xref: /netbsd-src/sys/compat/linux/common/linux_file.c (revision 181254a7b1bdde6873432bffef2d2decc4b5c22f)
1 /*	$NetBSD: linux_file.c,v 1.118 2020/05/23 23:42:41 ad 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 1.118 2020/05/23 23:42:41 ad 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/namei.h>
52 #include <sys/vnode.h>
53 #include <sys/tty.h>
54 #include <sys/socketvar.h>
55 #include <sys/conf.h>
56 #include <sys/pipe.h>
57 
58 #include <sys/syscallargs.h>
59 #include <sys/vfs_syscalls.h>
60 
61 #include <compat/linux/common/linux_types.h>
62 #include <compat/linux/common/linux_signal.h>
63 #include <compat/linux/common/linux_fcntl.h>
64 #include <compat/linux/common/linux_util.h>
65 #include <compat/linux/common/linux_machdep.h>
66 #include <compat/linux/common/linux_ipc.h>
67 #include <compat/linux/common/linux_sem.h>
68 
69 #include <compat/linux/linux_syscallargs.h>
70 
71 static int bsd_to_linux_ioflags(int);
72 #ifndef __amd64__
73 static void bsd_to_linux_stat(struct stat *, struct linux_stat *);
74 #endif
75 
76 conv_linux_flock(linux, flock)
77 
78 /*
79  * Some file-related calls are handled here. The usual flag conversion
80  * an structure conversion is done, and alternate emul path searching.
81  */
82 
83 /*
84  * The next two functions convert between the Linux and NetBSD values
85  * of the flags used in open(2) and fcntl(2).
86  */
87 int
88 linux_to_bsd_ioflags(int lflags)
89 {
90 	int res = 0;
91 
92 	res |= cvtto_bsd_mask(lflags, LINUX_O_WRONLY, O_WRONLY);
93 	res |= cvtto_bsd_mask(lflags, LINUX_O_RDONLY, O_RDONLY);
94 	res |= cvtto_bsd_mask(lflags, LINUX_O_RDWR, O_RDWR);
95 
96 	res |= cvtto_bsd_mask(lflags, LINUX_O_CREAT, O_CREAT);
97 	res |= cvtto_bsd_mask(lflags, LINUX_O_EXCL, O_EXCL);
98 	res |= cvtto_bsd_mask(lflags, LINUX_O_NOCTTY, O_NOCTTY);
99 	res |= cvtto_bsd_mask(lflags, LINUX_O_TRUNC, O_TRUNC);
100 	res |= cvtto_bsd_mask(lflags, LINUX_O_APPEND, O_APPEND);
101 	res |= cvtto_bsd_mask(lflags, LINUX_O_NONBLOCK, O_NONBLOCK);
102 	res |= cvtto_bsd_mask(lflags, LINUX_O_NDELAY, O_NDELAY);
103 	res |= cvtto_bsd_mask(lflags, LINUX_O_SYNC, O_FSYNC);
104 	res |= cvtto_bsd_mask(lflags, LINUX_FASYNC, O_ASYNC);
105 	res |= cvtto_bsd_mask(lflags, LINUX_O_DIRECT, O_DIRECT);
106 	res |= cvtto_bsd_mask(lflags, LINUX_O_DIRECTORY, O_DIRECTORY);
107 	res |= cvtto_bsd_mask(lflags, LINUX_O_NOFOLLOW, O_NOFOLLOW);
108 	res |= cvtto_bsd_mask(lflags, LINUX_O_CLOEXEC, O_CLOEXEC);
109 
110 	return res;
111 }
112 
113 static int
114 bsd_to_linux_ioflags(int bflags)
115 {
116 	int res = 0;
117 
118 	res |= cvtto_linux_mask(bflags, O_WRONLY, LINUX_O_WRONLY);
119 	res |= cvtto_linux_mask(bflags, O_RDONLY, LINUX_O_RDONLY);
