xref: /freebsd/sys/compat/linux/linux_file.c (revision 1e4896b176ff664dc9c2fce5426bf2fdf8017a7d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1994-1995 Søren Schmidt
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  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_compat.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/capsicum.h>
37 #include <sys/conf.h>
38 #include <sys/dirent.h>
39 #include <sys/fcntl.h>
40 #include <sys/file.h>
41 #include <sys/filedesc.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mman.h>
45 #include <sys/mount.h>
46 #include <sys/mutex.h>
47 #include <sys/namei.h>
48 #include <sys/proc.h>
49 #include <sys/stat.h>
50 #include <sys/sx.h>
51 #include <sys/syscallsubr.h>
52 #include <sys/sysproto.h>
53 #include <sys/tty.h>
54 #include <sys/unistd.h>
55 #include <sys/vnode.h>
56 
57 #ifdef COMPAT_LINUX32
58 #include <compat/freebsd32/freebsd32_misc.h>
59 #include <machine/../linux32/linux.h>
60 #include <machine/../linux32/linux32_proto.h>
61 #else
62 #include <machine/../linux/linux.h>
63 #include <machine/../linux/linux_proto.h>
64 #endif
65 #include <compat/linux/linux_misc.h>
66 #include <compat/linux/linux_util.h>
67 #include <compat/linux/linux_file.h>
68 
69 static int	linux_common_open(struct thread *, int, const char *, int, int,
70 		    enum uio_seg);
71 static int	linux_getdents_error(struct thread *, int, int);
72 
73 static struct bsd_to_linux_bitmap seal_bitmap[] = {
74 	BITMAP_1t1_LINUX(F_SEAL_SEAL),
75 	BITMAP_1t1_LINUX(F_SEAL_SHRINK),
76 	BITMAP_1t1_LINUX(F_SEAL_GROW),
77 	BITMAP_1t1_LINUX(F_SEAL_WRITE),
78 };
79 
80 #define	MFD_HUGETLB_ENTRY(_size)					\
81 	{								\
82 		.bsd_value = MFD_HUGE_##_size,				\
83 		.linux_value = LINUX_HUGETLB_FLAG_ENCODE_##_size	\
84 	}
85 static struct bsd_to_linux_bitmap mfd_bitmap[] = {
86 	BITMAP_1t1_LINUX(MFD_CLOEXEC),
87 	BITMAP_1t1_LINUX(MFD_ALLOW_SEALING),
88 	BITMAP_1t1_LINUX(MFD_HUGETLB),
89 	MFD_HUGETLB_ENTRY(64KB),
90 	MFD_HUGETLB_ENTRY(512KB),
91 	MFD_HUGETLB_ENTRY(1MB),
92 	MFD_HUGETLB_ENTRY(2MB),
93 	MFD_HUGETLB_ENTRY(8MB),
94 	MFD_HUGETLB_ENTRY(16MB),
95 	MFD_HUGETLB_ENTRY(32MB),
96 	MFD_HUGETLB_ENTRY(256MB),
97 	MFD_HUGETLB_ENTRY(512MB),
98 	MFD_HUGETLB_ENTRY(1GB),
99 	MFD_HUGETLB_ENTRY(2GB),
100 	MFD_HUGETLB_ENTRY(16GB),
101 };
102 #undef MFD_HUGETLB_ENTRY
103 
104 #ifdef LINUX_LEGACY_SYSCALLS
105 int
106 linux_creat(struct thread *td, struct linux_creat_args *args)
107 {
108 	char *path;
109 	int error;
110 
111 	if (!LUSECONVPATH(td)) {
112 		return (kern_openat(td, AT_FDCWD, args->path, UIO_USERSPACE,
113 		    O_WRONLY | O_CREAT | O_TRUNC, args->mode));
114 	}
115 	LCONVPATHEXIST(td, args->path, &path);
116 	error = kern_openat(td, AT_FDCWD, path, UIO_SYSSPACE,
117 	    O_WRONLY | O_CREAT | O_TRUNC, args->mode);
118 	LFREEPATH(path);
119 	return (error);
120 }
121 #endif
122 
123 static int
124 linux_common_openflags(int l_flags)
125 {
126 	int bsd_flags;
127 
128 	bsd_flags = 0;
129 	switch (l_flags & LINUX_O_ACCMODE) {
130 	case LINUX_O_WRONLY:
131 		bsd_flags |= O_WRONLY;
132 		break;
133 	case LINUX_O_RDWR:
134 		bsd_flags |= O_RDWR;
135 		break;
136 	default:
137 		bsd_flags |= O_RDONLY;
138 	}
139 	if (l_flags & LINUX_O_NDELAY)
140 		bsd_flags |= O_NONBLOCK;
141 	if (l_flags & LINUX_O_APPEND)
142 		bsd_flags |= O_APPEND;
143 	if (l_flags & LINUX_O_SYNC)
144 		bsd_flags |= O_FSYNC;
145 	if (l_flags & LINUX_O_CLOEXEC)
146 		bsd_flags |= O_CLOEXEC;
147 	if (l_flags & LINUX_O_NONBLOCK)
148 		bsd_flags |= O_NONBLOCK;
149 	if (l_flags & LINUX_O_ASYNC)
150 		bsd_flags |= O_ASYNC;
151 	if (l_flags & LINUX_O_CREAT)
152 		bsd_flags |= O_CREAT;
153 	if (l_flags & LINUX_O_TRUNC)
154 		bsd_flags |= O_TRUNC;
155 	if (l_flags & LINUX_O_EXCL)
156 		bsd_flags |= O_EXCL;
157 	if (l_flags & LINUX_O_NOCTTY)
158 		bsd_flags |= O_NOCTTY;
159 	if (l_flags & LINUX_O_DIRECT)
160 		bsd_flags |= O_DIRECT;
161 	if (l_flags & LINUX_O_NOFOLLOW)
162 		bsd_flags |= O_NOFOLLOW;
163 	if (l_flags & LINUX_O_DIRECTORY)
164 		bsd_flags |= O_DIRECTORY;
165 	/* XXX LINUX_O_NOATIME: unable to be easily implemented. */
166 	return (bsd_flags);
167 }
168 
169 static int
170 linux_common_open(struct thread *td, int dirfd, const char *path, int l_flags,
171     int mode, enum uio_seg seg)
172 {
173 	struct proc *p = td->td_proc;
174 	struct file *fp;
175 	int fd;
176 	int bsd_flags, error;
177 
178 	bsd_flags = linux_common_openflags(l_flags);
179 	error = kern_openat(td, dirfd, path, seg, bsd_flags, mode);
180 	if (error != 0) {
181 		if (error == EMLINK)
182 			error = ELOOP;
183 		goto done;
184 	}
185 	if (p->p_flag & P_CONTROLT)
186 		goto done;
187 	if (bsd_flags & O_NOCTTY)
188 		goto done;
189 
190 	/*
191 	 * XXX In between kern_openat() and fget(), another process
192 	 * having the same filedesc could use that fd without
193 	 * checking below.
194 	*/
195 	fd = td->td_retval[0];
196 	if (fget(td, fd, &cap_ioctl_rights, &fp) == 0) {
197 		if (fp->f_type != DTYPE_VNODE) {
198 			fdrop(fp, td);
199 			goto done;
200 		}
201 		sx_slock(&proctree_lock);
202 		PROC_LOCK(p);
203 		if (SESS_LEADER(p) && !(p->p_flag & P_CONTROLT)) {
204 			PROC_UNLOCK(p);
205 			sx_sunlock(&proctree_lock);
206 			/* XXXPJD: Verify if TIOCSCTTY is allowed. */
207 			(void) fo_ioctl(fp, TIOCSCTTY, (caddr_t) 0,
208 			    td->td_ucred, td);
209 		} else {
210 			PROC_UNLOCK(p);
211 			sx_sunlock(&proctree_lock);
212 		}
213 		fdrop(fp, td);
214 	}
215 
216 done:
217 	return (error);
218 }
219 
220 int
221 linux_openat(struct thread *td, struct linux_openat_args *args)
222 {
223 	char *path;
224 	int dfd, error;
225 
226 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
227 	if (!LUSECONVPATH(td)) {
228 		return (linux_common_open(td, dfd, args->filename, args->flags,
229 		    args->mode, UIO_USERSPACE));
230 	}
231 	if (args->flags & LINUX_O_CREAT)
232 		LCONVPATH_AT(td, args->filename, &path, 1, dfd);
233 	else
234 		LCONVPATH_AT(td, args->filename, &path, 0, dfd);
235 
236 	error = linux_common_open(td, dfd, path, args->flags, args->mode,
237 	    UIO_SYSSPACE);
238 	LFREEPATH(path);
239 	return (error);
240 }
241 
242 #ifdef LINUX_LEGACY_SYSCALLS
243 int
244 linux_open(struct thread *td, struct linux_open_args *args)
245 {
246 	char *path;
247 	int error;
248 
249 	if (!LUSECONVPATH(td)) {
250 		return (linux_common_open(td, AT_FDCWD, args->path, args->flags,
251 		    args->mode, UIO_USERSPACE));
252 	}
253 	if (args->flags & LINUX_O_CREAT)
254 		LCONVPATHCREAT(td, args->path, &path);
255 	else
256 		LCONVPATHEXIST(td, args->path, &path);
257 
258 	error = linux_common_open(td, AT_FDCWD, path, args->flags, args->mode,
259 	    UIO_SYSSPACE);
260 	LFREEPATH(path);
261 	return (error);
262 }
263 #endif
264 
265 int
266 linux_name_to_handle_at(struct thread *td,
267     struct linux_name_to_handle_at_args *args)
268 {
269 	static const l_int valid_flags = (LINUX_AT_SYMLINK_FOLLOW |
270 	    LINUX_AT_EMPTY_PATH);
271 	static const l_uint fh_size = sizeof(fhandle_t);
272 
273 	fhandle_t fh;
274 	l_uint fh_bytes;
275 	l_int mount_id;
276 	int error, fd, bsd_flags;
277 
278 	if (args->flags & ~valid_flags)
279 		return (EINVAL);
280 	if (args->flags & LINUX_AT_EMPTY_PATH)
281 		/* XXX: not supported yet */
282 		return (EOPNOTSUPP);
283 
284 	fd = args->dirfd;
285 	if (fd == LINUX_AT_FDCWD)
286 		fd = AT_FDCWD;
287 
288 	bsd_flags = 0;
289 	if (!(args->flags & LINUX_AT_SYMLINK_FOLLOW))
290 		bsd_flags |= AT_SYMLINK_NOFOLLOW;
291 
292 	if (!LUSECONVPATH(td)) {
293 		error = kern_getfhat(td, bsd_flags, fd, args->name,
294 		    UIO_USERSPACE, &fh, UIO_SYSSPACE);
295 	} else {
296 		char *path;
297 
298 		LCONVPATH_AT(td, args->name, &path, 0, fd);
299 		error = kern_getfhat(td, bsd_flags, fd, path, UIO_SYSSPACE,
300 		    &fh, UIO_SYSSPACE);
301 		LFREEPATH(path);
302 	}
303 	if (error != 0)
304 		return (error);
305 
306 	/* Emit mount_id -- required before EOVERFLOW case. */
307 	mount_id = (fh.fh_fsid.val[0] ^ fh.fh_fsid.val[1]);
308 	error = copyout(&mount_id, args->mnt_id, sizeof(mount_id));
309 	if (error != 0)
310 		return (error);
311 
312 	/* Check if there is room for handle. */
313 	error = copyin(&args->handle->handle_bytes, &fh_bytes,
314 	    sizeof(fh_bytes));
315 	if (error != 0)
316 		return (error);
317 
318 	if (fh_bytes < fh_size) {
319 		error = copyout(&fh_size, &args->handle->handle_bytes,
320 		    sizeof(fh_size));
321 		if (error == 0)
322 			error = EOVERFLOW;
323 		return (error);
324 	}
325 
326 	/* Emit handle. */
327 	mount_id = 0;
328 	/*
329 	 * We don't use handle_type for anything yet, but initialize a known
330 	 * value.
