xref: /netbsd-src/sys/compat/linux/common/linux_file.c (revision da39824b722dbd84beb9a1ab7e8de6710cc44d4b)
1 /*	$NetBSD: linux_file.c,v 1.108 2013/12/08 15:55:10 njoly Exp $	*/
2 
3 /*-
4  * Copyright (c) 1995, 1998, 2008 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Functions in multiarch:
34  *	linux_sys_llseek	: linux_llseek.c
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: linux_file.c,v 1.108 2013/12/08 15:55:10 njoly Exp $");
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
43 #include <sys/proc.h>
44 #include <sys/file.h>
45 #include <sys/fcntl.h>
46 #include <sys/stat.h>
47 #include <sys/filedesc.h>
48 #include <sys/ioctl.h>
49 #include <sys/kernel.h>
50 #include <sys/mount.h>
51 #include <sys/malloc.h>
52 #include <sys/namei.h>
53 #include <sys/vnode.h>
54 #include <sys/tty.h>
55 #include <sys/socketvar.h>
56 #include <sys/conf.h>
57 #include <sys/pipe.h>
58 
59 #include <sys/syscallargs.h>
60 #include <sys/vfs_syscalls.h>
61 
62 #include <compat/linux/common/linux_types.h>
63 #include <compat/linux/common/linux_signal.h>
64 #include <compat/linux/common/linux_fcntl.h>
65 #include <compat/linux/common/linux_util.h>
66 #include <compat/linux/common/linux_machdep.h>
67 #include <compat/linux/common/linux_ipc.h>
68 #include <compat/linux/common/linux_sem.h>
69 
70 #include <compat/linux/linux_syscallargs.h>
71 
72 static int linux_to_bsd_ioflags(int);
73 static int bsd_to_linux_ioflags(int);
74 #ifndef __amd64__
75 static void bsd_to_linux_stat(struct stat *, struct linux_stat *);
76 #endif
77 
78 conv_linux_flock(linux, flock)
79 
80 /*
81  * Some file-related calls are handled here. The usual flag conversion
82  * an structure conversion is done, and alternate emul path searching.
83  */
84 
85 /*
86  * The next two functions convert between the Linux and NetBSD values
87  * of the flags used in open(2) and fcntl(2).
88  */
89 static int
90 linux_to_bsd_ioflags(int lflags)
91 {
92 	int res = 0;
93 
94 	res |= cvtto_bsd_mask(lflags, LINUX_O_WRONLY, O_WRONLY);
95 	res |= cvtto_bsd_mask(lflags, LINUX_O_RDONLY, O_RDONLY);
96 	res |= cvtto_bsd_mask(lflags, LINUX_O_RDWR, O_RDWR);
97 	res |= cvtto_bsd_mask(lflags, LINUX_O_CREAT, O_CREAT);
98 	res |= cvtto_bsd_mask(lflags, LINUX_O_EXCL, O_EXCL);
99 	res |= cvtto_bsd_mask(lflags, LINUX_O_NOCTTY, O_NOCTTY);
100 	res |= cvtto_bsd_mask(lflags, LINUX_O_TRUNC, O_TRUNC);
101 	res |= cvtto_bsd_mask(lflags, LINUX_O_NDELAY, O_NDELAY);
102 	res |= cvtto_bsd_mask(lflags, LINUX_O_SYNC, O_FSYNC);
103 	res |= cvtto_bsd_mask(lflags, LINUX_FASYNC, O_ASYNC);
104 	res |= cvtto_bsd_mask(lflags, LINUX_O_APPEND, O_APPEND);
105 	res |= cvtto_bsd_mask(lflags, LINUX_O_DIRECTORY, O_DIRECTORY);
106 	res |= cvtto_bsd_mask(lflags, LINUX_O_CLOEXEC, O_CLOEXEC);
107 