120 	res |= cvtto_linux_mask(bflags, O_RDWR, LINUX_O_RDWR);
121 
122 	res |= cvtto_linux_mask(bflags, O_CREAT, LINUX_O_CREAT);
123 	res |= cvtto_linux_mask(bflags, O_EXCL, LINUX_O_EXCL);
124 	res |= cvtto_linux_mask(bflags, O_NOCTTY, LINUX_O_NOCTTY);
125 	res |= cvtto_linux_mask(bflags, O_TRUNC, LINUX_O_TRUNC);
126 	res |= cvtto_linux_mask(bflags, O_APPEND, LINUX_O_APPEND);
127 	res |= cvtto_linux_mask(bflags, O_NONBLOCK, LINUX_O_NONBLOCK);
128 	res |= cvtto_linux_mask(bflags, O_NDELAY, LINUX_O_NDELAY);
129 	res |= cvtto_linux_mask(bflags, O_FSYNC, LINUX_O_SYNC);
130 	res |= cvtto_linux_mask(bflags, O_ASYNC, LINUX_FASYNC);
131 	res |= cvtto_linux_mask(bflags, O_DIRECT, LINUX_O_DIRECT);
132 	res |= cvtto_linux_mask(bflags, O_DIRECTORY, LINUX_O_DIRECTORY);
133 	res |= cvtto_linux_mask(bflags, O_NOFOLLOW, LINUX_O_NOFOLLOW);
134 	res |= cvtto_linux_mask(bflags, O_CLOEXEC, LINUX_O_CLOEXEC);
135 
136 	return res;
137 }
138 
139 /*
140  * creat(2) is an obsolete function, but it's present as a Linux
141  * system call, so let's deal with it.
142  *
143  * Note: On the Alpha this doesn't really exist in Linux, but it's defined
144  * in syscalls.master anyway so this doesn't have to be special cased.
145  *
146  * Just call open(2) with the TRUNC, CREAT and WRONLY flags.
147  */
148 int
149 linux_sys_creat(struct lwp *l, const struct linux_sys_creat_args *uap, register_t *retval)
150 {
151 	/* {
152 		syscallarg(const char *) path;
153 		syscallarg(linux_umode_t) mode;
154 	} */
155 	struct sys_open_args oa;
156 
157 	SCARG(&oa, path) = SCARG(uap, path);
158 	SCARG(&oa, flags) = O_CREAT | O_TRUNC | O_WRONLY;
159 	SCARG(&oa, mode) = SCARG(uap, mode);
160 
161 	return sys_open(l, &oa, retval);
162 }
163 
164 static void
165 linux_open_ctty(struct lwp *l, int flags, int fd)
166 {
167 	struct proc *p = l->l_proc;
168 
169 	/*
170 	 * this bit from sunos_misc.c (and svr4_fcntl.c).
171 	 * If we are a session leader, and we don't have a controlling
172 	 * terminal yet, and the O_NOCTTY flag is not set, try to make
173 	 * this the controlling terminal.
174 	 */
175         if (!(flags & O_NOCTTY) && SESS_LEADER(p) && !(p->p_lflag & PL_CONTROLT)) {
176                 file_t *fp;
177 
178 		fp = fd_getfile(fd);
179 
180                 /* ignore any error, just give it a try */
181                 if (fp != NULL) {
182 			if (fp->f_type == DTYPE_VNODE) {
183 				(fp->f_ops->fo_ioctl) (fp, TIOCSCTTY, NULL);
184 			}
185 			fd_putfile(fd);
186 		}
187         }
188 }
189 
190 /*
191  * open(2). Take care of the different flag values, and let the
192  * NetBSD syscall do the real work. See if this operation
193  * gives the current process a controlling terminal.
194  * (XXX is this necessary?)
