, &boottime, &t);
499 	splx(s);
500 
501 	retval[0] = ((linux_clock_t)(CONVTCK(t)));
502 	return 0;
503 }
504 
505 /*
506  * Linux 'readdir' call. This code is mostly taken from the
507  * SunOS getdents call (see compat/sunos/sunos_misc.c), though
508  * an attempt has been made to keep it a little cleaner (failing
509  * miserably, because of the cruft needed if count 1 is passed).
510  *
511  * The d_off field should contain the offset of the next valid entry,
512  * but in Linux it has the offset of the entry itself. We emulate
513  * that bug here.
514  *
515  * Read in BSD-style entries, convert them, and copy them out.
516  *
517  * Note that this doesn't handle union-mounted filesystems.
518  */
519 int
520 linux_sys_getdents(p, v, retval)
521 	struct proc *p;
522 	void *v;
523 	register_t *retval;
524 {
525 	struct linux_sys_getdents_args /* {
526 		syscallarg(int) fd;
527 		syscallarg(struct linux_dirent *) dent;
528 		syscallarg(unsigned int) count;
529 	} */ *uap = v;
530 	register struct dirent *bdp;
531 	struct vnode *vp;
532 	caddr_t	inp, buf;		/* BSD-format */
533 	int len, reclen;		/* BSD-format */
534 	caddr_t outp;			/* Linux-format */
535 	int resid, linux_reclen = 0;	/* Linux-format */
536 	struct file *fp;
537 	struct uio auio;
538 	struct iovec aiov;
539 	struct linux_dirent idb;
540 	off_t off;		/* true file offset */
541 	int buflen, error, eofflag, nbytes, oldcall;
542 	struct vattr va;
543 	off_t *cookiebuf = NULL, *cookie;
544 	int ncookies;
545 
546 	if ((error = getvnode(p->p_fd, SCARG(uap, fd), &fp)) != 0)
547 		return (error);
548 
549 	if ((fp->f_flag & FREAD) == 0)
550 		return (EBADF);
551 
552 	vp = (struct vnode *)fp->f_data;
553 	if (vp->v_type != VDIR)
554 		return (EINVAL);
555 
556 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)))
557 		return error;
558 
559 	nbytes = SCARG(uap, count);
560 	if (nbytes == 1) {	/* emulating old, broken behaviour */
561 		nbytes = sizeof (struct linux_dirent);
562 		buflen = max(va.va_blocksize, nbytes);
563 		oldcall = 1;
564 	} else {
565 		buflen = min(MAXBSIZE, nbytes);
566 		if (buflen < va.va_blocksize)
567 			buflen = va.va_blocksize;
568 		oldcall = 0;
569 	}
570 	buf = malloc(buflen, M_TEMP, M_WAITOK);
571 
572 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
573 	off = fp->f_offset;
574 again:
575 	aiov.iov_base = buf;
576 	aiov.iov_len = buflen;
577 	auio.uio_iov = &aiov;
578 	auio.uio_iovcnt = 1;
579 	auio.uio_rw = UIO_READ;
580 	auio.uio_segflg = UIO_SYSSPACE;
581 	auio.uio_procp = p;
582 	auio.uio_resid = buflen;
583 	auio.uio_offset = off;
584 	/*
585          * First we read into the malloc'ed buffer, then
586          * we massage it into user space, one record at a time.
587          */
588 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, &cookiebuf,
589 	    &ncookies);
590 	if (error)
591 		goto out;
592 
593 	inp = buf;
594 	outp = (caddr_t)SCARG(uap, dent);
595 	resid = nbytes;
596 	if ((len = buflen - auio.uio_resid) == 0)
597 		goto eof;
598 
599 	for (cookie = cookiebuf; len > 0; len -= reclen) {
600 		bdp = (struct dirent *)inp;
601 		reclen = bdp->d_reclen;
602 		if (reclen & 3)
603 			panic("linux_readdir");
604 		if (bdp->d_fileno == 0) {
605 			inp += reclen;	/* it is a hole; squish it out */
606 			off = *cookie++;
607 			continue;
608 		}
609 		linux_reclen = LINUX_RECLEN(&idb, bdp->d_namlen);
610 		if (reclen > len || resid < linux_reclen) {
611 			/* entry too big for buffer, so just stop */
612 			outp++;
613 			break;
614 		}
615 		/*
616 		 * Massage in place to make a Linux-shaped dirent (otherwise
617 		 * we have to worry about touching user memory outside of
618 		 * the copyout() call).
619 		 */
620 		idb.d_ino = (linux_ino_t)bdp->d_fileno;
621 		/*
622 		 * The old readdir() call misuses the offset and reclen fields.
