xref: /freebsd/sys/compat/linux/linux_file.c (revision fa9012aef1ad596a99950f986cf1fb31111c034a)
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, char *, int, int);
70 static int	linux_getdents_error(struct thread *, int, int);
71 
72 static struct bsd_to_linux_bitmap seal_bitmap[] = {
73 	BITMAP_1t1_LINUX(F_SEAL_SEAL),
74 	BITMAP_1t1_LINUX(F_SEAL_SHRINK),
75 	BITMAP_1t1_LINUX(F_SEAL_GROW),
76 	BITMAP_1t1_LINUX(F_SEAL_WRITE),
77 };
78 
79 #define	MFD_HUGETLB_ENTRY(_size)					\
80 	{								\
81 		.bsd_value = MFD_HUGE_##_size,				\
82 		.linux_value = LINUX_HUGETLB_FLAG_ENCODE_##_size	\
83 	}
84 static struct bsd_to_linux_bitmap mfd_bitmap[] = {
85 	BITMAP_1t1_LINUX(MFD_CLOEXEC),
86 	BITMAP_1t1_LINUX(MFD_ALLOW_SEALING),
87 	BITMAP_1t1_LINUX(MFD_HUGETLB),
88 	MFD_HUGETLB_ENTRY(64KB),
89 	MFD_HUGETLB_ENTRY(512KB),
90 	MFD_HUGETLB_ENTRY(1MB),
91 	MFD_HUGETLB_ENTRY(2MB),
92 	MFD_HUGETLB_ENTRY(8MB),
93 	MFD_HUGETLB_ENTRY(16MB),
94 	MFD_HUGETLB_ENTRY(32MB),
95 	MFD_HUGETLB_ENTRY(256MB),
96 	MFD_HUGETLB_ENTRY(512MB),
97 	MFD_HUGETLB_ENTRY(1GB),
98 	MFD_HUGETLB_ENTRY(2GB),
99 	MFD_HUGETLB_ENTRY(16GB),
100 };
101 #undef MFD_HUGETLB_ENTRY
102 
103 #ifdef LINUX_LEGACY_SYSCALLS
104 int
105 linux_creat(struct thread *td, struct linux_creat_args *args)
106 {
107 	char *path;
108 	int error;
109 
110 	LCONVPATHEXIST(td, args->path, &path);
111 
112 	error = kern_openat(td, AT_FDCWD, path, UIO_SYSSPACE,
113 	    O_WRONLY | O_CREAT | O_TRUNC, args->mode);
114 	LFREEPATH(path);
115 	return (error);
116 }
117 #endif
118 
119 static int
120 linux_common_open(struct thread *td, int dirfd, char *path, int l_flags, int mode)
121 {
122 	struct proc *p = td->td_proc;
123 	struct file *fp;
124 	int fd;
125 	int bsd_flags, error;
126 
127 	bsd_flags = 0;
128 	switch (l_flags & LINUX_O_ACCMODE) {
129 	case LINUX_O_WRONLY:
130 		bsd_flags |= O_WRONLY;
131 		break;
132 	case LINUX_O_RDWR:
133 		bsd_flags |= O_RDWR;
134 		break;
135 	default:
136 		bsd_flags |= O_RDONLY;
137 	}
138 	if (l_flags & LINUX_O_NDELAY)
139 		bsd_flags |= O_NONBLOCK;
140 	if (l_flags & LINUX_O_APPEND)
141 		bsd_flags |= O_APPEND;
142 	if (l_flags & LINUX_O_SYNC)
143 		bsd_flags |= O_FSYNC;
144 	if (l_flags & LINUX_O_CLOEXEC)
145 		bsd_flags |= O_CLOEXEC;
146 	if (l_flags & LINUX_O_NONBLOCK)
147 		bsd_flags |= O_NONBLOCK;
148 	if (l_flags & LINUX_O_ASYNC)
149 		bsd_flags |= O_ASYNC;
150 	if (l_flags & LINUX_O_CREAT)
151 		bsd_flags |= O_CREAT;
152 	if (l_flags & LINUX_O_TRUNC)
153 		bsd_flags |= O_TRUNC;
154 	if (l_flags & LINUX_O_EXCL)
155 		bsd_flags |= O_EXCL;
156 	if (l_flags & LINUX_O_NOCTTY)
157 		bsd_flags |= O_NOCTTY;
158 	if (l_flags & LINUX_O_DIRECT)
159 		bsd_flags |= O_DIRECT;
160 	if (l_flags & LINUX_O_NOFOLLOW)
161 		bsd_flags |= O_NOFOLLOW;
162 	if (l_flags & LINUX_O_DIRECTORY)
163 		bsd_flags |= O_DIRECTORY;
164 	/* XXX LINUX_O_NOATIME: unable to be easily implemented. */
165 
166 	error = kern_openat(td, dirfd, path, UIO_SYSSPACE, bsd_flags, mode);
167 	if (error != 0) {
168 		if (error == EMLINK)
169 			error = ELOOP;
170 		goto done;
171 	}
172 	if (p->p_flag & P_CONTROLT)
173 		goto done;
174 	if (bsd_flags & O_NOCTTY)
175 		goto done;
176 
177 	/*
178 	 * XXX In between kern_openat() and fget(), another process
179 	 * having the same filedesc could use that fd without
180 	 * checking below.
181 	*/
182 	fd = td->td_retval[0];
183 	if (fget(td, fd, &cap_ioctl_rights, &fp) == 0) {
184 		if (fp->f_type != DTYPE_VNODE) {
185 			fdrop(fp, td);
186 			goto done;
187 		}
188 		sx_slock(&proctree_lock);
189 		PROC_LOCK(p);
190 		if (SESS_LEADER(p) && !(p->p_flag & P_CONTROLT)) {
191 			PROC_UNLOCK(p);
192 			sx_sunlock(&proctree_lock);
193 			/* XXXPJD: Verify if TIOCSCTTY is allowed. */
194 			(void) fo_ioctl(fp, TIOCSCTTY, (caddr_t) 0,
195 			    td->td_ucred, td);
196 		} else {
197 			PROC_UNLOCK(p);
198 			sx_sunlock(&proctree_lock);
199 		}
200 		fdrop(fp, td);
201 	}
202 
203 done:
204 	LFREEPATH(path);
205 	return (error);
206 }
207 
208 int
209 linux_openat(struct thread *td, struct linux_openat_args *args)
210 {
211 	char *path;
212 	int dfd;
213 
214 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
215 	if (args->flags & LINUX_O_CREAT)
216 		LCONVPATH_AT(td, args->filename, &path, 1, dfd);
217 	else
218 		LCONVPATH_AT(td, args->filename, &path, 0, dfd);
219 
220 	return (linux_common_open(td, dfd, path, args->flags, args->mode));
221 }
222 
223 #ifdef LINUX_LEGACY_SYSCALLS
224 int
225 linux_open(struct thread *td, struct linux_open_args *args)
226 {
227 	char *path;
228 
229 	if (args->flags & LINUX_O_CREAT)
230 		LCONVPATHCREAT(td, args->path, &path);
231 	else
232 		LCONVPATHEXIST(td, args->path, &path);
233 
234 	return (linux_common_open(td, AT_FDCWD, path, args->flags, args->mode));
235 }
236 #endif
237 
238 int
239 linux_lseek(struct thread *td, struct linux_lseek_args *args)
240 {
241 
242 	return (kern_lseek(td, args->fdes, args->off, args->whence));
243 }
244 
245 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
246 int
247 linux_llseek(struct thread *td, struct linux_llseek_args *args)
248 {
249 	int error;
250 	off_t off;
251 
252 	off = (args->olow) | (((off_t) args->ohigh) << 32);
253 
254 	error = kern_lseek(td, args->fd, off, args->whence);
255 	if (error != 0)
256 		return (error);
257 
258 	error = copyout(td->td_retval, args->res, sizeof(off_t));
259 	if (error != 0)
260 		return (error);
261 
262 	td->td_retval[0] = 0;
263 	return (0);
264 }
265 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
266 
267 /*
268  * Note that linux_getdents(2) and linux_getdents64(2) have the same
269  * arguments. They only differ in the definition of struct dirent they
270  * operate on.