195  */
196 int
197 linux_sys_open(struct lwp *l, const struct linux_sys_open_args *uap, register_t *retval)
198 {
199 	/* {
200 		syscallarg(const char *) path;
201 		syscallarg(int) flags;
202 		syscallarg(linux_umode_t) mode;
203 	} */
204 	int error, fl;
205 	struct sys_open_args boa;
206 
207 	fl = linux_to_bsd_ioflags(SCARG(uap, flags));
208 
209 	SCARG(&boa, path) = SCARG(uap, path);
210 	SCARG(&boa, flags) = fl;
211 	SCARG(&boa, mode) = SCARG(uap, mode);
212 
213 	if ((error = sys_open(l, &boa, retval)))
214 		return (error == EFTYPE) ? ELOOP : error;
215 
216 	linux_open_ctty(l, fl, *retval);
217 	return 0;
218 }
219 
220 int
221 linux_sys_openat(struct lwp *l, const struct linux_sys_openat_args *uap, register_t *retval)
222 {
223 	/* {
224 		syscallarg(int) fd;
225 		syscallarg(const char *) path;
226 		syscallarg(int) flags;
227 		syscallarg(linux_umode_t) mode;
228 	} */
229 	int error, fl;
230 	struct sys_openat_args boa;
231 
232 	fl = linux_to_bsd_ioflags(SCARG(uap, flags));
233 
234 	SCARG(&boa, fd) = SCARG(uap, fd);
235 	SCARG(&boa, path) = SCARG(uap, path);
236 	SCARG(&boa, oflags) = fl;
237 	SCARG(&boa, mode) = SCARG(uap, mode);
238 
239 	if ((error = sys_openat(l, &boa, retval)))
240 		return (error == EFTYPE) ? ELOOP : error;
241 
242 	linux_open_ctty(l, fl, *retval);
243 	return 0;
244 }
245 
246 /*
247  * Most actions in the fcntl() call are straightforward; simply
248  * pass control to the NetBSD system call. A few commands need
249  * conversions after the actual system call has done its work,
250  * because the flag values and lock structure are different.
251  */
252 int
253 linux_sys_fcntl(struct lwp *l, const struct linux_sys_fcntl_args *uap, register_t *retval)
254 {
255 	/* {
256 		syscallarg(int) fd;
257 		syscallarg(int) cmd;
258 		syscallarg(void *) arg;
259 	} */
260 	struct proc *p = l->l_proc;
261 	int fd, cmd, error;
262 	u_long val;
263 	void *arg;
264 	struct sys_fcntl_args fca;
265 	file_t *fp;
266 	struct vnode *vp;
267 	struct vattr va;
268 	long pgid;
269 	struct pgrp *pgrp;
270 	struct tty *tp;
271 
272 	fd = SCARG(uap, fd);
273 	cmd = SCARG(uap, cmd);
274 	arg = SCARG(uap, arg);
275 
276 	switch (cmd) {
277 
278 	case LINUX_F_DUPFD:
279 		cmd = F_DUPFD;
280 		break;
281 
282 	case LINUX_F_GETFD:
283 		cmd = F_GETFD;
284 		break;
285 
286 	case LINUX_F_SETFD:
287 		cmd = F_SETFD;
288 		break;
289 
290 	case LINUX_F_GETFL:
291 		SCARG(&fca, fd) = fd;
292 		SCARG(&fca, cmd) = F_GETFL;
293 		SCARG(&fca, arg) = arg;
294 		if ((error = sys_fcntl(l, &fca, retval)))
295 			return error;
296 		retval[0] = bsd_to_linux_ioflags(retval[0]);
297 		return 0;
298 
299 	case LINUX_F_SETFL: {
300 		file_t	*fp1 = NULL;
301 
302 		val = linux_to_bsd_ioflags((unsigned long)SCARG(uap, arg));
303 		/*
304 		 * Linux seems to have same semantics for sending SIGIO to the
305 		 * read side of socket, but slightly different semantics
306 		 * for SIGIO to the write side.  Rather than sending the SIGIO
307 		 * every time it's possible to write (directly) more data, it
308 		 * only sends SIGIO if last write(2) failed due to insufficient
309 		 * memory to hold the data. This is compatible enough
310 		 * with NetBSD semantics to not do anything about the
311 		 * difference.
312 		 *
313 		 * Linux does NOT send SIGIO for pipes. Deal with socketpair
314 		 * ones and DTYPE_PIPE ones. For these, we don't set
315 		 * the underlying flags (we don't pass O_ASYNC flag down
316 		 * to sys_fcntl()), but set the FASYNC flag for file descriptor,
317 		 * so that F_GETFL would report the ASYNC i/o is on.