623 		 */
624 		if (oldcall) {
625 			idb.d_off = (linux_off_t)linux_reclen;
626 			idb.d_reclen = (u_short)bdp->d_namlen;
627 		} else {
628 			if (sizeof (linux_off_t) < 4 && (off >> 32) != 0) {
629 				compat_offseterr(vp, "linux_getdents");
630 				error = EINVAL;
631 				goto out;
632 			}
633 			idb.d_off = (linux_off_t)off;
634 			idb.d_reclen = (u_short)linux_reclen;
635 		}
636 		strcpy(idb.d_name, bdp->d_name);
637 		if ((error = copyout((caddr_t)&idb, outp, linux_reclen)))
638 			goto out;
639 		/* advance past this real entry */
640 		inp += reclen;
641 		off = *cookie++;	/* each entry points to itself */
642 		/* advance output past Linux-shaped entry */
643 		outp += linux_reclen;
644 		resid -= linux_reclen;
645 		if (oldcall)
646 			break;
647 	}
648 
649 	/* if we squished out the whole block, try again */
650 	if (outp == (caddr_t)SCARG(uap, dent))
651 		goto again;
652 	fp->f_offset = off;	/* update the vnode offset */
653 
654 	if (oldcall)
655 		nbytes = resid + linux_reclen;
656 
657 eof:
658 	*retval = nbytes - resid;
659 out:
660 	VOP_UNLOCK(vp, 0);
661 	if (cookiebuf)
662 		free(cookiebuf, M_TEMP);
663 	free(buf, M_TEMP);
664 	return error;
665 }
666 
667 /*
668  * Even when just using registers to pass arguments to syscalls you can
669  * have 5 of them on the i386. So this newer version of select() does
670  * this.
671  */
672 int
673 linux_sys_select(p, v, retval)
674 	struct proc *p;
675 	void *v;
676 	register_t *retval;
677 {
678 	struct linux_sys_select_args /* {
679 		syscallarg(int) nfds;
680 		syscallarg(fd_set *) readfds;
681 		syscallarg(fd_set *) writefds;
682 		syscallarg(fd_set *) exceptfds;
683 		syscallarg(struct timeval *) timeout;
684 	} */ *uap = v;
685 
686 	return linux_select1(p, retval, SCARG(uap, nfds), SCARG(uap, readfds),
687 	    SCARG(uap, writefds), SCARG(uap, exceptfds), SCARG(uap, timeout));
688 }
689 
690 /*
691  * Common code for the old and new versions of select(). A couple of
692  * things are important:
693  * 1) return the amount of time left in the 'timeout' parameter
694  * 2) select never returns ERESTART on Linux, always return EINTR
695  */
696 int
697 linux_select1(p, retval, nfds, readfds, writefds, exceptfds, timeout)
698 	struct proc *p;
699 	register_t *retval;
700 	int nfds;
701 	fd_set *readfds, *writefds, *exceptfds;
702 	struct timeval *timeout;
703 {
704 	struct sys_select_args bsa;
705 	struct timeval tv0, tv1, utv, *tvp;
706 	caddr_t sg;
707 	int error;
708 
709 	SCARG(&bsa, nd) = nfds;
710 	SCARG(&bsa, in) = readfds;
711 	SCARG(&bsa, ou) = writefds;
712 	SCARG(&bsa, ex) = exceptfds;
713 	SCARG(&bsa, tv) = timeout;
714 
715 	/*
716 	 * Store current time for computation of the amount of
717 	 * time left.
718 	 */
719 	if (timeout) {
720 		if ((error = copyin(timeout, &utv, sizeof(utv))))
721 			return error;
722 		if (itimerfix(&utv)) {
723 			/*
724 			 * The timeval was invalid.  Convert it to something
725 			 * valid that will act as it does under Linux.
726 			 */
727 			sg = stackgap_init(p->p_emul);
728 			tvp = stackgap_alloc(&sg, sizeof(utv));
729 			utv.tv_sec += utv.tv_usec / 1000000;
730 			utv.tv_usec %= 1000000;
731 			if (utv.tv_usec < 0) {
732 				utv.tv_sec -= 1;
733 				utv.tv_usec += 1000000;
734 			}
735 			if (utv.tv_sec < 0)
736 				timerclear(&utv);
737 			if ((error = copyout(&utv, tvp, sizeof(utv))))
738 				return error;
739 			SCARG(&bsa, tv) = tvp;
740 		}
741 		microtime(&tv0);
742 	}
743 
744 	error = sys_select(p, &bsa, retval);
745 	if (error) {
746 		/*
747 		 * See fs/select.c in the Linux kernel.  Without this,
748 		 * Maelstrom doesn't work.
749 		 */
750 		if (error == ERESTART)
751 			error = EINTR;
752 		return error;
753 	}
754 
755 	if (timeout) {
756 		if (*retval) {
757 			/*
758 			 * Compute how much time was left of the timeout,
759 			 * by subtracting the current time and the time
760 			 * before we started the call, and subtracting
761 			 * that result from the user-supplied value.
