xref: /freebsd/sys/i386/linux/linux_machdep.c (revision 8aac90f18aef7c9eea906c3ff9a001ca7b94f375)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2000 Marcel Moolenaar
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 "opt_posix.h"
30 
31 #include <sys/param.h>
32 #include <sys/imgact_aout.h>
33 #include <sys/fcntl.h>
34 #include <sys/lock.h>
35 #include <sys/malloc.h>
36 #include <sys/mman.h>
37 #include <sys/mutex.h>
38 #include <sys/namei.h>
39 #include <sys/priv.h>
40 #include <sys/proc.h>
41 #include <sys/racct.h>
42 #include <sys/resource.h>
43 #include <sys/resourcevar.h>
44 #include <sys/syscallsubr.h>
45 #include <sys/sysproto.h>
46 #include <sys/vnode.h>
47 
48 #include <security/audit/audit.h>
49 #include <security/mac/mac_framework.h>
50 
51 #include <machine/frame.h>
52 #include <machine/pcb.h>			/* needed for pcb definition in linux_set_thread_area */
53 #include <machine/psl.h>
54 #include <machine/segments.h>
55 #include <machine/sysarch.h>
56 
57 #include <vm/pmap.h>
58 #include <vm/vm.h>
59 #include <vm/vm_extern.h>
60 #include <vm/vm_kern.h>
61 #include <vm/vm_map.h>
62 #include <vm/vm_param.h>
63 
64 #include <x86/reg.h>
65 
66 #include <i386/linux/linux.h>
67 #include <i386/linux/linux_proto.h>
68 #include <compat/linux/linux_emul.h>
69 #include <compat/linux/linux_fork.h>
70 #include <compat/linux/linux_ipc.h>
71 #include <compat/linux/linux_misc.h>
72 #include <compat/linux/linux_mmap.h>
73 #include <compat/linux/linux_signal.h>
74 #include <compat/linux/linux_util.h>
75 
76 
77 struct l_descriptor {
78 	l_uint		entry_number;
79 	l_ulong		base_addr;
80 	l_uint		limit;
81 	l_uint		seg_32bit:1;
82 	l_uint		contents:2;
83 	l_uint		read_exec_only:1;
84 	l_uint		limit_in_pages:1;
85 	l_uint		seg_not_present:1;
86 	l_uint		useable:1;
87 };
88 
89 struct l_old_select_argv {
90 	l_int		nfds;
91 	l_fd_set	*readfds;
92 	l_fd_set	*writefds;
93 	l_fd_set	*exceptfds;
94 	struct l_timeval	*timeout;
95 };
96 
97 struct l_ipc_kludge {
98 	struct l_msgbuf *msgp;
99 	l_long msgtyp;
100 };
101 
102 int
103 linux_ipc(struct thread *td, struct linux_ipc_args *args)
104 {
105 
106 	switch (args->what & 0xFFFF) {
107 	case LINUX_SEMOP: {
108 
109 		return (kern_semop(td, args->arg1, PTRIN(args->ptr),
110 		    args->arg2, NULL));
111 	}
112 	case LINUX_SEMGET: {
113 		struct linux_semget_args a;
114 
115 		a.key = args->arg1;
116 		a.nsems = args->arg2;
117 		a.semflg = args->arg3;
118 		return (linux_semget(td, &a));
119 	}
120 	case LINUX_SEMCTL: {
121 		struct linux_semctl_args a;
122 		int error;
123 
124 		a.semid = args->arg1;
125 		a.semnum = args->arg2;
126 		a.cmd = args->arg3;
127 		error = copyin(PTRIN(args->ptr), &a.arg, sizeof(a.arg));
128 		if (error)
129 			return (error);
130 		return (linux_semctl(td, &a));
131 	}
132 	case LINUX_SEMTIMEDOP: {
133 		struct linux_semtimedop_args a;
134 
135 		a.semid = args->arg1;
136 		a.tsops = PTRIN(args->ptr);
137 		a.nsops = args->arg2;
138 		a.timeout = PTRIN(args->arg5);
139 		return (linux_semtimedop(td, &a));
140 	}
141 	case LINUX_MSGSND: {
142 		struct linux_msgsnd_args a;
143 
144 		a.msqid = args->arg1;
145 		a.msgp = PTRIN(args->ptr);
146 		a.msgsz = args->arg2;