108 	return res;
109 }
110 
111 static int
112 bsd_to_linux_ioflags(int bflags)
113 {
114 	int res = 0;
115 
116 	res |= cvtto_linux_mask(bflags, O_WRONLY, LINUX_O_WRONLY);
117 	res |= cvtto_linux_mask(bflags, O_RDONLY, LINUX_O_RDONLY);
118 	res |= cvtto_linux_mask(bflags, O_RDWR, LINUX_O_RDWR);
119 	res |= cvtto_linux_mask(bflags, O_CREAT, LINUX_O_CREAT);
120 	res |= cvtto_linux_mask(bflags, O_EXCL, LINUX_O_EXCL);
121 	res |= cvtto_linux_mask(bflags, O_NOCTTY, LINUX_O_NOCTTY);
122 	res |= cvtto_linux_mask(bflags, O_TRUNC, LINUX_O_TRUNC);
123 	res |= cvtto_linux_mask(bflags, O_NDELAY, LINUX_O_NDELAY);
124 	res |= cvtto_linux_mask(bflags, O_FSYNC, LINUX_O_SYNC);
125 	res |= cvtto_linux_mask(bflags, O_ASYNC, LINUX_FASYNC);
126 	res |= cvtto_linux_mask(bflags, O_APPEND, LINUX_O_APPEND);
127 	res |= cvtto_linux_mask(bflags, O_DIRECTORY, LINUX_O_DIRECTORY);
128 	res |= cvtto_linux_mask(bflags, O_CLOEXEC, LINUX_O_CLOEXEC);
129 
130 	return res;
131 }
132 
133 /*
134  * creat(2) is an obsolete function, but it's present as a Linux
135  * system call, so let's deal with it.
136  *
137  * Note: On the Alpha this doesn't really exist in Linux, but it's defined
138  * in syscalls.master anyway so this doesn't have to be special cased.
139  *
140  * Just call open(2) with the TRUNC, CREAT and WRONLY flags.
141  */
142 int
143 linux_sys_creat(struct lwp *l, const struct linux_sys_creat_args *uap, register_t *retval)
144 {
145 	/* {
146 		syscallarg(const char *) path;
147 		syscallarg(linux_umode_t) mode;
148 	} */
149 	struct sys_open_args oa;
150 
151 	SCARG(&oa, path) = SCARG(uap, path);
152 	SCARG(&oa, flags) = O_CREAT | O_TRUNC | O_WRONLY;
153 	SCARG(&oa, mode) = SCARG(uap, mode);
154 
155 	return sys_open(l, &oa, retval);
156 }
157 
158 static void
159 linux_open_ctty(struct lwp *l, int flags, int fd)
160 {
161 	struct proc *p = l->l_proc;
162 
163 	/*
164 	 * this bit from sunos_misc.c (and svr4_fcntl.c).
165 	 * If we are a session leader, and we don't have a controlling
166 	 * terminal yet, and the O_NOCTTY flag is not set, try to make
167 	 * this the controlling terminal.
168 	 */
169         if (!(flags & O_NOCTTY) && SESS_LEADER(p) && !(p->p_lflag & PL_CONTROLT)) {
170                 file_t *fp;
171 
172 		fp = fd_getfile(fd);
173 
174                 /* ignore any error, just give it a try */
175                 if (fp != NULL) {
176 			if (fp->f_type == DTYPE_VNODE) {
177 				(fp->f_ops->fo_ioctl) (fp, TIOCSCTTY, NULL);
178 			}
179 			fd_putfile(fd);
180 		}
181         }
182 }
183 
184 /*
185  * open(2). Take care of the different flag values, and let the
186  * NetBSD syscall do the real work. See if this operation
187  * gives the current process a controlling terminal.
188  * (XXX is this necessary?)
