xref: /netbsd-src/sys/kern/kern_exec.c (revision c0179c282a5968435315a82f4128c61372c68fc3)
1 /*	$NetBSD: kern_exec.c,v 1.232 2006/11/22 02:02:51 elad Exp $	*/
2 
3 /*-
4  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
5  * Copyright (C) 1992 Wolfgang Solfrank.
6  * Copyright (C) 1992 TooLs GmbH.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by TooLs GmbH.
20  * 4. The name of TooLs GmbH may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
29  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
30  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
31  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
32  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.232 2006/11/22 02:02:51 elad Exp $");
37 
38 #include "opt_ktrace.h"
39 #include "opt_syscall_debug.h"
40 #include "opt_compat_netbsd.h"
41 #include "veriexec.h"
42 #include "opt_pax.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/filedesc.h>
47 #include <sys/kernel.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/malloc.h>
51 #include <sys/namei.h>
52 #include <sys/vnode.h>
53 #include <sys/file.h>
54 #include <sys/acct.h>
55 #include <sys/exec.h>
56 #include <sys/ktrace.h>
57 #include <sys/resourcevar.h>
58 #include <sys/wait.h>
59 #include <sys/mman.h>
60 #include <sys/ras.h>
61 #include <sys/signalvar.h>
62 #include <sys/stat.h>
63 #include <sys/syscall.h>
64 #include <sys/kauth.h>
65 
66 #include <sys/sa.h>
67 #include <sys/savar.h>
68 #include <sys/syscallargs.h>
69 #if NVERIEXEC > 0
70 #include <sys/verified_exec.h>
71 #endif /* NVERIEXEC > 0 */
72 
73 #ifdef SYSTRACE
74 #include <sys/systrace.h>
75 #endif /* SYSTRACE */
76 
77 #ifdef PAX_SEGVGUARD
78 #include <sys/pax.h>
79 #endif /* PAX_SEGVGUARD */
80 
81 #include <uvm/uvm_extern.h>
82 
83 #include <machine/cpu.h>
84 #include <machine/reg.h>
85 
86 static int exec_sigcode_map(struct proc *, const struct emul *);
87 
88 #ifdef DEBUG_EXEC
89 #define DPRINTF(a) uprintf a
90 #else
91 #define DPRINTF(a)
92 #endif /* DEBUG_EXEC */
93 
94 MALLOC_DEFINE(M_EXEC, "exec", "argument lists & other mem used by exec");
95 
96 /*
97  * Exec function switch:
98  *
99  * Note that each makecmds function is responsible for loading the
100  * exec package with the necessary functions for any exec-type-specific
101  * handling.
102  *
103  * Functions for specific exec types should be defined in their own
104  * header file.
105  */
106 extern const struct execsw	execsw_builtin[];
107 extern int			nexecs_builtin;
108 static const struct execsw	**execsw = NULL;
109 static int			nexecs;
110 
111 u_int	exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
112 
113 #ifdef LKM
114 /* list of supported emulations */
115 static
116 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
117 struct emul_entry {
118 	LIST_ENTRY(emul_entry)	el_list;
119 	const struct emul	*el_emul;
120 	int			ro_entry;
121 };
122 
123 /* list of dynamically loaded execsw entries */
124 static
125 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
126 struct exec_entry {
127 	LIST_ENTRY(exec_entry)	ex_list;
128 	const struct execsw	*es;
129 };
130 
131 /* structure used for building execw[] */
132 struct execsw_entry {
133 	struct execsw_entry	*next;
134 	const struct execsw	*es;
135 };
136 #endif /* LKM */
137 
138 #ifdef SYSCALL_DEBUG
139 extern const char * const syscallnames[];
140 #endif
141 
142 #ifdef COMPAT_16
143 extern char	sigcode[], esigcode[];
144 struct uvm_object *emul_netbsd_object;
145 #endif
146 
147 #ifndef __HAVE_SYSCALL_INTERN
148 void	syscall(void);
149 #endif
150 
151 static const struct sa_emul saemul_netbsd = {
152 	sizeof(ucontext_t),
153 	sizeof(struct sa_t),
154 	sizeof(struct sa_t *),
155 	NULL,
156 	NULL,
157 	cpu_upcall,
158 	(void (*)(struct lwp *, void *))getucontext,
159 	sa_ucsp
160 };
161 
162 /* NetBSD emul struct */
163 const struct emul emul_netbsd = {
164 	"netbsd",
165 	NULL,		/* emulation path */
166 #ifndef __HAVE_MINIMAL_EMUL
167 	EMUL_HAS_SYS___syscall,
168 	NULL,
169 	SYS_syscall,
170 	SYS_NSYSENT,
171 #endif
172 	sysent,
173 #ifdef SYSCALL_DEBUG
174 	syscallnames,
175 #else
176 	NULL,
177 #endif
178 	sendsig,
179 	trapsignal,
180 	NULL,
181 #ifdef COMPAT_16
182 	sigcode,
183 	esigcode,
184 	&emul_netbsd_object,
185 #else
186 	NULL,
187 	NULL,
188 	NULL,
189 #endif
190 	setregs,
191 	NULL,
192 	NULL,
193 	NULL,
194 	NULL,
195 	NULL,
196 #ifdef __HAVE_SYSCALL_INTERN
197 	syscall_intern,
198 #else
199 	syscall,
200 #endif
201 	NULL,
202 	NULL,
203 
204 	uvm_default_mapaddr,
205 	NULL,
206 	&saemul_netbsd,
207 };
208 
209 #ifdef LKM
210 /*
211  * Exec lock. Used to control access to execsw[] structures.