318 		 */
319 		if (val & O_ASYNC) {
320 			if (((fp1 = fd_getfile(fd)) == NULL))
321 			    return (EBADF);
322 			if (((fp1->f_type == DTYPE_SOCKET) && fp1->f_data
323 			      && ((struct socket *)fp1->f_data)->so_state & SS_ISAPIPE)
324 			    || (fp1->f_type == DTYPE_PIPE))
325 				val &= ~O_ASYNC;
326 			else {
327 				/* not a pipe, do not modify anything */
328 				fd_putfile(fd);
329 				fp1 = NULL;
330 			}
331 		}
332 
333 		SCARG(&fca, fd) = fd;
334 		SCARG(&fca, cmd) = F_SETFL;
335 		SCARG(&fca, arg) = (void *) val;
336 
337 		error = sys_fcntl(l, &fca, retval);
338 
339 		/* Now set the FASYNC flag for pipes */
340 		if (fp1) {
341 			if (!error) {
342 				mutex_enter(&fp1->f_lock);
343 				fp1->f_flag |= FASYNC;
344 				mutex_exit(&fp1->f_lock);
345 			}
346 			fd_putfile(fd);
347 		}
348 
349 		return (error);
350 	    }
351 
352 	case LINUX_F_GETLK:
353 		do_linux_getlk(fd, cmd, arg, linux, flock);
354 
355 	case LINUX_F_SETLK:
356 	case LINUX_F_SETLKW:
357 		do_linux_setlk(fd, cmd, arg, linux, flock, LINUX_F_SETLK);
358 
359 	case LINUX_F_SETOWN:
360 	case LINUX_F_GETOWN:
361 		/*
362 		 * We need to route fcntl() for tty descriptors around normal
363 		 * fcntl(), since NetBSD tty TIOC{G,S}PGRP semantics is too
364 		 * restrictive for Linux F_{G,S}ETOWN. For non-tty descriptors,
365 		 * this is not a problem.
366 		 */
367 		if ((fp = fd_getfile(fd)) == NULL)
368 			return EBADF;
369 
370 		/* Check it's a character device vnode */
371 		if (fp->f_type != DTYPE_VNODE
372 		    || (vp = (struct vnode *)fp->f_data) == NULL
373 		    || vp->v_type != VCHR) {
374 			fd_putfile(fd);
375 
376 	    not_tty:
377 			/* Not a tty, proceed with common fcntl() */
378 			cmd = cmd == LINUX_F_SETOWN ? F_SETOWN : F_GETOWN;
379 			break;
380 		}
381 
382 		vn_lock(vp, LK_SHARED | LK_RETRY);
383 		error = VOP_GETATTR(vp, &va, l->l_cred);
384 		VOP_UNLOCK(vp);
385 
386 		fd_putfile(fd);
387 
388 		if (error)
389 			return error;
390 
391 		if ((tp = cdev_tty(va.va_rdev)) == NULL)
392 			goto not_tty;
393 
394 		/* set tty pg_id appropriately */
395 		mutex_enter(&proc_lock);
396 		if (cmd == LINUX_F_GETOWN) {
397 			retval[0] = tp->t_pgrp ? tp->t_pgrp->pg_id : NO_PGID;
398 			mutex_exit(&proc_lock);
399 			return 0;
400 		}
401 		if ((long)arg <= 0) {
402 			pgid = -(long)arg;
403 		} else {
404 			struct proc *p1 = proc_find((long)arg);
405 			if (p1 == NULL) {
406 				mutex_exit(&proc_lock);
407 				return (ESRCH);
408 			}
409 			pgid = (long)p1->p_pgrp->pg_id;
410 		}
411 		pgrp = pgrp_find(pgid);
412 		if (pgrp == NULL || pgrp->pg_session != p->p_session) {
413 			mutex_exit(&proc_lock);
414 			return EPERM;
415 		}
416 		tp->t_pgrp = pgrp;
417 		mutex_exit(&proc_lock);
418 		return 0;
419 
420 	case LINUX_F_DUPFD_CLOEXEC:
421 		cmd = F_DUPFD_CLOEXEC;
422 		break;
423 
424 	default:
425 		return EOPNOTSUPP;
426 	}
427 
428 	SCARG(&fca, fd) = fd;
429 	SCARG(&fca, cmd) = cmd;
430 	SCARG(&fca, arg) = arg;
431 
432 	return sys_fcntl(l, &fca, retval);
433 }
434 
435 #if !defined(__amd64__)
436 /*
437  * Convert a NetBSD stat structure to a Linux stat structure.