189  */
190 int
191 linux_sys_open(struct lwp *l, const struct linux_sys_open_args *uap, register_t *retval)
192 {
193 	/* {
194 		syscallarg(const char *) path;
195 		syscallarg(int) flags;
196 		syscallarg(linux_umode_t) mode;
197 	} */
198 	int error, fl;
199 	struct sys_open_args boa;
200 
201 	fl = linux_to_bsd_ioflags(SCARG(uap, flags));
202 
203 	SCARG(&boa, path) = SCARG(uap, path);
204 	SCARG(&boa, flags) = fl;
205 	SCARG(&boa, mode) = SCARG(uap, mode);
206 
207 	if ((error = sys_open(l, &boa, retval)))
208 		return error;
209 
210 	linux_open_ctty(l, fl, *retval);
211 	return 0;
212 }
213 
214 int
215 linux_sys_openat(struct lwp *l, const struct linux_sys_openat_args *uap, register_t *retval)
216 {
217 	/* {
218 		syscallarg(int) fd;
219 		syscallarg(const char *) path;
220 		syscallarg(int) flags;
221 		syscallarg(linux_umode_t) mode;
222 	} */
223 	int error, fl;
224 	struct sys_openat_args boa;
225 
226 	fl = linux_to_bsd_ioflags(SCARG(uap, flags));
227 
228 	SCARG(&boa, fd) = SCARG(uap, fd);
229 	SCARG(&boa, path) = SCARG(uap, path);
230 	SCARG(&boa, oflags) = fl;
231 	SCARG(&boa, mode) = SCARG(uap, mode);
232 
233 	if ((error = sys_openat(l, &boa, retval)))
234 		return error;
235 
236 	linux_open_ctty(l, fl, *retval);
237 	return 0;
238 }
239 
240 /*
241  * Most actions in the fcntl() call are straightforward; simply
242  * pass control to the NetBSD system call. A few commands need
243  * conversions after the actual system call has done its work,
244  * because the flag values and lock structure are different.
245  */
246 int
247 linux_sys_fcntl(struct lwp *l, const struct linux_sys_fcntl_args *uap, register_t *retval)
248 {
249 	/* {
250 		syscallarg(int) fd;
251 		syscallarg(int) cmd;
252 		syscallarg(void *) arg;
253 	} */
254 	struct proc *p = l->l_proc;
255 	int fd, cmd, error;
256 	u_long val;
257 	void *arg;
258 	struct sys_fcntl_args fca;
259 	file_t *fp;
260 	struct vnode *vp;
261 	struct vattr va;
262 	long pgid;
263 	struct pgrp *pgrp;
264 	struct tty *tp;
265 
266 	fd = SCARG(uap, fd);
267 	cmd = SCARG(uap, cmd);
268 	arg = SCARG(uap, arg);
269 
270 	switch (cmd) {
271 
272 	case LINUX_F_DUPFD:
273 		cmd = F_DUPFD;
274 		break;
275 
276 	case LINUX_F_GETFD:
277 		cmd = F_GETFD;
278 		break;
279 
280 	case LINUX_F_SETFD:
281 		cmd = F_SETFD;
282 		break;
283 
284 	case LINUX_F_GETFL:
285 		SCARG(&fca, fd) = fd;
286 		SCARG(&fca, cmd) = F_GETFL;
287 		SCARG(&fca, arg) = arg;
288 		if ((error = sys_fcntl(l, &fca, retval)))
289 			return error;
290 		retval[0] = bsd_to_linux_ioflags(retval[0]);
291 		return 0;
292 
293 	case LINUX_F_SETFL: {
294 		file_t	*fp1 = NULL;
295 
296 		val = linux_to_bsd_ioflags((unsigned long)SCARG(uap, arg));
297 		/*
298 		 * Linux seems to have same semantics for sending SIGIO to the
299 		 * read side of socket, but slightly different semantics
300 		 * for SIGIO to the write side.  Rather than sending the SIGIO
301 		 * every time it's possible to write (directly) more data, it
302 		 * only sends SIGIO if last write(2) failed due to insufficient
303 		 * memory to hold the data. This is compatible enough
304 		 * with NetBSD semantics to not do anything about the
305 		 * difference.
306 		 *
307 		 * Linux does NOT send SIGIO for pipes. Deal with socketpair
308 		 * ones and DTYPE_PIPE ones. For these, we don't set
309 		 * the underlying flags (we don't pass O_ASYNC flag down
310 		 * to sys_fcntl()), but set the FASYNC flag for file descriptor,
311 		 * so that F_GETFL would report the ASYNC i/o is on.