271  * Note that linux_readdir(2) is a special case of linux_getdents(2)
272  * where count is always equals 1, meaning that the buffer is one
273  * dirent-structure in size and that the code can't handle more anyway.
274  * Note that linux_readdir(2) can't be implemented by means of linux_getdents(2)
275  * as in case when the *dent buffer size is equal to 1 linux_getdents(2) will
276  * trash user stack.
277  */
278 
279 static int
280 linux_getdents_error(struct thread *td, int fd, int err)
281 {
282 	struct vnode *vp;
283 	struct file *fp;
284 	int error;
285 
286 	/* Linux return ENOTDIR in case when fd is not a directory. */
287 	error = getvnode(td, fd, &cap_read_rights, &fp);
288 	if (error != 0)
289 		return (error);
290 	vp = fp->f_vnode;
291 	if (vp->v_type != VDIR) {
292 		fdrop(fp, td);
293 		return (ENOTDIR);
294 	}
295 	fdrop(fp, td);
296 	return (err);
297 }
298 
299 struct l_dirent {
300 	l_ulong		d_ino;
301 	l_off_t		d_off;
302 	l_ushort	d_reclen;
303 	char		d_name[LINUX_NAME_MAX + 1];
304 };
305 
306 struct l_dirent64 {
307 	uint64_t	d_ino;
308 	int64_t		d_off;
309 	l_ushort	d_reclen;
310 	u_char		d_type;
311 	char		d_name[LINUX_NAME_MAX + 1];
312 };
313 
314 /*
315  * Linux uses the last byte in the dirent buffer to store d_type,
316  * at least glibc-2.7 requires it. That is why l_dirent is padded with 2 bytes.
317  */
318 #define LINUX_RECLEN(namlen)						\
319     roundup(offsetof(struct l_dirent, d_name) + (namlen) + 2, sizeof(l_ulong))
320 
321 #define LINUX_RECLEN64(namlen)						\
322     roundup(offsetof(struct l_dirent64, d_name) + (namlen) + 1,		\
323     sizeof(uint64_t))
324 
325 #ifdef LINUX_LEGACY_SYSCALLS
326 int
327 linux_getdents(struct thread *td, struct linux_getdents_args *args)
328 {
329 	struct dirent *bdp;
330 	caddr_t inp, buf;		/* BSD-format */
331 	int len, reclen;		/* BSD-format */
332 	caddr_t outp;			/* Linux-format */
333 	int resid, linuxreclen;		/* Linux-format */
334 	caddr_t lbuf;			/* Linux-format */
335 	off_t base;
336 	struct l_dirent *linux_dirent;
337 	int buflen, error;
338 	size_t retval;
339 
340 	buflen = min(args->count, MAXBSIZE);
341 	buf = malloc(buflen, M_TEMP, M_WAITOK);
342 
343 	error = kern_getdirentries(td, args->fd, buf, buflen,
344 	    &base, NULL, UIO_SYSSPACE);
345 	if (error != 0) {
346 		error = linux_getdents_error(td, args->fd, error);
347 		goto out1;
348 	}
349 
350 	lbuf = malloc(LINUX_RECLEN(LINUX_NAME_MAX), M_TEMP, M_WAITOK | M_ZERO);
351 
352 	len = td->td_retval[0];
353 	inp = buf;
354 	outp = (caddr_t)args->dent;
355 	resid = args->count;
356 	retval = 0;
357 
358 	while (len > 0) {
359 		bdp = (struct dirent *) inp;
360 		reclen = bdp->d_reclen;
361 		linuxreclen = LINUX_RECLEN(bdp->d_namlen);
362 		/*
363 		 * No more space in the user supplied dirent buffer.
364 		 * Return EINVAL.
365 		 */
366 		if (resid < linuxreclen) {
367 			error = EINVAL;
368 			goto out;
369 		}
370 
371 		linux_dirent = (struct l_dirent*)lbuf;
372 		linux_dirent->d_ino = bdp->d_fileno;
373 		linux_dirent->d_off = base + reclen;
374 		linux_dirent->d_reclen = linuxreclen;
375 		/*
376 		 * Copy d_type to last byte of l_dirent buffer
377 		 */
378 		lbuf[linuxreclen - 1] = bdp->d_type;
379 		strlcpy(linux_dirent->d_name, bdp->d_name,
380 		    linuxreclen - offsetof(struct l_dirent, d_name)-1);
381 		error = copyout(linux_dirent, outp, linuxreclen);
382 		if (error != 0)
383 			goto out;
384 
385 		inp += reclen;
386 		base += reclen;
387 		len -= reclen;
388 
389 		retval += linuxreclen;
390 		outp += linuxreclen;
391 		resid -= linuxreclen;
392 	}
393 	td->td_retval[0] = retval;
394 
395 out:
396 	free(lbuf, M_TEMP);
397 out1:
398 	free(buf, M_TEMP);
399 	return (error);
400 }
401 #endif
402 
403 int
404 linux_getdents64(struct thread *td, struct linux_getdents64_args *args)
405 {
406 	struct dirent *bdp;
407 	caddr_t inp, buf;		/* BSD-format */
408 	int len, reclen;		/* BSD-format */
409 	caddr_t outp;			/* Linux-format */
410 	int resid, linuxreclen;		/* Linux-format */
411 	caddr_t lbuf;			/* Linux-format */
412 	off_t base;
413 	struct l_dirent64 *linux_dirent64;
414 	int buflen, error;
415 	size_t retval;
416 
417 	buflen = min(args->count, MAXBSIZE);
418 	buf = malloc(buflen, M_TEMP, M_WAITOK);
419 
420 	error = kern_getdirentries(td, args->fd, buf, buflen,
421 	    &base, NULL, UIO_SYSSPACE);
422 	if (error != 0) {
423 		error = linux_getdents_error(td, args->fd, error);
424 		goto out1;
425 	}
426 
427 	lbuf = malloc(LINUX_RECLEN64(LINUX_NAME_MAX), M_TEMP, M_WAITOK | M_ZERO);
428 
429 	len = td->td_retval[0];
430 	inp = buf;
431 	outp = (caddr_t)args->dirent;
432 	resid = args->count;
433 	retval = 0;
434 
435 	while (len > 0) {
436 		bdp = (struct dirent *) inp;
437 		reclen = bdp->d_reclen;
438 		linuxreclen = LINUX_RECLEN64(bdp->d_namlen);
439 		/*
440 		 * No more space in the user supplied dirent buffer.
441 		 * Return EINVAL.
