xref: /netbsd-src/sys/compat/linux/common/linux_misc.c (revision 7d62b00eb9ad855ffcd7da46b41e23feb5476fac)
1 /*	$NetBSD: linux_misc.c,v 1.256 2021/12/02 04:29:48 ryo Exp $	*/
2 
3 /*-
4  * Copyright (c) 1995, 1998, 1999, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Frank van der Linden and Eric Haszlakiewicz; by Jason R. Thorpe
9  * of the Numerical Aerospace Simulation Facility, NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Linux compatibility module. Try to deal with various Linux system calls.
35  */
36 
37 /*
38  * These functions have been moved to multiarch to allow
39  * selection of which machines include them to be
40  * determined by the individual files.linux_<arch> files.
41  *
42  * Function in multiarch:
43  *	linux_sys_break			: linux_break.c
44  *	linux_sys_alarm			: linux_misc_notalpha.c
45  *	linux_sys_getresgid		: linux_misc_notalpha.c
46  *	linux_sys_nice			: linux_misc_notalpha.c
47  *	linux_sys_readdir		: linux_misc_notalpha.c
48  *	linux_sys_setresgid		: linux_misc_notalpha.c
49  *	linux_sys_time			: linux_misc_notalpha.c
50  *	linux_sys_utime			: linux_misc_notalpha.c
51  *	linux_sys_waitpid		: linux_misc_notalpha.c
52  *	linux_sys_old_mmap		: linux_oldmmap.c
53  *	linux_sys_oldolduname		: linux_oldolduname.c
54  *	linux_sys_oldselect		: linux_oldselect.c
55  *	linux_sys_olduname		: linux_olduname.c
56  *	linux_sys_pipe			: linux_pipe.c
57  */
58 
59 #include <sys/cdefs.h>
60 __KERNEL_RCSID(0, "$NetBSD: linux_misc.c,v 1.256 2021/12/02 04:29:48 ryo Exp $");
61 
62 #include <sys/param.h>
63 #include <sys/systm.h>
64 #include <sys/namei.h>
65 #include <sys/proc.h>
66 #include <sys/dirent.h>
67 #include <sys/eventfd.h>
68 #include <sys/file.h>
69 #include <sys/stat.h>
70 #include <sys/filedesc.h>
71 #include <sys/ioctl.h>
72 #include <sys/kernel.h>
73 #include <sys/malloc.h>
74 #include <sys/mbuf.h>
75 #include <sys/mman.h>
76 #include <sys/mount.h>
77 #include <sys/poll.h>
78 #include <sys/prot.h>
79 #include <sys/reboot.h>
80 #include <sys/resource.h>
81 #include <sys/resourcevar.h>
82 #include <sys/select.h>
83 #include <sys/signal.h>
84 #include <sys/signalvar.h>
85 #include <sys/socket.h>
86 #include <sys/time.h>
87 #include <sys/times.h>
88 #include <sys/vnode.h>
89 #include <sys/uio.h>
90 #include <sys/wait.h>
91 #include <sys/utsname.h>
92 #include <sys/unistd.h>
93 #include <sys/vfs_syscalls.h>
94 #include <sys/swap.h>		/* for SWAP_ON */
95 #include <sys/sysctl.h>		/* for KERN_DOMAINNAME */
96 #include <sys/kauth.h>
97 #include <sys/futex.h>
98 
99 #include <sys/ptrace.h>
100 #include <machine/ptrace.h>
101 
102 #include <sys/syscall.h>
103 #include <sys/syscallargs.h>
104 
105 #include <compat/sys/resource.h>
106 
107 #include <compat/linux/common/linux_machdep.h>
108 #include <compat/linux/common/linux_types.h>
109 #include <compat/linux/common/linux_signal.h>
110 #include <compat/linux/common/linux_ipc.h>
111 #include <compat/linux/common/linux_sem.h>
112 
113 #include <compat/linux/common/linux_fcntl.h>
114 #include <compat/linux/common/linux_mmap.h>
115 #include <compat/linux/common/linux_dirent.h>
116 #include <compat/linux/common/linux_util.h>
117 #include <compat/linux/common/linux_misc.h>
118 #include <compat/linux/common/linux_statfs.h>
119 #include <compat/linux/common/linux_limit.h>
120 #include <compat/linux/common/linux_ptrace.h>
121 #include <compat/linux/common/linux_reboot.h>
122 #include <compat/linux/common/linux_emuldata.h>
123 #include <compat/linux/common/linux_sched.h>
124 
125 #include <compat/linux/linux_syscallargs.h>
126 
127 const int linux_ptrace_request_map[] = {
128 	LINUX_PTRACE_TRACEME,	PT_TRACE_ME,
129 	LINUX_PTRACE_PEEKTEXT,	PT_READ_I,
130 	LINUX_PTRACE_PEEKDATA,	PT_READ_D,
131 	LINUX_PTRACE_POKETEXT,	PT_WRITE_I,
132 	LINUX_PTRACE_POKEDATA,	PT_WRITE_D,
133 	LINUX_PTRACE_CONT,	PT_CONTINUE,
134 	LINUX_PTRACE_KILL,	PT_KILL,
135 	LINUX_PTRACE_ATTACH,	PT_ATTACH,
136 	LINUX_PTRACE_DETACH,	PT_DETACH,
137 # ifdef PT_STEP
138 	LINUX_PTRACE_SINGLESTEP,	PT_STEP,
139 # endif
140 	LINUX_PTRACE_SYSCALL,	PT_SYSCALL,
141 	-1
142 };
143 
144 const struct linux_mnttypes linux_fstypes[] = {
145 	{ MOUNT_FFS,		LINUX_DEFAULT_SUPER_MAGIC	},
146 	{ MOUNT_NFS,		LINUX_NFS_SUPER_MAGIC 		},
147 	{ MOUNT_MFS,		LINUX_DEFAULT_SUPER_MAGIC	},
148 	{ MOUNT_MSDOS,		LINUX_MSDOS_SUPER_MAGIC		},
149 	{ MOUNT_LFS,		LINUX_DEFAULT_SUPER_MAGIC	},
150 	{ MOUNT_FDESC,		LINUX_DEFAULT_SUPER_MAGIC	},
151 	{ MOUNT_NULL,		LINUX_DEFAULT_SUPER_MAGIC	},
152 	{ MOUNT_OVERLAY,	LINUX_DEFAULT_SUPER_MAGIC	},
153 	{ MOUNT_UMAP,		LINUX_DEFAULT_SUPER_MAGIC	},
154 	{ MOUNT_KERNFS,		LINUX_DEFAULT_SUPER_MAGIC	},
155 	{ MOUNT_PROCFS,		LINUX_PROC_SUPER_MAGIC		},
156 	{ MOUNT_AFS,		LINUX_DEFAULT_SUPER_MAGIC	},
157 	{ MOUNT_CD9660,		LINUX_ISOFS_SUPER_MAGIC		},
158 	{ MOUNT_UNION,		LINUX_DEFAULT_SUPER_MAGIC	},
159 	{ MOUNT_ADOSFS,		LINUX_ADFS_SUPER_MAGIC		},
160 	{ MOUNT_EXT2FS,		LINUX_EXT2_SUPER_MAGIC		},
161 	{ MOUNT_CFS,		LINUX_DEFAULT_SUPER_MAGIC	},
162 	{ MOUNT_CODA,		LINUX_CODA_SUPER_MAGIC		},
163 	{ MOUNT_FILECORE,	LINUX_DEFAULT_SUPER_MAGIC	},
164 	{ MOUNT_NTFS,		LINUX_DEFAULT_SUPER_MAGIC	},
165 	{ MOUNT_SMBFS,		LINUX_SMB_SUPER_MAGIC		},
166 	{ MOUNT_PTYFS,		LINUX_DEVPTS_SUPER_MAGIC	},
167 	{ MOUNT_TMPFS,		LINUX_TMPFS_SUPER_MAGIC		}
168 };
169 const int linux_fstypes_cnt = sizeof(linux_fstypes) / sizeof(linux_fstypes[0]);
170 
171 # ifdef DEBUG_LINUX
172 #define DPRINTF(a)	uprintf a
173 # else
174 #define DPRINTF(a)
175 # endif
176 
177 /* Local linux_misc.c functions: */
178 static void linux_to_bsd_mmap_args(struct sys_mmap_args *,
179     const struct linux_sys_mmap_args *);
180 static int linux_mmap(struct lwp *, const struct linux_sys_mmap_args *,
181     register_t *, off_t);
182 
183 
184 /*
185  * The information on a terminated (or stopped) process needs
186  * to be converted in order for Linux binaries to get a valid signal
187  * number out of it.