312 		 */
313 		if (val & O_ASYNC) {
314 			if (((fp1 = fd_getfile(fd)) == NULL))
315 			    return (EBADF);
316 			if (((fp1->f_type == DTYPE_SOCKET) && fp1->f_data
317 			      && ((struct socket *)fp1->f_data)->so_state & SS_ISAPIPE)
318 			    || (fp1->f_type == DTYPE_PIPE))
319 				val &= ~O_ASYNC;
320 			else {
321 				/* not a pipe, do not modify anything */
322 				fd_putfile(fd);
323 				fp1 = NULL;
324 			}
325 		}
326 
327 		SCARG(&fca, fd) = fd;
328 		SCARG(&fca, cmd) = F_SETFL;
329 		SCARG(&fca, arg) = (void *) val;
330 
331 		error = sys_fcntl(l, &fca, retval);
332 
333 		/* Now set the FASYNC flag for pipes */
334 		if (fp1) {
335 			if (!error) {
336 				mutex_enter(&fp1->f_lock);
337 				fp1->f_flag |= FASYNC;
338 				mutex_exit(&fp1->f_lock);
339 			}
340 			fd_putfile(fd);
341 		}
342 
343 		return (error);
344 	    }
345 
346 	case LINUX_F_GETLK:
347 		do_linux_getlk(fd, cmd, arg, linux, flock);
348 
349 	case LINUX_F_SETLK:
350 	case LINUX_F_SETLKW:
351 		do_linux_setlk(fd, cmd, arg, linux, flock, LINUX_F_SETLK);
352 
353 	case LINUX_F_SETOWN:
354 	case LINUX_F_GETOWN:
355 		/*
356 		 * We need to route fcntl() for tty descriptors around normal
357 		 * fcntl(), since NetBSD tty TIOC{G,S}PGRP semantics is too
358 		 * restrictive for Linux F_{G,S}ETOWN. For non-tty descriptors,
359 		 * this is not a problem.
360 		 */
361 		if ((fp = fd_getfile(fd)) == NULL)
362 			return EBADF;
363 
364 		/* Check it's a character device vnode */
365 		if (fp->f_type != DTYPE_VNODE
366 		    || (vp = (struct vnode *)fp->f_data) == NULL
367 		    || vp->v_type != VCHR) {
368 			fd_putfile(fd);
369 
370 	    not_tty:
371 			/* Not a tty, proceed with common fcntl() */
372 			cmd = cmd == LINUX_F_SETOWN ? F_SETOWN : F_GETOWN;
373 			break;
374 		}
375 
376 		vn_lock(vp, LK_SHARED | LK_RETRY);
377 		error = VOP_GETATTR(vp, &va, l->l_cred);
378 		VOP_UNLOCK(vp);
379 
380 		fd_putfile(fd);
381 
382 		if (error)
383 			return error;
384 
385 		if ((tp = cdev_tty(va.va_rdev)) == NULL)
386 			goto not_tty;
387 
388 		/* set tty pg_id appropriately */
389 		mutex_enter(proc_lock);
390 		if (cmd == LINUX_F_GETOWN) {
391 			retval[0] = tp->t_pgrp ? tp->t_pgrp->pg_id : NO_PGID;
392 			mutex_exit(proc_lock);
393 			return 0;
394 		}
395 		if ((long)arg <= 0) {
396 			pgid = -(long)arg;
397 		} else {
398 			struct proc *p1 = proc_find((long)arg);
399 			if (p1 == NULL) {
400 				mutex_exit(proc_lock);
401 				return (ESRCH);
402 			}
403 			pgid = (long)p1->p_pgrp->pg_id;
404 		}
405 		pgrp = pgrp_find(pgid);
406 		if (pgrp == NULL || pgrp->pg_session != p->p_session) {
407 			mutex_exit(proc_lock);
408 			return EPERM;
409 		}
410 		tp->t_pgrp = pgrp;
411 		mutex_exit(proc_lock);
412 		return 0;
413 
414 	default:
415 		return EOPNOTSUPP;
416 	}
417 
418 	SCARG(&fca, fd) = fd;
419 	SCARG(&fca, cmd) = cmd;
420 	SCARG(&fca, arg) = arg;
421 
422 	return sys_fcntl(l, &fca, retval);
423 }
424 
425 #if !defined(__amd64__)
426 /*
427  * Convert a NetBSD stat structure to a Linux stat structure.