442 		 */
443 		if (resid < linuxreclen) {
444 			error = EINVAL;
445 			goto out;
446 		}
447 
448 		linux_dirent64 = (struct l_dirent64*)lbuf;
449 		linux_dirent64->d_ino = bdp->d_fileno;
450 		linux_dirent64->d_off = base + reclen;
451 		linux_dirent64->d_reclen = linuxreclen;
452 		linux_dirent64->d_type = bdp->d_type;
453 		strlcpy(linux_dirent64->d_name, bdp->d_name,
454 		    linuxreclen - offsetof(struct l_dirent64, d_name));
455 		error = copyout(linux_dirent64, outp, linuxreclen);
456 		if (error != 0)
457 			goto out;
458 
459 		inp += reclen;
460 		base += reclen;
461 		len -= reclen;
462 
463 		retval += linuxreclen;
464 		outp += linuxreclen;
465 		resid -= linuxreclen;
466 	}
467 	td->td_retval[0] = retval;
468 
469 out:
470 	free(lbuf, M_TEMP);
471 out1:
472 	free(buf, M_TEMP);
473 	return (error);
474 }
475 
476 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
477 int
478 linux_readdir(struct thread *td, struct linux_readdir_args *args)
479 {
480 	struct dirent *bdp;
481 	caddr_t buf;			/* BSD-format */
482 	int linuxreclen;		/* Linux-format */
483 	caddr_t lbuf;			/* Linux-format */
484 	off_t base;
485 	struct l_dirent *linux_dirent;
486 	int buflen, error;
487 
488 	buflen = LINUX_RECLEN(LINUX_NAME_MAX);
489 	buf = malloc(buflen, M_TEMP, M_WAITOK);
490 
491 	error = kern_getdirentries(td, args->fd, buf, buflen,
492 	    &base, NULL, UIO_SYSSPACE);
493 	if (error != 0) {
494 		error = linux_getdents_error(td, args->fd, error);
495 		goto out;
496 	}
497 	if (td->td_retval[0] == 0)
498 		goto out;
499 
500 	lbuf = malloc(LINUX_RECLEN(LINUX_NAME_MAX), M_TEMP, M_WAITOK | M_ZERO);
501 
502 	bdp = (struct dirent *) buf;
503 	linuxreclen = LINUX_RECLEN(bdp->d_namlen);
504 
505 	linux_dirent = (struct l_dirent*)lbuf;
506 	linux_dirent->d_ino = bdp->d_fileno;
507 	linux_dirent->d_off = linuxreclen;
508 	linux_dirent->d_reclen = bdp->d_namlen;
509 	strlcpy(linux_dirent->d_name, bdp->d_name,
510 	    linuxreclen - offsetof(struct l_dirent, d_name));
511 	error = copyout(linux_dirent, args->dent, linuxreclen);
512 	if (error == 0)
513 		td->td_retval[0] = linuxreclen;
514 
515 	free(lbuf, M_TEMP);
516 out:
517 	free(buf, M_TEMP);
518 	return (error);
519 }
520 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
521 
522 
523 /*
524  * These exist mainly for hooks for doing /compat/linux translation.
525  */
526 
527 #ifdef LINUX_LEGACY_SYSCALLS
528 int
529 linux_access(struct thread *td, struct linux_access_args *args)
530 {
531 	char *path;
532 	int error;
533 
534 	/* Linux convention. */
535 	if (args->amode & ~(F_OK | X_OK | W_OK | R_OK))
536 		return (EINVAL);
537 
538 	LCONVPATHEXIST(td, args->path, &path);
539 
540 	error = kern_accessat(td, AT_FDCWD, path, UIO_SYSSPACE, 0,
541 	    args->amode);
542 	LFREEPATH(path);
543 
544 	return (error);
545 }
546 #endif
547 
548 int
549 linux_faccessat(struct thread *td, struct linux_faccessat_args *args)
550 {
551 	char *path;
552 	int error, dfd;
553 
554 	/* Linux convention. */
555 	if (args->amode & ~(F_OK | X_OK | W_OK | R_OK))
556 		return (EINVAL);
557 
558 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
559 	LCONVPATHEXIST_AT(td, args->filename, &path, dfd);
560 
561 	error = kern_accessat(td, dfd, path, UIO_SYSSPACE, 0, args->amode);
562 	LFREEPATH(path);
563 
564 	return (error);
565 }
566 
567 #ifdef LINUX_LEGACY_SYSCALLS
568 int
569 linux_unlink(struct thread *td, struct linux_unlink_args *args)
570 {
571 	char *path;
572 	int error;
573 	struct stat st;
574 
575 	LCONVPATHEXIST(td, args->path, &path);
576 
577 	error = kern_funlinkat(td, AT_FDCWD, path, FD_NONE, UIO_SYSSPACE, 0, 0);
578 	if (error == EPERM) {
579 		/* Introduce POSIX noncompliant behaviour of Linux */
580 		if (kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &st,
581 		    NULL) == 0) {
582 			if (S_ISDIR(st.st_mode))
583 				error = EISDIR;
584 		}
585 	}
586 	LFREEPATH(path);
587 	return (error);
588 }
589 #endif
590 
591 int
592 linux_unlinkat(struct thread *td, struct linux_unlinkat_args *args)
593 {
594 	char *path;
595 	int error, dfd;
596 	struct stat st;
597 
598 	if (args->flag & ~LINUX_AT_REMOVEDIR)
599 		return (EINVAL);
600 
601 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
602 	LCONVPATHEXIST_AT(td, args->pathname, &path, dfd);
603 
604 	if (args->flag & LINUX_AT_REMOVEDIR)
605 		error = kern_frmdirat(td, dfd, path, FD_NONE, UIO_SYSSPACE, 0);
606 	else
607 		error = kern_funlinkat(td, dfd, path, FD_NONE, UIO_SYSSPACE, 0,
608 		    0);
609 	if (error == EPERM && !(args->flag & LINUX_AT_REMOVEDIR)) {
610 		/* Introduce POSIX noncompliant behaviour of Linux */
611 		if (kern_statat(td, AT_SYMLINK_NOFOLLOW, dfd, path,
612 		    UIO_SYSSPACE, &st, NULL) == 0 && S_ISDIR(st.st_mode))
613 			error = EISDIR;
614 	}
615 	LFREEPATH(path);
616 	return (error);
617 }
618 int
619 linux_chdir(struct thread *td, struct linux_chdir_args *args)
620 {
621 	char *path;
622 	int error;
623 
624 	LCONVPATHEXIST(td, args->path, &path);
625 
626 	error = kern_chdir(td, path, UIO_SYSSPACE);
627 	LFREEPATH(path);
628 	return (error);
629 }
630 
631 #ifdef LINUX_LEGACY_SYSCALLS
632 int
633 linux_chmod(struct thread *td, struct linux_chmod_args *args)
634 {
635 	char *path;
636 	int error;
637 
638 	LCONVPATHEXIST(td, args->path, &path);
639 
640 	error = kern_fchmodat(td, AT_FDCWD, path, UIO_SYSSPACE,
641 	    args->mode, 0);
642 	LFREEPATH(path);
643 	return (error);
644 }
645 #endif
646 
647 int
648 linux_fchmodat(struct thread *td, struct linux_fchmodat_args *args)
649 {
650 	char *path;
651 	int error, dfd;
652 
653 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
654 	LCONVPATHEXIST_AT(td, args->filename, &path, dfd);
655 
656 	error = kern_fchmodat(td, dfd, path, UIO_SYSSPACE, args->mode, 0);
657 	LFREEPATH(path);
658 	return (error);
659 }
660 
661 #ifdef LINUX_LEGACY_SYSCALLS
662 int
663 linux_mkdir(struct thread *td, struct linux_mkdir_args *args)
664 {
665 	char *path;
666 	int error;
667 
668 	LCONVPATHCREAT(td, args->path, &path);
669 
670 	error = kern_mkdirat(td, AT_FDCWD, path, UIO_SYSSPACE, args->mode);
671 	LFREEPATH(path);
672 	return (error);
673 }
674 #endif
675 
676 int