438  * Only the order of the fields and the padding in the structure
439  * is different. linux_fakedev is a machine-dependent function
440  * which optionally converts device driver major/minor numbers
441  * (XXX horrible, but what can you do against code that compares
442  * things against constant major device numbers? sigh)
443  */
444 static void
445 bsd_to_linux_stat(struct stat *bsp, struct linux_stat *lsp)
446 {
447 
448 	lsp->lst_dev     = linux_fakedev(bsp->st_dev, 0);
449 	lsp->lst_ino     = bsp->st_ino;
450 	lsp->lst_mode    = (linux_mode_t)bsp->st_mode;
451 	if (bsp->st_nlink >= (1 << 15))
452 		lsp->lst_nlink = (1 << 15) - 1;
453 	else
454 		lsp->lst_nlink = (linux_nlink_t)bsp->st_nlink;
455 	lsp->lst_uid     = bsp->st_uid;
456 	lsp->lst_gid     = bsp->st_gid;
457 	lsp->lst_rdev    = linux_fakedev(bsp->st_rdev, 1);
458 	lsp->lst_size    = bsp->st_size;
459 	lsp->lst_blksize = bsp->st_blksize;
460 	lsp->lst_blocks  = bsp->st_blocks;
461 	lsp->lst_atime   = bsp->st_atime;
462 	lsp->lst_mtime   = bsp->st_mtime;
463 	lsp->lst_ctime   = bsp->st_ctime;
464 #ifdef LINUX_STAT_HAS_NSEC
465 	lsp->lst_atime_nsec   = bsp->st_atimensec;
466 	lsp->lst_mtime_nsec   = bsp->st_mtimensec;
467 	lsp->lst_ctime_nsec   = bsp->st_ctimensec;
468 #endif
469 }
470 
471 /*
472  * The stat functions below are plain sailing. stat and lstat are handled
473  * by one function to avoid code duplication.
474  */
475 int
476 linux_sys_fstat(struct lwp *l, const struct linux_sys_fstat_args *uap, register_t *retval)
477 {
478 	/* {
479 		syscallarg(int) fd;
480 		syscallarg(linux_stat *) sp;
481 	} */
482 	struct linux_stat tmplst;
483 	struct stat tmpst;
484 	int error;
485 
486 	error = do_sys_fstat(SCARG(uap, fd), &tmpst);
487 	if (error != 0)
488 		return error;
489 	bsd_to_linux_stat(&tmpst, &tmplst);
490 
491 	return copyout(&tmplst, SCARG(uap, sp), sizeof tmplst);
492 }
493 
494 static int
495 linux_stat1(const struct linux_sys_stat_args *uap, register_t *retval, int flags)
496 {
497 	struct linux_stat tmplst;
498 	struct stat tmpst;
499 	int error;
500 
501 	error = do_sys_stat(SCARG(uap, path), flags, &tmpst);
502 	if (error != 0)
503 		return error;
504 
505 	bsd_to_linux_stat(&tmpst, &tmplst);
506 
507 	return copyout(&tmplst, SCARG(uap, sp), sizeof tmplst);
508 }
509 
510 int
511 linux_sys_stat(struct lwp *l, const struct linux_sys_stat_args *uap, register_t *retval)
512 {
513 	/* {
514 		syscallarg(const char *) path;
515 		syscallarg(struct linux_stat *) sp;
516 	} */
517 
518 	return linux_stat1(uap, retval, FOLLOW);
519 }
520 
521 /* Note: this is "newlstat" in the Linux sources */
522 /*	(we don't bother with the old lstat currently) */
523 int
524 linux_sys_lstat(struct lwp *l, const struct linux_sys_lstat_args *uap, register_t *retval)
525 {
526 	/* {
527 		syscallarg(const char *) path;
528 		syscallarg(struct linux_stat *) sp;
529 	} */
530 
531 	return linux_stat1((const void *)uap, retval, NOFOLLOW);
532 }
533 #endif /* !__amd64__ */
534 
535 /*
536  * The following syscalls are mostly here because of the alternate path check.
