xref: /netbsd-src/sys/compat/linux/common/linux_break.c (revision 88b8e43a80ceb526a57d200b7e08df604247ca77)
1 /*	$NetBSD: linux_break.c,v 1.11 1995/08/14 01:27:53 mycroft Exp $	*/
2 
3 /*
4  * Copyright (c) 1995 Frank van der Linden
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  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *      This product includes software developed for the NetBSD Project
18  *      by Frank van der Linden
19  * 4. The name of the author may not be used to endorse or promote products
20  *    derived from this software without specific prior written permission
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 /*
35  * Linux compatibility module. Try to deal with various Linux system calls.
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/namei.h>
41 #include <sys/proc.h>
42 #include <sys/dir.h>
43 #include <sys/file.h>
44 #include <sys/stat.h>
45 #include <sys/filedesc.h>
46 #include <sys/ioctl.h>
47 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/mman.h>
51 #include <sys/mount.h>
52 #include <sys/ptrace.h>
53 #include <sys/resource.h>
54 #include <sys/resourcevar.h>
55 #include <sys/signal.h>
56 #include <sys/signalvar.h>
57 #include <sys/socket.h>
58 #include <sys/time.h>
59 #include <sys/times.h>
60 #include <sys/vnode.h>
61 #include <sys/uio.h>
62 #include <sys/wait.h>
63 #include <sys/utsname.h>
64 #include <sys/unistd.h>
65 
66 #include <sys/syscallargs.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_param.h>
70 
71 #include <compat/linux/linux_types.h>
72 #include <compat/linux/linux_fcntl.h>
73 #include <compat/linux/linux_mmap.h>
74 #include <compat/linux/linux_signal.h>
75 #include <compat/linux/linux_syscallargs.h>
76 #include <compat/linux/linux_util.h>
77 #include <compat/linux/linux_dirent.h>
78 
79 /*
80  * The information on a terminated (or stopped) process needs
81  * to be converted in order for Linux binaries to get a valid signal
82  * number out of it.
83  */
84 static int
85 bsd_to_linux_wstat(status)
86 	int *status;
87 {
88 	if (WIFSIGNALED(*status))
89 		*status = (*status & ~0177) |
90 		    bsd_to_linux_sig(WTERMSIG(*status));
91 	else if (WIFSTOPPED(*status))
92 		*status = (*status & ~0xff00) |
93 		    (bsd_to_linux_sig(WSTOPSIG(*status)) << 8);
94 }
95 
96 /*
97  * waitpid(2). Passed on to the NetBSD call, surrounded by code to
98  * reserve some space for a NetBSD-style wait status, and converting
99  * it to what Linux wants.
100  */
101 int
102 linux_waitpid(p, uap, retval)
103 	struct proc *p;
104 	struct linux_waitpid_args /* {
105 		syscallarg(int) pid;
106 		syscallarg(int *) status;
107 		syscallarg(int) options;
108 	} */ *uap;
109 	register_t *retval;
110 {
111 	struct wait4_args w4a;
112 	int error, *status, tstat;
113 	caddr_t sg;
114 
115 	sg = stackgap_init(p->p_emul);
116 	status = (int *) stackgap_alloc(&sg, sizeof status);
117 
118 	SCARG(&w4a, pid) = SCARG(uap, pid);
119 	SCARG(&w4a, status) = status;
120 	SCARG(&w4a, options) = SCARG(uap, options);
121 	SCARG(&w4a, rusage) = NULL;
122 
123 	if ((error = wait4(p, &w4a, retval)))
124 		return error;
125 
126 	if ((error = copyin(status, &tstat, sizeof tstat)))
127 		return error;
128 
129 	bsd_to_linux_wstat(&tstat);
130 
131 	return copyout(&tstat, SCARG(uap, status), sizeof tstat);
132 }
133 
134 /*
135  * This is very much the same as waitpid()
136  */
137 int
138 linux_wait4(p, uap, retval)
139 	struct proc *p;
140 	struct linux_wait4_args /* {