331 	 */
332 	error = copyout(&mount_id, &args->handle->handle_type,
333 	    sizeof(mount_id));
334 	if (error != 0)
335 		return (error);
336 
337 	error = copyout(&fh, &args->handle->f_handle,
338 	    sizeof(fh));
339 	return (error);
340 }
341 
342 int
343 linux_open_by_handle_at(struct thread *td,
344     struct linux_open_by_handle_at_args *args)
345 {
346 	l_uint fh_bytes;
347 	int bsd_flags, error;
348 
349 	error = copyin(&args->handle->handle_bytes, &fh_bytes,
350 	    sizeof(fh_bytes));
351 	if (error != 0)
352 		return (error);
353 
354 	if (fh_bytes < sizeof(fhandle_t))
355 		return (EINVAL);
356 
357 	bsd_flags = linux_common_openflags(args->flags);
358 	return (kern_fhopen(td, (void *)&args->handle->f_handle, bsd_flags));
359 }
360 
361 int
362 linux_lseek(struct thread *td, struct linux_lseek_args *args)
363 {
364 
365 	return (kern_lseek(td, args->fdes, args->off, args->whence));
366 }
367 
368 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
369 int
370 linux_llseek(struct thread *td, struct linux_llseek_args *args)
371 {
372 	int error;
373 	off_t off;
374 
375 	off = (args->olow) | (((off_t) args->ohigh) << 32);
376 
377 	error = kern_lseek(td, args->fd, off, args->whence);
378 	if (error != 0)
379 		return (error);
380 
381 	error = copyout(td->td_retval, args->res, sizeof(off_t));
382 	if (error != 0)
383 		return (error);
384 
385 	td->td_retval[0] = 0;
386 	return (0);
387 }
388 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
389 
390 /*
391  * Note that linux_getdents(2) and linux_getdents64(2) have the same
392  * arguments. They only differ in the definition of struct dirent they
393  * operate on.
394  * Note that linux_readdir(2) is a special case of linux_getdents(2)
395  * where count is always equals 1, meaning that the buffer is one
396  * dirent-structure in size and that the code can't handle more anyway.
397  * Note that linux_readdir(2) can't be implemented by means of linux_getdents(2)
398  * as in case when the *dent buffer size is equal to 1 linux_getdents(2) will
399  * trash user stack.
400  */
401 
402 static int
403 linux_getdents_error(struct thread *td, int fd, int err)
404 {
405 	struct vnode *vp;
406 	struct file *fp;
407 	int error;
408 
409 	/* Linux return ENOTDIR in case when fd is not a directory. */
410 	error = getvnode(td, fd, &cap_read_rights, &fp);
411 	if (error != 0)
412 		return (error);
413 	vp = fp->f_vnode;
414 	if (vp->v_type != VDIR) {
415 		fdrop(fp, td);
416 		return (ENOTDIR);
417 	}
418 	fdrop(fp, td);
419 	return (err);
420 }
421 
422 struct l_dirent {
423 	l_ulong		d_ino;
424 	l_off_t		d_off;
425 	l_ushort	d_reclen;
426 	char		d_name[LINUX_NAME_MAX + 1];
427 };
428 
429 struct l_dirent64 {
430 	uint64_t	d_ino;
431 	int64_t		d_off;
432 	l_ushort	d_reclen;
433 	u_char		d_type;
434 	char		d_name[LINUX_NAME_MAX + 1];
435 };
436 
437 /*
438  * Linux uses the last byte in the dirent buffer to store d_type,
439  * at least glibc-2.7 requires it. That is why l_dirent is padded with 2 bytes.
440  */
441 #define LINUX_RECLEN(namlen)						\
442     roundup(offsetof(struct l_dirent, d_name) + (namlen) + 2, sizeof(l_ulong))
443 
444 #define LINUX_RECLEN64(namlen)						\
445     roundup(offsetof(struct l_dirent64, d_name) + (namlen) + 1,		\
446     sizeof(uint64_t))
447 
448 #ifdef LINUX_LEGACY_SYSCALLS
449 int
450 linux_getdents(struct thread *td, struct linux_getdents_args *args)
451 {
452 	struct dirent *bdp;
453 	caddr_t inp, buf;		/* BSD-format */
454 	int len, reclen;		/* BSD-format */
455 	caddr_t outp;			/* Linux-format */
456 	int resid, linuxreclen;		/* Linux-format */
457 	caddr_t lbuf;			/* Linux-format */
458 	off_t base;
459 	struct l_dirent *linux_dirent;
460 	int buflen, error;
461 	size_t retval;
462 
463 	buflen = min(args->count, MAXBSIZE);
464 	buf = malloc(buflen, M_TEMP, M_WAITOK);
465 
466 	error = kern_getdirentries(td, args->fd, buf, buflen,
467 	    &base, NULL, UIO_SYSSPACE);
468 	if (error != 0) {
469 		error = linux_getdents_error(td, args->fd, error);
470 		goto out1;
471 	}
472 
473 	lbuf = malloc(LINUX_RECLEN(LINUX_NAME_MAX), M_TEMP, M_WAITOK | M_ZERO);
474 
475 	len = td->td_retval[0];
476 	inp = buf;
477 	outp = (caddr_t)args->dent;
478 	resid = args->count;
479 	retval = 0;
480 
481 	while (len > 0) {
482 		bdp = (struct dirent *) inp;
483 		reclen = bdp->d_reclen;
484 		linuxreclen = LINUX_RECLEN(bdp->d_namlen);
485 		/*
486 		 * No more space in the user supplied dirent buffer.
487 		 * Return EINVAL.
488 		 */
489 		if (resid < linuxreclen) {
490 			error = EINVAL;
491 			goto out;
492 		}
493 
494 		linux_dirent = (struct l_dirent*)lbuf;
495 		linux_dirent->d_ino = bdp->d_fileno;
496 		linux_dirent->d_off = base + reclen;
497 		linux_dirent->d_reclen = linuxreclen;
498 		/*
499 		 * Copy d_type to last byte of l_dirent buffer
500 		 */
501 		lbuf[linuxreclen - 1] = bdp->d_type;
502 		strlcpy(linux_dirent->d_name, bdp->d_name,
503 		    linuxreclen - offsetof(struct l_dirent, d_name)-1);
504 		error = copyout(linux_dirent, outp, linuxreclen);
505 		if (error != 0)
506 			goto out;
507 
508 		inp += reclen;
509 		base += reclen;
510 		len -= reclen;
511 
512 		retval += linuxreclen;
513 		outp += linuxreclen;
514 		resid -= linuxreclen;
515 	}
516 	td->td_retval[0] = retval;
517 
518 out:
519 	free(lbuf, M_TEMP);
520 out1:
521 	free(buf, M_TEMP);
522 	return (error);
523 }
524 #endif
525 
526 int
527 linux_getdents64(struct thread *td, struct linux_getdents64_args *args)
528 {
529 	struct dirent *bdp;
530 	caddr_t inp, buf;		/* BSD-format */
531 	int len, reclen;		/* BSD-format */
532 	caddr_t outp;			/* Linux-format */
533 	int resid, linuxreclen;		/* Linux-format */
534 	caddr_t lbuf;			/* Linux-format */
535 	off_t base;
536 	struct l_dirent64 *linux_dirent64;
537 	int buflen, error;
538 	size_t retval;
539 
540 	buflen = min(args->count, MAXBSIZE);
541 	buf = malloc(buflen, M_TEMP, M_WAITOK);
542 
543 	error = kern_getdirentries(td, args->fd, buf, buflen,
544 	    &base, NULL, UIO_SYSSPACE);
545 	if (error != 0) {
546 		error = linux_getdents_error(td, args->fd, error);
547 		goto out1;
548 	}
549 
550 	lbuf = malloc(LINUX_RECLEN64(LINUX_NAME_MAX), M_TEMP, M_WAITOK | M_ZERO);
551 
552 	len = td->td_retval[0];
553 	inp = buf;
554 	outp = (caddr_t)args->dirent;
555 	resid = args->count;
556 	retval = 0;
557 
558 	while (len > 0) {
559 		bdp = (struct dirent *) inp;
560 		reclen = bdp->d_reclen;
561 		linuxreclen = LINUX_RECLEN64(bdp->d_namlen);
562 		/*
563 		 * No more space in the user supplied dirent buffer.