762 			 */
763 			microtime(&tv1);
764 			timersub(&tv1, &tv0, &tv1);
765 			timersub(&utv, &tv1, &utv);
766 			if (utv.tv_sec < 0)
767 				timerclear(&utv);
768 		} else
769 			timerclear(&utv);
770 		if ((error = copyout(&utv, timeout, sizeof(utv))))
771 			return error;
772 	}
773 
774 	return 0;
775 }
776 
777 /*
778  * Get the process group of a certain process. Look it up
779  * and return the value.
780  */
781 int
782 linux_sys_getpgid(p, v, retval)
783 	struct proc *p;
784 	void *v;
785 	register_t *retval;
786 {
787 	struct linux_sys_getpgid_args /* {
788 		syscallarg(int) pid;
789 	} */ *uap = v;
790 	struct proc *targp;
791 
792 	if (SCARG(uap, pid) != 0 && SCARG(uap, pid) != p->p_pid) {
793 		if ((targp = pfind(SCARG(uap, pid))) == 0)
794 			return ESRCH;
795 	}
796 	else
797 		targp = p;
798 
799 	retval[0] = targp->p_pgid;
800 	return 0;
801 }
802 
803 /*
804  * Set the 'personality' (emulation mode) for the current process. Only
805  * accept the Linux personality here (0). This call is needed because
806  * the Linux ELF crt0 issues it in an ugly kludge to make sure that
807  * ELF binaries run in Linux mode, not SVR4 mode.
808  */
809 int
810 linux_sys_personality(p, v, retval)
811 	struct proc *p;
812 	void *v;
813 	register_t *retval;
814 {
815 	struct linux_sys_personality_args /* {
816 		syscallarg(int) per;
817 	} */ *uap = v;
818 
819 	if (SCARG(uap, per) != 0)
820 		return EINVAL;
821 	retval[0] = 0;
822 	return 0;
823 }
824 
825 /*
826  * The calls are here because of type conversions.
827  */
828 int
829 linux_sys_setreuid(p, v, retval)
830 	struct proc *p;
831 	void *v;
832 	register_t *retval;
833 {
834 	struct linux_sys_setreuid_args /* {
835 		syscallarg(int) ruid;
836 		syscallarg(int) euid;
837 	} */ *uap = v;
838 	struct sys_setreuid_args bsa;
839 
840 	SCARG(&bsa, ruid) = ((linux_uid_t)SCARG(uap, ruid) == (linux_uid_t)-1) ?
841 		(uid_t)-1 : SCARG(uap, ruid);
842 	SCARG(&bsa, euid) = ((linux_uid_t)SCARG(uap, euid) == (linux_uid_t)-1) ?
843 		(uid_t)-1 : SCARG(uap, euid);
844 
845 	return sys_setreuid(p, &bsa, retval);
846 }
847 
848 int
849 linux_sys_setregid(p, v, retval)
850 	struct proc *p;
851 	void *v;
852 	register_t *retval;
853 {
854 	struct linux_sys_setregid_args /* {
855 		syscallarg(int) rgid;
856 		syscallarg(int) egid;
857 	} */ *uap = v;
858 	struct sys_setregid_args bsa;
859 
860 	SCARG(&bsa, rgid) = ((linux_gid_t)SCARG(uap, rgid) == (linux_gid_t)-1) ?
861 		(uid_t)-1 : SCARG(uap, rgid);
862 	SCARG(&bsa, egid) = ((linux_gid_t)SCARG(uap, egid) == (linux_gid_t)-1) ?
863 		(uid_t)-1 : SCARG(uap, egid);
864 
865 	return sys_setregid(p, &bsa, retval);
866 }
867 
868 int
869 linux_sys___sysctl(p, v, retval)
870 	struct proc *p;
871 	void *v;
872 	register_t *retval;
873 {
874 	struct linux_sys___sysctl_args /* {
875 		syscallarg(struct linux___sysctl *) lsp;
876 	} */ *uap = v;
877 	struct linux___sysctl ls;
878 	struct sys___sysctl_args bsa;
879 	int error;
880 
881 	if ((error = copyin(SCARG(uap, lsp), &ls, sizeof ls)))
882 		return error;
883 	SCARG(&bsa, name) = ls.name;
884 	SCARG(&bsa, namelen) = ls.namelen;
885 	SCARG(&bsa, old) = ls.old;
886 	SCARG(&bsa, oldlenp) = ls.oldlenp;
887 	SCARG(&bsa, new) = ls.new;
888 	SCARG(&bsa, newlen) = ls.newlen;
889 
890 	return sys___sysctl(p, &bsa, retval);
891 }
892