141 		syscallarg(int) pid;
142 		syscallarg(int *) status;
143 		syscallarg(int) options;
144 		syscallarg(struct rusage *) rusage;
145 	} */ *uap;
146 	register_t *retval;
147 {
148 	struct wait4_args w4a;
149 	int error, *status, tstat;
150 	caddr_t sg;
151 
152 	sg = stackgap_init(p->p_emul);
153 	status = (int *) stackgap_alloc(&sg, sizeof status);
154 
155 	SCARG(&w4a, pid) = SCARG(uap, pid);
156 	SCARG(&w4a, status) = status;
157 	SCARG(&w4a, options) = SCARG(uap, options);
158 	SCARG(&w4a, rusage) = SCARG(uap, rusage);
159 
160 	if ((error = wait4(p, &w4a, retval)))
161 		return error;
162 
163 	if ((error = copyin(status, &tstat, sizeof tstat)))
164 		return error;
165 
166 	bsd_to_linux_wstat(&tstat);
167 
168 	return copyout(&tstat, SCARG(uap, status), sizeof tstat);
169 }
170 
171 /*
172  * This is the old brk(2) call. I don't think anything in the Linux
173  * world uses this anymore
174  */
175 int
176 linux_break(p, uap, retval)
177 	struct proc *p;
178 	struct linux_brk_args /* {
179 		syscallarg(char *) nsize;
180 	} */ *uap;
181 	register_t *retval;
182 {
183 	return ENOSYS;
184 }
185 
186 /*
187  * Linux brk(2). The check if the new address is >= the old one is
188  * done in the kernel in Linux. NetBSD does it in the library.
189  */
190 int
191 linux_brk(p, uap, retval)
192 	struct proc *p;
193 	struct linux_brk_args /* {
194 		syscallarg(char *) nsize;
195 	} */ *uap;
196 	register_t *retval;
197 {
198 	char *nbrk = SCARG(uap, nsize);
199 	struct obreak_args oba;
200 	struct vmspace *vm = p->p_vmspace;
201 	int error = 0;
202 	caddr_t oldbrk, newbrk;
203 
204 	oldbrk = vm->vm_daddr + ctob(vm->vm_dsize);
205 	/*
206 	 * XXX inconsistent.. Linux always returns at least the old
207 	 * brk value, but it will be page-aligned if this fails,
208 	 * and possibly not page aligned if it succeeds (the user
209 	 * supplied pointer is returned).
210 	 */
211 	SCARG(&oba, nsize) = nbrk;
212 
213 	if ((caddr_t) nbrk > vm->vm_daddr && obreak(p, &oba, retval) == 0)
214 		retval[0] = (register_t) nbrk;
215 	else
216 		retval[0] = (register_t) oldbrk;
217 
218 	return 0;
219 }
220 
221 /*
222  * I wonder why Linux has gettimeofday() _and_ time().. Still, we
223  * need to deal with it.
224  */
225 int
226 linux_time(p, uap, retval)
227 	struct proc *p;
228 	struct linux_time_args /* {
229 		linux_time_t *t;
230 	} */ *uap;
231 	register_t *retval;
232 {
233 	struct timeval atv;
234 	linux_time_t tt;
235 	int error;
236 
237 	microtime(&atv);
238 
239 	tt = atv.tv_sec;
240 	if (SCARG(uap, t) && (error = copyout(&tt, SCARG(uap, t), sizeof tt)))
241 		return error;
242 
243 	retval[0] = tt;
244 	return 0;
245 }
246 
247 /*
248  * Convert BSD statfs structure to Linux statfs structure.
249  * The Linux structure has less fields, and it also wants
250  * the length of a name in a dir entry in a field, which
251  * we fake (probably the wrong way).
252  */
253 static void
254 bsd_to_linux_statfs(bsp, lsp)
255 	struct statfs *bsp;
256 	struct linux_statfs *lsp;
257 {
258 	lsp->l_ftype = bsp->f_type;
259 	lsp->l_fbsize = bsp->f_bsize;
260 	lsp->l_fblocks = bsp->f_blocks;
261 	lsp->l_fbfree = bsp->f_bfree;
262 	lsp->l_fbavail = bsp->f_bavail;
263 	lsp->l_ffiles = bsp->f_files;
264 	lsp->l_fffree = bsp->f_ffree;
265 	lsp->l_ffsid.val[0] = bsp->f_fsid.val[0];
266 	lsp->l_ffsid.val[1] = bsp->f_fsid.val[1];
267 	lsp->l_fnamelen = MAXNAMLEN;	/* XXX */
268 }
269 
270 /*
271  * Implement the fs stat functions. Straightforward.