564 		 * Return EINVAL.
565 		 */
566 		if (resid < linuxreclen) {
567 			error = EINVAL;
568 			goto out;
569 		}
570 
571 		linux_dirent64 = (struct l_dirent64*)lbuf;
572 		linux_dirent64->d_ino = bdp->d_fileno;
573 		linux_dirent64->d_off = base + reclen;
574 		linux_dirent64->d_reclen = linuxreclen;
575 		linux_dirent64->d_type = bdp->d_type;
576 		strlcpy(linux_dirent64->d_name, bdp->d_name,
577 		    linuxreclen - offsetof(struct l_dirent64, d_name));
578 		error = copyout(linux_dirent64, outp, linuxreclen);
579 		if (error != 0)
580 			goto out;
581 
582 		inp += reclen;
583 		base += reclen;
584 		len -= reclen;
585 
586 		retval += linuxreclen;
587 		outp += linuxreclen;
588 		resid -= linuxreclen;
589 	}
590 	td->td_retval[0] = retval;
591 
592 out:
593 	free(lbuf, M_TEMP);
594 out1:
595 	free(buf, M_TEMP);
596 	return (error);
597 }
598 
599 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
600 int
601 linux_readdir(struct thread *td, struct linux_readdir_args *args)
602 {
603 	struct dirent *bdp;
604 	caddr_t buf;			/* BSD-format */
605 	int linuxreclen;		/* Linux-format */
606 	caddr_t lbuf;			/* Linux-format */
607 	off_t base;
608 	struct l_dirent *linux_dirent;
609 	int buflen, error;
610 
611 	buflen = LINUX_RECLEN(LINUX_NAME_MAX);
612 	buf = malloc(buflen, M_TEMP, M_WAITOK);
613 
614 	error = kern_getdirentries(td, args->fd, buf, buflen,
615 	    &base, NULL, UIO_SYSSPACE);
616 	if (error != 0) {
617 		error = linux_getdents_error(td, args->fd, error);
618 		goto out;
619 	}
620 	if (td->td_retval[0] == 0)
621 		goto out;
622 
623 	lbuf = malloc(LINUX_RECLEN(LINUX_NAME_MAX), M_TEMP, M_WAITOK | M_ZERO);
624 
625 	bdp = (struct dirent *) buf;
626 	linuxreclen = LINUX_RECLEN(bdp->d_namlen);
627 
628 	linux_dirent = (struct l_dirent*)lbuf;
629 	linux_dirent->d_ino = bdp->d_fileno;
630 	linux_dirent->d_off = linuxreclen;
631 	linux_dirent->d_reclen = bdp->d_namlen;
632 	strlcpy(linux_dirent->d_name, bdp->d_name,
633 	    linuxreclen - offsetof(struct l_dirent, d_name));
634 	error = copyout(linux_dirent, args->dent, linuxreclen);
635 	if (error == 0)
636 		td->td_retval[0] = linuxreclen;
637 
638 	free(lbuf, M_TEMP);
639 out:
640 	free(buf, M_TEMP);
641 	return (error);
642 }
643 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
644 
645 /*
646  * These exist mainly for hooks for doing /compat/linux translation.
647  */
648 
649 #ifdef LINUX_LEGACY_SYSCALLS
650 int
651 linux_access(struct thread *td, struct linux_access_args *args)
652 {
653 	char *path;
654 	int error;
655 
656 	/* Linux convention. */
657 	if (args->amode & ~(F_OK | X_OK | W_OK | R_OK))
658 		return (EINVAL);
659 
660 	if (!LUSECONVPATH(td)) {
661 		error = kern_accessat(td, AT_FDCWD, args->path, UIO_USERSPACE, 0,
662 		    args->amode);
663 	} else {
664 		LCONVPATHEXIST(td, args->path, &path);
665 		error = kern_accessat(td, AT_FDCWD, path, UIO_SYSSPACE, 0,
666 		    args->amode);
667 		LFREEPATH(path);
668 	}
669 
670 	return (error);
671 }
672 #endif
673 
674 int
675 linux_faccessat(struct thread *td, struct linux_faccessat_args *args)
676 {
677 	char *path;
678 	int error, dfd;
679 
680 	/* Linux convention. */
681 	if (args->amode & ~(F_OK | X_OK | W_OK | R_OK))
682 		return (EINVAL);
683 
684 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
685 	if (!LUSECONVPATH(td)) {
686 		error = kern_accessat(td, dfd, args->filename, UIO_USERSPACE, 0, args->amode);
687 	} else {
688 		LCONVPATHEXIST_AT(td, args->filename, &path, dfd);
689 		error = kern_accessat(td, dfd, path, UIO_SYSSPACE, 0, args->amode);
690 		LFREEPATH(path);
691 	}
692 
693 	return (error);
694 }
695 
696 #ifdef LINUX_LEGACY_SYSCALLS
697 int
698 linux_unlink(struct thread *td, struct linux_unlink_args *args)
699 {
700 	char *path;
701 	int error;
702 	struct stat st;
703 
704 	if (!LUSECONVPATH(td)) {
705 		error = kern_funlinkat(td, AT_FDCWD, args->path, FD_NONE,
706 		    UIO_USERSPACE, 0, 0);
707 		if (error == EPERM) {
708 			/* Introduce POSIX noncompliant behaviour of Linux */
709 			if (kern_statat(td, 0, AT_FDCWD, args->path,
710 			    UIO_SYSSPACE, &st, NULL) == 0) {
711 				if (S_ISDIR(st.st_mode))
712 					error = EISDIR;
713 			}
714 		}
715 	} else {
716 		LCONVPATHEXIST(td, args->path, &path);
717 		error = kern_funlinkat(td, AT_FDCWD, path, FD_NONE, UIO_SYSSPACE, 0, 0);
718 		if (error == EPERM) {
719 			/* Introduce POSIX noncompliant behaviour of Linux */
720 			if (kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &st,
721 			    NULL) == 0) {
722 				if (S_ISDIR(st.st_mode))
723 					error = EISDIR;
724 			}
725 		}
726 		LFREEPATH(path);
727 	}
728 
729 	return (error);
730 }
731 #endif
732 
733 static int
734 linux_unlinkat_impl(struct thread *td, enum uio_seg pathseg, const char *path,
735     int dfd, struct linux_unlinkat_args *args)
736 {
737 	struct stat st;
738 	int error;
739 
740 	if (args->flag & LINUX_AT_REMOVEDIR)
741 		error = kern_frmdirat(td, dfd, path, FD_NONE, pathseg, 0);
742 	else
743 		error = kern_funlinkat(td, dfd, path, FD_NONE, pathseg, 0, 0);
744 	if (error == EPERM && !(args->flag & LINUX_AT_REMOVEDIR)) {
745 		/* Introduce POSIX noncompliant behaviour of Linux */
746 		if (kern_statat(td, AT_SYMLINK_NOFOLLOW, dfd, path,
747 		    UIO_SYSSPACE, &st, NULL) == 0 && S_ISDIR(st.st_mode))
748 			error = EISDIR;
749 	}
750 	return (error);
751 }
752 
753 int
754 linux_unlinkat(struct thread *td, struct linux_unlinkat_args *args)
755 {
756 	char *path;
757 	int error, dfd;
758 
759 	if (args->flag & ~LINUX_AT_REMOVEDIR)
760 		return (EINVAL);
761 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
762 	if (!LUSECONVPATH(td)) {
763 		return (linux_unlinkat_impl(td, UIO_USERSPACE, args->pathname,
764 		    dfd, args));
765 	}
766 	LCONVPATHEXIST_AT(td, args->pathname, &path, dfd);
767 	error = linux_unlinkat_impl(td, UIO_SYSSPACE, path, dfd, args);
768 	LFREEPATH(path);
769 	return (error);
770 }
771 int
772 linux_chdir(struct thread *td, struct linux_chdir_args *args)
773 {
774 	char *path;
775 	int error;
776 
777 	if (!LUSECONVPATH(td)) {
778 		return (kern_chdir(td, args->path, UIO_USERSPACE));
779 	}
780 	LCONVPATHEXIST(td, args->path, &path);
781 	error = kern_chdir(td, path, UIO_SYSSPACE);
782 	LFREEPATH(path);
783 	return (error);
784 }
785 
786 #ifdef LINUX_LEGACY_SYSCALLS
787 int
788 linux_chmod(struct thread *td, struct linux_chmod_args *args)
789 {
790 	char *path;
791 	int error;
792 
793 	if (!LUSECONVPATH(td)) {
794 		return (kern_fchmodat(td, AT_FDCWD, args->path, UIO_USERSPACE,
795 		    args->mode, 0));
796 	}
797 	LCONVPATHEXIST(td, args->path, &path);
798 	error = kern_fchmodat(td, AT_FDCWD, path, UIO_SYSSPACE, args->mode, 0);
799 	LFREEPATH(path);
800 	return (error);
801 }
802 #endif
803 
804 int
805 linux_fchmodat(struct thread *td, struct linux_fchmodat_args *args)
806 {
807 	char *path;
808 	int error, dfd;
809 
810 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
811 	if (!LUSECONVPATH(td)) {
812 		return (kern_fchmodat(td, dfd, args->filename, UIO_USERSPACE,
813 		    args->mode, 0));
814 	}
815 	LCONVPATHEXIST_AT(td, args->filename, &path, dfd);
816 	error = kern_fchmodat(td, dfd, path, UIO_SYSSPACE, args->mode, 0);
817 	LFREEPATH(path);
818 	return (error);
819 }
820 
821 #ifdef LINUX_LEGACY_SYSCALLS
822 int
823 linux_mkdir(struct thread *td, struct linux_mkdir_args *args)
824 {
825 	char *path;
826 	int error;
827 
828 	if (!LUSECONVPATH(td)) {
829 		return (kern_mkdirat(td, AT_FDCWD, args->path, UIO_USERSPACE, args->mode));
830 	}
831 	LCONVPATHCREAT(td, args->path, &path);
832 	error = kern_mkdirat(td, AT_FDCWD, path, UIO_SYSSPACE, args->mode);