147 		a.msgflg = args->arg3;
148 		return (linux_msgsnd(td, &a));
149 	}
150 	case LINUX_MSGRCV: {
151 		struct linux_msgrcv_args a;
152 
153 		a.msqid = args->arg1;
154 		a.msgsz = args->arg2;
155 		a.msgflg = args->arg3;
156 		if ((args->what >> 16) == 0) {
157 			struct l_ipc_kludge tmp;
158 			int error;
159 
160 			if (args->ptr == 0)
161 				return (EINVAL);
162 			error = copyin(PTRIN(args->ptr), &tmp, sizeof(tmp));
163 			if (error)
164 				return (error);
165 			a.msgp = PTRIN(tmp.msgp);
166 			a.msgtyp = tmp.msgtyp;
167 		} else {
168 			a.msgp = PTRIN(args->ptr);
169 			a.msgtyp = args->arg5;
170 		}
171 		return (linux_msgrcv(td, &a));
172 	}
173 	case LINUX_MSGGET: {
174 		struct linux_msgget_args a;
175 
176 		a.key = args->arg1;
177 		a.msgflg = args->arg2;
178 		return (linux_msgget(td, &a));
179 	}
180 	case LINUX_MSGCTL: {
181 		struct linux_msgctl_args a;
182 
183 		a.msqid = args->arg1;
184 		a.cmd = args->arg2;
185 		a.buf = PTRIN(args->ptr);
186 		return (linux_msgctl(td, &a));
187 	}
188 	case LINUX_SHMAT: {
189 		struct linux_shmat_args a;
190 		l_uintptr_t addr;
191 		int error;
192 
193 		a.shmid = args->arg1;
194 		a.shmaddr = PTRIN(args->ptr);
195 		a.shmflg = args->arg2;
196 		error = linux_shmat(td, &a);
197 		if (error != 0)
198 			return (error);
199 		addr = td->td_retval[0];
200 		error = copyout(&addr, PTRIN(args->arg3), sizeof(addr));
201 		td->td_retval[0] = 0;
202 		return (error);
203 	}
204 	case LINUX_SHMDT: {
205 		struct linux_shmdt_args a;
206 
207 		a.shmaddr = PTRIN(args->ptr);
208 		return (linux_shmdt(td, &a));
209 	}
210 	case LINUX_SHMGET: {
211 		struct linux_shmget_args a;
212 
213 		a.key = args->arg1;
214 		a.size = args->arg2;
215 		a.shmflg = args->arg3;
216 		return (linux_shmget(td, &a));
217 	}
218 	case LINUX_SHMCTL: {
219 		struct linux_shmctl_args a;
220 
221 		a.shmid = args->arg1;
222 		a.cmd = args->arg2;
223 		a.buf = PTRIN(args->ptr);
224 		return (linux_shmctl(td, &a));
225 	}
226 	default:
227 		break;
228 	}
229 
230 	return (EINVAL);
231 }
232 
233 int
234 linux_old_select(struct thread *td, struct linux_old_select_args *args)
235 {
236 	struct l_old_select_argv linux_args;
237 	struct linux_select_args newsel;
238 	int error;
239 
240 	error = copyin(args->ptr, &linux_args, sizeof(linux_args));
241 	if (error)
242 		return (error);
243 
244 	newsel.nfds = linux_args.nfds;
245 	newsel.readfds = linux_args.readfds;
246 	newsel.writefds = linux_args.writefds;
247 	newsel.exceptfds = linux_args.exceptfds;
248 	newsel.timeout = linux_args.timeout;
249 	return (linux_select(td, &newsel));
250 }
251 
252 int
253 linux_set_cloned_tls(struct thread *td, void *desc)
254 {
255 	struct segment_descriptor sd;
256 	struct l_user_desc info;
257 	int idx, error;
258 	int a[2];
259 
260 	error = copyin(desc, &info, sizeof(struct l_user_desc));
261 	if (error) {
262 		linux_msg(td, "set_cloned_tls copyin failed!");
263 	} else {
264 		idx = info.entry_number;
265 
266 		/*
267 		 * looks like we're getting the idx we returned
268 		 * in the set_thread_area() syscall
269 		 */
270 		if (idx != 6 && idx != 3) {
271 			linux_msg(td, "set_cloned_tls resetting idx!");
272 			idx = 3;
273 		}
274 