212  * This must not be static so that netbsd32 can access it, too.
213  */
214 struct lock exec_lock;
215 
216 static void link_es(struct execsw_entry **, const struct execsw *);
217 #endif /* LKM */
218 
219 /*
220  * check exec:
221  * given an "executable" described in the exec package's namei info,
222  * see what we can do with it.
223  *
224  * ON ENTRY:
225  *	exec package with appropriate namei info
226  *	lwp pointer of exec'ing lwp
227  *      if verified exec enabled then flag indicating a direct exec or
228  *        an indirect exec (i.e. for a shell script interpreter)
229  *	NO SELF-LOCKED VNODES
230  *
231  * ON EXIT:
232  *	error:	nothing held, etc.  exec header still allocated.
233  *	ok:	filled exec package, executable's vnode (unlocked).
234  *
235  * EXEC SWITCH ENTRY:
236  * 	Locked vnode to check, exec package, proc.
237  *
238  * EXEC SWITCH EXIT:
239  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
240  *	error:	destructive:
241  *			everything deallocated execept exec header.
242  *		non-destructive:
243  *			error code, executable's vnode (unlocked),
244  *			exec header unmodified.
245  */
246 int
247 /*ARGSUSED*/
248 check_exec(struct lwp *l, struct exec_package *epp, int flag)
249 {
250 	int		error, i;
251 	struct vnode	*vp;
252 	struct nameidata *ndp;
253 	size_t		resid;
254 
255 	ndp = epp->ep_ndp;
256 	ndp->ni_cnd.cn_nameiop = LOOKUP;
257 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
258 	/* first get the vnode */
259 	if ((error = namei(ndp)) != 0)
260 		return error;
261 	epp->ep_vp = vp = ndp->ni_vp;
262 
263 	/* check access and type */
264 	if (vp->v_type != VREG) {
265 		error = EACCES;
266 		goto bad1;
267 	}
268 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred, l)) != 0)
269 		goto bad1;
270 
271 	/* get attributes */
272 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred, l)) != 0)
273 		goto bad1;
274 
275 	/* Check mount point */
276 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
277 		error = EACCES;
278 		goto bad1;
279 	}
280 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
281 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
282 
283 	/* try to open it */
284 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred, l)) != 0)
285 		goto bad1;
286 
287 	/* unlock vp, since we need it unlocked from here on out. */
288 	VOP_UNLOCK(vp, 0);
289 
290 #if NVERIEXEC > 0
291         if ((error = veriexec_verify(l, vp, epp->ep_ndp->ni_dirp, flag,
292 	    NULL)) != 0)
293                 goto bad2;
294 #endif /* NVERIEXEC > 0 */
295 
296 #ifdef PAX_SEGVGUARD
297 	if (pax_segvguard(l, vp, epp->ep_ndp->ni_dirp, FALSE))
298 		return (EPERM);
299 #endif /* PAX_SEGVGUARD */
300 
301 	/* now we have the file, get the exec header */
302 	uvn_attach(vp, VM_PROT_READ);
303 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
304 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
305 	if (error)
306 		goto bad2;
307 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
308 
309 	/*
310 	 * Set up default address space limits.  Can be overridden
311 	 * by individual exec packages.
312 	 *
313 	 * XXX probably should be all done in the exec pakages.
314 	 */
315 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
316 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
317 	/*
318 	 * set up the vmcmds for creation of the process
319 	 * address space
320 	 */
321 	error = ENOEXEC;
322 	for (i = 0; i < nexecs && error != 0; i++) {
323 		int newerror;
324 
325 		epp->ep_esch = execsw[i];
326 		newerror = (*execsw[i]->es_makecmds)(l, epp);
327 		/* make sure the first "interesting" error code is saved. */
328 		if (!newerror || error == ENOEXEC)
329 			error = newerror;
330 
331 		/* if es_makecmds call was successful, update epp->ep_es */
332 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
333 			epp->ep_es = execsw[i];
334 
335 		if (epp->ep_flags & EXEC_DESTR && error != 0)
336 			return error;
337 	}
338 	if (!error) {
339 		/* check that entry point is sane */
340 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
341 			error = ENOEXEC;
342 
343 		/* check limits */
344 		if ((epp->ep_tsize > MAXTSIZ) ||
345 		    (epp->ep_dsize >
346 		     (u_quad_t)l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur))
347 			error = ENOMEM;
348 
349 		if (!error)
350 			return (0);
351 	}
352 
353 	/*
354 	 * free any vmspace-creation commands,
355 	 * and release their references
356 	 */
357 	kill_vmcmds(&epp->ep_vmcmds);
358 
359 bad2:
360 	/*
361 	 * close and release the vnode, restore the old one, free the
362 	 * pathname buf, and punt.
363 	 */
364 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
365 	VOP_CLOSE(vp, FREAD, l->l_cred, l);
366 	vput(vp);
367 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
368 	return error;
369 
370 bad1:
371 	/*
372 	 * free the namei pathname buffer, and put the vnode
373 	 * (which we don't yet have open).