833 	LFREEPATH(path);
834 	return (error);
835 }
836 #endif
837 
838 int
839 linux_mkdirat(struct thread *td, struct linux_mkdirat_args *args)
840 {
841 	char *path;
842 	int error, dfd;
843 
844 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
845 	if (!LUSECONVPATH(td)) {
846 		return (kern_mkdirat(td, dfd, args->pathname, UIO_USERSPACE, args->mode));
847 	}
848 	LCONVPATHCREAT_AT(td, args->pathname, &path, dfd);
849 	error = kern_mkdirat(td, dfd, path, UIO_SYSSPACE, args->mode);
850 	LFREEPATH(path);
851 	return (error);
852 }
853 
854 #ifdef LINUX_LEGACY_SYSCALLS
855 int
856 linux_rmdir(struct thread *td, struct linux_rmdir_args *args)
857 {
858 	char *path;
859 	int error;
860 
861 	if (!LUSECONVPATH(td)) {
862 		return (kern_frmdirat(td, AT_FDCWD, args->path, FD_NONE,
863 		    UIO_USERSPACE, 0));
864 	}
865 	LCONVPATHEXIST(td, args->path, &path);
866 	error = kern_frmdirat(td, AT_FDCWD, path, FD_NONE, UIO_SYSSPACE, 0);
867 	LFREEPATH(path);
868 	return (error);
869 }
870 
871 int
872 linux_rename(struct thread *td, struct linux_rename_args *args)
873 {
874 	char *from, *to;
875 	int error;
876 
877 	if (!LUSECONVPATH(td)) {
878 		return (kern_renameat(td, AT_FDCWD, args->from, AT_FDCWD,
879 		    args->to, UIO_USERSPACE));
880 	}
881 	LCONVPATHEXIST(td, args->from, &from);
882 	/* Expand LCONVPATHCREATE so that `from' can be freed on errors */
883 	error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD);
884 	if (to == NULL) {
885 		LFREEPATH(from);
886 		return (error);
887 	}
888 	error = kern_renameat(td, AT_FDCWD, from, AT_FDCWD, to, UIO_SYSSPACE);
889 	LFREEPATH(from);
890 	LFREEPATH(to);
891 	return (error);
892 }
893 #endif
894 
895 int
896 linux_renameat(struct thread *td, struct linux_renameat_args *args)
897 {
898 	struct linux_renameat2_args renameat2_args = {
899 	    .olddfd = args->olddfd,
900 	    .oldname = args->oldname,
901 	    .newdfd = args->newdfd,
902 	    .newname = args->newname,
903 	    .flags = 0
904 	};
905 
906 	return (linux_renameat2(td, &renameat2_args));
907 }
908 
909 int
910 linux_renameat2(struct thread *td, struct linux_renameat2_args *args)
911 {
912 	char *from, *to;
913 	int error, olddfd, newdfd;
914 
915 	if (args->flags != 0) {
916 		if (args->flags & ~(LINUX_RENAME_EXCHANGE |
917 		    LINUX_RENAME_NOREPLACE | LINUX_RENAME_WHITEOUT))
918 			return (EINVAL);
919 		if (args->flags & LINUX_RENAME_EXCHANGE &&
920 		    args->flags & (LINUX_RENAME_NOREPLACE |
921 		    LINUX_RENAME_WHITEOUT))
922 			return (EINVAL);
923 #if 0
924 		/*
925 		 * This spams the console on Ubuntu Focal.
926 		 *
927 		 * What's needed here is a general mechanism to let users know
928 		 * about missing features without hogging the system.
929 		 */
930 		linux_msg(td, "renameat2 unsupported flags 0x%x",
931 		    args->flags);
932 #endif
933 		return (EINVAL);
934 	}
935 
936 	olddfd = (args->olddfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->olddfd;
937 	newdfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd;
938 	if (!LUSECONVPATH(td)) {
939 		return (kern_renameat(td, olddfd, args->oldname, newdfd,
940 		    args->newname, UIO_USERSPACE));
941 	}
942 	LCONVPATHEXIST_AT(td, args->oldname, &from, olddfd);
943 	/* Expand LCONVPATHCREATE so that `from' can be freed on errors */
944 	error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, newdfd);
945 	if (to == NULL) {
946 		LFREEPATH(from);
947 		return (error);
948 	}
949 	error = kern_renameat(td, olddfd, from, newdfd, to, UIO_SYSSPACE);
950 	LFREEPATH(from);
951 	LFREEPATH(to);
952 	return (error);
953 }
954 
955 #ifdef LINUX_LEGACY_SYSCALLS
956 int
957 linux_symlink(struct thread *td, struct linux_symlink_args *args)
958 {
959 	char *path, *to;
960 	int error;
961 
962 	if (!LUSECONVPATH(td)) {
963 		return (kern_symlinkat(td, args->path, AT_FDCWD, args->to,
964 		    UIO_USERSPACE));
965 	}
966 	LCONVPATHEXIST(td, args->path, &path);
967 	/* Expand LCONVPATHCREATE so that `path' can be freed on errors */
968 	error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD);
969 	if (to == NULL) {
970 		LFREEPATH(path);
971 		return (error);
972 	}
973 	error = kern_symlinkat(td, path, AT_FDCWD, to, UIO_SYSSPACE);
974 	LFREEPATH(path);
975 	LFREEPATH(to);
976 	return (error);
977 }
978 #endif
979 
980 int
981 linux_symlinkat(struct thread *td, struct linux_symlinkat_args *args)
982 {
983 	char *path, *to;
984 	int error, dfd;
985 
986 	dfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd;
987 	if (!LUSECONVPATH(td)) {
988 		return (kern_symlinkat(td, args->oldname, dfd, args->newname,
989 		    UIO_USERSPACE));
990 	}
991 	LCONVPATHEXIST(td, args->oldname, &path);
992 	/* Expand LCONVPATHCREATE so that `path' can be freed on errors */
993 	error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, dfd);
994 	if (to == NULL) {
995 		LFREEPATH(path);
996 		return (error);
997 	}
998 	error = kern_symlinkat(td, path, dfd, to, UIO_SYSSPACE);
999 	LFREEPATH(path);
1000 	LFREEPATH(to);
1001 	return (error);
1002 }
1003 
1004 #ifdef LINUX_LEGACY_SYSCALLS
1005 int
1006 linux_readlink(struct thread *td, struct linux_readlink_args *args)
1007 {
1008 	char *name;
1009 	int error;
1010 
1011 	if (!LUSECONVPATH(td)) {
1012 		return (kern_readlinkat(td, AT_FDCWD, args->name, UIO_USERSPACE,
1013 		    args->buf, UIO_USERSPACE, args->count));
1014 	}
1015 	LCONVPATHEXIST(td, args->name, &name);
1016 	error = kern_readlinkat(td, AT_FDCWD, name, UIO_SYSSPACE,
1017 	    args->buf, UIO_USERSPACE, args->count);
1018 	LFREEPATH(name);
1019 	return (error);
1020 }
1021 #endif
1022 
1023 int
1024 linux_readlinkat(struct thread *td, struct linux_readlinkat_args *args)
1025 {
1026 	char *name;
1027 	int error, dfd;
1028 
1029 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
1030 	if (!LUSECONVPATH(td)) {
1031 		return (kern_readlinkat(td, dfd, args->path, UIO_USERSPACE,
1032 		    args->buf, UIO_USERSPACE, args->bufsiz));
1033 	}
1034 	LCONVPATHEXIST_AT(td, args->path, &name, dfd);
1035 	error = kern_readlinkat(td, dfd, name, UIO_SYSSPACE, args->buf,
1036 	    UIO_USERSPACE, args->bufsiz);
1037 	LFREEPATH(name);
1038 	return (error);
1039 }
1040 
1041 int
1042 linux_truncate(struct thread *td, struct linux_truncate_args *args)
1043 {
1044 	char *path;
1045 	int error;
1046 
1047 	if (!LUSECONVPATH(td)) {
1048 		return (kern_truncate(td, args->path, UIO_USERSPACE, args->length));
1049 	}
1050 	LCONVPATHEXIST(td, args->path, &path);
1051 	error = kern_truncate(td, path, UIO_SYSSPACE, args->length);
1052 	LFREEPATH(path);
1053 	return (error);
1054 }
1055 
1056 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1057 int
1058 linux_truncate64(struct thread *td, struct linux_truncate64_args *args)
1059 {
1060 	char *path;
1061 	off_t length;
1062 	int error;
1063 
1064 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1065 	length = PAIR32TO64(off_t, args->length);
1066 #else
1067 	length = args->length;
1068 #endif
1069 
1070 	if (!LUSECONVPATH(td)) {
1071 		return (kern_truncate(td, args->path, UIO_USERSPACE, length));
1072 	}
1073 	LCONVPATHEXIST(td, args->path, &path);
1074 	error = kern_truncate(td, path, UIO_SYSSPACE, length);
1075 	LFREEPATH(path);
1076 	return (error);
1077 }
1078 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1079 
1080 int
1081 linux_ftruncate(struct thread *td, struct linux_ftruncate_args *args)
1082 {
1083 
1084 	return (kern_ftruncate(td, args->fd, args->length));
1085 }
1086 
1087 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1088 int
1089 linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args)
1090 {
1091 	off_t length;
1092 