272  */
273 int
274 linux_statfs(p, uap, retval)
275 	struct proc *p;
276 	struct linux_statfs_args /* {
277 		syscallarg(char *) path;
278 		syscallarg(struct linux_statfs *) sp;
279 	} */ *uap;
280 	register_t *retval;
281 {
282 	struct statfs btmp, *bsp;
283 	struct linux_statfs ltmp;
284 	struct statfs_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 	LINUX_CHECK_ALT_EXIST(p, &sg, SCARG(uap, path));
292 
293 	SCARG(&bsa, path) = SCARG(uap, path);
294 	SCARG(&bsa, buf) = bsp;
295 
296 	if ((error = statfs(p, &bsa, retval)))
297 		return error;
298 
299 	if ((error = copyin((caddr_t) bsp, (caddr_t) &btmp, sizeof btmp)))
300 		return error;
301 
302 	bsd_to_linux_statfs(&btmp, &ltmp);
303 
304 	return copyout((caddr_t) &ltmp, (caddr_t) SCARG(uap, sp), sizeof ltmp);
305 }
306 
307 int
308 linux_fstatfs(p, uap, retval)
309 	struct proc *p;
310 	struct linux_fstatfs_args /* {
311 		syscallarg(int) fd;
312 		syscallarg(struct linux_statfs *) sp;
313 	} */ *uap;
314 	register_t *retval;
315 {
316 	struct statfs btmp, *bsp;
317 	struct linux_statfs ltmp;
318 	struct fstatfs_args bsa;
319 	caddr_t sg;
320 	int error;
321 
322 	sg = stackgap_init(p->p_emul);
323 	bsp = (struct statfs *) stackgap_alloc(&sg, sizeof (struct statfs));
324 
325 	SCARG(&bsa, fd) = SCARG(uap, fd);
326 	SCARG(&bsa, buf) = bsp;
327 
328 	if ((error = statfs(p, &bsa, retval)))
329 		return error;
330 
331 	if ((error = copyin((caddr_t) bsp, (caddr_t) &btmp, sizeof btmp)))
332 		return error;
333 
334 	bsd_to_linux_statfs(&btmp, &ltmp);
335 
336 	return copyout((caddr_t) &ltmp, (caddr_t) SCARG(uap, sp), sizeof ltmp);
337 }
338 
339 /*
340  * uname(). Just copy the info from the various strings stored in the
341  * kernel, and put it in the Linux utsname structure. That structure
342  * is almost the same as the NetBSD one, only it has fields 65 characters
343  * long, and an extra domainname field.
344  */
345 int
346 linux_uname(p, uap, retval)
347 	struct proc *p;
348 	struct linux_uname_args /* {
349 		syscallarg(struct linux_utsname *) up;
350 	} */ *uap;
351 	register_t *retval;
352 {
353 	extern char ostype[], osrelease[], version[], hostname[], domainname[];
354 	extern char machine[];
355 	struct linux_utsname tluts;
356 	int len;
357 	char *cp;
358 
359 	strncpy(tluts.l_sysname, ostype, sizeof (tluts.l_sysname));
360 	strncpy(tluts.l_nodename, hostname, sizeof (tluts.l_nodename));
361 	strncpy(tluts.l_release, osrelease, sizeof (tluts.l_release));
362 	strncpy(tluts.l_machine, machine, sizeof (tluts.l_machine));
363 	strncpy(tluts.l_domainname, domainname, sizeof (tluts.l_domainname));
364 	strncpy(tluts.l_version, version, sizeof (tluts.l_version));
365 
366 	/* This part taken from the the uname() in libc */
367 	len = sizeof (tluts.l_version);
368 	for (cp = tluts.l_version; len--; ++cp)
369 		if (*cp == '\n' || *cp == '\t')
370 			if (len > 1)
371 				*cp = ' ';
372 			else
373 				*cp = '\0';
374 
375 	return copyout(&tluts, SCARG(uap, up), sizeof tluts);
376 }
377 
378 /*
379  * Linux wants to pass everything to a syscall in registers. However,
380  * mmap() has 6 of them. Oops: out of register error. They just pass
381  * everything in a structure.