537  */
538 
539 int
540 linux_sys_linkat(struct lwp *l, const struct linux_sys_linkat_args *uap, register_t *retval)
541 {
542 	/* {
543 		syscallarg(int) fd1;
544 		syscallarg(const char *) name1;
545 		syscallarg(int) fd2;
546 		syscallarg(const char *) name2;
547 		syscallarg(int) flags;
548 	} */
549 	int fd1 = SCARG(uap, fd1);
550 	const char *name1 = SCARG(uap, name1);
551 	int fd2 = SCARG(uap, fd2);
552 	const char *name2 = SCARG(uap, name2);
553 	int follow;
554 
555 	follow = SCARG(uap, flags) & LINUX_AT_SYMLINK_FOLLOW;
556 
557 	return do_sys_linkat(l, fd1, name1, fd2, name2, follow, retval);
558 }
559 
560 static int
561 linux_unlink_dircheck(const char *path)
562 {
563 	struct nameidata nd;
564 	struct pathbuf *pb;
565 	int error;
566 
567 	/*
568 	 * Linux returns EISDIR if unlink(2) is called on a directory.
569 	 * We return EPERM in such cases. To emulate correct behaviour,
570 	 * check if the path points to directory and return EISDIR if this
571 	 * is the case.
572 	 *
573 	 * XXX this should really not copy in the path buffer twice...
574 	 */
575 	error = pathbuf_copyin(path, &pb);
576 	if (error) {
577 		return error;
578 	}
579 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
580 	if (namei(&nd) == 0) {
581 		struct stat sb;
582 
583 		if (vn_stat(nd.ni_vp, &sb) == 0
584 		    && S_ISDIR(sb.st_mode))
585 			error = EISDIR;
586 
587 		vput(nd.ni_vp);
588 	}
589 	pathbuf_destroy(pb);
590 	return error ? error : EPERM;
591 }
592 
593 int
594 linux_sys_unlink(struct lwp *l, const struct linux_sys_unlink_args *uap, register_t *retval)
595 {
596 	/* {
597 		syscallarg(const char *) path;
598 	} */
599 	int error;
600 
601 	error = sys_unlink(l, (const void *)uap, retval);
602 	if (error == EPERM)
603 		error = linux_unlink_dircheck(SCARG(uap, path));
604 
605 	return error;
606 }
607 
608 int
609 linux_sys_unlinkat(struct lwp *l, const struct linux_sys_unlinkat_args *uap, register_t *retval)
610 {
611 	/* {
612 		syscallarg(int) fd;
613 		syscallarg(const char *) path;
614 		syscallarg(int) flag;
615 	} */
616 	struct sys_unlinkat_args ua;
617 	int error;
618 
619 	SCARG(&ua, fd) = SCARG(uap, fd);
620 	SCARG(&ua, path) = SCARG(uap, path);
621 	SCARG(&ua, flag) = linux_to_bsd_atflags(SCARG(uap, flag));
622 
623 	error = sys_unlinkat(l, &ua, retval);
624 	if (error == EPERM)
625 		error = linux_unlink_dircheck(SCARG(uap, path));
626 
627 	return error;
628 }
629 
630 int
631 linux_sys_mknod(struct lwp *l, const struct linux_sys_mknod_args *uap, register_t *retval)
632 {
633 	/* {
634 		syscallarg(const char *) path;
635 		syscallarg(linux_umode_t) mode;
636 		syscallarg(unsigned) dev;
637 	} */
638 	struct linux_sys_mknodat_args ua;
639 
640 	SCARG(&ua, fd) = LINUX_AT_FDCWD;
641 	SCARG(&ua, path) = SCARG(uap, path);
642 	SCARG(&ua, mode) = SCARG(uap, mode);
643 	SCARG(&ua, dev) = SCARG(uap, dev);
644 
645 	return linux_sys_mknodat(l, &ua, retval);
646 }
647 
648 int
649 linux_sys_mknodat(struct lwp *l, const struct linux_sys_mknodat_args *uap, register_t *retval)
650 {
651 	/* {
652 		syscallarg(int) fd;
653 		syscallarg(const char *) path;
654 		syscallarg(linux_umode_t) mode;
655 		syscallarg(unsigned) dev;
656 	} */
657 
658 	/*
659 	 * BSD handles FIFOs separately
660 	 */
661 	if (S_ISFIFO(SCARG(uap, mode))) {
662 		struct sys_mkfifoat_args bma;
663 
664 		SCARG(&bma, fd) = SCARG(uap, fd);
665 		SCARG(&bma, path) = SCARG(uap, path);
666 		SCARG(&bma, mode) = SCARG(uap, mode);
667 		return sys_mkfifoat(l, &bma, retval);
668 	} else {
669 
670 		/*
671 		 * Linux device numbers uses 8 bits for minor and 8 bits
672 		 * for major. Due to how we map our major and minor,
673 		 * this just fits into our dev_t. Just mask off the
674 		 * upper 16bit to remove any random junk.