1093 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1094 	length = PAIR32TO64(off_t, args->length);
1095 #else
1096 	length = args->length;
1097 #endif
1098 
1099 	return (kern_ftruncate(td, args->fd, length));
1100 }
1101 #endif
1102 
1103 #ifdef LINUX_LEGACY_SYSCALLS
1104 int
1105 linux_link(struct thread *td, struct linux_link_args *args)
1106 {
1107 	char *path, *to;
1108 	int error;
1109 
1110 	if (!LUSECONVPATH(td)) {
1111 		return (kern_linkat(td, AT_FDCWD, AT_FDCWD, args->path, args->to,
1112 		    UIO_USERSPACE, FOLLOW));
1113 	}
1114 	LCONVPATHEXIST(td, args->path, &path);
1115 	/* Expand LCONVPATHCREATE so that `path' can be freed on errors */
1116 	error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD);
1117 	if (to == NULL) {
1118 		LFREEPATH(path);
1119 		return (error);
1120 	}
1121 	error = kern_linkat(td, AT_FDCWD, AT_FDCWD, path, to, UIO_SYSSPACE,
1122 	    FOLLOW);
1123 	LFREEPATH(path);
1124 	LFREEPATH(to);
1125 	return (error);
1126 }
1127 #endif
1128 
1129 int
1130 linux_linkat(struct thread *td, struct linux_linkat_args *args)
1131 {
1132 	char *path, *to;
1133 	int error, olddfd, newdfd, follow;
1134 
1135 	if (args->flag & ~LINUX_AT_SYMLINK_FOLLOW)
1136 		return (EINVAL);
1137 
1138 	olddfd = (args->olddfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->olddfd;
1139 	newdfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd;
1140 	follow = (args->flag & LINUX_AT_SYMLINK_FOLLOW) == 0 ? NOFOLLOW :
1141 	    FOLLOW;
1142 	if (!LUSECONVPATH(td)) {
1143 		return (kern_linkat(td, olddfd, newdfd, args->oldname,
1144 		    args->newname, UIO_USERSPACE, follow));
1145 	}
1146 	LCONVPATHEXIST_AT(td, args->oldname, &path, olddfd);
1147 	/* Expand LCONVPATHCREATE so that `path' can be freed on errors */
1148 	error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, newdfd);
1149 	if (to == NULL) {
1150 		LFREEPATH(path);
1151 		return (error);
1152 	}
1153 	error = kern_linkat(td, olddfd, newdfd, path, to, UIO_SYSSPACE, follow);
1154 	LFREEPATH(path);
1155 	LFREEPATH(to);
1156 	return (error);
1157 }
1158 
1159 int
1160 linux_fdatasync(struct thread *td, struct linux_fdatasync_args *uap)
1161 {
1162 
1163 	return (kern_fsync(td, uap->fd, false));
1164 }
1165 
1166 int
1167 linux_sync_file_range(struct thread *td, struct linux_sync_file_range_args *uap)
1168 {
1169 	off_t nbytes, offset;
1170 
1171 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1172 	nbytes = PAIR32TO64(off_t, uap->nbytes);
1173 	offset = PAIR32TO64(off_t, uap->offset);
1174 #else
1175 	nbytes = uap->nbytes;
1176 	offset = uap->offset;
1177 #endif
1178 
1179 	if (offset < 0 || nbytes < 0 ||
1180 	    (uap->flags & ~(LINUX_SYNC_FILE_RANGE_WAIT_BEFORE |
1181 	    LINUX_SYNC_FILE_RANGE_WRITE |
1182 	    LINUX_SYNC_FILE_RANGE_WAIT_AFTER)) != 0) {
1183 		return (EINVAL);
1184 	}
1185 
1186 	return (kern_fsync(td, uap->fd, false));
1187 }
1188 
1189 int
1190 linux_pread(struct thread *td, struct linux_pread_args *uap)
1191 {
1192 	struct vnode *vp;
1193 	off_t offset;
1194 	int error;
1195 
1196 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1197 	offset = PAIR32TO64(off_t, uap->offset);
1198 #else
1199 	offset = uap->offset;
1200 #endif
1201 
1202 	error = kern_pread(td, uap->fd, uap->buf, uap->nbyte, offset);
1203 	if (error == 0) {
1204 		/* This seems to violate POSIX but Linux does it. */
1205 		error = fgetvp(td, uap->fd, &cap_pread_rights, &vp);
1206 		if (error != 0)
1207 			return (error);
1208 		if (vp->v_type == VDIR)
1209 			error = EISDIR;
1210 		vrele(vp);
1211 	}
1212 	return (error);
1213 }
1214 
1215 int
1216 linux_pwrite(struct thread *td, struct linux_pwrite_args *uap)
1217 {
1218 	off_t offset;
1219 
1220 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1221 	offset = PAIR32TO64(off_t, uap->offset);
1222 #else
1223 	offset = uap->offset;
1224 #endif
1225 
1226 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, offset));
1227 }
1228 
1229 int
1230 linux_preadv(struct thread *td, struct linux_preadv_args *uap)
1231 {
1232 	struct uio *auio;
1233 	int error;
1234 	off_t offset;
1235 
1236 	/*
1237 	 * According http://man7.org/linux/man-pages/man2/preadv.2.html#NOTES
1238 	 * pos_l and pos_h, respectively, contain the
1239 	 * low order and high order 32 bits of offset.
1240 	 */
1241 	offset = (((off_t)uap->pos_h << (sizeof(offset) * 4)) <<
1242 	    (sizeof(offset) * 4)) | uap->pos_l;
1243 	if (offset < 0)
1244 		return (EINVAL);
1245 #ifdef COMPAT_LINUX32
1246 	error = linux32_copyinuio(PTRIN(uap->vec), uap->vlen, &auio);
1247 #else
1248 	error = copyinuio(uap->vec, uap->vlen, &auio);
1249 #endif
1250 	if (error != 0)
1251 		return (error);
1252 	error = kern_preadv(td, uap->fd, auio, offset);
1253 	free(auio, M_IOV);
1254 	return (error);
1255 }
1256 
1257 int
1258 linux_pwritev(struct thread *td, struct linux_pwritev_args *uap)
1259 {
1260 	struct uio *auio;
1261 	int error;
1262 	off_t offset;
1263 
1264 	/*
1265 	 * According http://man7.org/linux/man-pages/man2/pwritev.2.html#NOTES
1266 	 * pos_l and pos_h, respectively, contain the
1267 	 * low order and high order 32 bits of offset.
1268 	 */
1269 	offset = (((off_t)uap->pos_h << (sizeof(offset) * 4)) <<
1270 	    (sizeof(offset) * 4)) | uap->pos_l;
1271 	if (offset < 0)
1272 		return (EINVAL);
1273 #ifdef COMPAT_LINUX32
1274 	error = linux32_copyinuio(PTRIN(uap->vec), uap->vlen, &auio);
1275 #else
1276 	error = copyinuio(uap->vec, uap->vlen, &auio);
1277 #endif
1278 	if (error != 0)
1279 		return (error);
1280 	error = kern_pwritev(td, uap->fd, auio, offset);
1281 	free(auio, M_IOV);
1282 	return (error);
1283 }
1284 
1285 int
1286 linux_mount(struct thread *td, struct linux_mount_args *args)
1287 {
1288 	struct mntarg *ma = NULL;
1289 	char *fstypename, *mntonname, *mntfromname, *data;
1290 	int error, fsflags;
1291 
1292 	fstypename = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1293 	mntonname = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1294 	mntfromname = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1295 	data = NULL;
1296 	error = copyinstr(args->filesystemtype, fstypename, MNAMELEN - 1,
1297 	    NULL);
1298 	if (error != 0)
1299 		goto out;
1300 	if (args->specialfile != NULL) {
1301 		error = copyinstr(args->specialfile, mntfromname, MNAMELEN - 1, NULL);
1302 		if (error != 0)
1303 			goto out;
1304 	} else {
1305 		mntfromname[0] = '\0';
1306 	}
1307 	error = copyinstr(args->dir, mntonname, MNAMELEN - 1, NULL);
1308 	if (error != 0)
1309 		goto out;
1310 
1311 	if (strcmp(fstypename, "ext2") == 0) {
1312 		strcpy(fstypename, "ext2fs");
1313 	} else if (strcmp(fstypename, "proc") == 0) {
1314 		strcpy(fstypename, "linprocfs");
1315 	} else if (strcmp(fstypename, "vfat") == 0) {
1316 		strcpy(fstypename, "msdosfs");
1317 	} else if (strcmp(fstypename, "fuse") == 0) {
1318 		char *fuse_options, *fuse_option, *fuse_name;
1319 
1320 		if (strcmp(mntfromname, "fuse") == 0)
1321 			strcpy(mntfromname, "/dev/fuse");
1322 
1323 		strcpy(fstypename, "fusefs");
1324 		data = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1325 		error = copyinstr(args->data, data, MNAMELEN - 1, NULL);
1326 		if (error != 0)
1327 			goto out;
1328 
1329 		fuse_options = data;
1330 		while ((fuse_option = strsep(&fuse_options, ",")) != NULL) {
1331 			fuse_name = strsep(&fuse_option, "=");
1332 			if (fuse_name == NULL || fuse_option == NULL)
1333 				goto out;
1334 			ma = mount_arg(ma, fuse_name, fuse_option, -1);
1335 		}
1336 
1337 		/*
1338 		 * The FUSE server uses Linux errno values instead of FreeBSD
1339 		 * ones; add a flag to tell fuse(4) to do errno translation.