428  * Only the order of the fields and the padding in the structure
429  * is different. linux_fakedev is a machine-dependent function
430  * which optionally converts device driver major/minor numbers
431  * (XXX horrible, but what can you do against code that compares
432  * things against constant major device numbers? sigh)
433  */
434 static void
435 bsd_to_linux_stat(struct stat *bsp, struct linux_stat *lsp)
436 {
437 
438 	lsp->lst_dev     = linux_fakedev(bsp->st_dev, 0);
439 	lsp->lst_ino     = bsp->st_ino;
440 	lsp->lst_mode    = (linux_mode_t)bsp->st_mode;
441 	if (bsp->st_nlink >= (1 << 15))
442 		lsp->lst_nlink = (1 << 15) - 1;
443 	else
444 		lsp->lst_nlink = (linux_nlink_t)bsp->st_nlink;
445 	lsp->lst_uid     = bsp->st_uid;
446 	lsp->lst_gid     = bsp->st_gid;
447 	lsp->lst_rdev    = linux_fakedev(bsp->st_rdev, 1);
448 	lsp->lst_size    = bsp->st_size;
449 	lsp->lst_blksize = bsp->st_blksize;
450 	lsp->lst_blocks  = bsp->st_blocks;
451 	lsp->lst_atime   = bsp->st_atime;
452 	lsp->lst_mtime   = bsp->st_mtime;
453 	lsp->lst_ctime   = bsp->st_ctime;
454 #ifdef LINUX_STAT_HAS_NSEC
455 	lsp->lst_atime_nsec   = bsp->st_atimensec;
456 	lsp->lst_mtime_nsec   = bsp->st_mtimensec;
457 	lsp->lst_ctime_nsec   = bsp->st_ctimensec;
458 #endif
459 }
460 
461 /*
462  * The stat functions below are plain sailing. stat and lstat are handled
463  * by one function to avoid code duplication.
464  */
465 int
466 linux_sys_fstat(struct lwp *l, const struct linux_sys_fstat_args *uap, register_t *retval)
467 {
468 	/* {
469 		syscallarg(int) fd;
470 		syscallarg(linux_stat *) sp;
471 	} */
472 	struct linux_stat tmplst;
473 	struct stat tmpst;
474 	int error;
475 
476 	error = do_sys_fstat(SCARG(uap, fd), &tmpst);
477 	if (error != 0)
478 		return error;
479 	bsd_to_linux_stat(&tmpst, &tmplst);
480 
481 	return copyout(&tmplst, SCARG(uap, sp), sizeof tmplst);
482 }
483 
484 static int
485 linux_stat1(const struct linux_sys_stat_args *uap, register_t *retval, int flags)
486 {
487 	struct linux_stat tmplst;
488 	struct stat tmpst;
489 	int error;
490 
491 	error = do_sys_stat(SCARG(uap, path), flags, &tmpst);
492 	if (error != 0)
493 		return error;
494 
495 	bsd_to_linux_stat(&tmpst, &tmplst);
496 
497 	return copyout(&tmplst, SCARG(uap, sp), sizeof tmplst);
498 }
499 
500 int
501 linux_sys_stat(struct lwp *l, const struct linux_sys_stat_args *uap, register_t *retval)
502 {
503 	/* {
504 		syscallarg(const char *) path;
505 		syscallarg(struct linux_stat *) sp;
506 	} */
507 
508 	return linux_stat1(uap, retval, FOLLOW);
509 }
510 
511 /* Note: this is "newlstat" in the Linux sources */
512 /*	(we don't bother with the old lstat currently) */
513 int
514 linux_sys_lstat(struct lwp *l, const struct linux_sys_lstat_args *uap, register_t *retval)
515 {
516 	/* {
517 		syscallarg(const char *) path;
518 		syscallarg(struct linux_stat *) sp;
519 	} */
520 
521 	return linux_stat1((const void *)uap, retval, NOFOLLOW);
522 }
523 #endif /* !__amd64__ */
524 
525 /*
526  * The following syscalls are mostly here because of the alternate path check.