275 		/* this doesnt happen in practice */
276 		if (idx == 6) {
277 			/* we might copy out the entry_number as 3 */
278 			info.entry_number = 3;
279 			error = copyout(&info, desc, sizeof(struct l_user_desc));
280 			if (error)
281 				linux_msg(td, "set_cloned_tls copyout failed!");
282 		}
283 
284 		a[0] = LINUX_LDT_entry_a(&info);
285 		a[1] = LINUX_LDT_entry_b(&info);
286 
287 		memcpy(&sd, &a, sizeof(a));
288 		/* set %gs */
289 		td->td_pcb->pcb_gsd = sd;
290 		td->td_pcb->pcb_gs = GSEL(GUGS_SEL, SEL_UPL);
291 	}
292 
293 	return (error);
294 }
295 
296 int
297 linux_set_upcall(struct thread *td, register_t stack)
298 {
299 
300 	if (stack)
301 		td->td_frame->tf_esp = stack;
302 
303 	/*
304 	 * The newly created Linux thread returns
305 	 * to the user space by the same path that a parent do.
306 	 */
307 	td->td_frame->tf_eax = 0;
308 	return (0);
309 }
310 
311 int
312 linux_mmap(struct thread *td, struct linux_mmap_args *args)
313 {
314 	int error;
315 	struct l_mmap_argv linux_args;
316 
317 	error = copyin(args->ptr, &linux_args, sizeof(linux_args));
318 	if (error)
319 		return (error);
320 
321 	return (linux_mmap_common(td, linux_args.addr, linux_args.len,
322 	    linux_args.prot, linux_args.flags, linux_args.fd,
323 	    (uint32_t)linux_args.pgoff));
324 }
325 
326 int
327 linux_ioperm(struct thread *td, struct linux_ioperm_args *args)
328 {
329 	int error;
330 	struct i386_ioperm_args iia;
331 
332 	iia.start = args->start;
333 	iia.length = args->length;
334 	iia.enable = args->enable;
335 	error = i386_set_ioperm(td, &iia);
336 	return (error);
337 }
338 
339 int
340 linux_iopl(struct thread *td, struct linux_iopl_args *args)
341 {
342 	int error;
343 
344 	if (args->level < 0 || args->level > 3)
345 		return (EINVAL);
346 	if ((error = priv_check(td, PRIV_IO)) != 0)
347 		return (error);
348 	if ((error = securelevel_gt(td->td_ucred, 0)) != 0)
349 		return (error);
350 	td->td_frame->tf_eflags = (td->td_frame->tf_eflags & ~PSL_IOPL) |
351 	    (args->level * (PSL_IOPL / 3));
352 	return (0);
353 }
354 
355 int
356 linux_modify_ldt(struct thread *td, struct linux_modify_ldt_args *uap)
357 {
358 	int error;
359 	struct i386_ldt_args ldt;
360 	struct l_descriptor ld;
361 	union descriptor desc;
362 	int size, written;
363 
364 	switch (uap->func) {
365 	case 0x00: /* read_ldt */
366 		ldt.start = 0;
367 		ldt.descs = uap->ptr;
368 		ldt.num = uap->bytecount / sizeof(union descriptor);
369 		error = i386_get_ldt(td, &ldt);
370 		td->td_retval[0] *= sizeof(union descriptor);
371 		break;
372 	case 0x02: /* read_default_ldt = 0 */
373 		size = 5*sizeof(struct l_desc_struct);
374 		if (size > uap->bytecount)
375 			size = uap->bytecount;
376 		for (written = error = 0; written < size && error == 0; written++)
377 			error = subyte((char *)uap->ptr + written, 0);
378 		td->td_retval[0] = written;
379 		break;
380 	case 0x01: /* write_ldt */
381 	case 0x11: /* write_ldt */
382 		if (uap->bytecount != sizeof(ld))
383 			return (EINVAL);
384 
385 		error = copyin(uap->ptr, &ld, sizeof(ld));
386 		if (error)
387 			return (error);
388 
389 		ldt.start = ld.entry_number;
390 		ldt.descs = &desc;
391 		ldt.num = 1;