374 	 */
375 	vput(vp);				/* was still locked */
376 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
377 	return error;
378 }
379 
380 #ifdef __MACHINE_STACK_GROWS_UP
381 #define STACK_PTHREADSPACE NBPG
382 #else
383 #define STACK_PTHREADSPACE 0
384 #endif
385 
386 static int
387 execve_fetch_element(char * const *array, size_t index, char **value)
388 {
389 	return copyin(array + index, value, sizeof(*value));
390 }
391 
392 /*
393  * exec system call
394  */
395 /* ARGSUSED */
396 int
397 sys_execve(struct lwp *l, void *v, register_t *retval)
398 {
399 	struct sys_execve_args /* {
400 		syscallarg(const char *)	path;
401 		syscallarg(char * const *)	argp;
402 		syscallarg(char * const *)	envp;
403 	} */ *uap = v;
404 
405 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
406 	    SCARG(uap, envp), execve_fetch_element);
407 }
408 
409 int
410 execve1(struct lwp *l, const char *path, char * const *args,
411     char * const *envs, execve_fetch_element_t fetch_element)
412 {
413 	int			error;
414 	u_int			i;
415 	struct exec_package	pack;
416 	struct nameidata	nid;
417 	struct vattr		attr;
418 	struct proc		*p;
419 	char			*argp;
420 	char			*dp, *sp;
421 	long			argc, envc;
422 	size_t			len;
423 	char			*stack;
424 	struct ps_strings	arginfo;
425 	struct ps_strings	*aip = &arginfo;
426 	struct vmspace		*vm;
427 	char			**tmpfap;
428 	int			szsigcode;
429 	struct exec_vmcmd	*base_vcp;
430 	int			oldlwpflags;
431 #ifdef SYSTRACE
432 	int			wassugid = ISSET(p->p_flag, P_SUGID);
433 	char			pathbuf[MAXPATHLEN];
434 	size_t			pathbuflen;
435 #endif /* SYSTRACE */
436 
437 	/* Disable scheduler activation upcalls. */
438 	oldlwpflags = l->l_flag & (L_SA | L_SA_UPCALL);
439 	if (l->l_flag & L_SA)
440 		l->l_flag &= ~(L_SA | L_SA_UPCALL);
441 
442 	p = l->l_proc;
443 	/*
444 	 * Lock the process and set the P_INEXEC flag to indicate that
445 	 * it should be left alone until we're done here.  This is
446 	 * necessary to avoid race conditions - e.g. in ptrace() -
447 	 * that might allow a local user to illicitly obtain elevated
448 	 * privileges.
449 	 */
450 	p->p_flag |= P_INEXEC;
451 
452 	base_vcp = NULL;
453 	/*
454 	 * Init the namei data to point the file user's program name.
455 	 * This is done here rather than in check_exec(), so that it's
456 	 * possible to override this settings if any of makecmd/probe
457 	 * functions call check_exec() recursively - for example,
458 	 * see exec_script_makecmds().
459 	 */
460 #ifdef SYSTRACE
461 	if (ISSET(p->p_flag, P_SYSTRACE))
462 		systrace_execve0(p);
463 
464 	error = copyinstr(path, pathbuf, sizeof(pathbuf),
465 			  &pathbuflen);
466 	if (error)
467 		goto clrflg;
468 
469 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_SYSSPACE, pathbuf, l);
470 #else
471 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, path, l);
472 #endif /* SYSTRACE */
473 
474 	/*
475 	 * initialize the fields of the exec package.
476 	 */
477 #ifdef SYSTRACE
478 	pack.ep_name = pathbuf;
479 #else
480 	pack.ep_name = path;
481 #endif /* SYSTRACE */
482 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
483 	pack.ep_hdrlen = exec_maxhdrsz;
484 	pack.ep_hdrvalid = 0;
485 	pack.ep_ndp = &nid;
486 	pack.ep_emul_arg = NULL;
487 	pack.ep_vmcmds.evs_cnt = 0;
488 	pack.ep_vmcmds.evs_used = 0;
489 	pack.ep_vap = &attr;
490 	pack.ep_flags = 0;
491 
492 #ifdef LKM
493 	lockmgr(&exec_lock, LK_SHARED, NULL);
494 #endif
495 
496 	/* see if we can run it. */
497 #if NVERIEXEC > 0
498         if ((error = check_exec(l, &pack, VERIEXEC_DIRECT)) != 0)
499 #else
500         if ((error = check_exec(l, &pack, 0)) != 0)
501 #endif /* NVERIEXEC > 0 */
502 		goto freehdr;
503 
504 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
505 
506 	/* allocate an argument buffer */
507 	argp = (char *) uvm_km_alloc(exec_map, NCARGS, 0,
508 	    UVM_KMF_PAGEABLE|UVM_KMF_WAITVA);
509 #ifdef DIAGNOSTIC
510 	if (argp == NULL)
511 		panic("execve: argp == NULL");
512 #endif
513 	dp = argp;
514 	argc = 0;
515 
516 	/* copy the fake args list, if there's one, freeing it as we go */
517 	if (pack.ep_flags & EXEC_HASARGL) {
518 		tmpfap = pack.ep_fa;
519 		while (*tmpfap != NULL) {
520 			char *cp;
521 
522 			cp = *tmpfap;
523 			while (*cp)
524 				*dp++ = *cp++;
525 			dp++;
526 
527 			FREE(*tmpfap, M_EXEC);
528 			tmpfap++; argc++;
529 		}
530 		FREE(pack.ep_fa, M_EXEC);
531 		pack.ep_flags &= ~EXEC_HASARGL;
532 	}
533 
534 	/* Now get argv & environment */
535 	if (args == NULL) {
536 		error = EINVAL;
537 		goto bad;
538 	}
539 	/* 'i' will index the argp/envp element to be retrieved */
540 	i = 0;
541 	if (pack.ep_flags & EXEC_SKIPARG)
542 		i++;
543 
544 	while (1) {
545 		len = argp + ARG_MAX - dp;
546 		if ((error = (*fetch_element)(args, i, &sp)) != 0)
547 			goto bad;
548 		if (!sp)
549 			break;
550 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
551 			if (error == ENAMETOOLONG)
552 				error = E2BIG;
553 			goto bad;
554 		}
555 #ifdef KTRACE
556 		if (KTRPOINT(p, KTR_EXEC_ARG))
557 			ktrkmem(l, KTR_EXEC_ARG, dp, len - 1);
558 #endif