677 linux_mkdirat(struct thread *td, struct linux_mkdirat_args *args)
678 {
679 	char *path;
680 	int error, dfd;
681 
682 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
683 	LCONVPATHCREAT_AT(td, args->pathname, &path, dfd);
684 
685 	error = kern_mkdirat(td, dfd, path, UIO_SYSSPACE, args->mode);
686 	LFREEPATH(path);
687 	return (error);
688 }
689 
690 #ifdef LINUX_LEGACY_SYSCALLS
691 int
692 linux_rmdir(struct thread *td, struct linux_rmdir_args *args)
693 {
694 	char *path;
695 	int error;
696 
697 	LCONVPATHEXIST(td, args->path, &path);
698 
699 	error = kern_frmdirat(td, AT_FDCWD, path, FD_NONE, UIO_SYSSPACE, 0);
700 	LFREEPATH(path);
701 	return (error);
702 }
703 
704 int
705 linux_rename(struct thread *td, struct linux_rename_args *args)
706 {
707 	char *from, *to;
708 	int error;
709 
710 	LCONVPATHEXIST(td, args->from, &from);
711 	/* Expand LCONVPATHCREATE so that `from' can be freed on errors */
712 	error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD);
713 	if (to == NULL) {
714 		LFREEPATH(from);
715 		return (error);
716 	}
717 
718 	error = kern_renameat(td, AT_FDCWD, from, AT_FDCWD, to, UIO_SYSSPACE);
719 	LFREEPATH(from);
720 	LFREEPATH(to);
721 	return (error);
722 }
723 #endif
724 
725 int
726 linux_renameat(struct thread *td, struct linux_renameat_args *args)
727 {
728 	struct linux_renameat2_args renameat2_args = {
729 	    .olddfd = args->olddfd,
730 	    .oldname = args->oldname,
731 	    .newdfd = args->newdfd,
732 	    .newname = args->newname,
733 	    .flags = 0
734 	};
735 
736 	return (linux_renameat2(td, &renameat2_args));
737 }
738 
739 int
740 linux_renameat2(struct thread *td, struct linux_renameat2_args *args)
741 {
742 	char *from, *to;
743 	int error, olddfd, newdfd;
744 
745 	if (args->flags != 0) {
746 		if (args->flags & ~(LINUX_RENAME_EXCHANGE |
747 		    LINUX_RENAME_NOREPLACE | LINUX_RENAME_WHITEOUT))
748 			return (EINVAL);
749 		if (args->flags & LINUX_RENAME_EXCHANGE &&
750 		    args->flags & (LINUX_RENAME_NOREPLACE |
751 		    LINUX_RENAME_WHITEOUT))
752 			return (EINVAL);
753 		linux_msg(td, "renameat2 unsupported flags 0x%x",
754 		    args->flags);
755 		return (EINVAL);
756 	}
757 
758 	olddfd = (args->olddfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->olddfd;
759 	newdfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd;
760 	LCONVPATHEXIST_AT(td, args->oldname, &from, olddfd);
761 	/* Expand LCONVPATHCREATE so that `from' can be freed on errors */
762 	error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, newdfd);
763 	if (to == NULL) {
764 		LFREEPATH(from);
765 		return (error);
766 	}
767 
768 	error = kern_renameat(td, olddfd, from, newdfd, to, UIO_SYSSPACE);
769 	LFREEPATH(from);
770 	LFREEPATH(to);
771 	return (error);
772 }
773 
774 #ifdef LINUX_LEGACY_SYSCALLS
775 int
776 linux_symlink(struct thread *td, struct linux_symlink_args *args)
777 {
778 	char *path, *to;
779 	int error;
780 
781 	LCONVPATHEXIST(td, args->path, &path);
782 	/* Expand LCONVPATHCREATE so that `path' can be freed on errors */
783 	error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD);
784 	if (to == NULL) {
785 		LFREEPATH(path);
786 		return (error);
787 	}
788 
789 	error = kern_symlinkat(td, path, AT_FDCWD, to, UIO_SYSSPACE);
790 	LFREEPATH(path);
791 	LFREEPATH(to);
792 	return (error);
793 }
794 #endif
795 
796 int
797 linux_symlinkat(struct thread *td, struct linux_symlinkat_args *args)
798 {
799 	char *path, *to;
800 	int error, dfd;
801 
802 	dfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd;
803 	LCONVPATHEXIST(td, args->oldname, &path);
804 	/* Expand LCONVPATHCREATE so that `path' can be freed on errors */
805 	error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, dfd);
806 	if (to == NULL) {
807 		LFREEPATH(path);
808 		return (error);
809 	}
810 
811 	error = kern_symlinkat(td, path, dfd, to, UIO_SYSSPACE);
812 	LFREEPATH(path);
813 	LFREEPATH(to);
814 	return (error);
815 }
816 
817 #ifdef LINUX_LEGACY_SYSCALLS
818 int
819 linux_readlink(struct thread *td, struct linux_readlink_args *args)
820 {
821 	char *name;
822 	int error;
823 
824 	LCONVPATHEXIST(td, args->name, &name);
825 
826 	error = kern_readlinkat(td, AT_FDCWD, name, UIO_SYSSPACE,
827 	    args->buf, UIO_USERSPACE, args->count);
828 	LFREEPATH(name);
829 	return (error);
830 }
831 #endif
832 
833 int
834 linux_readlinkat(struct thread *td, struct linux_readlinkat_args *args)
835 {
836 	char *name;
837 	int error, dfd;
838 
839 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd;
840 	LCONVPATHEXIST_AT(td, args->path, &name, dfd);
841 
842 	error = kern_readlinkat(td, dfd, name, UIO_SYSSPACE, args->buf,
843 	    UIO_USERSPACE, args->bufsiz);
844 	LFREEPATH(name);
845 	return (error);
846 }
847 
848 int
849 linux_truncate(struct thread *td, struct linux_truncate_args *args)
850 {
851 	char *path;
852 	int error;
853 
854 	LCONVPATHEXIST(td, args->path, &path);
855 	error = kern_truncate(td, path, UIO_SYSSPACE, args->length);
856 	LFREEPATH(path);
857 	return (error);
858 }
859 
860 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
861 int
862 linux_truncate64(struct thread *td, struct linux_truncate64_args *args)
863 {
864 	char *path;
865 	off_t length;
866 	int error;
867 
868 #if defined(__amd64__) && defined(COMPAT_LINUX32)
869 	length = PAIR32TO64(off_t, args->length);
870 #else
871 	length = args->length;
872 #endif
873 
874 	LCONVPATHEXIST(td, args->path, &path);
875 	error = kern_truncate(td, path, UIO_SYSSPACE, length);
876 	LFREEPATH(path);
877 	return (error);
878 }
879 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
880 
881 int
882 linux_ftruncate(struct thread *td, struct linux_ftruncate_args *args)
883 {
884 
885 	return (kern_ftruncate(td, args->fd, args->length));
886 }
887 
888 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
889 int
890 linux_ftruncate64(struct thread *td, struct linux_ftruncate64_args *args)
891 {
892 	off_t length;
893 
894 #if defined(__amd64__) && defined(COMPAT_LINUX32)
895 	length = PAIR32TO64(off_t, args->length);
896 #else
897 	length = args->length;
898 #endif
899 
900 	return (kern_ftruncate(td, args->fd, length));
901 }
902 #endif
903 
904 #ifdef LINUX_LEGACY_SYSCALLS