188  */
189 int
190 bsd_to_linux_wstat(int st)
191 {
192 
193 	int sig;
194 
195 	if (WIFSIGNALED(st)) {
196 		sig = WTERMSIG(st);
197 		if (sig >= 0 && sig < NSIG)
198 			st= (st & ~0177) | native_to_linux_signo[sig];
199 	} else if (WIFSTOPPED(st)) {
200 		sig = WSTOPSIG(st);
201 		if (sig >= 0 && sig < NSIG)
202 			st = (st & ~0xff00) |
203 			    (native_to_linux_signo[sig] << 8);
204 	}
205 	return st;
206 }
207 
208 /*
209  * wait4(2).  Passed on to the NetBSD call, surrounded by code to
210  * reserve some space for a NetBSD-style wait status, and converting
211  * it to what Linux wants.
212  */
213 int
214 linux_sys_wait4(struct lwp *l, const struct linux_sys_wait4_args *uap, register_t *retval)
215 {
216 	/* {
217 		syscallarg(int) pid;
218 		syscallarg(int *) status;
219 		syscallarg(int) options;
220 		syscallarg(struct rusage50 *) rusage;
221 	} */
222 	int error, status, options, linux_options, pid = SCARG(uap, pid);
223 	struct rusage50 ru50;
224 	struct rusage ru;
225 	proc_t *p;
226 
227 	linux_options = SCARG(uap, options);
228 	if (linux_options & ~(LINUX_WAIT4_KNOWNFLAGS))
229 		return (EINVAL);
230 
231 	options = 0;
232 	if (linux_options & LINUX_WAIT4_WNOHANG)
233 		options |= WNOHANG;
234 	if (linux_options & LINUX_WAIT4_WUNTRACED)
235 		options |= WUNTRACED;
236 	if (linux_options & LINUX_WAIT4_WCONTINUED)
237 		options |= WCONTINUED;
238 	if (linux_options & LINUX_WAIT4_WALL)
239 		options |= WALLSIG;
240 	if (linux_options & LINUX_WAIT4_WCLONE)
241 		options |= WALTSIG;
242 # ifdef DIAGNOSTIC
243 	if (linux_options & LINUX_WAIT4_WNOTHREAD)
244 		printf("WARNING: %s: linux process %d.%d called "
245 		       "waitpid with __WNOTHREAD set!\n",
246 		       __FILE__, l->l_proc->p_pid, l->l_lid);
247 
248 # endif
249 
250 	error = do_sys_wait(&pid, &status, options,
251 	    SCARG(uap, rusage) != NULL ? &ru : NULL);
252 
253 	retval[0] = pid;
254 	if (pid == 0)
255 		return error;
256 
257 	p = curproc;
258 	mutex_enter(p->p_lock);
259 	sigdelset(&p->p_sigpend.sp_set, SIGCHLD); /* XXXAD ksiginfo leak */
260 	mutex_exit(p->p_lock);
261 
262 	if (SCARG(uap, rusage) != NULL) {
263 		rusage_to_rusage50(&ru, &ru50);
264 		error = copyout(&ru, SCARG(uap, rusage), sizeof(ru));
265 	}
266 
267 	if (error == 0 && SCARG(uap, status) != NULL) {
268 		status = bsd_to_linux_wstat(status);
269 		error = copyout(&status, SCARG(uap, status), sizeof status);
270 	}
271 
272 	return error;
273 }
274 
275 /*
276  * Linux brk(2).  Like native, but always return the new break value.
277  */
278 int
279 linux_sys_brk(struct lwp *l, const struct linux_sys_brk_args *uap, register_t *retval)
280 {
281 	/* {
282 		syscallarg(char *) nsize;
283 	} */
284 	struct proc *p = l->l_proc;
285 	struct vmspace *vm = p->p_vmspace;
286 	struct sys_obreak_args oba;
287 
288 	SCARG(&oba, nsize) = SCARG(uap, nsize);
289 
290 	(void) sys_obreak(l, &oba, retval);
291 	retval[0] = (register_t)((char *)vm->vm_daddr + ptoa(vm->vm_dsize));
292 	return 0;
293 }
294 
295 /*
296  * Implement the fs stat functions. Straightforward.
297  */
298 int
299 linux_sys_statfs(struct lwp *l, const struct linux_sys_statfs_args *uap, register_t *retval)
300 {
301 	/* {
302 		syscallarg(const char *) path;
303 		syscallarg(struct linux_statfs *) sp;
304 	} */
305 	struct statvfs *sb;
306 	struct linux_statfs ltmp;
307 	int error;
308 
309 	sb = STATVFSBUF_GET();
310 	error = do_sys_pstatvfs(l, SCARG(uap, path), ST_WAIT, sb);
311 	if (error == 0) {
312 		bsd_to_linux_statfs(sb, &ltmp);
313 		error = copyout(&ltmp, SCARG(uap, sp), sizeof ltmp);
314 	}
315 	STATVFSBUF_PUT(sb);
316 
317 	return error;
318 }
319 
320 int
321 linux_sys_fstatfs(struct lwp *l, const struct linux_sys_fstatfs_args *uap, register_t *retval)
322 {
323 	/* {
324 		syscallarg(int) fd;
325 		syscallarg(struct linux_statfs *) sp;
326 	} */
327 	struct statvfs *sb;
328 	struct linux_statfs ltmp;
329 	int error;
330 
331 	sb = STATVFSBUF_GET();
332 	error = do_sys_fstatvfs(l, SCARG(uap, fd), ST_WAIT, sb);
333 	if (error == 0) {
334 		bsd_to_linux_statfs(sb, &ltmp);
335 		error = copyout(&ltmp, SCARG(uap, sp), sizeof ltmp);
336 	}
337 	STATVFSBUF_PUT(sb);
338 
339 	return error;
340 }
341 
342 /*
343  * uname(). Just copy the info from the various strings stored in the
344  * kernel, and put it in the Linux utsname structure. That structure
345  * is almost the same as the NetBSD one, only it has fields 65 characters
346  * long, and an extra domainname field.
347  */
348 int
349 linux_sys_uname(struct lwp *l, const struct linux_sys_uname_args *uap, register_t *retval)
350 {
351 	/* {
352 		syscallarg(struct linux_utsname *) up;
353 	} */
354 	struct linux_utsname luts;
355 
356 	memset(&luts, 0, sizeof(luts));
357 	strlcpy(luts.l_sysname, linux_sysname, sizeof(luts.l_sysname));
358 	strlcpy(luts.l_nodename, hostname, sizeof(luts.l_nodename));
359 	strlcpy(luts.l_release, linux_release, sizeof(luts.l_release));
360 	strlcpy(luts.l_version, linux_version, sizeof(luts.l_version));
361 	strlcpy(luts.l_machine, LINUX_UNAME_ARCH, sizeof(luts.l_machine));
362 	strlcpy(luts.l_domainname, domainname, sizeof(luts.l_domainname));
363 
364 	return copyout(&luts, SCARG(uap, up), sizeof(luts));
365 }
366 
367 /* Used directly on: alpha, mips, ppc, sparc, sparc64 */
368 /* Used indirectly on: arm, i386, m68k */
369 
370 /*
371  * New type Linux mmap call.
372  * Only called directly on machines with >= 6 free regs.
373  */
374 int
375 linux_sys_mmap(struct lwp *l, const struct linux_sys_mmap_args *uap, register_t *retval)
376 {
377 	/* {
378 		syscallarg(unsigned long) addr;
379 		syscallarg(size_t) len;
380 		syscallarg(int) prot;
381 		syscallarg(int) flags;
382 		syscallarg(int) fd;
383 		syscallarg(linux_off_t) offset;
384 	} */
385 
386 	if (SCARG(uap, offset) & PAGE_MASK)
387 		return EINVAL;
388 
389 	return linux_mmap(l, uap, retval, SCARG(uap, offset));
390 }
391 
392 /*
393  * Guts of most architectures' mmap64() implementations.  This shares
394  * its list of arguments with linux_sys_mmap().
395  *
396  * The difference in linux_sys_mmap2() is that "offset" is actually
397  * (offset / pagesize), not an absolute byte count.  This translation
398  * to pagesize offsets is done inside glibc between the mmap64() call
399  * point, and the actual syscall.