559 		dp += len;
560 		i++;
561 		argc++;
562 	}
563 
564 	envc = 0;
565 	/* environment need not be there */
566 	if (envs != NULL) {
567 		i = 0;
568 		while (1) {
569 			len = argp + ARG_MAX - dp;
570 			if ((error = (*fetch_element)(envs, i, &sp)) != 0)
571 				goto bad;
572 			if (!sp)
573 				break;
574 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
575 				if (error == ENAMETOOLONG)
576 					error = E2BIG;
577 				goto bad;
578 			}
579 #ifdef KTRACE
580 			if (KTRPOINT(p, KTR_EXEC_ENV))
581 				ktrkmem(l, KTR_EXEC_ENV, dp, len - 1);
582 #endif
583 			dp += len;
584 			i++;
585 			envc++;
586 		}
587 	}
588 
589 	dp = (char *) ALIGN(dp);
590 
591 	szsigcode = pack.ep_es->es_emul->e_esigcode -
592 	    pack.ep_es->es_emul->e_sigcode;
593 
594 	/* Now check if args & environ fit into new stack */
595 	if (pack.ep_flags & EXEC_32)
596 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
597 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
598 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
599 		    - argp;
600 	else
601 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
602 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
603 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
604 		    - argp;
605 
606 	len = ALIGN(len);	/* make the stack "safely" aligned */
607 
608 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
609 		error = ENOMEM;
610 		goto bad;
611 	}
612 
613 	/* Get rid of other LWPs/ */
614 	p->p_flag |= P_WEXIT; /* XXX hack. lwp-exit stuff wants to see it. */
615 	exit_lwps(l);
616 	p->p_flag &= ~P_WEXIT;
617 	KDASSERT(p->p_nlwps == 1);
618 
619 	/* This is now LWP 1 */
620 	l->l_lid = 1;
621 	p->p_nlwpid = 1;
622 
623 	/* Release any SA state. */
624 	if (p->p_sa)
625 		sa_release(p);
626 
627 	/* Remove POSIX timers */
628 	timers_free(p, TIMERS_POSIX);
629 
630 	/* adjust "active stack depth" for process VSZ */
631 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
632 
633 	/*
634 	 * Do whatever is necessary to prepare the address space
635 	 * for remapping.  Note that this might replace the current
636 	 * vmspace with another!
637 	 */
638 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
639 
640 	/* record proc's vnode, for use by procfs and others */
641         if (p->p_textvp)
642                 vrele(p->p_textvp);
643 	VREF(pack.ep_vp);
644 	p->p_textvp = pack.ep_vp;
645 
646 	/* Now map address space */
647 	vm = p->p_vmspace;
648 	vm->vm_taddr = (caddr_t) pack.ep_taddr;
649 	vm->vm_tsize = btoc(pack.ep_tsize);
650 	vm->vm_daddr = (caddr_t) pack.ep_daddr;
651 	vm->vm_dsize = btoc(pack.ep_dsize);
652 	vm->vm_ssize = btoc(pack.ep_ssize);
653 	vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr;
654 	vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr;
655 
656 	/* create the new process's VM space by running the vmcmds */
657 #ifdef DIAGNOSTIC
658 	if (pack.ep_vmcmds.evs_used == 0)
659 		panic("execve: no vmcmds");
660 #endif
661 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
662 		struct exec_vmcmd *vcp;
663 
664 		vcp = &pack.ep_vmcmds.evs_cmds[i];
665 		if (vcp->ev_flags & VMCMD_RELATIVE) {
666 #ifdef DIAGNOSTIC
667 			if (base_vcp == NULL)
668 				panic("execve: relative vmcmd with no base");
669 			if (vcp->ev_flags & VMCMD_BASE)
670 				panic("execve: illegal base & relative vmcmd");
671 #endif
672 			vcp->ev_addr += base_vcp->ev_addr;
673 		}
674 		error = (*vcp->ev_proc)(l, vcp);
675 #ifdef DEBUG_EXEC
676 		if (error) {
677 			int j;
678 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
679 			for (j = 0; j <= i; j++)
680 				uprintf(
681 			    "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
682 				    j, vp[j].ev_addr, vp[j].ev_len,
683 				    vp[j].ev_offset, vp[j].ev_prot,
684 				    vp[j].ev_flags);
685 		}
686 #endif /* DEBUG_EXEC */
687 		if (vcp->ev_flags & VMCMD_BASE)
688 			base_vcp = vcp;
689 	}
690 
691 	/* free the vmspace-creation commands, and release their references */
692 	kill_vmcmds(&pack.ep_vmcmds);
693 
694 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
695 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred, l);
696 	vput(pack.ep_vp);
697 
698 	/* if an error happened, deallocate and punt */
699 	if (error) {
700 		DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error));
701 		goto exec_abort;
702 	}
703 
704 	/* remember information about the process */
705 	arginfo.ps_nargvstr = argc;
706 	arginfo.ps_nenvstr = envc;
707 
708 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
709 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
710 		len - (sizeof(struct ps_strings) + szsigcode));
711 #ifdef __MACHINE_STACK_GROWS_UP
712 	/*
713 	 * The copyargs call always copies into lower addresses
714 	 * first, moving towards higher addresses, starting with
715 	 * the stack pointer that we give.  When the stack grows
716 	 * down, this puts argc/argv/envp very shallow on the
717 	 * stack, right at the first user stack pointer, and puts
718 	 * STACKGAPLEN very deep in the stack.  When the stack
719 	 * grows up, the situation is reversed.