1340 		 */
1341 		ma = mount_arg(ma, "linux_errnos", "1", -1);
1342 	}
1343 
1344 	fsflags = 0;
1345 
1346 	/*
1347 	 * Linux SYNC flag is not included; the closest equivalent
1348 	 * FreeBSD has is !ASYNC, which is our default.
1349 	 */
1350 	if (args->rwflag & LINUX_MS_RDONLY)
1351 		fsflags |= MNT_RDONLY;
1352 	if (args->rwflag & LINUX_MS_NOSUID)
1353 		fsflags |= MNT_NOSUID;
1354 	if (args->rwflag & LINUX_MS_NOEXEC)
1355 		fsflags |= MNT_NOEXEC;
1356 	if (args->rwflag & LINUX_MS_REMOUNT)
1357 		fsflags |= MNT_UPDATE;
1358 
1359 	ma = mount_arg(ma, "fstype", fstypename, -1);
1360 	ma = mount_arg(ma, "fspath", mntonname, -1);
1361 	ma = mount_arg(ma, "from", mntfromname, -1);
1362 	error = kernel_mount(ma, fsflags);
1363 out:
1364 	free(fstypename, M_TEMP);
1365 	free(mntonname, M_TEMP);
1366 	free(mntfromname, M_TEMP);
1367 	free(data, M_TEMP);
1368 	return (error);
1369 }
1370 
1371 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1372 int
1373 linux_oldumount(struct thread *td, struct linux_oldumount_args *args)
1374 {
1375 
1376 	return (kern_unmount(td, args->path, 0));
1377 }
1378 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1379 
1380 #ifdef LINUX_LEGACY_SYSCALLS
1381 int
1382 linux_umount(struct thread *td, struct linux_umount_args *args)
1383 {
1384 	int flags;
1385 
1386 	flags = 0;
1387 	if ((args->flags & LINUX_MNT_FORCE) != 0) {
1388 		args->flags &= ~LINUX_MNT_FORCE;
1389 		flags |= MNT_FORCE;
1390 	}
1391 	if (args->flags != 0) {
1392 		linux_msg(td, "unsupported umount2 flags %#x", args->flags);
1393 		return (EINVAL);
1394 	}
1395 
1396 	return (kern_unmount(td, args->path, flags));
1397 }
1398 #endif
1399 
1400 /*
1401  * fcntl family of syscalls
1402  */
1403 
1404 struct l_flock {
1405 	l_short		l_type;
1406 	l_short		l_whence;
1407 	l_off_t		l_start;
1408 	l_off_t		l_len;
1409 	l_pid_t		l_pid;
1410 }
1411 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1412 __packed
1413 #endif
1414 ;
1415 
1416 static void
1417 linux_to_bsd_flock(struct l_flock *linux_flock, struct flock *bsd_flock)
1418 {
1419 	switch (linux_flock->l_type) {
1420 	case LINUX_F_RDLCK:
1421 		bsd_flock->l_type = F_RDLCK;
1422 		break;
1423 	case LINUX_F_WRLCK:
1424 		bsd_flock->l_type = F_WRLCK;
1425 		break;
1426 	case LINUX_F_UNLCK:
1427 		bsd_flock->l_type = F_UNLCK;
1428 		break;
1429 	default:
1430 		bsd_flock->l_type = -1;
1431 		break;
1432 	}
1433 	bsd_flock->l_whence = linux_flock->l_whence;
1434 	bsd_flock->l_start = (off_t)linux_flock->l_start;
1435 	bsd_flock->l_len = (off_t)linux_flock->l_len;
1436 	bsd_flock->l_pid = (pid_t)linux_flock->l_pid;
1437 	bsd_flock->l_sysid = 0;
1438 }
1439 
1440 static void
1441 bsd_to_linux_flock(struct flock *bsd_flock, struct l_flock *linux_flock)
1442 {
1443 	switch (bsd_flock->l_type) {
1444 	case F_RDLCK:
1445 		linux_flock->l_type = LINUX_F_RDLCK;
1446 		break;
1447 	case F_WRLCK:
1448 		linux_flock->l_type = LINUX_F_WRLCK;
1449 		break;
1450 	case F_UNLCK:
1451 		linux_flock->l_type = LINUX_F_UNLCK;
1452 		break;
1453 	}
1454 	linux_flock->l_whence = bsd_flock->l_whence;
1455 	linux_flock->l_start = (l_off_t)bsd_flock->l_start;
1456 	linux_flock->l_len = (l_off_t)bsd_flock->l_len;
1457 	linux_flock->l_pid = (l_pid_t)bsd_flock->l_pid;
1458 }
1459 
1460 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1461 struct l_flock64 {
1462 	l_short		l_type;
1463 	l_short		l_whence;
1464 	l_loff_t	l_start;
1465 	l_loff_t	l_len;
1466 	l_pid_t		l_pid;
1467 }
1468 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1469 __packed
1470 #endif
1471 ;
1472 
1473 static void
1474 linux_to_bsd_flock64(struct l_flock64 *linux_flock, struct flock *bsd_flock)
1475 {
1476 	switch (linux_flock->l_type) {
1477 	case LINUX_F_RDLCK:
1478 		bsd_flock->l_type = F_RDLCK;
1479 		break;
1480 	case LINUX_F_WRLCK:
1481 		bsd_flock->l_type = F_WRLCK;
1482 		break;
1483 	case LINUX_F_UNLCK:
1484 		bsd_flock->l_type = F_UNLCK;
1485 		break;
1486 	default:
1487 		bsd_flock->l_type = -1;
1488 		break;
1489 	}
1490 	bsd_flock->l_whence = linux_flock->l_whence;
1491 	bsd_flock->l_start = (off_t)linux_flock->l_start;
1492 	bsd_flock->l_len = (off_t)linux_flock->l_len;
1493 	bsd_flock->l_pid = (pid_t)linux_flock->l_pid;
1494 	bsd_flock->l_sysid = 0;
1495 }
1496 
1497 static void
1498 bsd_to_linux_flock64(struct flock *bsd_flock, struct l_flock64 *linux_flock)
1499 {
1500 	switch (bsd_flock->l_type) {
1501 	case F_RDLCK:
1502 		linux_flock->l_type = LINUX_F_RDLCK;
1503 		break;
1504 	case F_WRLCK:
1505 		linux_flock->l_type = LINUX_F_WRLCK;
1506 		break;
1507 	case F_UNLCK:
1508 		linux_flock->l_type = LINUX_F_UNLCK;
1509 		break;
1510 	}
1511 	linux_flock->l_whence = bsd_flock->l_whence;
1512 	linux_flock->l_start = (l_loff_t)bsd_flock->l_start;
1513 	linux_flock->l_len = (l_loff_t)bsd_flock->l_len;
1514 	linux_flock->l_pid = (l_pid_t)bsd_flock->l_pid;
1515 }
1516 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1517 
1518 static int
1519 fcntl_common(struct thread *td, struct linux_fcntl_args *args)
1520 {
1521 	struct l_flock linux_flock;
1522 	struct flock bsd_flock;
1523 	struct file *fp;
1524 	long arg;
1525 	int error, result;
1526 
1527 	switch (args->cmd) {
1528 	case LINUX_F_DUPFD:
1529 		return (kern_fcntl(td, args->fd, F_DUPFD, args->arg));
1530 
1531 	case LINUX_F_GETFD:
1532 		return (kern_fcntl(td, args->fd, F_GETFD, 0));
1533 
1534 	case LINUX_F_SETFD:
1535 		return (kern_fcntl(td, args->fd, F_SETFD, args->arg));
1536 
1537 	case LINUX_F_GETFL:
1538 		error = kern_fcntl(td, args->fd, F_GETFL, 0);
1539 		result = td->td_retval[0];
1540 		td->td_retval[0] = 0;
1541 		if (result & O_RDONLY)
1542 			td->td_retval[0] |= LINUX_O_RDONLY;
1543 		if (result & O_WRONLY)
1544 			td->td_retval[0] |= LINUX_O_WRONLY;
1545 		if (result & O_RDWR)
1546 			td->td_retval[0] |= LINUX_O_RDWR;
1547 		if (result & O_NDELAY)
1548 			td->td_retval[0] |= LINUX_O_NONBLOCK;
1549 		if (result & O_APPEND)
1550 			td->td_retval[0] |= LINUX_O_APPEND;
1551 		if (result & O_FSYNC)
1552 			td->td_retval[0] |= LINUX_O_SYNC;
1553 		if (result & O_ASYNC)
1554 			td->td_retval[0] |= LINUX_O_ASYNC;
1555 #ifdef LINUX_O_NOFOLLOW
1556 		if (result & O_NOFOLLOW)
1557 			td->td_retval[0] |= LINUX_O_NOFOLLOW;
1558 #endif
1559 #ifdef LINUX_O_DIRECT
1560 		if (result & O_DIRECT)
1561 			td->td_retval[0] |= LINUX_O_DIRECT;
1562 #endif
1563 		return (error);
1564 
1565 	case LINUX_F_SETFL:
1566 		arg = 0;
1567 		if (args->arg & LINUX_O_NDELAY)
1568 			arg |= O_NONBLOCK;
1569 		if (args->arg & LINUX_O_APPEND)
1570 			arg |= O_APPEND;
1571 		if (args->arg & LINUX_O_SYNC)
1572 			arg |= O_FSYNC;
1573 		if (args->arg & LINUX_O_ASYNC)
1574 			arg |= O_ASYNC;
1575 #ifdef LINUX_O_NOFOLLOW
1576 		if (args->arg & LINUX_O_NOFOLLOW)
1577 			arg |= O_NOFOLLOW;
1578 #endif