400  */
401 int
402 linux_sys_mmap2(struct lwp *l, const struct linux_sys_mmap2_args *uap, register_t *retval)
403 {
404 	/* {
405 		syscallarg(unsigned long) addr;
406 		syscallarg(size_t) len;
407 		syscallarg(int) prot;
408 		syscallarg(int) flags;
409 		syscallarg(int) fd;
410 		syscallarg(linux_off_t) offset;
411 	} */
412 
413 	return linux_mmap(l, uap, retval,
414 	    ((off_t)SCARG(uap, offset)) << PAGE_SHIFT);
415 }
416 
417 /*
418  * Massage arguments and call system mmap(2).
419  */
420 static int
421 linux_mmap(struct lwp *l, const struct linux_sys_mmap_args *uap, register_t *retval, off_t offset)
422 {
423 	struct sys_mmap_args cma;
424 	int error;
425 	size_t mmoff=0;
426 
427 	linux_to_bsd_mmap_args(&cma, uap);
428 	SCARG(&cma, pos) = offset;
429 
430 	if (SCARG(uap, flags) & LINUX_MAP_GROWSDOWN) {
431 		/*
432 		 * Request for stack-like memory segment. On linux, this
433 		 * works by mmap()ping (small) segment, which is automatically
434 		 * extended when page fault happens below the currently
435 		 * allocated area. We emulate this by allocating (typically
436 		 * bigger) segment sized at current stack size limit, and
437 		 * offsetting the requested and returned address accordingly.
438 		 * Since physical pages are only allocated on-demand, this
439 		 * is effectively identical.
440 		 */
441 		rlim_t ssl = l->l_proc->p_rlimit[RLIMIT_STACK].rlim_cur;
442 
443 		if (SCARG(&cma, len) < ssl) {
444 			/* Compute the address offset */
445 			mmoff = round_page(ssl) - SCARG(uap, len);
446 
447 			if (SCARG(&cma, addr))
448 				SCARG(&cma, addr) = (char *)SCARG(&cma, addr) - mmoff;
449 
450 			SCARG(&cma, len) = (size_t) ssl;
451 		}
452 	}
453 
454 	error = sys_mmap(l, &cma, retval);
455 	if (error)
456 		return (error);
457 
458 	/* Shift the returned address for stack-like segment if necessary */
459 	retval[0] += mmoff;
460 
461 	return (0);
462 }
463 
464 static void
465 linux_to_bsd_mmap_args(struct sys_mmap_args *cma, const struct linux_sys_mmap_args *uap)
466 {
467 	int flags = MAP_TRYFIXED, fl = SCARG(uap, flags);
468 
469 	flags |= cvtto_bsd_mask(fl, LINUX_MAP_SHARED, MAP_SHARED);
470 	flags |= cvtto_bsd_mask(fl, LINUX_MAP_PRIVATE, MAP_PRIVATE);
471 	flags |= cvtto_bsd_mask(fl, LINUX_MAP_FIXED, MAP_FIXED);
472 	flags |= cvtto_bsd_mask(fl, LINUX_MAP_ANON, MAP_ANON);
473 	flags |= cvtto_bsd_mask(fl, LINUX_MAP_LOCKED, MAP_WIRED);
474 	/* XXX XAX ERH: Any other flags here?  There are more defined... */
475 
476 	SCARG(cma, addr) = (void *)SCARG(uap, addr);
477 	SCARG(cma, len) = SCARG(uap, len);
478 	SCARG(cma, prot) = SCARG(uap, prot);
479 	if (SCARG(cma, prot) & VM_PROT_WRITE) /* XXX */
480 		SCARG(cma, prot) |= VM_PROT_READ;
481 	SCARG(cma, flags) = flags;
482 	SCARG(cma, fd) = flags & MAP_ANON ? -1 : SCARG(uap, fd);
483 	SCARG(cma, PAD) = 0;
484 }
485 
486 #define	LINUX_MREMAP_MAYMOVE	1
487 #define	LINUX_MREMAP_FIXED	2
488 
489 int
490 linux_sys_mremap(struct lwp *l, const struct linux_sys_mremap_args *uap, register_t *retval)
491 {
492 	/* {
493 		syscallarg(void *) old_address;
494 		syscallarg(size_t) old_size;
495 		syscallarg(size_t) new_size;
496 		syscallarg(u_long) flags;
497 	} */
498 
499 	struct proc *p;
500 	struct vm_map *map;
501 	vaddr_t oldva;
502 	vaddr_t newva;
503 	size_t oldsize;
504 	size_t newsize;
505 	int flags;
506 	int uvmflags;
507 	int error;
508 
509 	flags = SCARG(uap, flags);
510 	oldva = (vaddr_t)SCARG(uap, old_address);
511 	oldsize = round_page(SCARG(uap, old_size));
512 	newsize = round_page(SCARG(uap, new_size));
513 	if ((flags & ~(LINUX_MREMAP_FIXED|LINUX_MREMAP_MAYMOVE)) != 0) {
514 		error = EINVAL;
515 		goto done;
516 	}
517 	if ((flags & LINUX_MREMAP_FIXED) != 0) {
518 		if ((flags & LINUX_MREMAP_MAYMOVE) == 0) {
519 			error = EINVAL;
520 			goto done;
521 		}
522 #if 0 /* notyet */
523 		newva = SCARG(uap, new_address);
524 		uvmflags = MAP_FIXED;
525 #else /* notyet */
526 		error = EOPNOTSUPP;
527 		goto done;
528 #endif /* notyet */
529 	} else if ((flags & LINUX_MREMAP_MAYMOVE) != 0) {
530 		uvmflags = 0;
531 	} else {
532 		newva = oldva;
533 		uvmflags = MAP_FIXED;
534 	}
535 	p = l->l_proc;
536 	map = &p->p_vmspace->vm_map;
537 	error = uvm_mremap(map, oldva, oldsize, map, &newva, newsize, p,
538 	    uvmflags);
539 
540 done:
541 	*retval = (error != 0) ? 0 : (register_t)newva;
542 	return error;
543 }
544 
545 #ifdef USRSTACK
546 int
547 linux_sys_mprotect(struct lwp *l, const struct linux_sys_mprotect_args *uap, register_t *retval)
548 {
549 	/* {
550 		syscallarg(const void *) start;
551 		syscallarg(unsigned long) len;
552 		syscallarg(int) prot;
553 	} */
554 	struct vm_map_entry *entry;
555 	struct vm_map *map;
556 	struct proc *p;
557 	vaddr_t end, start, len, stacklim;
558 	int prot, grows;
559 
560 	start = (vaddr_t)SCARG(uap, start);
561 	len = round_page(SCARG(uap, len));
562 	prot = SCARG(uap, prot);
563 	grows = prot & (LINUX_PROT_GROWSDOWN | LINUX_PROT_GROWSUP);
564 	prot &= ~grows;
565 	end = start + len;
566 
567 	if (start & PAGE_MASK)
568 		return EINVAL;
569 	if (end < start)
570 		return EINVAL;
571 	if (end == start)
572 		return 0;
573 
574 	if (prot & ~(PROT_READ | PROT_WRITE | PROT_EXEC))
575 		return EINVAL;
576 	if (grows == (LINUX_PROT_GROWSDOWN | LINUX_PROT_GROWSUP))
577 		return EINVAL;
578 
579 	p = l->l_proc;
580 	map = &p->p_vmspace->vm_map;
581 	vm_map_lock(map);
582 # ifdef notdef
583 	VM_MAP_RANGE_CHECK(map, start, end);
584 # endif
585 	if (!uvm_map_lookup_entry(map, start, &entry) || entry->start > start) {
586 		vm_map_unlock(map);
587 		return ENOMEM;
588 	}
589 
590 	/*
591 	 * Approximate the behaviour of PROT_GROWS{DOWN,UP}.