392 		desc.sd.sd_lolimit = (ld.limit & 0x0000ffff);
393 		desc.sd.sd_hilimit = (ld.limit & 0x000f0000) >> 16;
394 		desc.sd.sd_lobase = (ld.base_addr & 0x00ffffff);
395 		desc.sd.sd_hibase = (ld.base_addr & 0xff000000) >> 24;
396 		desc.sd.sd_type = SDT_MEMRO | ((ld.read_exec_only ^ 1) << 1) |
397 			(ld.contents << 2);
398 		desc.sd.sd_dpl = 3;
399 		desc.sd.sd_p = (ld.seg_not_present ^ 1);
400 		desc.sd.sd_xx = 0;
401 		desc.sd.sd_def32 = ld.seg_32bit;
402 		desc.sd.sd_gran = ld.limit_in_pages;
403 		error = i386_set_ldt(td, &ldt, &desc);
404 		break;
405 	default:
406 		error = ENOSYS;
407 		break;
408 	}
409 
410 	if (error == EOPNOTSUPP) {
411 		linux_msg(td, "modify_ldt needs kernel option USER_LDT");
412 		error = ENOSYS;
413 	}
414 
415 	return (error);
416 }
417 
418 int
419 linux_sigaction(struct thread *td, struct linux_sigaction_args *args)
420 {
421 	l_osigaction_t osa;
422 	l_sigaction_t act, oact;
423 	int error;
424 
425 	if (args->nsa != NULL) {
426 		error = copyin(args->nsa, &osa, sizeof(l_osigaction_t));
427 		if (error)
428 			return (error);
429 		act.lsa_handler = osa.lsa_handler;
430 		act.lsa_flags = osa.lsa_flags;
431 		act.lsa_restorer = osa.lsa_restorer;
432 		LINUX_SIGEMPTYSET(act.lsa_mask);
433 		act.lsa_mask.__mask = osa.lsa_mask;
434 	}
435 
436 	error = linux_do_sigaction(td, args->sig, args->nsa ? &act : NULL,
437 	    args->osa ? &oact : NULL);
438 
439 	if (args->osa != NULL && !error) {
440 		osa.lsa_handler = oact.lsa_handler;
441 		osa.lsa_flags = oact.lsa_flags;
442 		osa.lsa_restorer = oact.lsa_restorer;
443 		osa.lsa_mask = oact.lsa_mask.__mask;
444 		error = copyout(&osa, args->osa, sizeof(l_osigaction_t));
445 	}
446 
447 	return (error);
448 }
449 
450 /*
451  * Linux has two extra args, restart and oldmask.  We dont use these,
452  * but it seems that "restart" is actually a context pointer that
453  * enables the signal to happen with a different register set.
454  */
455 int
456 linux_sigsuspend(struct thread *td, struct linux_sigsuspend_args *args)
457 {
458 	sigset_t sigmask;
459 	l_sigset_t mask;
460 
461 	LINUX_SIGEMPTYSET(mask);
462 	mask.__mask = args->mask;
463 	linux_to_bsd_sigset(&mask, &sigmask);
464 	return (kern_sigsuspend(td, sigmask));
465 }
466 
467 int
468 linux_pause(struct thread *td, struct linux_pause_args *args)
469 {
470 	struct proc *p = td->td_proc;
471 	sigset_t sigmask;
472 
473 	PROC_LOCK(p);
474 	sigmask = td->td_sigmask;
475 	PROC_UNLOCK(p);
476 	return (kern_sigsuspend(td, sigmask));
477 }
478 
479 int
480 linux_set_thread_area(struct thread *td, struct linux_set_thread_area_args *args)
481 {
482 	struct l_user_desc info;
483 	int error;
484 	int idx;
485 	int a[2];
486 	struct segment_descriptor sd;
487 
488 	error = copyin(args->desc, &info, sizeof(struct l_user_desc));
489 	if (error)
490 		return (error);
491 
492 	idx = info.entry_number;
493 	/*
494 	 * Semantics of Linux version: every thread in the system has array of
495 	 * 3 tls descriptors. 1st is GLIBC TLS, 2nd is WINE, 3rd unknown. This
496 	 * syscall loads one of the selected tls descriptors with a value and
497 	 * also loads GDT descriptors 6, 7 and 8 with the content of the
498 	 * per-thread descriptors.