382  */
383 int
384 linux_mmap(p, uap, retval)
385 	struct proc *p;
386 	struct linux_mmap_args /* {
387 		syscallarg(struct linux_mmap *) lmp;
388 	} */ *uap;
389 	register_t *retval;
390 {
391 	struct linux_mmap lmap;
392 	struct mmap_args cma;
393 	int error, flags;
394 
395 	if ((error = copyin(SCARG(uap, lmp), &lmap, sizeof lmap)))
396 		return error;
397 
398 	flags = 0;
399 	flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_SHARED, MAP_SHARED);
400 	flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_PRIVATE, MAP_PRIVATE);
401 	flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_FIXED, MAP_FIXED);
402 	flags |= cvtto_bsd_mask(lmap.lm_flags, LINUX_MAP_ANON, MAP_ANON);
403 
404 	SCARG(&cma,addr) = lmap.lm_addr;
405 	SCARG(&cma,len) = lmap.lm_len;
406  	SCARG(&cma,prot) = lmap.lm_prot;
407 	SCARG(&cma,flags) = flags;
408 	SCARG(&cma,fd) = lmap.lm_fd;
409 	SCARG(&cma,pad) = 0;
410 	SCARG(&cma,pos) = lmap.lm_pos;
411 
412 	return mmap(p, &cma, retval);
413 }
414 
415 /*
416  * Linux doesn't use the retval[1] value to determine whether
417  * we are the child or parent.
418  */
419 int
420 linux_fork(p, uap, retval)
421 	struct proc *p;
422 	void *uap;
423 	register_t *retval;
424 {
425 	int error;
426 
427 	if ((error = fork(p, uap, retval)))
428 		return error;
429 
430 	if (retval[1] == 1)
431 		retval[0] = 0;
432 
433 	return 0;
434 }
435 
436 /*
437  * This code is partly stolen from src/lib/libc/compat-43/times.c
438  * XXX - CLK_TCK isn't declared in /sys, just in <time.h>, done here
439  */
440 
441 #define CLK_TCK 100
442 #define	CONVTCK(r)	(r.tv_sec * CLK_TCK + r.tv_usec / (1000000 / CLK_TCK))
443 
444 int
445 linux_times(p, uap, retval)
446 	struct proc *p;
447 	struct linux_times_args /* {
448 		syscallarg(struct times *) tms;
449 	} */ *uap;
450 	register_t *retval;
451 {
452 	struct timeval t;
453 	struct linux_tms ltms;
454 	struct rusage ru;
455 	int error, s;
456 
457 	calcru(p, &ru.ru_utime, &ru.ru_stime, NULL);
458 	ltms.ltms_utime = CONVTCK(ru.ru_utime);
459 	ltms.ltms_stime = CONVTCK(ru.ru_stime);
460 
461 	ltms.ltms_cutime = CONVTCK(p->p_stats->p_cru.ru_utime);
462 	ltms.ltms_cstime = CONVTCK(p->p_stats->p_cru.ru_stime);
463 
464 	if ((error = copyout(&ltms, SCARG(uap, tms), sizeof ltms)))
465 		return error;
466 
467 	s = splclock();
468 	timersub(&time, &boottime, &t);
469 	splx(s);
470 
471 	retval[0] = ((linux_clock_t)(CONVTCK(t)));
472 	return 0;
473 }
474 
475 /*
476  * NetBSD passes fd[0] in retval[0], and fd[1] in retval[1].
477  * Linux directly passes the pointer.
478  */
479 int
480 linux_pipe(p, uap, retval)
481 	struct proc *p;
482 	struct linux_pipe_args /* {
483 		syscallarg(int *) pfds;
484 	} */ *uap;
485 	register_t *retval;
486 {
487 	int error;
488 
489 	if ((error = pipe(p, 0, retval)))
490 		return error;
491 
492 	/* Assumes register_t is an int */
493 
494 	if ((error = copyout(retval, SCARG(uap, pfds), 2 * sizeof (int))))
495 		return error;
496 
497 	retval[0] = 0;
498 	return 0;
499 }
500 
501 /*
502  * Alarm. This is a libc call which used setitimer(2) in NetBSD.