527  */
528 
529 int
530 linux_sys_linkat(struct lwp *l, const struct linux_sys_linkat_args *uap, register_t *retval)
531 {
532 	/* {
533 		syscallarg(int) fd1;
534 		syscallarg(const char *) name1;
535 		syscallarg(int) fd2;
536 		syscallarg(const char *) name2;
537 		syscallarg(int) flags;
538 	} */
539 	int fd1 = SCARG(uap, fd1);
540 	const char *name1 = SCARG(uap, name1);
541 	int fd2 = SCARG(uap, fd2);
542 	const char *name2 = SCARG(uap, name2);
543 	int follow;
544 
545 	follow = SCARG(uap, flags) & LINUX_AT_SYMLINK_FOLLOW;
546 
547 	return do_sys_linkat(l, fd1, name1, fd2, name2, follow, retval);
548 }
549 
550 static int
551 linux_unlink_dircheck(const char *path)
552 {
553 	struct nameidata nd;
554 	struct pathbuf *pb;
555 	int error;
556 
557 	/*
558 	 * Linux returns EISDIR if unlink(2) is called on a directory.
559 	 * We return EPERM in such cases. To emulate correct behaviour,
560 	 * check if the path points to directory and return EISDIR if this
561 	 * is the case.
562 	 *
563 	 * XXX this should really not copy in the path buffer twice...
564 	 */
565 	error = pathbuf_copyin(path, &pb);
566 	if (error) {
567 		return error;
568 	}
569 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
570 	if (namei(&nd) == 0) {
571 		struct stat sb;
572 
573 		if (vn_stat(nd.ni_vp, &sb) == 0
574 		    && S_ISDIR(sb.st_mode))
575 			error = EISDIR;
576 
577 		vput(nd.ni_vp);
578 	}
579 	pathbuf_destroy(pb);
580 	return error ? error : EPERM;
581 }
582 
583 int
584 linux_sys_unlink(struct lwp *l, const struct linux_sys_unlink_args *uap, register_t *retval)
585 {
586 	/* {
587 		syscallarg(const char *) path;
588 	} */
589 	int error;
590 
591 	error = sys_unlink(l, (const void *)uap, retval);
592 	if (error == EPERM)
593 		error = linux_unlink_dircheck(SCARG(uap, path));
594 
595 	return error;
596 }
597 
598 int
599 linux_sys_unlinkat(struct lwp *l, const struct linux_sys_unlinkat_args *uap, register_t *retval)
600 {
601 	/* {
602 		syscallarg(int) fd;
603 		syscallarg(const char *) path;
604 		syscallarg(int) flag;
605 	} */
606 	struct sys_unlinkat_args ua;
607 	int error;
608 
609 	SCARG(&ua, fd) = SCARG(uap, fd);
610 	SCARG(&ua, path) = SCARG(uap, path);
611 	SCARG(&ua, flag) = linux_to_bsd_atflags(SCARG(uap, flag));
612 
613 	error = sys_unlinkat(l, &ua, retval);
614 	if (error == EPERM)
615 		error = linux_unlink_dircheck(SCARG(uap, path));
616 
617 	return error;
618 }
619 
620 int
621 linux_sys_mknod(struct lwp *l, const struct linux_sys_mknod_args *uap, register_t *retval)
622 {
623 	/* {
624 		syscallarg(const char *) path;
625 		syscallarg(linux_umode_t) mode;
626 		syscallarg(unsigned) dev;
627 	} */
628 	struct linux_sys_mknodat_args ua;
629 
630 	SCARG(&ua, fd) = LINUX_AT_FDCWD;
631 	SCARG(&ua, path) = SCARG(uap, path);
632 	SCARG(&ua, mode) = SCARG(uap, mode);