1579 #ifdef LINUX_O_DIRECT
1580 		if (args->arg & LINUX_O_DIRECT)
1581 			arg |= O_DIRECT;
1582 #endif
1583 		return (kern_fcntl(td, args->fd, F_SETFL, arg));
1584 
1585 	case LINUX_F_GETLK:
1586 		error = copyin((void *)args->arg, &linux_flock,
1587 		    sizeof(linux_flock));
1588 		if (error)
1589 			return (error);
1590 		linux_to_bsd_flock(&linux_flock, &bsd_flock);
1591 		error = kern_fcntl(td, args->fd, F_GETLK, (intptr_t)&bsd_flock);
1592 		if (error)
1593 			return (error);
1594 		bsd_to_linux_flock(&bsd_flock, &linux_flock);
1595 		return (copyout(&linux_flock, (void *)args->arg,
1596 		    sizeof(linux_flock)));
1597 
1598 	case LINUX_F_SETLK:
1599 		error = copyin((void *)args->arg, &linux_flock,
1600 		    sizeof(linux_flock));
1601 		if (error)
1602 			return (error);
1603 		linux_to_bsd_flock(&linux_flock, &bsd_flock);
1604 		return (kern_fcntl(td, args->fd, F_SETLK,
1605 		    (intptr_t)&bsd_flock));
1606 
1607 	case LINUX_F_SETLKW:
1608 		error = copyin((void *)args->arg, &linux_flock,
1609 		    sizeof(linux_flock));
1610 		if (error)
1611 			return (error);
1612 		linux_to_bsd_flock(&linux_flock, &bsd_flock);
1613 		return (kern_fcntl(td, args->fd, F_SETLKW,
1614 		     (intptr_t)&bsd_flock));
1615 
1616 	case LINUX_F_GETOWN:
1617 		return (kern_fcntl(td, args->fd, F_GETOWN, 0));
1618 
1619 	case LINUX_F_SETOWN:
1620 		/*
1621 		 * XXX some Linux applications depend on F_SETOWN having no
1622 		 * significant effect for pipes (SIGIO is not delivered for
1623 		 * pipes under Linux-2.2.35 at least).
1624 		 */
1625 		error = fget(td, args->fd,
1626 		    &cap_fcntl_rights, &fp);
1627 		if (error)
1628 			return (error);
1629 		if (fp->f_type == DTYPE_PIPE) {
1630 			fdrop(fp, td);
1631 			return (EINVAL);
1632 		}
1633 		fdrop(fp, td);
1634 
1635 		return (kern_fcntl(td, args->fd, F_SETOWN, args->arg));
1636 
1637 	case LINUX_F_DUPFD_CLOEXEC:
1638 		return (kern_fcntl(td, args->fd, F_DUPFD_CLOEXEC, args->arg));
1639 	/*
1640 	 * Our F_SEAL_* values match Linux one for maximum compatibility.  So we
1641 	 * only needed to account for different values for fcntl(2) commands.
1642 	 */
1643 	case LINUX_F_GET_SEALS:
1644 		error = kern_fcntl(td, args->fd, F_GET_SEALS, 0);
1645 		if (error != 0)
1646 			return (error);
1647 		td->td_retval[0] = bsd_to_linux_bits(td->td_retval[0],
1648 		    seal_bitmap, 0);
1649 		return (0);
1650 
1651 	case LINUX_F_ADD_SEALS:
1652 		return (kern_fcntl(td, args->fd, F_ADD_SEALS,
1653 		    linux_to_bsd_bits(args->arg, seal_bitmap, 0)));
1654 	default:
1655 		linux_msg(td, "unsupported fcntl cmd %d\n", args->cmd);
1656 		return (EINVAL);
1657 	}
1658 }
1659 
1660 int
1661 linux_fcntl(struct thread *td, struct linux_fcntl_args *args)
1662 {
1663 
1664 	return (fcntl_common(td, args));
1665 }
1666 
1667 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1668 int
1669 linux_fcntl64(struct thread *td, struct linux_fcntl64_args *args)
1670 {
1671 	struct l_flock64 linux_flock;
1672 	struct flock bsd_flock;
1673 	struct linux_fcntl_args fcntl_args;
1674 	int error;
1675 
1676 	switch (args->cmd) {
1677 	case LINUX_F_GETLK64:
1678 		error = copyin((void *)args->arg, &linux_flock,
1679 		    sizeof(linux_flock));
1680 		if (error)
1681 			return (error);
1682 		linux_to_bsd_flock64(&linux_flock, &bsd_flock);
1683 		error = kern_fcntl(td, args->fd, F_GETLK, (intptr_t)&bsd_flock);
1684 		if (error)
1685 			return (error);
1686 		bsd_to_linux_flock64(&bsd_flock, &linux_flock);
1687 		return (copyout(&linux_flock, (void *)args->arg,
1688 			    sizeof(linux_flock)));
1689 
1690 	case LINUX_F_SETLK64:
1691 		error = copyin((void *)args->arg, &linux_flock,
1692 		    sizeof(linux_flock));
1693 		if (error)
1694 			return (error);
1695 		linux_to_bsd_flock64(&linux_flock, &bsd_flock);
1696 		return (kern_fcntl(td, args->fd, F_SETLK,
1697 		    (intptr_t)&bsd_flock));
1698 
1699 	case LINUX_F_SETLKW64:
1700 		error = copyin((void *)args->arg, &linux_flock,
1701 		    sizeof(linux_flock));
1702 		if (error)
1703 			return (error);
1704 		linux_to_bsd_flock64(&linux_flock, &bsd_flock);
1705 		return (kern_fcntl(td, args->fd, F_SETLKW,
1706 		    (intptr_t)&bsd_flock));
1707 	}
1708 
1709 	fcntl_args.fd = args->fd;
1710 	fcntl_args.cmd = args->cmd;
1711 	fcntl_args.arg = args->arg;
1712 	return (fcntl_common(td, &fcntl_args));
1713 }
1714 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1715 
1716 #ifdef LINUX_LEGACY_SYSCALLS
1717 int
1718 linux_chown(struct thread *td, struct linux_chown_args *args)
1719 {
1720 	char *path;
1721 	int error;
1722 
1723 	if (!LUSECONVPATH(td)) {
1724 		return (kern_fchownat(td, AT_FDCWD, args->path, UIO_USERSPACE,
1725 		    args->uid, args->gid, 0));
1726 	}
1727 	LCONVPATHEXIST(td, args->path, &path);
1728 	error = kern_fchownat(td, AT_FDCWD, path, UIO_SYSSPACE, args->uid,
1729 	    args->gid, 0);
1730 	LFREEPATH(path);
1731 	return (error);
1732 }
1733 #endif
1734 
1735 int
1736 linux_fchownat(struct thread *td, struct linux_fchownat_args *args)
1737 {
1738 	char *path;
1739 	int error, dfd, flag;
1740 
1741 	if (args->flag & ~LINUX_AT_SYMLINK_NOFOLLOW)
1742 		return (EINVAL);
1743 
1744 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD :  args->dfd;
1745 	flag = (args->flag & LINUX_AT_SYMLINK_NOFOLLOW) == 0 ? 0 :
1746 	    AT_SYMLINK_NOFOLLOW;
1747 	if (!LUSECONVPATH(td)) {
1748 		return (kern_fchownat(td, dfd, args->filename, UIO_USERSPACE,
1749 		    args->uid, args->gid, flag));
1750 	}
1751 	LCONVPATHEXIST_AT(td, args->filename, &path, dfd);
1752 	error = kern_fchownat(td, dfd, path, UIO_SYSSPACE, args->uid, args->gid,
1753 	    flag);
1754 	LFREEPATH(path);
1755 	return (error);
1756 }
1757 
1758 #ifdef LINUX_LEGACY_SYSCALLS
1759 int
1760 linux_lchown(struct thread *td, struct linux_lchown_args *args)
1761 {
1762 	char *path;
1763 	int error;
1764 
1765 	if (!LUSECONVPATH(td)) {
1766 		return (kern_fchownat(td, AT_FDCWD, args->path, UIO_USERSPACE, args->uid,
1767 		    args->gid, AT_SYMLINK_NOFOLLOW));
1768 	}
1769 	LCONVPATHEXIST(td, args->path, &path);
1770 	error = kern_fchownat(td, AT_FDCWD, path, UIO_SYSSPACE, args->uid, args->gid,
1771 	    AT_SYMLINK_NOFOLLOW);
1772 	LFREEPATH(path);
1773 	return (error);
1774 }
1775 #endif
1776 
1777 static int
1778 convert_fadvice(int advice)
1779 {
1780 	switch (advice) {
1781 	case LINUX_POSIX_FADV_NORMAL:
1782 		return (POSIX_FADV_NORMAL);
1783 	case LINUX_POSIX_FADV_RANDOM:
1784 		return (POSIX_FADV_RANDOM);
1785 	case LINUX_POSIX_FADV_SEQUENTIAL:
1786 		return (POSIX_FADV_SEQUENTIAL);
1787 	case LINUX_POSIX_FADV_WILLNEED:
1788 		return (POSIX_FADV_WILLNEED);
1789 	case LINUX_POSIX_FADV_DONTNEED:
1790 		return (POSIX_FADV_DONTNEED);
1791 	case LINUX_POSIX_FADV_NOREUSE:
1792 		return (POSIX_FADV_NOREUSE);
1793 	default:
1794 		return (-1);
1795 	}
1796 }
1797 
1798 int
1799 linux_fadvise64(struct thread *td, struct linux_fadvise64_args *args)
1800 {
1801 	off_t offset;
1802 	int advice;
1803 
1804 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1805 	offset = PAIR32TO64(off_t, args->offset);
1806 #else
1807 	offset = args->offset;
1808 #endif
1809 
1810 	advice = convert_fadvice(args->advice);
1811 	if (advice == -1)
1812 		return (EINVAL);
1813 	return (kern_posix_fadvise(td, args->fd, offset, args->len, advice));
1814 }
1815 
1816 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1817 int
1818 linux_fadvise64_64(struct thread *td, struct linux_fadvise64_64_args *args)
1819 {
1820 	off_t len, offset;
1821 	int advice;
1822 
1823 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1824 	len = PAIR32TO64(off_t, args->len);
1825 	offset = PAIR32TO64(off_t, args->offset);
1826 #else
1827 	len = args->len;
1828 	offset = args->offset;
1829 #endif
1830 
1831 	advice = convert_fadvice(args->advice);
1832 	if (advice == -1)
1833 		return (EINVAL);
1834 	return (kern_posix_fadvise(td, args->fd, offset, len, advice));
1835 }
1836 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1837 
1838 #ifdef LINUX_LEGACY_SYSCALLS
1839 int
1840 linux_pipe(struct thread *td, struct linux_pipe_args *args)
1841 {
1842 	int fildes[2];
1843 	int error;
1844 
1845 	error = kern_pipe(td, fildes, 0, NULL, NULL);
1846 	if (error != 0)
1847 		return (error);
1848 
1849 	error = copyout(fildes, args->pipefds, sizeof(fildes));
1850 	if (error != 0) {
1851 		(void)kern_close(td, fildes[0]);
1852 		(void)kern_close(td, fildes[1]);
1853 	}
1854 
1855 	return (error);
1856 }
1857 #endif
1858 
1859 int
1860 linux_pipe2(struct thread *td, struct linux_pipe2_args *args)
1861 {
1862 	int fildes[2];
1863 	int error, flags;
1864 
1865 	if ((args->flags & ~(LINUX_O_NONBLOCK | LINUX_O_CLOEXEC)) != 0)
1866 		return (EINVAL);
1867 
1868 	flags = 0;
1869 	if ((args->flags & LINUX_O_NONBLOCK) != 0)
1870 		flags |= O_NONBLOCK;
1871 	if ((args->flags & LINUX_O_CLOEXEC) != 0)
1872 		flags |= O_CLOEXEC;
1873 	error = kern_pipe(td, fildes, flags, NULL, NULL);
1874 	if (error != 0)
1875 		return (error);
1876 
1877 	error = copyout(fildes, args->pipefds, sizeof(fildes));
1878 	if (error != 0) {
1879 		(void)kern_close(td, fildes[0]);
1880 		(void)kern_close(td, fildes[1]);
1881 	}
1882 
1883 	return (error);
1884 }
1885 
1886 int
1887 linux_dup3(struct thread *td, struct linux_dup3_args *args)
1888 {
1889 	int cmd;
1890 	intptr_t newfd;
1891 
1892 	if (args->oldfd == args->newfd)
1893 		return (EINVAL);
1894 	if ((args->flags & ~LINUX_O_CLOEXEC) != 0)
1895 		return (EINVAL);
1896 	if (args->flags & LINUX_O_CLOEXEC)
1897 		cmd = F_DUP2FD_CLOEXEC;
1898 	else
1899 		cmd = F_DUP2FD;
1900 
1901 	newfd = args->newfd;
1902 	return (kern_fcntl(td, args->oldfd, cmd, newfd));
1903 }
1904 
1905 int
1906 linux_fallocate(struct thread *td, struct linux_fallocate_args *args)
1907 {
1908 	off_t len, offset;
1909 
1910 	/*
1911 	 * We emulate only posix_fallocate system call for which
1912 	 * mode should be 0.
1913 	 */
1914 	if (args->mode != 0)
1915 		return (EOPNOTSUPP);
1916 
1917 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1918 	len = PAIR32TO64(off_t, args->len);
1919 	offset = PAIR32TO64(off_t, args->offset);
1920 #else
1921 	len = args->len;
1922 	offset = args->offset;
1923 #endif
1924 
1925 	return (kern_posix_fallocate(td, args->fd, offset, len));
1926 }
1927 
1928 int
1929 linux_copy_file_range(struct thread *td, struct linux_copy_file_range_args
1930     *args)
1931 {
1932 	l_loff_t inoff, outoff, *inoffp, *outoffp;
1933 	int error, flags;
1934 
1935 	/*
1936 	 * copy_file_range(2) on Linux doesn't define any flags (yet), so is
1937 	 * the native implementation.  Enforce it.
1938 	 */
1939 	if (args->flags != 0) {
1940 		linux_msg(td, "copy_file_range unsupported flags 0x%x",
1941 		    args->flags);
1942 		return (EINVAL);
1943 	}
1944 	flags = 0;
1945 	inoffp = outoffp = NULL;
1946 	if (args->off_in != NULL) {
1947 		error = copyin(args->off_in, &inoff, sizeof(l_loff_t));
1948 		if (error != 0)
1949 			return (error);
1950 		inoffp = &inoff;
1951 	}
1952 	if (args->off_out != NULL) {
1953 		error = copyin(args->off_out, &outoff, sizeof(l_loff_t));
1954 		if (error != 0)
1955 			return (error);
1956 		outoffp = &outoff;
1957 	}
1958 
1959 	error = kern_copy_file_range(td, args->fd_in, inoffp, args->fd_out,
1960 	    outoffp, args->len, flags);
1961 	if (error == 0 && args->off_in != NULL)
1962 		error = copyout(inoffp, args->off_in, sizeof(l_loff_t));
1963 	if (error == 0 && args->off_out != NULL)
1964 		error = copyout(outoffp, args->off_out, sizeof(l_loff_t));
1965 	return (error);
1966 }
1967 
1968 #define	LINUX_MEMFD_PREFIX	"memfd:"
1969 
1970 int
1971 linux_memfd_create(struct thread *td, struct linux_memfd_create_args *args)
1972 {
1973 	char memfd_name[LINUX_NAME_MAX + 1];
1974 	int error, flags, shmflags, oflags;
1975 
1976 	/*
1977 	 * This is our clever trick to avoid the heap allocation to copy in the
1978 	 * uname.  We don't really need to go this far out of our way, but it
1979 	 * does keep the rest of this function fairly clean as they don't have
1980 	 * to worry about cleanup on the way out.
1981 	 */
1982 	error = copyinstr(args->uname_ptr,
1983 	    memfd_name + sizeof(LINUX_MEMFD_PREFIX) - 1,
1984 	    LINUX_NAME_MAX - sizeof(LINUX_MEMFD_PREFIX) - 1, NULL);
1985 	if (error != 0) {
1986 		if (error == ENAMETOOLONG)
1987 			error = EINVAL;
1988 		return (error);
1989 	}
1990 
1991 	memcpy(memfd_name, LINUX_MEMFD_PREFIX, sizeof(LINUX_MEMFD_PREFIX) - 1);
1992 	flags = linux_to_bsd_bits(args->flags, mfd_bitmap, 0);
1993 	if ((flags & ~(MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB |
1994 	    MFD_HUGE_MASK)) != 0)
1995 		return (EINVAL);
1996 	/* Size specified but no HUGETLB. */
1997 	if ((flags & MFD_HUGE_MASK) != 0 && (flags & MFD_HUGETLB) == 0)
1998 		return (EINVAL);
1999 	/* We don't actually support HUGETLB. */
2000 	if ((flags & MFD_HUGETLB) != 0)
2001 		return (ENOSYS);
2002 	oflags = O_RDWR;
2003 	shmflags = SHM_GROW_ON_WRITE;
2004 	if ((flags & MFD_CLOEXEC) != 0)
2005 		oflags |= O_CLOEXEC;
2006 	if ((flags & MFD_ALLOW_SEALING) != 0)
2007 		shmflags |= SHM_ALLOW_SEALING;
2008 	return (kern_shm_open2(td, SHM_ANON, oflags, 0, shmflags, NULL,
2009 	    memfd_name));
2010 }
2011 
2012 int
2013 linux_splice(struct thread *td, struct linux_splice_args *args)
2014 {
2015 
2016 	linux_msg(td, "syscall splice not really implemented");
2017 
2018 	/*
2019 	 * splice(2) is documented to return EINVAL in various circumstances;
2020 	 * returning it instead of ENOSYS should hint the caller to use fallback
2021 	 * instead.
2022 	 */
2023 	return (EINVAL);
2024 }
2025