592 	 */
593 
594 	stacklim = (vaddr_t)p->p_limit->pl_rlimit[RLIMIT_STACK].rlim_cur;
595 	if (grows & LINUX_PROT_GROWSDOWN) {
596 		if (USRSTACK - stacklim <= start && start < USRSTACK) {
597 			start = USRSTACK - stacklim;
598 		} else {
599 			start = entry->start;
600 		}
601 	} else if (grows & LINUX_PROT_GROWSUP) {
602 		if (USRSTACK <= end && end < USRSTACK + stacklim) {
603 			end = USRSTACK + stacklim;
604 		} else {
605 			end = entry->end;
606 		}
607 	}
608 	vm_map_unlock(map);
609 	return uvm_map_protect_user(l, start, end, prot);
610 }
611 #endif /* USRSTACK */
612 
613 /*
614  * This code is partly stolen from src/lib/libc/compat-43/times.c
615  */
616 
617 #define	CONVTCK(r)	(r.tv_sec * hz + r.tv_usec / (1000000 / hz))
618 
619 int
620 linux_sys_times(struct lwp *l, const struct linux_sys_times_args *uap, register_t *retval)
621 {
622 	/* {
623 		syscallarg(struct times *) tms;
624 	} */
625 	struct proc *p = l->l_proc;
626 	struct timeval t;
627 	int error;
628 
629 	if (SCARG(uap, tms)) {
630 		struct linux_tms ltms;
631 		struct rusage ru;
632 
633 		memset(&ltms, 0, sizeof(ltms));
634 
635 		mutex_enter(p->p_lock);
636 		calcru(p, &ru.ru_utime, &ru.ru_stime, NULL, NULL);
637 		ltms.ltms_utime = CONVTCK(ru.ru_utime);
638 		ltms.ltms_stime = CONVTCK(ru.ru_stime);
639 		ltms.ltms_cutime = CONVTCK(p->p_stats->p_cru.ru_utime);
640 		ltms.ltms_cstime = CONVTCK(p->p_stats->p_cru.ru_stime);
641 		mutex_exit(p->p_lock);
642 
643 		if ((error = copyout(&ltms, SCARG(uap, tms), sizeof ltms)))
644 			return error;
645 	}
646 
647 	getmicrouptime(&t);
648 
649 	retval[0] = ((linux_clock_t)(CONVTCK(t)));
650 	return 0;
651 }
652 
653 #undef CONVTCK
654 
655 #if !defined(__aarch64__)
656 /*
657  * Linux 'readdir' call. This code is mostly taken from the
658  * SunOS getdents call (see compat/sunos/sunos_misc.c), though
659  * an attempt has been made to keep it a little cleaner (failing
660  * miserably, because of the cruft needed if count 1 is passed).
661  *
662  * The d_off field should contain the offset of the next valid entry,
663  * but in Linux it has the offset of the entry itself. We emulate
664  * that bug here.
665  *
666  * Read in BSD-style entries, convert them, and copy them out.
667  *
668  * Note that this doesn't handle union-mounted filesystems.
669  */
670 int
671 linux_sys_getdents(struct lwp *l, const struct linux_sys_getdents_args *uap, register_t *retval)
672 {
673 	/* {
674 		syscallarg(int) fd;
675 		syscallarg(struct linux_dirent *) dent;
676 		syscallarg(unsigned int) count;
677 	} */
678 	struct dirent *bdp;
679 	struct vnode *vp;
680 	char *inp, *tbuf;		/* BSD-format */
681 	int len, reclen;		/* BSD-format */
682 	char *outp;			/* Linux-format */
683 	int resid, linux_reclen = 0;	/* Linux-format */
684 	struct file *fp;
685 	struct uio auio;
686 	struct iovec aiov;
687 	struct linux_dirent idb;
688 	off_t off;		/* true file offset */
689 	int buflen, error, eofflag, nbytes, oldcall;
690 	struct vattr va;
691 	off_t *cookiebuf = NULL, *cookie;
692 	int ncookies;
693 
694 	/* fd_getvnode() will use the descriptor for us */
695 	if ((error = fd_getvnode(SCARG(uap, fd), &fp)) != 0)
696 		return (error);
697 
698 	if ((fp->f_flag & FREAD) == 0) {
699 		error = EBADF;
700 		goto out1;
701 	}
702 
703 	vp = (struct vnode *)fp->f_data;
704 	if (vp->v_type != VDIR) {
705 		error = ENOTDIR;
706 		goto out1;
707 	}
708 
709 	vn_lock(vp, LK_SHARED | LK_RETRY);
710 	error = VOP_GETATTR(vp, &va, l->l_cred);
711 	VOP_UNLOCK(vp);
712 	if (error)
713 		goto out1;
714 
715 	nbytes = SCARG(uap, count);
716 	if (nbytes == 1) {	/* emulating old, broken behaviour */
717 		nbytes = sizeof (idb);
718 		buflen = uimax(va.va_blocksize, nbytes);
719 		oldcall = 1;
720 	} else {
721 		buflen = uimin(MAXBSIZE, nbytes);
722 		if (buflen < va.va_blocksize)
723 			buflen = va.va_blocksize;
724 		oldcall = 0;
725 	}
726 	tbuf = malloc(buflen, M_TEMP, M_WAITOK);
727 
728 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
729 	off = fp->f_offset;
730 again:
731 	aiov.iov_base = tbuf;
732 	aiov.iov_len = buflen;
733 	auio.uio_iov = &aiov;
734 	auio.uio_iovcnt = 1;
735 	auio.uio_rw = UIO_READ;
736 	auio.uio_resid = buflen;
737 	auio.uio_offset = off;
738 	UIO_SETUP_SYSSPACE(&auio);
739 	/*
740          * First we read into the malloc'ed buffer, then
741          * we massage it into user space, one record at a time.
742          */
743 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, &cookiebuf,
744 	    &ncookies);
745 	if (error)
746 		goto out;
747 
748 	inp = tbuf;
749 	outp = (void *)SCARG(uap, dent);
750 	resid = nbytes;
751 	if ((len = buflen - auio.uio_resid) == 0)
752 		goto eof;
753 
754 	for (cookie = cookiebuf; len > 0; len -= reclen) {
755 		bdp = (struct dirent *)inp;
756 		reclen = bdp->d_reclen;
757 		if (reclen & 3) {
758 			error = EIO;
759 			goto out;
760 		}
761 		if (bdp->d_fileno == 0) {
762 			inp += reclen;	/* it is a hole; squish it out */
763 			if (cookie)
764 				off = *cookie++;
765 			else
766 				off += reclen;
767 			continue;
768 		}
769 		linux_reclen = LINUX_RECLEN(&idb, bdp->d_namlen);
770 		if (reclen > len || resid < linux_reclen) {
771 			/* entry too big for buffer, so just stop */
772 			outp++;
773 			break;
774 		}
775 		/*
776 		 * Massage in place to make a Linux-shaped dirent (otherwise
777 		 * we have to worry about touching user memory outside of
778 		 * the copyout() call).
779 		 */
780 		memset(&idb, 0, sizeof(idb));
781 		idb.d_ino = bdp->d_fileno;
782 		/*
783 		 * The old readdir() call misuses the offset and reclen fields.
784 		 */
785 		if (oldcall) {
786 			idb.d_off = (linux_off_t)linux_reclen;
787 			idb.d_reclen = (u_short)bdp->d_namlen;
788 		} else {
789 			if (sizeof (idb.d_off) <= 4 && (off >> 32) != 0) {
790 				compat_offseterr(vp, "linux_getdents");
791 				error = EINVAL;
792 				goto out;
793 			}
794 			idb.d_off = (linux_off_t)off;
795 			idb.d_reclen = (u_short)linux_reclen;
796 			/* Linux puts d_type at the end of each record */
797 			*((char *)&idb + idb.d_reclen - 1) = bdp->d_type;
798 		}
799 		memcpy(idb.d_name, bdp->d_name,
800 		    MIN(sizeof(idb.d_name), bdp->d_namlen + 1));
801 		if ((error = copyout((void *)&idb, outp, linux_reclen)))
802 			goto out;
803 		/* advance past this real entry */
804 		inp += reclen;
805 		if (cookie)
806 			off = *cookie++; /* each entry points to itself */
807 		else
808 			off += reclen;
809 		/* advance output past Linux-shaped entry */
810 		outp += linux_reclen;
811 		resid -= linux_reclen;
812 		if (oldcall)
813 			break;
814 	}
815 
816 	/* if we squished out the whole block, try again */
817 	if (outp == (void *)SCARG(uap, dent)) {
818 		if (cookiebuf)
819 			free(cookiebuf, M_TEMP);
820 		cookiebuf = NULL;
821 		goto again;
822 	}
823 	fp->f_offset = off;	/* update the vnode offset */
824 
825 	if (oldcall)
826 		nbytes = resid + linux_reclen;
827 
828 eof:
829 	*retval = nbytes - resid;
830 out:
831 	VOP_UNLOCK(vp);
832 	if (cookiebuf)
833 		free(cookiebuf, M_TEMP);
834 	free(tbuf, M_TEMP);
835 out1:
836 	fd_putfile(SCARG(uap, fd));
837 	return error;
838 }
839 #endif
840 
841 #if !defined(__aarch64__)
842 /*
843  * Even when just using registers to pass arguments to syscalls you can
844  * have 5 of them on the i386. So this newer version of select() does
845  * this.