503  * Fiddle with the timers to make it work.
504  */
505 int
506 linux_alarm(p, uap, retval)
507 	struct proc *p;
508 	struct linux_alarm_args /* {
509 		syscallarg(unsigned int) secs;
510 	} */ *uap;
511 	register_t *retval;
512 {
513 	int error, s;
514 	struct itimerval *itp, it;
515 
516 	itp = &p->p_realtimer;
517 	s = splclock();
518 	/*
519 	 * Clear any pending timer alarms.
520 	 */
521 	untimeout(realitexpire, p);
522 	timerclear(&itp->it_interval);
523 	if (timerisset(&itp->it_value) &&
524 	    timercmp(&itp->it_value, &time, >))
525 		timersub(&itp->it_value, &time, &itp->it_value);
526 	/*
527 	 * Return how many seconds were left (rounded up)
528 	 */
529 	retval[0] = itp->it_value.tv_sec;
530 	if (itp->it_value.tv_usec)
531 		retval[0]++;
532 
533 	/*
534 	 * alarm(0) just resets the timer.
535 	 */
536 	if (SCARG(uap, secs) == 0) {
537 		timerclear(&itp->it_value);
538 		splx(s);
539 		return 0;
540 	}
541 
542 	/*
543 	 * Check the new alarm time for sanity, and set it.
544 	 */
545 	timerclear(&it.it_interval);
546 	it.it_value.tv_sec = SCARG(uap, secs);
547 	it.it_value.tv_usec = 0;
548 	if (itimerfix(&it.it_value) || itimerfix(&it.it_interval)) {
549 		splx(s);
550 		return (EINVAL);
551 	}
552 
553 	if (timerisset(&it.it_value)) {
554 		timeradd(&it.it_value, &time, &it.it_value);
555 		timeout(realitexpire, p, hzto(&it.it_value));
556 	}
557 	p->p_realtimer = it;
558 	splx(s);
559 
560 	return 0;
561 }
562 
563 /*
564  * utime(). Do conversion to things that utimes() understands,
565  * and pass it on.
566  */
567 int
568 linux_utime(p, uap, retval)
569 	struct proc *p;
570 	struct linux_utime_args /* {
571 		syscallarg(char *) path;
572 		syscallarg(struct linux_utimbuf *)times;
573 	} */ *uap;
574 	register_t *retval;
575 {
576 	caddr_t sg;
577 	int error;
578 	struct utimes_args ua;
579 	struct timeval tv[2], *tvp;
580 	struct linux_utimbuf lut;
581 
582 	sg = stackgap_init(p->p_emul);
583 	LINUX_CHECK_ALT_EXIST(p, &sg, SCARG(uap, path));
584 
585 	SCARG(&ua, path) = SCARG(uap, path);
586 
587 	if (SCARG(uap, times) != NULL) {
588 		if ((error = copyin(SCARG(uap, times), &lut, sizeof lut)))
589 			return error;
590 		tv[0].tv_usec = tv[1].tv_usec = 0;
591 		tv[0].tv_sec = lut.l_actime;
592 		tv[1].tv_sec = lut.l_modtime;
593 		tvp = (struct timeval *) stackgap_alloc(&sg, sizeof(tv));
594 		if ((error = copyout(tv, tvp, sizeof tv)))
595 			return error;
596 		SCARG(&ua, tptr) = tvp;
597 	}
598 	else
599 		SCARG(&ua, tptr) = NULL;
600 
601 	return utimes(p, uap, retval);
602 }
603 
604 /*
605  * Linux 'readdir' call. This code is mostly taken from the
606  * SunOS getdents call (see compat/sunos/sunos_misc.c), though
607  * an attempt has been made to keep it a little cleaner (failing
608  * miserably, because of the cruft needed if count 1 is passed).
609  *
610  * Read in BSD-style entries, convert them, and copy them out.
611  * Note that the Linux d_reclen is actually the name length,
612  * and d_off is the reclen.
613  *
614  * Note that this doesn't handle union-mounted filesystems.