675 		 */
676 
677 		return do_sys_mknodat(l, SCARG(uap, fd), SCARG(uap, path),
678 		    SCARG(uap, mode), SCARG(uap, dev) & 0xffff, UIO_USERSPACE);
679 	}
680 }
681 
682 int
683 linux_sys_fchmodat(struct lwp *l, const struct linux_sys_fchmodat_args *uap, register_t *retval)
684 {
685 	/* {
686 		syscallarg(int) fd;
687 		syscallarg(const char *) path;
688 		syscallarg(linux_umode_t) mode;
689 	} */
690 
691 	return do_sys_chmodat(l, SCARG(uap, fd), SCARG(uap, path),
692 			      SCARG(uap, mode), AT_SYMLINK_FOLLOW);
693 }
694 
695 int
696 linux_sys_fchownat(struct lwp *l, const struct linux_sys_fchownat_args *uap, register_t *retval)
697 {
698 	/* {
699 		syscallarg(int) fd;
700 		syscallarg(const char *) path;
701 		syscallarg(uid_t) owner;
702 		syscallarg(gid_t) group;
703 		syscallarg(int) flag;
704 	} */
705 	int flag;
706 
707 	flag = linux_to_bsd_atflags(SCARG(uap, flag));
708 	return do_sys_chownat(l, SCARG(uap, fd), SCARG(uap, path),
709 			      SCARG(uap, owner), SCARG(uap, group), flag);
710 }
711 
712 int
713 linux_sys_faccessat(struct lwp *l, const struct linux_sys_faccessat_args *uap, register_t *retval)
714 {
715 	/* {
716 		syscallarg(int) fd;
717 		syscallarg(const char *) path;
718 		syscallarg(int) amode;
719 	} */
720 
721 	return do_sys_accessat(l, SCARG(uap, fd), SCARG(uap, path),
722 	     SCARG(uap, amode), AT_SYMLINK_FOLLOW);
723 }
724 
725 /*
726  * This is just fsync() for now (just as it is in the Linux kernel)
727  * Note: this is not implemented under Linux on Alpha and Arm
728  *	but should still be defined in our syscalls.master.
729  *	(syscall #148 on the arm)
730  */
731 int
732 linux_sys_fdatasync(struct lwp *l, const struct linux_sys_fdatasync_args *uap, register_t *retval)
733 {
734 	/* {
735 		syscallarg(int) fd;
736 	} */
737 
738 	return sys_fsync(l, (const void *)uap, retval);
739 }
740 
741 /*
742  * pread(2).