720 	 *
721 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
722 	 * function expects to be called with a single pointer to
723 	 * a region that has a few words it can stash values into,
724 	 * followed by argc/argv/envp.  When the stack grows down,
725 	 * it's easy to decrement the stack pointer a little bit to
726 	 * allocate the space for these few words and pass the new
727 	 * stack pointer to _rtld.  When the stack grows up, however,
728 	 * a few words before argc is part of the signal trampoline, XXX
729 	 * so we have a problem.
730 	 *
731 	 * Instead of changing how _rtld works, we take the easy way
732 	 * out and steal 32 bytes before we call copyargs.  This
733 	 * space is effectively stolen from STACKGAPLEN.
734 	 */
735 	stack += 32;
736 #endif /* __MACHINE_STACK_GROWS_UP */
737 
738 	/* Now copy argc, args & environ to new stack */
739 	error = (*pack.ep_es->es_copyargs)(l, &pack, &arginfo, &stack, argp);
740 	if (error) {
741 		DPRINTF(("execve: copyargs failed %d\n", error));
742 		goto exec_abort;
743 	}
744 	/* Move the stack back to original point */
745 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
746 
747 	/* fill process ps_strings info */
748 	p->p_psstr = (struct ps_strings *)
749 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
750 	    sizeof(struct ps_strings));
751 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
752 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
753 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
754 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
755 
756 	/* copy out the process's ps_strings structure */
757 	if ((error = copyout(aip, (char *)p->p_psstr,
758 	    sizeof(arginfo))) != 0) {
759 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
760 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
761 		goto exec_abort;
762 	}
763 
764 	stopprofclock(p);	/* stop profiling */
765 	fdcloseexec(l);		/* handle close on exec */
766 	execsigs(p);		/* reset catched signals */
767 
768 	l->l_ctxlink = NULL;	/* reset ucontext link */
769 
770 	/* set command name & other accounting info */
771 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
772 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
773 	p->p_comm[len] = 0;
774 	p->p_acflag &= ~AFORK;
775 
776 	p->p_flag |= P_EXEC;
777 	if (p->p_flag & P_PPWAIT) {
778 		p->p_flag &= ~P_PPWAIT;
779 		wakeup((caddr_t) p->p_pptr);
780 	}
781 
782 	/*
783 	 * deal with set[ug]id.
784 	 * MNT_NOSUID has already been used to disable s[ug]id.
785 	 */
786 	if ((p->p_flag & P_TRACED) == 0 &&
787 
788 	    (((attr.va_mode & S_ISUID) != 0 &&
789 	      kauth_cred_geteuid(l->l_cred) != attr.va_uid) ||
790 
791 	     ((attr.va_mode & S_ISGID) != 0 &&
792 	      kauth_cred_getegid(l->l_cred) != attr.va_gid))) {
793 		/*
794 		 * Mark the process as SUGID before we do
795 		 * anything that might block.
796 		 */
797 		p_sugid(p);
798 
799 		/* Make sure file descriptors 0..2 are in use. */
800 		if ((error = fdcheckstd(l)) != 0) {
801 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
802 			goto exec_abort;
803 		}
804 
805 		/*
806 		 * Copy the credential so other references don't see our
807 		 * changes.
808 		 */
809 		l->l_cred = kauth_cred_copy(l->l_cred);
810 #ifdef KTRACE
811 		/*
812 		 * If process is being ktraced, turn off - unless
813 		 * root set it.
814 		 */
815 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
816 			ktrderef(p);
817 #endif
818 		if (attr.va_mode & S_ISUID)
819 			kauth_cred_seteuid(l->l_cred, attr.va_uid);
820 		if (attr.va_mode & S_ISGID)
821 			kauth_cred_setegid(l->l_cred, attr.va_gid);
822 	} else {
823 		if (kauth_cred_geteuid(l->l_cred) ==
824 		    kauth_cred_getuid(l->l_cred) &&
825 		    kauth_cred_getegid(l->l_cred) ==
826 		    kauth_cred_getgid(l->l_cred))
827 			p->p_flag &= ~P_SUGID;
828 	}
829 
830 	/*
831 	 * Copy the credential so other references don't see our changes.