905 int
906 linux_link(struct thread *td, struct linux_link_args *args)
907 {
908 	char *path, *to;
909 	int error;
910 
911 	LCONVPATHEXIST(td, args->path, &path);
912 	/* Expand LCONVPATHCREATE so that `path' can be freed on errors */
913 	error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD);
914 	if (to == NULL) {
915 		LFREEPATH(path);
916 		return (error);
917 	}
918 
919 	error = kern_linkat(td, AT_FDCWD, AT_FDCWD, path, to, UIO_SYSSPACE,
920 	    FOLLOW);
921 	LFREEPATH(path);
922 	LFREEPATH(to);
923 	return (error);
924 }
925 #endif
926 
927 int
928 linux_linkat(struct thread *td, struct linux_linkat_args *args)
929 {
930 	char *path, *to;
931 	int error, olddfd, newdfd, follow;
932 
933 	if (args->flag & ~LINUX_AT_SYMLINK_FOLLOW)
934 		return (EINVAL);
935 
936 	olddfd = (args->olddfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->olddfd;
937 	newdfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd;
938 	LCONVPATHEXIST_AT(td, args->oldname, &path, olddfd);
939 	/* Expand LCONVPATHCREATE so that `path' can be freed on errors */
940 	error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, newdfd);
941 	if (to == NULL) {
942 		LFREEPATH(path);
943 		return (error);
944 	}
945 
946 	follow = (args->flag & LINUX_AT_SYMLINK_FOLLOW) == 0 ? NOFOLLOW :
947 	    FOLLOW;
948 	error = kern_linkat(td, olddfd, newdfd, path, to, UIO_SYSSPACE, follow);
949 	LFREEPATH(path);
950 	LFREEPATH(to);
951 	return (error);
952 }
953 
954 int
955 linux_fdatasync(struct thread *td, struct linux_fdatasync_args *uap)
956 {
957 
958 	return (kern_fsync(td, uap->fd, false));
959 }
960 
961 int
962 linux_sync_file_range(struct thread *td, struct linux_sync_file_range_args *uap)
963 {
964 	off_t nbytes, offset;
965 
966 #if defined(__amd64__) && defined(COMPAT_LINUX32)
967 	nbytes = PAIR32TO64(off_t, uap->nbytes);
968 	offset = PAIR32TO64(off_t, uap->offset);
969 #else
970 	nbytes = uap->nbytes;
971 	offset = uap->offset;
972 #endif
973 
974 	if (offset < 0 || nbytes < 0 ||
975 	    (uap->flags & ~(LINUX_SYNC_FILE_RANGE_WAIT_BEFORE |
976 	    LINUX_SYNC_FILE_RANGE_WRITE |
977 	    LINUX_SYNC_FILE_RANGE_WAIT_AFTER)) != 0) {
978 		return (EINVAL);
979 	}
980 
981 	return (kern_fsync(td, uap->fd, false));
982 }
983 
984 int
985 linux_pread(struct thread *td, struct linux_pread_args *uap)
986 {
987 	struct vnode *vp;
988 	off_t offset;
989 	int error;
990 
991 #if defined(__amd64__) && defined(COMPAT_LINUX32)
992 	offset = PAIR32TO64(off_t, uap->offset);
993 #else
994 	offset = uap->offset;
995 #endif
996 
997 	error = kern_pread(td, uap->fd, uap->buf, uap->nbyte, offset);
998 	if (error == 0) {
999 		/* This seems to violate POSIX but Linux does it. */
1000 		error = fgetvp(td, uap->fd, &cap_pread_rights, &vp);
1001 		if (error != 0)
1002 			return (error);
1003 		if (vp->v_type == VDIR)
1004 			error = EISDIR;
1005 		vrele(vp);
1006 	}
1007 	return (error);
1008 }
1009 
1010 int
1011 linux_pwrite(struct thread *td, struct linux_pwrite_args *uap)
1012 {
1013 	off_t offset;
1014 
1015 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1016 	offset = PAIR32TO64(off_t, uap->offset);
1017 #else
1018 	offset = uap->offset;
1019 #endif
1020 
1021 	return (kern_pwrite(td, uap->fd, uap->buf, uap->nbyte, offset));
1022 }
1023 
1024 int
1025 linux_preadv(struct thread *td, struct linux_preadv_args *uap)
1026 {
1027 	struct uio *auio;
1028 	int error;
1029 	off_t offset;
1030 
1031 	/*
1032 	 * According http://man7.org/linux/man-pages/man2/preadv.2.html#NOTES
1033 	 * pos_l and pos_h, respectively, contain the
1034 	 * low order and high order 32 bits of offset.
1035 	 */
1036 	offset = (((off_t)uap->pos_h << (sizeof(offset) * 4)) <<
1037 	    (sizeof(offset) * 4)) | uap->pos_l;
1038 	if (offset < 0)
1039 		return (EINVAL);
1040 #ifdef COMPAT_LINUX32
1041 	error = linux32_copyinuio(PTRIN(uap->vec), uap->vlen, &auio);
1042 #else
1043 	error = copyinuio(uap->vec, uap->vlen, &auio);
1044 #endif
1045 	if (error != 0)
1046 		return (error);
1047 	error = kern_preadv(td, uap->fd, auio, offset);
1048 	free(auio, M_IOV);
1049 	return (error);
1050 }
1051 
1052 int
1053 linux_pwritev(struct thread *td, struct linux_pwritev_args *uap)
1054 {
1055 	struct uio *auio;
1056 	int error;
1057 	off_t offset;
1058 
1059 	/*
1060 	 * According http://man7.org/linux/man-pages/man2/pwritev.2.html#NOTES
1061 	 * pos_l and pos_h, respectively, contain the
1062 	 * low order and high order 32 bits of offset.
1063 	 */
1064 	offset = (((off_t)uap->pos_h << (sizeof(offset) * 4)) <<
1065 	    (sizeof(offset) * 4)) | uap->pos_l;
1066 	if (offset < 0)
1067 		return (EINVAL);
1068 #ifdef COMPAT_LINUX32
1069 	error = linux32_copyinuio(PTRIN(uap->vec), uap->vlen, &auio);
1070 #else
1071 	error = copyinuio(uap->vec, uap->vlen, &auio);
1072 #endif
1073 	if (error != 0)
1074 		return (error);
1075 	error = kern_pwritev(td, uap->fd, auio, offset);
1076 	free(auio, M_IOV);
1077 	return (error);
1078 }
1079 
1080 int
1081 linux_mount(struct thread *td, struct linux_mount_args *args)
1082 {
1083 	char fstypename[MFSNAMELEN];
1084 	char *mntonname, *mntfromname;
1085 	int error, fsflags;
1086 
1087 	mntonname = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1088 	mntfromname = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1089 	error = copyinstr(args->filesystemtype, fstypename, MFSNAMELEN - 1,
1090 	    NULL);
1091 	if (error != 0)
1092 		goto out;
1093 	if (args->specialfile != NULL) {
1094 		error = copyinstr(args->specialfile, mntfromname, MNAMELEN - 1, NULL);
1095 		if (error != 0)
1096 			goto out;
1097 	} else {
1098 		mntfromname[0] = '\0';
1099 	}
1100 	error = copyinstr(args->dir, mntonname, MNAMELEN - 1, NULL);
1101 	if (error != 0)
1102 		goto out;
1103 
1104 	if (strcmp(fstypename, "ext2") == 0) {
1105 		strcpy(fstypename, "ext2fs");
1106 	} else if (strcmp(fstypename, "proc") == 0) {
1107 		strcpy(fstypename, "linprocfs");
1108 	} else if (strcmp(fstypename, "vfat") == 0) {
1109 		strcpy(fstypename, "msdosfs");
1110 	}
1111 
1112 	fsflags = 0;
1113 
1114 	/*
1115 	 * Linux SYNC flag is not included; the closest equivalent
1116 	 * FreeBSD has is !ASYNC, which is our default.