615  */
616 int
617 linux_readdir(p, uap, retval)
618 	struct proc *p;
619 	struct linux_readdir_args /* {
620 		syscallarg(int) fd;
621 		syscallarg(struct linux_dirent *) dent;
622 		syscallarg(unsigned int) count;
623 	} */ *uap;
624 	register_t *retval;
625 {
626 	register struct dirent *bdp;
627 	struct vnode *vp;
628 	caddr_t	inp, buf;	/* BSD-format */
629 	int len, reclen;	/* BSD-format */
630 	caddr_t outp;		/* Linux-format */
631 	int resid, linuxreclen;	/* Linux-format */
632 	struct file *fp;
633 	struct uio auio;
634 	struct iovec aiov;
635 	struct linux_dirent idb;
636 	off_t off;		/* true file offset */
637 	linux_off_t soff;	/* Linux file offset */
638 	int buflen, error, eofflag, nbytes, justone;
639 	struct vattr va;
640 
641 	if ((error = getvnode(p->p_fd, SCARG(uap, fd), &fp)) != 0)
642 		return (error);
643 
644 	if ((fp->f_flag & FREAD) == 0)
645 		return (EBADF);
646 
647 	vp = (struct vnode *)fp->f_data;
648 
649 	if (vp->v_type != VDIR)	/* XXX  vnode readdir op should do this */
650 		return (EINVAL);
651 
652 	if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)))
653 		return error;
654 
655 	nbytes = SCARG(uap, count);
656 	if (nbytes == 1) {	/* Need this for older Linux libs, apparently */
657 		nbytes = sizeof (struct linux_dirent);
658 		buflen = max(va.va_blocksize, nbytes);
659 		justone = 1;
660 	} else {
661 		buflen = min(MAXBSIZE, nbytes);
662 		justone = 0;
663 	}
664 	buf = malloc(buflen, M_TEMP, M_WAITOK);
665 	VOP_LOCK(vp);
666 	off = fp->f_offset;
667 again:
668 	aiov.iov_base = buf;
669 	aiov.iov_len = buflen;
670 	auio.uio_iov = &aiov;
671 	auio.uio_iovcnt = 1;
672 	auio.uio_rw = UIO_READ;
673 	auio.uio_segflg = UIO_SYSSPACE;
674 	auio.uio_procp = p;
675 	auio.uio_resid = buflen;
676 	auio.uio_offset = off;
677 	/*
678          * First we read into the malloc'ed buffer, then
679          * we massage it into user space, one record at a time.
680          */
681 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, (u_long *)0, 0);
682 	if (error)
683 		goto out;
684 
685 	inp = buf;
686 	outp = (caddr_t) SCARG(uap, dent);
687 	resid = nbytes;
688 	if ((len = buflen - auio.uio_resid) == 0)
689 		goto eof;
690 
691 	for (; len > 0; len -= reclen) {
692 		bdp = (struct dirent *)inp;
693 		reclen = bdp->d_reclen;
694 		if (reclen & 3)
695 			panic("linux_readdir");
696 		off += reclen;	/* each entry points to next */
697 		if (bdp->d_fileno == 0) {
698 			inp += reclen;	/* it is a hole; squish it out */
699 			continue;
700 		}
701 		linuxreclen = LINUX_RECLEN(&idb, bdp->d_namlen);
702 		if (reclen > len || resid < linuxreclen) {
703 			/* entry too big for buffer, so just stop */
704 			outp++;
705 			break;
706 		}
707 		/*
708 		 * Massage in place to make a Linux-shaped dirent (otherwise
709 		 * we have to worry about touching user memory outside of
710 		 * the copyout() call).