846  */
847 int
848 linux_sys_select(struct lwp *l, const struct linux_sys_select_args *uap, register_t *retval)
849 {
850 	/* {
851 		syscallarg(int) nfds;
852 		syscallarg(fd_set *) readfds;
853 		syscallarg(fd_set *) writefds;
854 		syscallarg(fd_set *) exceptfds;
855 		syscallarg(struct timeval50 *) timeout;
856 	} */
857 
858 	return linux_select1(l, retval, SCARG(uap, nfds), SCARG(uap, readfds),
859 	    SCARG(uap, writefds), SCARG(uap, exceptfds),
860 	    (struct linux_timeval *)SCARG(uap, timeout));
861 }
862 
863 /*
864  * Common code for the old and new versions of select(). A couple of
865  * things are important:
866  * 1) return the amount of time left in the 'timeout' parameter
867  * 2) select never returns ERESTART on Linux, always return EINTR
868  */
869 int
870 linux_select1(struct lwp *l, register_t *retval, int nfds, fd_set *readfds,
871     fd_set *writefds, fd_set *exceptfds, struct linux_timeval *timeout)
872 {
873 	struct timespec ts0, ts1, uts, *ts = NULL;
874 	struct linux_timeval ltv;
875 	int error;
876 
877 	/*
878 	 * Store current time for computation of the amount of
879 	 * time left.
880 	 */
881 	if (timeout) {
882 		if ((error = copyin(timeout, &ltv, sizeof(ltv))))
883 			return error;
884 		uts.tv_sec = ltv.tv_sec;
885 		uts.tv_nsec = (long)((unsigned long)ltv.tv_usec * 1000);
886 		if (itimespecfix(&uts)) {
887 			/*
888 			 * The timeval was invalid.  Convert it to something
889 			 * valid that will act as it does under Linux.
890 			 */
891 			uts.tv_sec += uts.tv_nsec / 1000000000;
892 			uts.tv_nsec %= 1000000000;
893 			if (uts.tv_nsec < 0) {
894 				uts.tv_sec -= 1;
895 				uts.tv_nsec += 1000000000;
896 			}
897 			if (uts.tv_sec < 0)
898 				timespecclear(&uts);
899 		}
900 		ts = &uts;
901 		nanotime(&ts0);
902 	}
903 
904 	error = selcommon(retval, nfds, readfds, writefds, exceptfds, ts, NULL);
905 
906 	if (error) {
907 		/*
908 		 * See fs/select.c in the Linux kernel.  Without this,
909 		 * Maelstrom doesn't work.
910 		 */
911 		if (error == ERESTART)
912 			error = EINTR;
913 		return error;
914 	}
915 
916 	if (timeout) {
917 		if (*retval) {
918 			/*
919 			 * Compute how much time was left of the timeout,
920 			 * by subtracting the current time and the time
921 			 * before we started the call, and subtracting
922 			 * that result from the user-supplied value.
923 			 */
924 			nanotime(&ts1);
925 			timespecsub(&ts1, &ts0, &ts1);
926 			timespecsub(&uts, &ts1, &uts);
927 			if (uts.tv_sec < 0)
928 				timespecclear(&uts);
929 		} else
930 			timespecclear(&uts);
931 		ltv.tv_sec = uts.tv_sec;
932 		ltv.tv_usec = uts.tv_nsec / 1000;
933 		if ((error = copyout(&ltv, timeout, sizeof(ltv))))
934 			return error;
935 	}
936 
937 	return 0;
938 }
939 #endif
940 
941 /*
942  * Derived from FreeBSD's sys/compat/linux/linux_misc.c:linux_pselect6()
943  * which was contributed by Dmitry Chagin
944  * https://svnweb.freebsd.org/base?view=revision&revision=283403
945  */
946 int
947 linux_sys_pselect6(struct lwp *l,
948 	const struct linux_sys_pselect6_args *uap, register_t *retval)
949 {
950 	/* {
951 		syscallarg(int) nfds;
952 		syscallarg(fd_set *) readfds;
953 		syscallarg(fd_set *) writefds;
954 		syscallarg(fd_set *) exceptfds;
955 		syscallarg(struct timespec *) timeout;
956 		syscallarg(linux_sized_sigset_t *) ss;
957 	} */
958 	struct timespec uts, ts0, ts1, *tsp;
959 	linux_sized_sigset_t lsss;
960 	struct linux_timespec lts;
961 	linux_sigset_t lss;
962 	sigset_t *ssp;
963 	sigset_t ss;
964 	int error;
965 
966 	ssp = NULL;
967 	if (SCARG(uap, ss) != NULL) {
968 		if ((error = copyin(SCARG(uap, ss), &lsss, sizeof(lsss))) != 0)
969 			return (error);
970 		if (lsss.ss_len != sizeof(lss))
971 			return (EINVAL);
972 		if (lsss.ss != NULL) {
973 			if ((error = copyin(lsss.ss, &lss, sizeof(lss))) != 0)
974 				return (error);
975 			linux_to_native_sigset(&ss, &lss);
976 			ssp = &ss;
977 		}
978 	}
979 
980 	if (SCARG(uap, timeout) != NULL) {
981 		error = copyin(SCARG(uap, timeout), &lts, sizeof(lts));
982 		if (error != 0)
983 			return (error);
984 		linux_to_native_timespec(&uts, &lts);
985 
986 		if (itimespecfix(&uts))
987 			return (EINVAL);
988 
989 		nanotime(&ts0);
990 		tsp = &uts;
991 	} else {
992 		tsp = NULL;
993 	}
994 
995 	error = selcommon(retval, SCARG(uap, nfds), SCARG(uap, readfds),
996 	    SCARG(uap, writefds), SCARG(uap, exceptfds), tsp, ssp);
997 
998 	if (error == 0 && tsp != NULL) {
999 		if (retval != 0) {
1000 			/*
1001 			 * Compute how much time was left of the timeout,
1002 			 * by subtracting the current time and the time
1003 			 * before we started the call, and subtracting
1004 			 * that result from the user-supplied value.
1005 			 */
1006 			nanotime(&ts1);
1007 			timespecsub(&ts1, &ts0, &ts1);
1008 			timespecsub(&uts, &ts1, &uts);
1009 			if (uts.tv_sec < 0)
1010 				timespecclear(&uts);
1011 		} else {
1012 			timespecclear(&uts);
1013 		}
1014 
1015 		native_to_linux_timespec(&lts, &uts);
1016 		error = copyout(&lts, SCARG(uap, timeout), sizeof(lts));
1017 	}
1018 
1019 	return (error);
1020 }
1021 
1022 int
1023 linux_sys_ppoll(struct lwp *l,
1024 	const struct linux_sys_ppoll_args *uap, register_t *retval)
1025 {
1026 	/* {
1027 		syscallarg(struct pollfd *) fds;
1028 		syscallarg(u_int) nfds;
1029 		syscallarg(struct linux_timespec *) timeout;
1030 		syscallarg(linux_sigset_t *) sigset;
1031 	} */
1032 	struct linux_timespec lts0, *lts;
1033 	struct timespec ts0, *ts = NULL;
1034 	linux_sigset_t lsigmask0, *lsigmask;
1035 	sigset_t sigmask0, *sigmask = NULL;
1036 	int error;
1037 
1038 	lts = SCARG(uap, timeout);
1039 	if (lts) {
1040 		if ((error = copyin(lts, &lts0, sizeof(lts0))) != 0)
1041 			return error;
1042 		linux_to_native_timespec(&ts0, &lts0);
1043 		ts = &ts0;
1044 	}
1045 
1046 	lsigmask = SCARG(uap, sigset);
1047 	if (lsigmask) {
1048 		if ((error = copyin(lsigmask, &lsigmask0, sizeof(lsigmask0))))
1049 			return error;
1050 		linux_to_native_sigset(&sigmask0, &lsigmask0);
1051 		sigmask = &sigmask0;
1052 	}
1053 
1054 	return pollcommon(retval, SCARG(uap, fds), SCARG(uap, nfds),
1055 	    ts, sigmask);
1056 }
1057 
1058 /*
1059  * Set the 'personality' (emulation mode) for the current process. Only
1060  * accept the Linux personality here (0). This call is needed because
1061  * the Linux ELF crt0 issues it in an ugly kludge to make sure that
1062  * ELF binaries run in Linux mode, not SVR4 mode.