832 	 * Test to see if this is necessary first, since in the common case
833 	 * we won't need a private reference.
834 	 */
835 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
836 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
837 		l->l_cred = kauth_cred_copy(l->l_cred);
838 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
839 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
840 	}
841 
842 	/* Update the master credentials. */
843 	if (l->l_cred != p->p_cred) {
844 		kauth_cred_t ocred;
845 
846 		kauth_cred_hold(l->l_cred);
847 		simple_lock(&p->p_lock);
848 		ocred = p->p_cred;
849 		p->p_cred = l->l_cred;
850 		simple_unlock(&p->p_lock);
851 		kauth_cred_free(ocred);
852 	}
853 
854 #if defined(__HAVE_RAS)
855 	/*
856 	 * Remove all RASs from the address space.
857 	 */
858 	ras_purgeall(p);
859 #endif
860 
861 	doexechooks(p);
862 
863 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
864 
865 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
866 
867 	/* notify others that we exec'd */
868 	KNOTE(&p->p_klist, NOTE_EXEC);
869 
870 	/* setup new registers and do misc. setup. */
871 	(*pack.ep_es->es_emul->e_setregs)(l, &pack, (u_long) stack);
872 	if (pack.ep_es->es_setregs)
873 		(*pack.ep_es->es_setregs)(l, &pack, (u_long) stack);
874 
875 	/* map the process's signal trampoline code */
876 	if (exec_sigcode_map(p, pack.ep_es->es_emul)) {
877 		DPRINTF(("execve: map sigcode failed %d\n", error));
878 		goto exec_abort;
879 	}
880 
881 	if ((p->p_flag & (P_TRACED|P_SYSCALL)) == P_TRACED)
882 		psignal(p, SIGTRAP);
883 
884 	free(pack.ep_hdr, M_EXEC);
885 
886 	/*
887 	 * Call emulation specific exec hook. This can setup per-process
888 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
889 	 *
890 	 * If we are executing process of different emulation than the
891 	 * original forked process, call e_proc_exit() of the old emulation
892 	 * first, then e_proc_exec() of new emulation. If the emulation is
893 	 * same, the exec hook code should deallocate any old emulation
894 	 * resources held previously by this process.
895 	 */
896 	if (p->p_emul && p->p_emul->e_proc_exit
897 	    && p->p_emul != pack.ep_es->es_emul)
898 		(*p->p_emul->e_proc_exit)(p);
899 
900 	/*
901 	 * Call exec hook. Emulation code may NOT store reference to anything
902 	 * from &pack.
903 	 */
904         if (pack.ep_es->es_emul->e_proc_exec)
905                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
906 
907 	/* update p_emul, the old value is no longer needed */
908 	p->p_emul = pack.ep_es->es_emul;
909 
910 	/* ...and the same for p_execsw */
911 	p->p_execsw = pack.ep_es;
912 
913 #ifdef __HAVE_SYSCALL_INTERN
914 	(*p->p_emul->e_syscall_intern)(p);
915 #endif
916 #ifdef KTRACE
917 	if (KTRPOINT(p, KTR_EMUL))
918 		ktremul(l);
919 #endif
920 
921 #ifdef LKM
922 	lockmgr(&exec_lock, LK_RELEASE, NULL);
923 #endif
924 	p->p_flag &= ~P_INEXEC;
925 
926 	if (p->p_flag & P_STOPEXEC) {
927 		int s;
928 
929 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
930 		SCHED_LOCK(s);
931 		p->p_pptr->p_nstopchild++;
932 		p->p_stat = SSTOP;
933 		l->l_stat = LSSTOP;
934 		p->p_nrlwps--;
935 		mi_switch(l, NULL);
936 		SCHED_ASSERT_UNLOCKED();
937 		splx(s);
938 	}
939 
940 #ifdef SYSTRACE
941 	if (ISSET(p->p_flag, P_SYSTRACE) &&
942 	    wassugid && !ISSET(p->p_flag, P_SUGID))
943 		systrace_execve1(pathbuf, p);
944 #endif /* SYSTRACE */
945 
946 	return (EJUSTRETURN);
947 
948  bad:
949 	p->p_flag &= ~P_INEXEC;
950 	/* free the vmspace-creation commands, and release their references */
951 	kill_vmcmds(&pack.ep_vmcmds);
952 	/* kill any opened file descriptor, if necessary */
953 	if (pack.ep_flags & EXEC_HASFD) {
954 		pack.ep_flags &= ~EXEC_HASFD;
955 		(void) fdrelease(l, pack.ep_fd);
956 	}
957 	/* close and put the exec'd file */
958 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
959 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred, l);
960 	vput(pack.ep_vp);
961 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
962 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
963 
964  freehdr:
965 	free(pack.ep_hdr, M_EXEC);
966 
967 #ifdef SYSTRACE
968  clrflg:
969 #endif /* SYSTRACE */
970 	l->l_flag |= oldlwpflags;
971 	p->p_flag &= ~P_INEXEC;
972 #ifdef LKM
973 	lockmgr(&exec_lock, LK_RELEASE, NULL);
974 #endif
975 
976 	return error;
977 
978  exec_abort:
979 	p->p_flag &= ~P_INEXEC;
980 #ifdef LKM
981 	lockmgr(&exec_lock, LK_RELEASE, NULL);
982 #endif
983 
984 	/*
985 	 * the old process doesn't exist anymore.  exit gracefully.