499 	 *
500 	 * Semantics of FreeBSD version: I think we can ignore that Linux has 3
501 	 * per-thread descriptors and use just the 1st one. The tls_array[]
502 	 * is used only in set/get-thread_area() syscalls and for loading the
503 	 * GDT descriptors. In FreeBSD we use just one GDT descriptor for TLS
504 	 * so we will load just one.
505 	 *
506 	 * XXX: this doesn't work when a user space process tries to use more
507 	 * than 1 TLS segment. Comment in the Linux sources says wine might do
508 	 * this.
509 	 */
510 
511 	/*
512 	 * we support just GLIBC TLS now
513 	 * we should let 3 proceed as well because we use this segment so
514 	 * if code does two subsequent calls it should succeed
515 	 */
516 	if (idx != 6 && idx != -1 && idx != 3)
517 		return (EINVAL);
518 
519 	/*
520 	 * we have to copy out the GDT entry we use
521 	 * FreeBSD uses GDT entry #3 for storing %gs so load that
522 	 *
523 	 * XXX: what if a user space program doesn't check this value and tries
524 	 * to use 6, 7 or 8?
525 	 */
526 	idx = info.entry_number = 3;
527 	error = copyout(&info, args->desc, sizeof(struct l_user_desc));
528 	if (error)
529 		return (error);
530 
531 	if (LINUX_LDT_empty(&info)) {
532 		a[0] = 0;
533 		a[1] = 0;
534 	} else {
535 		a[0] = LINUX_LDT_entry_a(&info);
536 		a[1] = LINUX_LDT_entry_b(&info);
537 	}
538 
539 	memcpy(&sd, &a, sizeof(a));
540 	/* this is taken from i386 version of cpu_set_user_tls() */
541 	critical_enter();
542 	/* set %gs */
543 	td->td_pcb->pcb_gsd = sd;
544 	PCPU_GET(fsgs_gdt)[1] = sd;
545 	load_gs(GSEL(GUGS_SEL, SEL_UPL));
546 	critical_exit();
547 
548 	return (0);
549 }
550 
551 int
552 linux_get_thread_area(struct thread *td, struct linux_get_thread_area_args *args)
553 {
554 
555 	struct l_user_desc info;
556 	int error;
557 	int idx;
558 	struct l_desc_struct desc;
559 	struct segment_descriptor sd;
560 
561 	error = copyin(args->desc, &info, sizeof(struct l_user_desc));
562 	if (error)
563 		return (error);
564 
565 	idx = info.entry_number;
566 	/* XXX: I am not sure if we want 3 to be allowed too. */
567 	if (idx != 6 && idx != 3)
568 		return (EINVAL);
569 
570 	idx = 3;
571 
572 	memset(&info, 0, sizeof(info));
573 
574 	sd = PCPU_GET(fsgs_gdt)[1];
575 
576 	memcpy(&desc, &sd, sizeof(desc));
577 
578 	info.entry_number = idx;
579 	info.base_addr = LINUX_GET_BASE(&desc);
580 	info.limit = LINUX_GET_LIMIT(&desc);
581 	info.seg_32bit = LINUX_GET_32BIT(&desc);
582 	info.contents = LINUX_GET_CONTENTS(&desc);
583 	info.read_exec_only = !LINUX_GET_WRITABLE(&desc);
584 	info.limit_in_pages = LINUX_GET_LIMIT_PAGES(&desc);
585 	info.seg_not_present = !LINUX_GET_PRESENT(&desc);
586 	info.useable = LINUX_GET_USEABLE(&desc);
587 
588 	error = copyout(&info, args->desc, sizeof(struct l_user_desc));
589 	if (error)
590 		return (EFAULT);
591 
592 	return (0);
593 }
594 
595 /* XXX: this wont work with module - convert it */
596 int