633 	SCARG(&ua, dev) = SCARG(uap, dev);
634 
635 	return linux_sys_mknodat(l, &ua, retval);
636 }
637 
638 int
639 linux_sys_mknodat(struct lwp *l, const struct linux_sys_mknodat_args *uap, register_t *retval)
640 {
641 	/* {
642 		syscallarg(int) fd;
643 		syscallarg(const char *) path;
644 		syscallarg(linux_umode_t) mode;
645 		syscallarg(unsigned) dev;
646 	} */
647 
648 	/*
649 	 * BSD handles FIFOs separately
650 	 */
651 	if (S_ISFIFO(SCARG(uap, mode))) {
652 		struct sys_mkfifoat_args bma;
653 
654 		SCARG(&bma, fd) = SCARG(uap, fd);
655 		SCARG(&bma, path) = SCARG(uap, path);
656 		SCARG(&bma, mode) = SCARG(uap, mode);
657 		return sys_mkfifoat(l, &bma, retval);
658 	} else {
659 
660 		/*
661 		 * Linux device numbers uses 8 bits for minor and 8 bits
662 		 * for major. Due to how we map our major and minor,
663 		 * this just fits into our dev_t. Just mask off the
664 		 * upper 16bit to remove any random junk.
665 		 */
666 
667 		return do_sys_mknodat(l, SCARG(uap, fd), SCARG(uap, path),
668 		    SCARG(uap, mode), SCARG(uap, dev) & 0xffff, retval,
669 		    UIO_USERSPACE);
670 	}
671 }
672 
673 int
674 linux_sys_fchmodat(struct lwp *l, const struct linux_sys_fchmodat_args *uap, register_t *retval)
675 {
676 	/* {
677 		syscallarg(int) fd;
678 		syscallarg(const char *) path;
679 		syscallarg(linux_umode_t) mode;
680 	} */
681 
682 	return do_sys_chmodat(l, SCARG(uap, fd), SCARG(uap, path),
683 			      SCARG(uap, mode), AT_SYMLINK_FOLLOW);
684 }
685 
686 int
687 linux_sys_fchownat(struct lwp *l, const struct linux_sys_fchownat_args *uap, register_t *retval)
688 {
689 	/* {
690 		syscallarg(int) fd;
691 		syscallarg(const char *) path;
692 		syscallarg(uid_t) owner;
693 		syscallarg(gid_t) group;
694 		syscallarg(int) flag;
695 	} */
696 	int flag;
697 
698 	flag = linux_to_bsd_atflags(SCARG(uap, flag));
699 	return do_sys_chownat(l, SCARG(uap, fd), SCARG(uap, path),
700 			      SCARG(uap, owner), SCARG(uap, group), flag);
701 }
702 
703 int
704 linux_sys_faccessat(struct lwp *l, const struct linux_sys_faccessat_args *uap, register_t *retval)
705 {
706 	/* {
707 		syscallarg(int) fd;
708 		syscallarg(const char *) path;
709 		syscallarg(int) amode;
710 	} */
711 
712 	return do_sys_accessat(l, SCARG(uap, fd), SCARG(uap, path),
713 	     SCARG(uap, amode), AT_SYMLINK_FOLLOW);
714 }
715 
716 /*
717  * This is just fsync() for now (just as it is in the Linux kernel)
718  * Note: this is not implemented under Linux on Alpha and Arm
719  *	but should still be defined in our syscalls.master.
720  *	(syscall #148 on the arm)
721  */
722 int
723 linux_sys_fdatasync(struct lwp *l, const struct linux_sys_fdatasync_args *uap, register_t *retval)
724 {
725 	/* {
726 		syscallarg(int) fd;
727 	} */
728 
729 	return sys_fsync(l, (const void *)uap, retval);
730 }
731 
732 /*
733  * pread(2).