986 	 * get rid of the (new) address space we have created, if any, get rid
987 	 * of our namei data and vnode, and exit noting failure
988 	 */
989 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
990 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
991 	if (pack.ep_emul_arg)
992 		FREE(pack.ep_emul_arg, M_TEMP);
993 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
994 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
995 	free(pack.ep_hdr, M_EXEC);
996 	exit1(l, W_EXITCODE(error, SIGABRT));
997 
998 	/* NOTREACHED */
999 	return 0;
1000 }
1001 
1002 
1003 int
1004 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1005     char **stackp, void *argp)
1006 {
1007 	char	**cpp, *dp, *sp;
1008 	size_t	len;
1009 	void	*nullp;
1010 	long	argc, envc;
1011 	int	error;
1012 
1013 	cpp = (char **)*stackp;
1014 	nullp = NULL;
1015 	argc = arginfo->ps_nargvstr;
1016 	envc = arginfo->ps_nenvstr;
1017 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
1018 		return error;
1019 
1020 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
1021 	sp = argp;
1022 
1023 	/* XXX don't copy them out, remap them! */
1024 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1025 
1026 	for (; --argc >= 0; sp += len, dp += len)
1027 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1028 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1029 			return error;
1030 
1031 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1032 		return error;
1033 
1034 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1035 
1036 	for (; --envc >= 0; sp += len, dp += len)
1037 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1038 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1039 			return error;
1040 
1041 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1042 		return error;
1043 
1044 	*stackp = (char *)cpp;
1045 	return 0;
1046 }
1047 
1048 #ifdef LKM
1049 /*
1050  * Find an emulation of given name in list of emulations.
1051  * Needs to be called with the exec_lock held.
1052  */
1053 const struct emul *
1054 emul_search(const char *name)
1055 {
1056 	struct emul_entry *it;
1057 
1058 	LIST_FOREACH(it, &el_head, el_list) {
1059 		if (strcmp(name, it->el_emul->e_name) == 0)
1060 			return it->el_emul;
1061 	}
1062 
1063 	return NULL;
1064 }
1065 
1066 /*
1067  * Add an emulation to list, if it's not there already.
1068  */
1069 int
1070 emul_register(const struct emul *emul, int ro_entry)
1071 {
1072 	struct emul_entry	*ee;
1073 	int			error;
1074 
1075 	error = 0;
1076 	lockmgr(&exec_lock, LK_SHARED, NULL);
1077 
1078 	if (emul_search(emul->e_name)) {
1079 		error = EEXIST;
1080 		goto out;
1081 	}
1082 
1083 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
1084 		M_EXEC, M_WAITOK);
1085 	ee->el_emul = emul;
1086 	ee->ro_entry = ro_entry;
1087 	LIST_INSERT_HEAD(&el_head, ee, el_list);
1088 
1089  out:
1090 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1091 	return error;
1092 }
1093 
1094 /*
1095  * Remove emulation with name 'name' from list of supported emulations.
1096  */
1097 int
1098 emul_unregister(const char *name)
1099 {
1100 	const struct proclist_desc *pd;
1101 	struct emul_entry	*it;
1102 	int			i, error;
1103 	struct proc		*ptmp;
1104 
1105 	error = 0;
1106 	lockmgr(&exec_lock, LK_SHARED, NULL);
1107 
1108 	LIST_FOREACH(it, &el_head, el_list) {
1109 		if (strcmp(it->el_emul->e_name, name) == 0)
1110 			break;
1111 	}
1112 
1113 	if (!it) {
1114 		error = ENOENT;
1115 		goto out;
1116 	}
1117 
1118 	if (it->ro_entry) {
1119 		error = EBUSY;
1120 		goto out;
1121 	}
1122 
1123 	/* test if any execw[] entry is still using this */
1124 	for(i=0; i < nexecs; i++) {
1125 		if (execsw[i]->es_emul == it->el_emul) {
1126 			error = EBUSY;
1127 			goto out;
1128 		}
1129 	}
1130 
1131 	/*
1132 	 * Test if any process is running under this emulation - since
1133 	 * emul_unregister() is running quite sendomly, it's better
1134 	 * to do expensive check here than to use any locking.
1135 	 */
1136 	proclist_lock_read();
1137 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
1138 		PROCLIST_FOREACH(ptmp, pd->pd_list) {
1139 			if (ptmp->p_emul == it->el_emul) {
1140 				error = EBUSY;
1141 				break;
1142 			}
1143 		}
1144 	}
1145 	proclist_unlock_read();
1146 
1147 	if (error)
1148 		goto out;
1149 
1150 
1151 	/* entry is not used, remove it */
1152 	LIST_REMOVE(it, el_list);
1153 	FREE(it, M_EXEC);
1154 
1155  out:
1156 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1157 	return error;
1158 }
1159 
1160 /*
1161  * Add execsw[] entry.