711 		 */
712 		idb.d_ino = (long)bdp->d_fileno;
713 		idb.d_off = (linux_off_t)linuxreclen;
714 		idb.d_reclen = (u_short)bdp->d_namlen;
715 		strcpy(idb.d_name, bdp->d_name);
716 		if ((error = copyout((caddr_t)&idb, outp, linuxreclen)))
717 			goto out;
718 		/* advance past this real entry */
719 		inp += reclen;
720 		/* advance output past Linux-shaped entry */
721 		outp += linuxreclen;
722 		resid -= linuxreclen;
723 		if (justone)
724 			break;
725 	}
726 
727 	/* if we squished out the whole block, try again */
728 	if (outp == (caddr_t) SCARG(uap, dent))
729 		goto again;
730 	fp->f_offset = off;	/* update the vnode offset */
731 
732 	if (justone)
733 		nbytes = resid + linuxreclen;
734 
735 eof:
736 	*retval = nbytes - resid;
737 out:
738 	VOP_UNLOCK(vp);
739 	free(buf, M_TEMP);
740 	return error;
741 }
742 
743 /*
744  * Out of register error once more.. Also, Linux copies the amount of
745  * time left into the user-supplied timeval structure.
746  */
747 int
748 linux_select(p, uap, retval)
749 	struct proc *p;
750 	struct linux_select_args /* {
751 		syscallarg(struct linux_select *) lsp;
752 	} */ *uap;
753 	register_t *retval;
754 {
755 	struct linux_select ls;
756 	struct select_args bsa;
757 	struct timeval tv0, tv1, utv;
758 	int error;
759 
760 	if ((error = copyin(SCARG(uap, lsp), (caddr_t)&ls, sizeof ls)))
761 		return error;
762 
763 	SCARG(&bsa, nd) = ls.nfds;
764 	SCARG(&bsa, in) = ls.readfds;
765 	SCARG(&bsa, ou) = ls.writefds;
766 	SCARG(&bsa, ex) = ls.exceptfds;
767 	SCARG(&bsa, tv) = ls.timeout;
768 
769 	/*
770 	 * Store current time for computation of the amount of
771 	 * time left.
772 	 */
773 	if (ls.timeout)
774 		microtime(&tv0);
775 
776 	error = select(p, &bsa, retval);
777 	if (error) {
778 		/*
779 		 * See fs/select.c in the Linux kernel.  Without this,
780 		 * Maelstrom doesn't work.
781 		 */
782 		if (error == ERESTART)
783 			error = EINTR;
784 		return error;
785 	}
786 
787 	if (ls.timeout) {
788 		if (!*retval) {
789 			utv.tv_sec = 0;
790 			utv.tv_usec = 0;
791 		} else {
792 			/*
793 			 * Compute how many time was left of the timeout,
794 			 * by subtracting the current time and the time
795 			 * before we started the call, and subtracting
796 			 * that result from the user-supplied value.
797 			 */
798 			microtime(&tv1);
799 			if ((error = copyin((caddr_t)ls.timeout, (caddr_t)&utv,
800 			    sizeof utv)))
801 				return error;
802 			timersub(&tv1, &tv0, &tv1);
803 			timersub(&utv, &tv1, &utv);
804 		}
805 		if ((error = copyout((caddr_t)&utv, (caddr_t)ls.timeout,
806 		    sizeof utv)))
807 			return error;
808 	}
809 	return 0;
810 }
811 
812 /*
813  * Get the process group of a certain process. Look it up
814  * and return the value.
815  */
816 int
817 linux_getpgid(p, uap, retval)
818 	struct proc *p;
819 	struct linux_getpgid_args /* {
820 		syscallarg(int) pid;
821 	} */ *uap;
822 	register_t *retval;
823 {
824 	struct proc *targp;
825 
826 	if (SCARG(uap, pid) != 0 && SCARG(uap, pid) != p->p_pid)
827 		if ((targp = pfind(SCARG(uap, pid))) == 0)
828 			return ESRCH;
829 	else
830 		targp = p;
831 
832 	retval[0] = targp->p_pgid;
833 	return 0;
834 }
835 
836 /*
837  * Set the 'personality' (emulation mode) for the current process. Only
838  * accept the Linux personality here (0). This call is needed because
839  * the Linux ELF crt0 issues it in an ugly kludge to make sure that
840  * ELF binaries run in Linux mode, not SVR4 mode.
841  */
842 int
843 linux_personality(p, uap, retval)
844 	struct proc *p;
845 	struct linux_personality_args /* P
846 		syscallarg(int) per;
847 	} */ *uap;
848 	register_t *retval;
849 {
850 	if (SCARG(uap, per) != 0)
851 		return EINVAL;
852 	retval[0] = 0;
853 	return 0;
854 }
855