1063  */
1064 int
1065 linux_sys_personality(struct lwp *l, const struct linux_sys_personality_args *uap, register_t *retval)
1066 {
1067 	/* {
1068 		syscallarg(unsigned long) per;
1069 	} */
1070 	struct linux_emuldata *led;
1071 	int per;
1072 
1073 	per = SCARG(uap, per);
1074 	led = l->l_emuldata;
1075 	if (per == LINUX_PER_QUERY) {
1076 		retval[0] = led->led_personality;
1077 		return 0;
1078 	}
1079 
1080 	switch (per & LINUX_PER_MASK) {
1081 	case LINUX_PER_LINUX:
1082 	case LINUX_PER_LINUX32:
1083 		led->led_personality = per;
1084 		break;
1085 
1086 	default:
1087 		return EINVAL;
1088 	}
1089 
1090 	retval[0] = per;
1091 	return 0;
1092 }
1093 
1094 /*
1095  * We have nonexistent fsuid equal to uid.
1096  * If modification is requested, refuse.
1097  */
1098 int
1099 linux_sys_setfsuid(struct lwp *l, const struct linux_sys_setfsuid_args *uap, register_t *retval)
1100 {
1101 	 /* {
1102 		 syscallarg(uid_t) uid;
1103 	 } */
1104 	 uid_t uid;
1105 
1106 	 uid = SCARG(uap, uid);
1107 	 if (kauth_cred_getuid(l->l_cred) != uid)
1108 		 return sys_nosys(l, uap, retval);
1109 
1110 	 *retval = uid;
1111 	 return 0;
1112 }
1113 
1114 int
1115 linux_sys_setfsgid(struct lwp *l, const struct linux_sys_setfsgid_args *uap, register_t *retval)
1116 {
1117 	/* {
1118 		syscallarg(gid_t) gid;
1119 	} */
1120 	gid_t gid;
1121 
1122 	gid = SCARG(uap, gid);
1123 	if (kauth_cred_getgid(l->l_cred) != gid)
1124 		return sys_nosys(l, uap, retval);
1125 
1126 	*retval = gid;
1127 	return 0;
1128 }
1129 
1130 int
1131 linux_sys_setresuid(struct lwp *l, const struct linux_sys_setresuid_args *uap, register_t *retval)
1132 {
1133 	/* {
1134 		syscallarg(uid_t) ruid;
1135 		syscallarg(uid_t) euid;
1136 		syscallarg(uid_t) suid;
1137 	} */
1138 
1139 	/*
1140 	 * Note: These checks are a little different than the NetBSD
1141 	 * setreuid(2) call performs.  This precisely follows the
1142 	 * behavior of the Linux kernel.
1143 	 */
1144 
1145 	return do_setresuid(l, SCARG(uap, ruid), SCARG(uap, euid),
1146 			    SCARG(uap, suid),
1147 			    ID_R_EQ_R | ID_R_EQ_E | ID_R_EQ_S |
1148 			    ID_E_EQ_R | ID_E_EQ_E | ID_E_EQ_S |
1149 			    ID_S_EQ_R | ID_S_EQ_E | ID_S_EQ_S );
1150 }
1151 
1152 int
1153 linux_sys_getresuid(struct lwp *l, const struct linux_sys_getresuid_args *uap, register_t *retval)
1154 {
1155 	/* {
1156 		syscallarg(uid_t *) ruid;
1157 		syscallarg(uid_t *) euid;
1158 		syscallarg(uid_t *) suid;
1159 	} */
1160 	kauth_cred_t pc = l->l_cred;
1161 	int error;
1162 	uid_t uid;
1163 
1164 	/*
1165 	 * Linux copies these values out to userspace like so:
1166 	 *
1167 	 *	1. Copy out ruid.
1168 	 *	2. If that succeeds, copy out euid.
1169 	 *	3. If both of those succeed, copy out suid.
1170 	 */
1171 	uid = kauth_cred_getuid(pc);
1172 	if ((error = copyout(&uid, SCARG(uap, ruid), sizeof(uid_t))) != 0)
1173 		return (error);
1174 
1175 	uid = kauth_cred_geteuid(pc);
1176 	if ((error = copyout(&uid, SCARG(uap, euid), sizeof(uid_t))) != 0)
1177 		return (error);
1178 
1179 	uid = kauth_cred_getsvuid(pc);
1180 
1181 	return (copyout(&uid, SCARG(uap, suid), sizeof(uid_t)));
1182 }
1183 
1184 int
1185 linux_sys_ptrace(struct lwp *l, const struct linux_sys_ptrace_args *uap, register_t *retval)
1186 {
1187 	/* {
1188 		i386, m68k, powerpc: T=int
1189 		alpha, amd64: T=long
1190 		syscallarg(T) request;
1191 		syscallarg(T) pid;
1192 		syscallarg(T) addr;
1193 		syscallarg(T) data;
1194 	} */
1195 	const int *ptr;
1196 	int request;
1197 	int error;
1198 
1199 	ptr = linux_ptrace_request_map;
1200 	request = SCARG(uap, request);
1201 	while (*ptr != -1)
1202 		if (*ptr++ == request) {
1203 			struct sys_ptrace_args pta;
1204 
1205 			SCARG(&pta, req) = *ptr;
1206 			SCARG(&pta, pid) = SCARG(uap, pid);
1207 			SCARG(&pta, addr) = (void *)SCARG(uap, addr);
1208 			SCARG(&pta, data) = SCARG(uap, data);
1209 
1210 			/*
1211 			 * Linux ptrace(PTRACE_CONT, pid, 0, 0) means actually
1212 			 * to continue where the process left off previously.
1213  			 * The same thing is achieved by addr == (void *) 1
1214 			 * on NetBSD, so rewrite 'addr' appropriately.
1215 			 */
1216 			if (request == LINUX_PTRACE_CONT && SCARG(uap, addr)==0)
1217 				SCARG(&pta, addr) = (void *) 1;
1218 
1219 			error = sysent[SYS_ptrace].sy_call(l, &pta, retval);
1220 			if (error)
1221 				return error;
1222 			switch (request) {
1223 			case LINUX_PTRACE_PEEKTEXT:
1224 			case LINUX_PTRACE_PEEKDATA:
1225 				error = copyout (retval,
1226 				    (void *)SCARG(uap, data),
1227 				    sizeof *retval);
1228 				*retval = SCARG(uap, data);
1229 				break;
1230 			default:
1231 				break;
1232 			}
1233 			return error;
1234 		}
1235 		else
1236 			ptr++;
1237 
1238 	return LINUX_SYS_PTRACE_ARCH(l, uap, retval);
1239 }
1240 
1241 int
1242 linux_sys_reboot(struct lwp *l, const struct linux_sys_reboot_args *uap, register_t *retval)
1243 {
1244 	/* {
1245 		syscallarg(int) magic1;
1246 		syscallarg(int) magic2;
1247 		syscallarg(int) cmd;
1248 		syscallarg(void *) arg;
1249 	} */
1250 	struct sys_reboot_args /* {
1251 		syscallarg(int) opt;
1252 		syscallarg(char *) bootstr;
1253 	} */ sra;
1254 	int error;
1255 
1256 	if ((error = kauth_authorize_system(l->l_cred,
1257 	    KAUTH_SYSTEM_REBOOT, 0, NULL, NULL, NULL)) != 0)
1258 		return(error);
1259 
1260 	if (SCARG(uap, magic1) != LINUX_REBOOT_MAGIC1)
1261 		return(EINVAL);
1262 	if (SCARG(uap, magic2) != LINUX_REBOOT_MAGIC2 &&
1263 	    SCARG(uap, magic2) != LINUX_REBOOT_MAGIC2A &&
1264 	    SCARG(uap, magic2) != LINUX_REBOOT_MAGIC2B)
1265 		return(EINVAL);
1266 
1267 	switch ((unsigned long)SCARG(uap, cmd)) {
1268 	case LINUX_REBOOT_CMD_RESTART:
1269 		SCARG(&sra, opt) = RB_AUTOBOOT;
1270 		break;
1271 	case LINUX_REBOOT_CMD_HALT:
1272 		SCARG(&sra, opt) = RB_HALT;
1273 		break;
1274 	case LINUX_REBOOT_CMD_POWER_OFF:
1275 		SCARG(&sra, opt) = RB_HALT|RB_POWERDOWN;
1276 		break;
1277 	case LINUX_REBOOT_CMD_RESTART2:
1278 		/* Reboot with an argument. */
1279 		SCARG(&sra, opt) = RB_AUTOBOOT|RB_STRING;
1280 		SCARG(&sra, bootstr) = SCARG(uap, arg);
1281 		break;
1282 	case LINUX_REBOOT_CMD_CAD_ON:
1283 		return(EINVAL);	/* We don't implement ctrl-alt-delete */
1284 	case LINUX_REBOOT_CMD_CAD_OFF:
1285 		return(0);
1286 	default:
1287 		return(EINVAL);
1288 	}
1289 
1290 	return(sys_reboot(l, &sra, retval));
1291 }
1292 
1293 /*
1294  * Copy of compat_12_sys_swapon().