743  */
744 int
745 linux_sys_pread(struct lwp *l, const struct linux_sys_pread_args *uap, register_t *retval)
746 {
747 	/* {
748 		syscallarg(int) fd;
749 		syscallarg(void *) buf;
750 		syscallarg(size_t) nbyte;
751 		syscallarg(off_t) offset;
752 	} */
753 	struct sys_pread_args pra;
754 
755 	SCARG(&pra, fd) = SCARG(uap, fd);
756 	SCARG(&pra, buf) = SCARG(uap, buf);
757 	SCARG(&pra, nbyte) = SCARG(uap, nbyte);
758 	SCARG(&pra, PAD) = 0;
759 	SCARG(&pra, offset) = SCARG(uap, offset);
760 
761 	return sys_pread(l, &pra, retval);
762 }
763 
764 /*
765  * pwrite(2).
766  */
767 int
768 linux_sys_pwrite(struct lwp *l, const struct linux_sys_pwrite_args *uap, register_t *retval)
769 {
770 	/* {
771 		syscallarg(int) fd;
772 		syscallarg(void *) buf;
773 		syscallarg(size_t) nbyte;
774 		syscallarg(off_t) offset;
775 	} */
776 	struct sys_pwrite_args pra;
777 
778 	SCARG(&pra, fd) = SCARG(uap, fd);
779 	SCARG(&pra, buf) = SCARG(uap, buf);
780 	SCARG(&pra, nbyte) = SCARG(uap, nbyte);
781 	SCARG(&pra, PAD) = 0;
782 	SCARG(&pra, offset) = SCARG(uap, offset);
783 
784 	return sys_pwrite(l, &pra, retval);
785 }
786 
787 int
788 linux_sys_dup3(struct lwp *l, const struct linux_sys_dup3_args *uap,
789     register_t *retval)
790 {
791 	/* {
792 		syscallarg(int) from;
793 		syscallarg(int) to;
794 		syscallarg(int) flags;
795 	} */
796 	int flags;
797 
798 	flags = linux_to_bsd_ioflags(SCARG(uap, flags));
799 	if ((flags & ~O_CLOEXEC) != 0)
800 		return EINVAL;
801 
802 	if (SCARG(uap, from) == SCARG(uap, to))
803 		return EINVAL;
804 
805 	return dodup(l, SCARG(uap, from), SCARG(uap, to), flags, retval);
806 }
807 
808 
809 int
810 linux_to_bsd_atflags(int lflags)
811 {
812 	int bflags = 0;
813 
814 	if (lflags & LINUX_AT_SYMLINK_NOFOLLOW)
815 		bflags |= AT_SYMLINK_NOFOLLOW;
816 	if (lflags & LINUX_AT_REMOVEDIR)
817 		bflags |= AT_REMOVEDIR;
818 	if (lflags & LINUX_AT_SYMLINK_FOLLOW)
819 		bflags |= AT_SYMLINK_FOLLOW;
820 
821 	return bflags;
822 }
823 
824 
825 #define LINUX_NOT_SUPPORTED(fun) \
826 int \
827 fun(struct lwp *l, const struct fun##_args *uap, register_t *retval) \
828 { \
829 	return EOPNOTSUPP; \
830 }
831 
832 LINUX_NOT_SUPPORTED(linux_sys_setxattr)
833 LINUX_NOT_SUPPORTED(linux_sys_lsetxattr)
834 LINUX_NOT_SUPPORTED(linux_sys_fsetxattr)
835 
836 LINUX_NOT_SUPPORTED(linux_sys_getxattr)
837 LINUX_NOT_SUPPORTED(linux_sys_lgetxattr)
838 LINUX_NOT_SUPPORTED(linux_sys_fgetxattr)
839 
840 LINUX_NOT_SUPPORTED(linux_sys_listxattr)
841 LINUX_NOT_SUPPORTED(linux_sys_llistxattr)
842 LINUX_NOT_SUPPORTED(linux_sys_flistxattr)
843 
844 LINUX_NOT_SUPPORTED(linux_sys_removexattr)
845 LINUX_NOT_SUPPORTED(linux_sys_lremovexattr)
846 LINUX_NOT_SUPPORTED(linux_sys_fremovexattr)
847 
848 /*
849  * For now just return EOPNOTSUPP, this makes glibc posix_fallocate()
850  * to fallback to emulation.
851  * XXX Right now no filesystem actually implements fallocate support,
852  * so no need for mapping.
853  */
854 LINUX_NOT_SUPPORTED(linux_sys_fallocate)
855