1117 	 */
1118 	if (args->rwflag & LINUX_MS_RDONLY)
1119 		fsflags |= MNT_RDONLY;
1120 	if (args->rwflag & LINUX_MS_NOSUID)
1121 		fsflags |= MNT_NOSUID;
1122 	if (args->rwflag & LINUX_MS_NOEXEC)
1123 		fsflags |= MNT_NOEXEC;
1124 	if (args->rwflag & LINUX_MS_REMOUNT)
1125 		fsflags |= MNT_UPDATE;
1126 
1127 	error = kernel_vmount(fsflags,
1128 	    "fstype", fstypename,
1129 	    "fspath", mntonname,
1130 	    "from", mntfromname,
1131 	    NULL);
1132 out:
1133 	free(mntonname, M_TEMP);
1134 	free(mntfromname, M_TEMP);
1135 	return (error);
1136 }
1137 
1138 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1139 int
1140 linux_oldumount(struct thread *td, struct linux_oldumount_args *args)
1141 {
1142 
1143 	return (kern_unmount(td, args->path, 0));
1144 }
1145 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1146 
1147 #ifdef LINUX_LEGACY_SYSCALLS
1148 int
1149 linux_umount(struct thread *td, struct linux_umount_args *args)
1150 {
1151 	int flags;
1152 
1153 	flags = 0;
1154 	if ((args->flags & LINUX_MNT_FORCE) != 0) {
1155 		args->flags &= ~LINUX_MNT_FORCE;
1156 		flags |= MNT_FORCE;
1157 	}
1158 	if (args->flags != 0) {
1159 		linux_msg(td, "unsupported umount2 flags %#x", args->flags);
1160 		return (EINVAL);
1161 	}
1162 
1163 	return (kern_unmount(td, args->path, flags));
1164 }
1165 #endif
1166 
1167 /*
1168  * fcntl family of syscalls
1169  */
1170 
1171 struct l_flock {
1172 	l_short		l_type;
1173 	l_short		l_whence;
1174 	l_off_t		l_start;
1175 	l_off_t		l_len;
1176 	l_pid_t		l_pid;
1177 }
1178 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1179 __packed
1180 #endif
1181 ;
1182 
1183 static void
1184 linux_to_bsd_flock(struct l_flock *linux_flock, struct flock *bsd_flock)
1185 {
1186 	switch (linux_flock->l_type) {
1187 	case LINUX_F_RDLCK:
1188 		bsd_flock->l_type = F_RDLCK;
1189 		break;
1190 	case LINUX_F_WRLCK:
1191 		bsd_flock->l_type = F_WRLCK;
1192 		break;
1193 	case LINUX_F_UNLCK:
1194 		bsd_flock->l_type = F_UNLCK;
1195 		break;
1196 	default:
1197 		bsd_flock->l_type = -1;
1198 		break;
1199 	}
1200 	bsd_flock->l_whence = linux_flock->l_whence;
1201 	bsd_flock->l_start = (off_t)linux_flock->l_start;
1202 	bsd_flock->l_len = (off_t)linux_flock->l_len;
1203 	bsd_flock->l_pid = (pid_t)linux_flock->l_pid;
1204 	bsd_flock->l_sysid = 0;
1205 }
1206 
1207 static void
1208 bsd_to_linux_flock(struct flock *bsd_flock, struct l_flock *linux_flock)
1209 {
1210 	switch (bsd_flock->l_type) {
1211 	case F_RDLCK:
1212 		linux_flock->l_type = LINUX_F_RDLCK;
1213 		break;
1214 	case F_WRLCK:
1215 		linux_flock->l_type = LINUX_F_WRLCK;
1216 		break;
1217 	case F_UNLCK:
1218 		linux_flock->l_type = LINUX_F_UNLCK;
1219 		break;
1220 	}
1221 	linux_flock->l_whence = bsd_flock->l_whence;
1222 	linux_flock->l_start = (l_off_t)bsd_flock->l_start;
1223 	linux_flock->l_len = (l_off_t)bsd_flock->l_len;
1224 	linux_flock->l_pid = (l_pid_t)bsd_flock->l_pid;
1225 }
1226 
1227 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1228 struct l_flock64 {
1229 	l_short		l_type;
1230 	l_short		l_whence;
1231 	l_loff_t	l_start;
1232 	l_loff_t	l_len;
1233 	l_pid_t		l_pid;
1234 }
1235 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1236 __packed
1237 #endif
1238 ;
1239 
1240 static void
1241 linux_to_bsd_flock64(struct l_flock64 *linux_flock, struct flock *bsd_flock)
1242 {
1243 	switch (linux_flock->l_type) {
1244 	case LINUX_F_RDLCK:
1245 		bsd_flock->l_type = F_RDLCK;
1246 		break;
1247 	case LINUX_F_WRLCK:
1248 		bsd_flock->l_type = F_WRLCK;
1249 		break;
1250 	case LINUX_F_UNLCK:
1251 		bsd_flock->l_type = F_UNLCK;
1252 		break;
1253 	default:
1254 		bsd_flock->l_type = -1;
1255 		break;
1256 	}
1257 	bsd_flock->l_whence = linux_flock->l_whence;
1258 	bsd_flock->l_start = (off_t)linux_flock->l_start;
1259 	bsd_flock->l_len = (off_t)linux_flock->l_len;
1260 	bsd_flock->l_pid = (pid_t)linux_flock->l_pid;
1261 	bsd_flock->l_sysid = 0;
1262 }
1263 
1264 static void
1265 bsd_to_linux_flock64(struct flock *bsd_flock, struct l_flock64 *linux_flock)
1266 {
1267 	switch (bsd_flock->l_type) {
1268 	case F_RDLCK:
1269 		linux_flock->l_type = LINUX_F_RDLCK;
1270 		break;
1271 	case F_WRLCK:
1272 		linux_flock->l_type = LINUX_F_WRLCK;
1273 		break;
1274 	case F_UNLCK:
1275 		linux_flock->l_type = LINUX_F_UNLCK;
1276 		break;
1277 	}
1278 	linux_flock->l_whence = bsd_flock->l_whence;
1279 	linux_flock->l_start = (l_loff_t)bsd_flock->l_start;
1280 	linux_flock->l_len = (l_loff_t)bsd_flock->l_len;
1281 	linux_flock->l_pid = (l_pid_t)bsd_flock->l_pid;
1282 }
1283 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1284 
1285 static int
1286 fcntl_common(struct thread *td, struct linux_fcntl_args *args)
1287 {
1288 	struct l_flock linux_flock;
1289 	struct flock bsd_flock;
1290 	struct file *fp;
1291 	long arg;
1292 	int error, result;
1293 
1294 	switch (args->cmd) {
1295 	case LINUX_F_DUPFD:
1296 		return (kern_fcntl(td, args->fd, F_DUPFD, args->arg));
1297 
1298 	case LINUX_F_GETFD:
1299 		return (kern_fcntl(td, args->fd, F_GETFD, 0));
1300 
1301 	case LINUX_F_SETFD:
1302 		return (kern_fcntl(td, args->fd, F_SETFD, args->arg));
1303 
1304 	case LINUX_F_GETFL:
1305 		error = kern_fcntl(td, args->fd, F_GETFL, 0);
1306 		result = td->td_retval[0];
1307 		td->td_retval[0] = 0;
1308 		if (result & O_RDONLY)
1309 			td->td_retval[0] |= LINUX_O_RDONLY;
1310 		if (result & O_WRONLY)
1311 			td->td_retval[0] |= LINUX_O_WRONLY;
1312 		if (result & O_RDWR)
1313 			td->td_retval[0] |= LINUX_O_RDWR;
1314 		if (result & O_NDELAY)
1315 			td->td_retval[0] |= LINUX_O_NONBLOCK;
1316 		if (result & O_APPEND)
1317 			td->td_retval[0] |= LINUX_O_APPEND;
1318 		if (result & O_FSYNC)
1319 			td->td_retval[0] |= LINUX_O_SYNC;
1320 		if (result & O_ASYNC)
1321 			td->td_retval[0] |= LINUX_O_ASYNC;
1322 #ifdef LINUX_O_NOFOLLOW
1323 		if (result & O_NOFOLLOW)
1324 			td->td_retval[0] |= LINUX_O_NOFOLLOW;
1325 #endif
1326 #ifdef LINUX_O_DIRECT
1327 		if (result & O_DIRECT)
1328 			td->td_retval[0] |= LINUX_O_DIRECT;
1329 #endif
1330 		return (error);
1331 
1332 	case LINUX_F_SETFL:
1333 		arg = 0;
1334 		if (args->arg & LINUX_O_NDELAY)
1335 			arg |= O_NONBLOCK;
1336 		if (args->arg & LINUX_O_APPEND)
1337 			arg |= O_APPEND;
1338 		if (args->arg & LINUX_O_SYNC)
1339 			arg |= O_FSYNC;
1340 		if (args->arg & LINUX_O_ASYNC)
1341 			arg |= O_ASYNC;
1342 #ifdef LINUX_O_NOFOLLOW
1343 		if (args->arg & LINUX_O_NOFOLLOW)
1344 			arg |= O_NOFOLLOW;
1345 #endif
1346 #ifdef LINUX_O_DIRECT
1347 		if (args->arg & LINUX_O_DIRECT)
1348 			arg |= O_DIRECT;
1349 #endif
1350 		return (kern_fcntl(td, args->fd, F_SETFL, arg));
1351 
1352 	case LINUX_F_GETLK:
1353 		error = copyin((void *)args->arg, &linux_flock,
1354 		    sizeof(linux_flock));
1355 		if (error)
1356 			return (error);
1357 		linux_to_bsd_flock(&linux_flock, &bsd_flock);
1358 		error = kern_fcntl(td, args->fd, F_GETLK, (intptr_t)&bsd_flock);
1359 		if (error)
1360 			return (error);
1361 		bsd_to_linux_flock(&bsd_flock, &linux_flock);
1362 		return (copyout(&linux_flock, (void *)args->arg,
1363 		    sizeof(linux_flock)));
1364 
1365 	case LINUX_F_SETLK:
1366 		error = copyin((void *)args->arg, &linux_flock,
1367 		    sizeof(linux_flock));
1368 		if (error)
1369 			return (error);
1370 		linux_to_bsd_flock(&linux_flock, &bsd_flock);
1371 		return (kern_fcntl(td, args->fd, F_SETLK,
1372 		    (intptr_t)&bsd_flock));
1373 
1374 	case LINUX_F_SETLKW:
1375 		error = copyin((void *)args->arg, &linux_flock,
1376 		    sizeof(linux_flock));
1377 		if (error)
1378 			return (error);
1379 		linux_to_bsd_flock(&linux_flock, &bsd_flock);
1380 		return (kern_fcntl(td, args->fd, F_SETLKW,
1381 		     (intptr_t)&bsd_flock));
1382 
1383 	case LINUX_F_GETOWN:
1384 		return (kern_fcntl(td, args->fd, F_GETOWN, 0));
1385 
1386 	case LINUX_F_SETOWN:
1387 		/*
1388 		 * XXX some Linux applications depend on F_SETOWN having no
1389 		 * significant effect for pipes (SIGIO is not delivered for
1390 		 * pipes under Linux-2.2.35 at least).