597 linux_mq_open(struct thread *td, struct linux_mq_open_args *args)
598 {
599 #ifdef P1003_1B_MQUEUE
600 	return (sys_kmq_open(td, (struct kmq_open_args *)args));
601 #else
602 	return (ENOSYS);
603 #endif
604 }
605 
606 int
607 linux_mq_unlink(struct thread *td, struct linux_mq_unlink_args *args)
608 {
609 #ifdef P1003_1B_MQUEUE
610 	return (sys_kmq_unlink(td, (struct kmq_unlink_args *)args));
611 #else
612 	return (ENOSYS);
613 #endif
614 }
615 
616 int
617 linux_mq_timedsend(struct thread *td, struct linux_mq_timedsend_args *args)
618 {
619 #ifdef P1003_1B_MQUEUE
620 	return (sys_kmq_timedsend(td, (struct kmq_timedsend_args *)args));
621 #else
622 	return (ENOSYS);
623 #endif
624 }
625 
626 int
627 linux_mq_timedreceive(struct thread *td, struct linux_mq_timedreceive_args *args)
628 {
629 #ifdef P1003_1B_MQUEUE
630 	return (sys_kmq_timedreceive(td, (struct kmq_timedreceive_args *)args));
631 #else
632 	return (ENOSYS);
633 #endif
634 }
635 
636 int
637 linux_mq_notify(struct thread *td, struct linux_mq_notify_args *args)
638 {
639 #ifdef P1003_1B_MQUEUE
640 	return (sys_kmq_notify(td, (struct kmq_notify_args *)args));
641 #else
642 	return (ENOSYS);
643 #endif
644 }
645 
646 int
647 linux_mq_getsetattr(struct thread *td, struct linux_mq_getsetattr_args *args)
648 {
649 #ifdef P1003_1B_MQUEUE
650 	return (sys_kmq_setattr(td, (struct kmq_setattr_args *)args));
651 #else
652 	return (ENOSYS);
653 #endif
654 }
655 
656 void
657 bsd_to_linux_regset(const struct reg *b_reg,
658     struct linux_pt_regset *l_regset)
659 {
660 
661 	l_regset->ebx = b_reg->r_ebx;
662 	l_regset->ecx = b_reg->r_ecx;
663 	l_regset->edx = b_reg->r_edx;
664 	l_regset->esi = b_reg->r_esi;
665 	l_regset->edi = b_reg->r_edi;
666 	l_regset->ebp = b_reg->r_ebp;
667 	l_regset->eax = b_reg->r_eax;
668 	l_regset->ds = b_reg->r_ds;
669 	l_regset->es = b_reg->r_es;
670 	l_regset->fs = b_reg->r_fs;
671 	l_regset->gs = b_reg->r_gs;
672 	l_regset->orig_eax = b_reg->r_eax;
673 	l_regset->eip = b_reg->r_eip;
674 	l_regset->cs = b_reg->r_cs;
675 	l_regset->eflags = b_reg->r_eflags;
676 	l_regset->esp = b_reg->r_esp;
677 	l_regset->ss = b_reg->r_ss;
678 }
679 
680 int
681 linux_uselib(struct thread *td, struct linux_uselib_args *args)
682 {
683 	struct nameidata ni;
684 	struct vnode *vp;
685 	struct exec *a_out;
686 	vm_map_t map;
687 	vm_map_entry_t entry;
688 	struct vattr attr;
689 	vm_offset_t vmaddr;
690 	unsigned long file_offset;
691 	unsigned long bss_size;
692 	ssize_t aresid;
693 	int error;
694 	bool locked, opened, textset;
695 
696 	a_out = NULL;
697 	vp = NULL;
698 	locked = false;
699 	textset = false;
700 	opened = false;
701 
702 	NDINIT(&ni, LOOKUP, ISOPEN | FOLLOW | LOCKLEAF | AUDITVNODE1,
703 	    UIO_USERSPACE, args->library);
704 	error = namei(&ni);
705 	if (error)
706 		goto cleanup;
707 
708 	vp = ni.ni_vp;
709 	NDFREE_PNBUF(&ni);
710 
711 	/*
712 	 * From here on down, we have a locked vnode that must be unlocked.