1295  */
1296 int
1297 linux_sys_swapon(struct lwp *l, const struct linux_sys_swapon_args *uap, register_t *retval)
1298 {
1299 	/* {
1300 		syscallarg(const char *) name;
1301 	} */
1302 	struct sys_swapctl_args ua;
1303 
1304 	SCARG(&ua, cmd) = SWAP_ON;
1305 	SCARG(&ua, arg) = (void *)__UNCONST(SCARG(uap, name));
1306 	SCARG(&ua, misc) = 0;	/* priority */
1307 	return (sys_swapctl(l, &ua, retval));
1308 }
1309 
1310 /*
1311  * Stop swapping to the file or block device specified by path.
1312  */
1313 int
1314 linux_sys_swapoff(struct lwp *l, const struct linux_sys_swapoff_args *uap, register_t *retval)
1315 {
1316 	/* {
1317 		syscallarg(const char *) path;
1318 	} */
1319 	struct sys_swapctl_args ua;
1320 
1321 	SCARG(&ua, cmd) = SWAP_OFF;
1322 	SCARG(&ua, arg) = __UNCONST(SCARG(uap, path)); /*XXXUNCONST*/
1323 	return (sys_swapctl(l, &ua, retval));
1324 }
1325 
1326 /*
1327  * Copy of compat_09_sys_setdomainname()
1328  */
1329 /* ARGSUSED */
1330 int
1331 linux_sys_setdomainname(struct lwp *l, const struct linux_sys_setdomainname_args *uap, register_t *retval)
1332 {
1333 	/* {
1334 		syscallarg(char *) domainname;
1335 		syscallarg(int) len;
1336 	} */
1337 	int name[2];
1338 
1339 	name[0] = CTL_KERN;
1340 	name[1] = KERN_DOMAINNAME;
1341 	return (old_sysctl(&name[0], 2, 0, 0, SCARG(uap, domainname),
1342 			    SCARG(uap, len), l));
1343 }
1344 
1345 /*
1346  * sysinfo()
1347  */
1348 /* ARGSUSED */
1349 int
1350 linux_sys_sysinfo(struct lwp *l, const struct linux_sys_sysinfo_args *uap, register_t *retval)
1351 {
1352 	/* {
1353 		syscallarg(struct linux_sysinfo *) arg;
1354 	} */
1355 	struct linux_sysinfo si;
1356 	struct loadavg *la;
1357 	int64_t filepg;
1358 
1359 	memset(&si, 0, sizeof(si));
1360 	si.uptime = time_uptime;
1361 	la = &averunnable;
1362 	si.loads[0] = la->ldavg[0] * LINUX_SYSINFO_LOADS_SCALE / la->fscale;
1363 	si.loads[1] = la->ldavg[1] * LINUX_SYSINFO_LOADS_SCALE / la->fscale;
1364 	si.loads[2] = la->ldavg[2] * LINUX_SYSINFO_LOADS_SCALE / la->fscale;
1365 	si.totalram = ctob((u_long)physmem);
1366 	/* uvm_availmem() may sync the counters. */
1367 	si.freeram = (u_long)uvm_availmem(true) * uvmexp.pagesize;
1368 	filepg = cpu_count_get(CPU_COUNT_FILECLEAN) +
1369 	    cpu_count_get(CPU_COUNT_FILEDIRTY) +
1370 	    cpu_count_get(CPU_COUNT_FILEUNKNOWN) -
1371 	    cpu_count_get(CPU_COUNT_EXECPAGES);
1372 	si.sharedram = 0;	/* XXX */
1373 	si.bufferram = (u_long)(filepg * uvmexp.pagesize);
1374 	si.totalswap = (u_long)uvmexp.swpages * uvmexp.pagesize;
1375 	si.freeswap =
1376 	    (u_long)(uvmexp.swpages - uvmexp.swpginuse) * uvmexp.pagesize;
1377 	si.procs = atomic_load_relaxed(&nprocs);
1378 
1379 	/* The following are only present in newer Linux kernels. */
1380 	si.totalbig = 0;
1381 	si.freebig = 0;
1382 	si.mem_unit = 1;
1383 
1384 	return (copyout(&si, SCARG(uap, arg), sizeof si));
1385 }
1386 
1387 int
1388 linux_sys_getrlimit(struct lwp *l, const struct linux_sys_getrlimit_args *uap, register_t *retval)
1389 {
1390 	/* {
1391 		syscallarg(int) which;
1392 # ifdef LINUX_LARGEFILE64
1393 		syscallarg(struct rlimit *) rlp;
1394 # else
1395 		syscallarg(struct orlimit *) rlp;
1396 # endif
1397 	} */
1398 # ifdef LINUX_LARGEFILE64
1399 	struct rlimit orl;
1400 # else
1401 	struct orlimit orl;
1402 # endif
1403 	int which;
1404 
1405 	which = linux_to_bsd_limit(SCARG(uap, which));
1406 	if (which < 0)
1407 		return -which;
1408 
1409 	memset(&orl, 0, sizeof(orl));
1410 	bsd_to_linux_rlimit(&orl, &l->l_proc->p_rlimit[which]);
1411 
1412 	return copyout(&orl, SCARG(uap, rlp), sizeof(orl));
1413 }
1414 
1415 int
1416 linux_sys_setrlimit(struct lwp *l, const struct linux_sys_setrlimit_args *uap, register_t *retval)
1417 {
1418 	/* {
1419 		syscallarg(int) which;
1420 # ifdef LINUX_LARGEFILE64
1421 		syscallarg(struct rlimit *) rlp;
1422 # else
1423 		syscallarg(struct orlimit *) rlp;
1424 # endif
1425 	} */
1426 	struct rlimit rl;
1427 # ifdef LINUX_LARGEFILE64
1428 	struct rlimit orl;
1429 # else
1430 	struct orlimit orl;
1431 # endif
1432 	int error;
1433 	int which;
1434 
1435 	if ((error = copyin(SCARG(uap, rlp), &orl, sizeof(orl))) != 0)
1436 		return error;
1437 
1438 	which = linux_to_bsd_limit(SCARG(uap, which));
1439 	if (which < 0)
1440 		return -which;
1441 
1442 	linux_to_bsd_rlimit(&rl, &orl);
1443 	return dosetrlimit(l, l->l_proc, which, &rl);
1444 }
1445 
1446 # if !defined(__aarch64__) && !defined(__mips__) && !defined(__amd64__)
1447 /* XXX: this doesn't look 100% common, at least mips doesn't have it */
1448 int
1449 linux_sys_ugetrlimit(struct lwp *l, const struct linux_sys_ugetrlimit_args *uap, register_t *retval)
1450 {
1451 	return linux_sys_getrlimit(l, (const void *)uap, retval);
1452 }
1453 # endif
1454 
1455 int
1456 linux_sys_prlimit64(struct lwp *l, const struct linux_sys_prlimit64_args *uap, register_t *retval)
1457 {
1458 	/* {
1459 		syscallarg(pid_t) pid;
1460 		syscallarg(int) witch;
1461 		syscallarg(struct rlimit *) new_rlp;
1462 		syscallarg(struct rlimit *) old_rlp;
1463 	}; */
1464 	struct rlimit rl, nrl, orl;
1465 	struct rlimit *p;
1466 	int which;
1467 	int error;
1468 
1469 	/* XXX: Cannot operate any process other than its own */
1470 	if (SCARG(uap, pid) != 0)
1471 		return EPERM;
1472 
1473 	which = linux_to_bsd_limit(SCARG(uap, which));
1474 	if (which < 0)
1475 		return -which;
1476 
1477 	p = SCARG(uap, old_rlp);
1478 	if (p != NULL) {
1479 		memset(&orl, 0, sizeof(orl));
1480 		bsd_to_linux_rlimit64(&orl, &l->l_proc->p_rlimit[which]);
1481 		if ((error = copyout(&orl, p, sizeof(orl))) != 0)
1482 			return error;
1483 	}
1484 
1485 	p = SCARG(uap, new_rlp);
1486 	if (p != NULL) {
1487 		if ((error = copyin(p, &nrl, sizeof(nrl))) != 0)
1488 			return error;
1489 
1490 		linux_to_bsd_rlimit(&rl, &nrl);
1491 		return dosetrlimit(l, l->l_proc, which, &rl);
1492 	}
1493 
1494 	return 0;
1495 }
1496 
1497 /*
1498  * This gets called for unsupported syscalls. The difference to sys_nosys()
1499  * is that process does not get SIGSYS, the call just returns with ENOSYS.