1391 		 */
1392 		error = fget(td, args->fd,
1393 		    &cap_fcntl_rights, &fp);
1394 		if (error)
1395 			return (error);
1396 		if (fp->f_type == DTYPE_PIPE) {
1397 			fdrop(fp, td);
1398 			return (EINVAL);
1399 		}
1400 		fdrop(fp, td);
1401 
1402 		return (kern_fcntl(td, args->fd, F_SETOWN, args->arg));
1403 
1404 	case LINUX_F_DUPFD_CLOEXEC:
1405 		return (kern_fcntl(td, args->fd, F_DUPFD_CLOEXEC, args->arg));
1406 	/*
1407 	 * Our F_SEAL_* values match Linux one for maximum compatibility.  So we
1408 	 * only needed to account for different values for fcntl(2) commands.
1409 	 */
1410 	case LINUX_F_GET_SEALS:
1411 		error = kern_fcntl(td, args->fd, F_GET_SEALS, 0);
1412 		if (error != 0)
1413 			return (error);
1414 		td->td_retval[0] = bsd_to_linux_bits(td->td_retval[0],
1415 		    seal_bitmap, 0);
1416 		return (0);
1417 
1418 	case LINUX_F_ADD_SEALS:
1419 		return (kern_fcntl(td, args->fd, F_ADD_SEALS,
1420 		    linux_to_bsd_bits(args->arg, seal_bitmap, 0)));
1421 	default:
1422 		linux_msg(td, "unsupported fcntl cmd %d\n", args->cmd);
1423 		return (EINVAL);
1424 	}
1425 }
1426 
1427 int
1428 linux_fcntl(struct thread *td, struct linux_fcntl_args *args)
1429 {
1430 
1431 	return (fcntl_common(td, args));
1432 }
1433 
1434 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1435 int
1436 linux_fcntl64(struct thread *td, struct linux_fcntl64_args *args)
1437 {
1438 	struct l_flock64 linux_flock;
1439 	struct flock bsd_flock;
1440 	struct linux_fcntl_args fcntl_args;
1441 	int error;
1442 
1443 	switch (args->cmd) {
1444 	case LINUX_F_GETLK64:
1445 		error = copyin((void *)args->arg, &linux_flock,
1446 		    sizeof(linux_flock));
1447 		if (error)
1448 			return (error);
1449 		linux_to_bsd_flock64(&linux_flock, &bsd_flock);
1450 		error = kern_fcntl(td, args->fd, F_GETLK, (intptr_t)&bsd_flock);
1451 		if (error)
1452 			return (error);
1453 		bsd_to_linux_flock64(&bsd_flock, &linux_flock);
1454 		return (copyout(&linux_flock, (void *)args->arg,
1455 			    sizeof(linux_flock)));
1456 
1457 	case LINUX_F_SETLK64:
1458 		error = copyin((void *)args->arg, &linux_flock,
1459 		    sizeof(linux_flock));
1460 		if (error)
1461 			return (error);
1462 		linux_to_bsd_flock64(&linux_flock, &bsd_flock);
1463 		return (kern_fcntl(td, args->fd, F_SETLK,
1464 		    (intptr_t)&bsd_flock));
1465 
1466 	case LINUX_F_SETLKW64:
1467 		error = copyin((void *)args->arg, &linux_flock,
1468 		    sizeof(linux_flock));
1469 		if (error)
1470 			return (error);
1471 		linux_to_bsd_flock64(&linux_flock, &bsd_flock);
1472 		return (kern_fcntl(td, args->fd, F_SETLKW,
1473 		    (intptr_t)&bsd_flock));
1474 	}
1475 
1476 	fcntl_args.fd = args->fd;
1477 	fcntl_args.cmd = args->cmd;
1478 	fcntl_args.arg = args->arg;
1479 	return (fcntl_common(td, &fcntl_args));
1480 }
1481 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1482 
1483 #ifdef LINUX_LEGACY_SYSCALLS
1484 int
1485 linux_chown(struct thread *td, struct linux_chown_args *args)
1486 {
1487 	char *path;
1488 	int error;
1489 
1490 	LCONVPATHEXIST(td, args->path, &path);
1491 
1492 	error = kern_fchownat(td, AT_FDCWD, path, UIO_SYSSPACE, args->uid,
1493 	    args->gid, 0);
1494 	LFREEPATH(path);
1495 	return (error);
1496 }
1497 #endif
1498 
1499 int
1500 linux_fchownat(struct thread *td, struct linux_fchownat_args *args)
1501 {
1502 	char *path;
1503 	int error, dfd, flag;
1504 
1505 	if (args->flag & ~LINUX_AT_SYMLINK_NOFOLLOW)
1506 		return (EINVAL);
1507 
1508 	dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD :  args->dfd;
1509 	LCONVPATHEXIST_AT(td, args->filename, &path, dfd);
1510 
1511 	flag = (args->flag & LINUX_AT_SYMLINK_NOFOLLOW) == 0 ? 0 :
1512 	    AT_SYMLINK_NOFOLLOW;
1513 	error = kern_fchownat(td, dfd, path, UIO_SYSSPACE, args->uid, args->gid,
1514 	    flag);
1515 	LFREEPATH(path);
1516 	return (error);
1517 }
1518 
1519 #ifdef LINUX_LEGACY_SYSCALLS
1520 int
1521 linux_lchown(struct thread *td, struct linux_lchown_args *args)
1522 {
1523 	char *path;
1524 	int error;
1525 
1526 	LCONVPATHEXIST(td, args->path, &path);
1527 
1528 	error = kern_fchownat(td, AT_FDCWD, path, UIO_SYSSPACE, args->uid,
1529 	    args->gid, AT_SYMLINK_NOFOLLOW);
1530 	LFREEPATH(path);
1531 	return (error);
1532 }
1533 #endif
1534 
1535 static int
1536 convert_fadvice(int advice)
1537 {
1538 	switch (advice) {
1539 	case LINUX_POSIX_FADV_NORMAL:
1540 		return (POSIX_FADV_NORMAL);
1541 	case LINUX_POSIX_FADV_RANDOM:
1542 		return (POSIX_FADV_RANDOM);
1543 	case LINUX_POSIX_FADV_SEQUENTIAL:
1544 		return (POSIX_FADV_SEQUENTIAL);
1545 	case LINUX_POSIX_FADV_WILLNEED:
1546 		return (POSIX_FADV_WILLNEED);
1547 	case LINUX_POSIX_FADV_DONTNEED:
1548 		return (POSIX_FADV_DONTNEED);
1549 	case LINUX_POSIX_FADV_NOREUSE:
1550 		return (POSIX_FADV_NOREUSE);
1551 	default:
1552 		return (-1);
1553 	}
1554 }
1555 
1556 int
1557 linux_fadvise64(struct thread *td, struct linux_fadvise64_args *args)
1558 {
1559 	off_t offset;
1560 	int advice;
1561 
1562 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1563 	offset = PAIR32TO64(off_t, args->offset);
1564 #else
1565 	offset = args->offset;
1566 #endif
1567 
1568 	advice = convert_fadvice(args->advice);
1569 	if (advice == -1)
1570 		return (EINVAL);
1571 	return (kern_posix_fadvise(td, args->fd, offset, args->len, advice));
1572 }
1573 
1574 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
1575 int
1576 linux_fadvise64_64(struct thread *td, struct linux_fadvise64_64_args *args)
1577 {
1578 	off_t len, offset;
1579 	int advice;
1580 
1581 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1582 	len = PAIR32TO64(off_t, args->len);