734  */
735 int
736 linux_sys_pread(struct lwp *l, const struct linux_sys_pread_args *uap, register_t *retval)
737 {
738 	/* {
739 		syscallarg(int) fd;
740 		syscallarg(void *) buf;
741 		syscallarg(size_t) nbyte;
742 		syscallarg(linux_off_t) offset;
743 	} */
744 	struct sys_pread_args pra;
745 
746 	SCARG(&pra, fd) = SCARG(uap, fd);
747 	SCARG(&pra, buf) = SCARG(uap, buf);
748 	SCARG(&pra, nbyte) = SCARG(uap, nbyte);
749 	SCARG(&pra, offset) = SCARG(uap, offset);
750 
751 	return sys_pread(l, &pra, retval);
752 }
753 
754 /*
755  * pwrite(2).
756  */
757 int
758 linux_sys_pwrite(struct lwp *l, const struct linux_sys_pwrite_args *uap, register_t *retval)
759 {
760 	/* {
761 		syscallarg(int) fd;
762 		syscallarg(void *) buf;
763 		syscallarg(size_t) nbyte;
764 		syscallarg(linux_off_t) offset;
765 	} */
766 	struct sys_pwrite_args pra;
767 
768 	SCARG(&pra, fd) = SCARG(uap, fd);
769 	SCARG(&pra, buf) = SCARG(uap, buf);
770 	SCARG(&pra, nbyte) = SCARG(uap, nbyte);
771 	SCARG(&pra, offset) = SCARG(uap, offset);
772 
773 	return sys_pwrite(l, &pra, retval);
774 }
775 
776 int
777 linux_sys_dup3(struct lwp *l, const struct linux_sys_dup3_args *uap,
778     register_t *retval)
779 {
780 	/* {
781 		syscallarg(int) from;
782 		syscallarg(int) to;
783 		syscallarg(int) flags;
784 	} */
785 	int error;
786 	if ((error = sys_dup2(l, (const struct sys_dup2_args *)uap, retval)))
787 		return error;
788 
789 	if (SCARG(uap, flags) & LINUX_O_CLOEXEC)
790 		fd_set_exclose(l, SCARG(uap, to), true);
791 
792 	return 0;
793 }
794 
795 
796 int
797 linux_to_bsd_atflags(int lflags)
798 {
799 	int bflags = 0;
800 
801 	if (lflags & LINUX_AT_SYMLINK_NOFOLLOW)
802 		bflags |= AT_SYMLINK_NOFOLLOW;
803 	if (lflags & LINUX_AT_REMOVEDIR)
804 		bflags |= AT_REMOVEDIR;
805 	if (lflags & LINUX_AT_SYMLINK_FOLLOW)
806 		bflags |= AT_SYMLINK_FOLLOW;
807 
808 	return bflags;
809 }
810 
811 
812 #define LINUX_NOT_SUPPORTED(fun) \
813 int \
814 fun(struct lwp *l, const struct fun##_args *uap, register_t *retval) \
815 { \
816 	return EOPNOTSUPP; \
817 }
818 
819 LINUX_NOT_SUPPORTED(linux_sys_setxattr)
820 LINUX_NOT_SUPPORTED(linux_sys_lsetxattr)
821 LINUX_NOT_SUPPORTED(linux_sys_fsetxattr)
822 
823 LINUX_NOT_SUPPORTED(linux_sys_getxattr)
824 LINUX_NOT_SUPPORTED(linux_sys_lgetxattr)
825 LINUX_NOT_SUPPORTED(linux_sys_fgetxattr)
826 
827 LINUX_NOT_SUPPORTED(linux_sys_listxattr)
828 LINUX_NOT_SUPPORTED(linux_sys_llistxattr)
829 LINUX_NOT_SUPPORTED(linux_sys_flistxattr)
830 
831 LINUX_NOT_SUPPORTED(linux_sys_removexattr)
832 LINUX_NOT_SUPPORTED(linux_sys_lremovexattr)
833 LINUX_NOT_SUPPORTED(linux_sys_fremovexattr)
834