1500  * This is the way Linux does it and glibc depends on this behaviour.
1501  */
1502 int
1503 linux_sys_nosys(struct lwp *l, const void *v, register_t *retval)
1504 {
1505 	return (ENOSYS);
1506 }
1507 
1508 int
1509 linux_sys_getpriority(struct lwp *l, const struct linux_sys_getpriority_args *uap, register_t *retval)
1510 {
1511         /* {
1512                 syscallarg(int) which;
1513                 syscallarg(int) who;
1514         } */
1515         struct sys_getpriority_args bsa;
1516         int error;
1517 
1518         SCARG(&bsa, which) = SCARG(uap, which);
1519         SCARG(&bsa, who) = SCARG(uap, who);
1520 
1521         if ((error = sys_getpriority(l, &bsa, retval)))
1522                 return error;
1523 
1524         *retval = NZERO - *retval;
1525 
1526         return 0;
1527 }
1528 
1529 int
1530 linux_do_sys_utimensat(struct lwp *l, int fd, const char *path, struct timespec *tsp, int flags, register_t *retval)
1531 {
1532 	int follow, error;
1533 
1534 	follow = (flags & LINUX_AT_SYMLINK_NOFOLLOW) ? NOFOLLOW : FOLLOW;
1535 
1536 	if (path == NULL && fd != AT_FDCWD) {
1537 		file_t *fp;
1538 
1539 		/* fd_getvnode() will use the descriptor for us */
1540 		if ((error = fd_getvnode(fd, &fp)) != 0)
1541 			return error;
1542 		error = do_sys_utimensat(l, AT_FDCWD, fp->f_data, NULL, 0,
1543 		    tsp, UIO_SYSSPACE);
1544 		fd_putfile(fd);
1545 		return error;
1546 	}
1547 
1548 	return do_sys_utimensat(l, fd, NULL, path, follow, tsp, UIO_SYSSPACE);
1549 }
1550 
1551 int
1552 linux_sys_utimensat(struct lwp *l, const struct linux_sys_utimensat_args *uap,
1553 	register_t *retval)
1554 {
1555 	/* {
1556 		syscallarg(int) fd;
1557 		syscallarg(const char *) path;
1558 		syscallarg(const struct linux_timespec *) times;
1559 		syscallarg(int) flag;
1560 	} */
1561 	int error;
1562 	struct linux_timespec lts[2];
1563 	struct timespec *tsp = NULL, ts[2];
1564 
1565 	if (SCARG(uap, times)) {
1566 		error = copyin(SCARG(uap, times), &lts, sizeof(lts));
1567 		if (error != 0)
1568 			return error;
1569 		linux_to_native_timespec(&ts[0], &lts[0]);
1570 		linux_to_native_timespec(&ts[1], &lts[1]);
1571 		tsp = ts;
1572 	}
1573 
1574 	return linux_do_sys_utimensat(l, SCARG(uap, fd), SCARG(uap, path),
1575 	    tsp, SCARG(uap, flag), retval);
1576 }
1577 
1578 int
1579 linux_sys_futex(struct lwp *l, const struct linux_sys_futex_args *uap,
1580 	register_t *retval)
1581 {
1582 	/* {
1583 		syscallarg(int *) uaddr;
1584 		syscallarg(int) op;
1585 		syscallarg(int) val;
1586 		syscallarg(const struct linux_timespec *) timeout;
1587 		syscallarg(int *) uaddr2;
1588 		syscallarg(int) val3;
1589 	} */
1590 	struct linux_timespec lts;
1591 	struct timespec ts, *tsp = NULL;
1592 	int val2 = 0;
1593 	int error;
1594 
1595 	/*
1596 	 * Linux overlays the "timeout" field and the "val2" field.
1597 	 * "timeout" is only valid for FUTEX_WAIT and FUTEX_WAIT_BITSET
1598 	 * on Linux.
1599 	 */
1600 	const int op = (SCARG(uap, op) & FUTEX_CMD_MASK);
1601 	if ((op == FUTEX_WAIT || op == FUTEX_WAIT_BITSET) &&
1602 	    SCARG(uap, timeout) != NULL) {
1603 		if ((error = copyin(SCARG(uap, timeout),
1604 		    &lts, sizeof(lts))) != 0) {
1605 			return error;
1606 		}
1607 		linux_to_native_timespec(&ts, &lts);
1608 		tsp = &ts;
1609 	} else {
1610 		val2 = (int)(uintptr_t)SCARG(uap, timeout);
1611 	}
1612 
1613 	return linux_do_futex(SCARG(uap, uaddr), SCARG(uap, op),
1614 	    SCARG(uap, val), tsp, SCARG(uap, uaddr2), val2,
1615 	    SCARG(uap, val3), retval);
1616 }
1617 
1618 int
1619 linux_do_futex(int *uaddr, int op, int val, struct timespec *timeout,
1620     int *uaddr2, int val2, int val3, register_t *retval)
1621 {
1622 	/*
1623 	 * Always clear FUTEX_PRIVATE_FLAG for Linux processes.
1624 	 * NetBSD-native futexes exist in different namespace
1625 	 * depending on FUTEX_PRIVATE_FLAG.  This appears not
1626 	 * to be the case in Linux, and some futex users will
1627 	 * mix private and non-private ops on the same futex
1628 	 * object.
1629 	 */
1630 	return do_futex(uaddr, op & ~FUTEX_PRIVATE_FLAG,
1631 			val, timeout, uaddr2, val2, val3, retval);
1632 }
1633 
1634 #define	LINUX_EFD_SEMAPHORE	0x0001
1635 #define	LINUX_EFD_CLOEXEC	LINUX_O_CLOEXEC
1636 #define	LINUX_EFD_NONBLOCK	LINUX_O_NONBLOCK
1637 
1638 static int
1639 linux_do_eventfd2(struct lwp *l, unsigned int initval, int flags,
1640     register_t *retval)
1641 {
1642 	int nflags = 0;
1643 
1644 	if (flags & ~(LINUX_EFD_SEMAPHORE | LINUX_EFD_CLOEXEC |
1645 		      LINUX_EFD_NONBLOCK)) {
1646 		return EINVAL;
1647 	}
1648 	if (flags & LINUX_EFD_SEMAPHORE) {
1649 		nflags |= EFD_SEMAPHORE;
1650 	}
1651 	if (flags & LINUX_EFD_CLOEXEC) {
1652 		nflags |= EFD_CLOEXEC;
1653 	}
1654 	if (flags & LINUX_EFD_NONBLOCK) {
1655 		nflags |= EFD_NONBLOCK;
1656 	}
1657 
1658 	return do_eventfd(l, initval, nflags, retval);
1659 }
1660 
1661 int
1662 linux_sys_eventfd(struct lwp *l, const struct linux_sys_eventfd_args *uap,
1663     register_t *retval)
1664 {
1665 	/* {
1666 		syscallarg(unsigned int) initval;
1667 	} */
1668 
1669 	return linux_do_eventfd2(l, SCARG(uap, initval), 0, retval);
1670 }
1671 
1672 int
1673 linux_sys_eventfd2(struct lwp *l, const struct linux_sys_eventfd2_args *uap,
1674     register_t *retval)
1675 {
1676 	/* {
1677 		syscallarg(unsigned int) initval;
1678 		syscallarg(int) flags;
1679 	} */
1680 
1681 	return linux_do_eventfd2(l, SCARG(uap, initval), SCARG(uap, flags),
1682 				 retval);
1683 }
1684