1583 	offset = PAIR32TO64(off_t, args->offset);
1584 #else
1585 	len = args->len;
1586 	offset = args->offset;
1587 #endif
1588 
1589 	advice = convert_fadvice(args->advice);
1590 	if (advice == -1)
1591 		return (EINVAL);
1592 	return (kern_posix_fadvise(td, args->fd, offset, len, advice));
1593 }
1594 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
1595 
1596 #ifdef LINUX_LEGACY_SYSCALLS
1597 int
1598 linux_pipe(struct thread *td, struct linux_pipe_args *args)
1599 {
1600 	int fildes[2];
1601 	int error;
1602 
1603 	error = kern_pipe(td, fildes, 0, NULL, NULL);
1604 	if (error != 0)
1605 		return (error);
1606 
1607 	error = copyout(fildes, args->pipefds, sizeof(fildes));
1608 	if (error != 0) {
1609 		(void)kern_close(td, fildes[0]);
1610 		(void)kern_close(td, fildes[1]);
1611 	}
1612 
1613 	return (error);
1614 }
1615 #endif
1616 
1617 int
1618 linux_pipe2(struct thread *td, struct linux_pipe2_args *args)
1619 {
1620 	int fildes[2];
1621 	int error, flags;
1622 
1623 	if ((args->flags & ~(LINUX_O_NONBLOCK | LINUX_O_CLOEXEC)) != 0)
1624 		return (EINVAL);
1625 
1626 	flags = 0;
1627 	if ((args->flags & LINUX_O_NONBLOCK) != 0)
1628 		flags |= O_NONBLOCK;
1629 	if ((args->flags & LINUX_O_CLOEXEC) != 0)
1630 		flags |= O_CLOEXEC;
1631 	error = kern_pipe(td, fildes, flags, NULL, NULL);
1632 	if (error != 0)
1633 		return (error);
1634 
1635 	error = copyout(fildes, args->pipefds, sizeof(fildes));
1636 	if (error != 0) {
1637 		(void)kern_close(td, fildes[0]);
1638 		(void)kern_close(td, fildes[1]);
1639 	}
1640 
1641 	return (error);
1642 }
1643 
1644 int
1645 linux_dup3(struct thread *td, struct linux_dup3_args *args)
1646 {
1647 	int cmd;
1648 	intptr_t newfd;
1649 
1650 	if (args->oldfd == args->newfd)
1651 		return (EINVAL);
1652 	if ((args->flags & ~LINUX_O_CLOEXEC) != 0)
1653 		return (EINVAL);
1654 	if (args->flags & LINUX_O_CLOEXEC)
1655 		cmd = F_DUP2FD_CLOEXEC;
1656 	else
1657 		cmd = F_DUP2FD;
1658 
1659 	newfd = args->newfd;
1660 	return (kern_fcntl(td, args->oldfd, cmd, newfd));
1661 }
1662 
1663 int
1664 linux_fallocate(struct thread *td, struct linux_fallocate_args *args)
1665 {
1666 	off_t len, offset;
1667 
1668 	/*
1669 	 * We emulate only posix_fallocate system call for which
1670 	 * mode should be 0.
1671 	 */
1672 	if (args->mode != 0)
1673 		return (ENOSYS);
1674 
1675 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1676 	len = PAIR32TO64(off_t, args->len);
1677 	offset = PAIR32TO64(off_t, args->offset);
1678 #else
1679 	len = args->len;
1680 	offset = args->offset;
1681 #endif
1682 
1683 	return (kern_posix_fallocate(td, args->fd, offset, len));
1684 }
1685 
1686 int
1687 linux_copy_file_range(struct thread *td, struct linux_copy_file_range_args
1688     *args)
1689 {
1690 	l_loff_t inoff, outoff, *inoffp, *outoffp;
1691 	int error, flags;
1692 
1693 	/*
1694 	 * copy_file_range(2) on Linux doesn't define any flags (yet), so is
1695 	 * the native implementation.  Enforce it.
1696 	 */
1697 	if (args->flags != 0) {
1698 		linux_msg(td, "copy_file_range unsupported flags 0x%x",
1699 		    args->flags);
1700 		return (EINVAL);
1701 	}
1702 	flags = 0;
1703 	inoffp = outoffp = NULL;
1704 	if (args->off_in != NULL) {
1705 		error = copyin(args->off_in, &inoff, sizeof(l_loff_t));
1706 		if (error != 0)
1707 			return (error);
1708 		inoffp = &inoff;
1709 	}
1710 	if (args->off_out != NULL) {
1711 		error = copyin(args->off_out, &outoff, sizeof(l_loff_t));
1712 		if (error != 0)
1713 			return (error);
1714 		outoffp = &outoff;
1715 	}
1716 
1717 	error = kern_copy_file_range(td, args->fd_in, inoffp, args->fd_out,
1718 	    outoffp, args->len, flags);
1719 	if (error == 0 && args->off_in != NULL)
1720 		error = copyout(inoffp, args->off_in, sizeof(l_loff_t));
1721 	if (error == 0 && args->off_out != NULL)
1722 		error = copyout(outoffp, args->off_out, sizeof(l_loff_t));
1723 	return (error);
1724 }
1725 
1726 #define	LINUX_MEMFD_PREFIX	"memfd:"
1727 
1728 int
1729 linux_memfd_create(struct thread *td, struct linux_memfd_create_args *args)
1730 {
1731 	char memfd_name[LINUX_NAME_MAX + 1];
1732 	int error, flags, shmflags, oflags;
1733 
1734 	/*
1735 	 * This is our clever trick to avoid the heap allocation to copy in the
1736 	 * uname.  We don't really need to go this far out of our way, but it
1737 	 * does keep the rest of this function fairly clean as they don't have
1738 	 * to worry about cleanup on the way out.
1739 	 */
1740 	error = copyinstr(args->uname_ptr,
1741 	    memfd_name + sizeof(LINUX_MEMFD_PREFIX) - 1,
1742 	    LINUX_NAME_MAX - sizeof(LINUX_MEMFD_PREFIX) - 1, NULL);
1743 	if (error != 0) {
1744 		if (error == ENAMETOOLONG)
1745 			error = EINVAL;
1746 		return (error);
1747 	}
1748 
1749 	memcpy(memfd_name, LINUX_MEMFD_PREFIX, sizeof(LINUX_MEMFD_PREFIX) - 1);
1750 	flags = linux_to_bsd_bits(args->flags, mfd_bitmap, 0);
1751 	if ((flags & ~(MFD_CLOEXEC | MFD_ALLOW_SEALING | MFD_HUGETLB |
1752 	    MFD_HUGE_MASK)) != 0)
1753 		return (EINVAL);
1754 	/* Size specified but no HUGETLB. */
1755 	if ((flags & MFD_HUGE_MASK) != 0 && (flags & MFD_HUGETLB) == 0)
1756 		return (EINVAL);
1757 	/* We don't actually support HUGETLB. */
1758 	if ((flags & MFD_HUGETLB) != 0)
1759 		return (ENOSYS);
1760 	oflags = O_RDWR;
1761 	shmflags = 0;
1762 	if ((flags & MFD_CLOEXEC) != 0)
1763 		oflags |= O_CLOEXEC;
1764 	if ((flags & MFD_ALLOW_SEALING) != 0)
1765 		shmflags |= SHM_ALLOW_SEALING;
1766 	return (kern_shm_open2(td, SHM_ANON, oflags, 0, shmflags, NULL,
1767 	    memfd_name));
1768 }
1769