713 	 * XXX: The code below largely duplicates exec_check_permissions().
714 	 */
715 	locked = true;
716 
717 	/* Executable? */
718 	error = VOP_GETATTR(vp, &attr, td->td_ucred);
719 	if (error)
720 		goto cleanup;
721 
722 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
723 	    ((attr.va_mode & 0111) == 0) || (attr.va_type != VREG)) {
724 		/* EACCESS is what exec(2) returns. */
725 		error = ENOEXEC;
726 		goto cleanup;
727 	}
728 
729 	/* Sensible size? */
730 	if (attr.va_size == 0) {
731 		error = ENOEXEC;
732 		goto cleanup;
733 	}
734 
735 	/* Can we access it? */
736 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
737 	if (error)
738 		goto cleanup;
739 
740 	/*
741 	 * XXX: This should use vn_open() so that it is properly authorized,
742 	 * and to reduce code redundancy all over the place here.
743 	 * XXX: Not really, it duplicates far more of exec_check_permissions()
744 	 * than vn_open().
745 	 */
746 #ifdef MAC
747 	error = mac_vnode_check_open(td->td_ucred, vp, VREAD);
748 	if (error)
749 		goto cleanup;
750 #endif
751 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
752 	if (error)
753 		goto cleanup;
754 	opened = true;
755 
756 	/* Pull in executable header into exec_map */
757 	error = vm_mmap(exec_map, (vm_offset_t *)&a_out, PAGE_SIZE,
758 	    VM_PROT_READ, VM_PROT_READ, 0, OBJT_VNODE, vp, 0);
759 	if (error)
760 		goto cleanup;
761 
762 	/* Is it a Linux binary ? */
763 	if (((a_out->a_magic >> 16) & 0xff) != 0x64) {
764 		error = ENOEXEC;
765 		goto cleanup;
766 	}
767 
768 	/*
769 	 * While we are here, we should REALLY do some more checks
770 	 */
771 
772 	/* Set file/virtual offset based on a.out variant. */
773 	switch ((int)(a_out->a_magic & 0xffff)) {
774 	case 0413:			/* ZMAGIC */
775 		file_offset = 1024;
776 		break;
777 	case 0314:			/* QMAGIC */
778 		file_offset = 0;
779 		break;
780 	default:
781 		error = ENOEXEC;
782 		goto cleanup;
783 	}
784 
785 	bss_size = round_page(a_out->a_bss);
786 
787 	/* Check various fields in header for validity/bounds. */
788 	if (a_out->a_text & PAGE_MASK || a_out->a_data & PAGE_MASK) {
789 		error = ENOEXEC;
790 		goto cleanup;
791 	}
792 
793 	/* text + data can't exceed file size */
794 	if (a_out->a_data + a_out->a_text > attr.va_size) {
795 		error = EFAULT;
796 		goto cleanup;
797 	}
798 
799 	/*
800 	 * text/data/bss must not exceed limits
801 	 * XXX - this is not complete. it should check current usage PLUS
802 	 * the resources needed by this library.
803 	 */
804 	PROC_LOCK(td->td_proc);
805 	if (a_out->a_text > maxtsiz ||
806 	    a_out->a_data + bss_size > lim_cur_proc(td->td_proc, RLIMIT_DATA) ||
807 	    racct_set(td->td_proc, RACCT_DATA, a_out->a_data +
808 	    bss_size) != 0) {
809 		PROC_UNLOCK(td->td_proc);
810 		error = ENOMEM;
811 		goto cleanup;
812 	}
813 	PROC_UNLOCK(td->td_proc);
814 
815 	/*
816 	 * Prevent more writers.
817 	 */
818 	error = VOP_SET_TEXT(vp);
819 	if (error != 0)
820 		goto cleanup;
821 	textset = true;
822 
823 	/*
824 	 * Lock no longer needed
825 	 */
826 	locked = false;
827 	VOP_UNLOCK(vp);
828 
829 	/*
830 	 * Check if file_offset page aligned. Currently we cannot handle
831 	 * misalinged file offsets, and so we read in the entire image
832 	 * (what a waste).
833 	 */
834 	if (file_offset & PAGE_MASK) {
835 		/* Map text+data read/write/execute */
836 
837 		/* a_entry is the load address and is page aligned */
838 		vmaddr = trunc_page(a_out->a_entry);
839 
840 		/* get anon user mapping, read+write+execute */
841 		error = vm_map_find(&td->td_proc->p_vmspace->vm_map, NULL, 0,
842 		    &vmaddr, a_out->a_text + a_out->a_data, 0, VMFS_NO_SPACE,
843 		    VM_PROT_ALL, VM_PROT_ALL, 0);
844 		if (error)
845 			goto cleanup;
846 
847 		error = vn_rdwr(UIO_READ, vp, (void *)vmaddr, file_offset,
848 		    a_out->a_text + a_out->a_data, UIO_USERSPACE, 0,
849 		    td->td_ucred, NOCRED, &aresid, td);
850 		if (error != 0)
851 			goto cleanup;
852 		if (aresid != 0) {
853 			error = ENOEXEC;
854 			goto cleanup;
855 		}
856 	} else {
857 		/*
858 		 * for QMAGIC, a_entry is 20 bytes beyond the load address
859 		 * to skip the executable header
860 		 */
861 		vmaddr = trunc_page(a_out->a_entry);
862 
863 		/*
864 		 * Map it all into the process's space as a single
865 		 * copy-on-write "data" segment.
866 		 */
867 		map = &td->td_proc->p_vmspace->vm_map;
868 		error = vm_mmap(map, &vmaddr,
869 		    a_out->a_text + a_out->a_data, VM_PROT_ALL, VM_PROT_ALL,
870 		    MAP_PRIVATE | MAP_FIXED, OBJT_VNODE, vp, file_offset);
871 		if (error)
872 			goto cleanup;
873 		vm_map_lock(map);
874 		if (!vm_map_lookup_entry(map, vmaddr, &entry)) {
875 			vm_map_unlock(map);
876 			error = EDOOFUS;
877 			goto cleanup;
878 		}
879 		entry->eflags |= MAP_ENTRY_VN_EXEC;
880 		vm_map_unlock(map);
881 		textset = false;
882 	}
883 
884 	if (bss_size != 0) {
885 		/* Calculate BSS start address */
886 		vmaddr = trunc_page(a_out->a_entry) + a_out->a_text +
887 		    a_out->a_data;
888 
889 		/* allocate some 'anon' space */
890 		error = vm_map_find(&td->td_proc->p_vmspace->vm_map, NULL, 0,
891 		    &vmaddr, bss_size, 0, VMFS_NO_SPACE, VM_PROT_ALL,
892 		    VM_PROT_ALL, 0);
893 		if (error)
894 			goto cleanup;
895 	}
896 
897 cleanup:
898 	if (opened) {
899 		if (locked)
900 			VOP_UNLOCK(vp);
901 		locked = false;
902 		VOP_CLOSE(vp, FREAD, td->td_ucred, td);
903 	}
904 	if (textset) {
905 		if (!locked) {
906 			locked = true;
907 			VOP_LOCK(vp, LK_SHARED | LK_RETRY);
908 		}
909 		VOP_UNSET_TEXT_CHECKED(vp);
910 	}
911 	if (locked)
912 		VOP_UNLOCK(vp);
913 
914 	/* Release the temporary mapping. */
915 	if (a_out)
916 		kmap_free_wakeup(exec_map, (vm_offset_t)a_out, PAGE_SIZE);
917 
918 	return (error);
919 }
920