1162  */
1163 int
1164 exec_add(struct execsw *esp, const char *e_name)
1165 {
1166 	struct exec_entry	*it;
1167 	int			error;
1168 
1169 	error = 0;
1170 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
1171 
1172 	if (!esp->es_emul) {
1173 		esp->es_emul = emul_search(e_name);
1174 		if (!esp->es_emul) {
1175 			error = ENOENT;
1176 			goto out;
1177 		}
1178 	}
1179 
1180 	LIST_FOREACH(it, &ex_head, ex_list) {
1181 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1182 		if (it->es->es_makecmds == esp->es_makecmds
1183 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1184 		    && it->es->es_emul == esp->es_emul) {
1185 			error = EEXIST;
1186 			goto out;
1187 		}
1188 	}
1189 
1190 	/* if we got here, the entry doesn't exist yet */
1191 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
1192 		M_EXEC, M_WAITOK);
1193 	it->es = esp;
1194 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
1195 
1196 	/* update execsw[] */
1197 	exec_init(0);
1198 
1199  out:
1200 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1201 	return error;
1202 }
1203 
1204 /*
1205  * Remove execsw[] entry.
1206  */
1207 int
1208 exec_remove(const struct execsw *esp)
1209 {
1210 	struct exec_entry	*it;
1211 	int			error;
1212 
1213 	error = 0;
1214 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
1215 
1216 	LIST_FOREACH(it, &ex_head, ex_list) {
1217 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1218 		if (it->es->es_makecmds == esp->es_makecmds
1219 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1220 		    && it->es->es_emul == esp->es_emul)
1221 			break;
1222 	}
1223 	if (!it) {
1224 		error = ENOENT;
1225 		goto out;
1226 	}
1227 
1228 	/* remove item from list and free resources */
1229 	LIST_REMOVE(it, ex_list);
1230 	FREE(it, M_EXEC);
1231 
1232 	/* update execsw[] */
1233 	exec_init(0);
1234 
1235  out:
1236 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1237 	return error;
1238 }
1239 
1240 static void
1241 link_es(struct execsw_entry **listp, const struct execsw *esp)
1242 {
1243 	struct execsw_entry *et, *e1;
1244 
1245 	et = (struct execsw_entry *) malloc(sizeof(struct execsw_entry),
1246 			M_TEMP, M_WAITOK);
1247 	et->next = NULL;
1248 	et->es = esp;
1249 	if (*listp == NULL) {
1250 		*listp = et;
1251 		return;
1252 	}
1253 
1254 	switch(et->es->es_prio) {
1255 	case EXECSW_PRIO_FIRST:
1256 		/* put new entry as the first */
1257 		et->next = *listp;
1258 		*listp = et;
1259 		break;
1260 	case EXECSW_PRIO_ANY:
1261 		/* put new entry after all *_FIRST and *_ANY entries */
1262 		for(e1 = *listp; e1->next
1263 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
1264 			e1 = e1->next);
1265 		et->next = e1->next;
1266 		e1->next = et;
1267 		break;
1268 	case EXECSW_PRIO_LAST:
1269 		/* put new entry as the last one */
1270 		for(e1 = *listp; e1->next; e1 = e1->next);
1271 		e1->next = et;
1272 		break;
1273 	default:
1274 #ifdef DIAGNOSTIC
1275 		panic("execw[] entry with unknown priority %d found",
1276 			et->es->es_prio);
1277 #else
1278 		free(et, M_TEMP);
1279 #endif
1280 		break;
1281 	}
1282 }
1283 
1284 /*
1285  * Initialize exec structures. If init_boot is true, also does necessary
1286  * one-time initialization (it's called from main() that way).
1287  * Once system is multiuser, this should be called with exec_lock held,
1288  * i.e. via exec_{add|remove}().
1289  */
1290 int
1291 exec_init(int init_boot)
1292 {
1293 	const struct execsw	**new_es, * const *old_es;
1294 	struct execsw_entry	*list, *e1;
1295 	struct exec_entry	*e2;
1296 	int			i, es_sz;
1297 
1298 	if (init_boot) {
1299 		/* do one-time initializations */
1300 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1301 
1302 		/* register compiled-in emulations */
1303 		for(i=0; i < nexecs_builtin; i++) {
1304 			if (execsw_builtin[i].es_emul)
1305 				emul_register(execsw_builtin[i].es_emul, 1);
1306 		}
1307 #ifdef DIAGNOSTIC
1308 		if (i == 0)
1309 			panic("no emulations found in execsw_builtin[]");
1310 #endif
1311 	}
1312 
1313 	/*
1314 	 * Build execsw[] array from builtin entries and entries added
1315 	 * at runtime.
1316 	 */
1317 	list = NULL;
1318 	for(i=0; i < nexecs_builtin; i++)
1319 		link_es(&list, &execsw_builtin[i]);
1320 
1321 	/* Add dynamically loaded entries */
1322 	es_sz = nexecs_builtin;
1323 	LIST_FOREACH(e2, &ex_head, ex_list) {
1324 		link_es(&list, e2->es);
1325 		es_sz++;
1326 	}
1327 
1328 	/*
1329 	 * Now that we have sorted all execw entries, create new execsw[]
1330 	 * and free no longer needed memory in the process.
1331 	 */
1332 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
1333 	for(i=0; list; i++) {
1334 		new_es[i] = list->es;
1335 		e1 = list->next;
1336 		free(list, M_TEMP);
1337 		list = e1;
1338 	}
1339 
1340 	/*
1341 	 * New execsw[] array built, now replace old execsw[] and free
1342 	 * used memory.
1343 	 */
1344 	old_es = execsw;
1345 	execsw = new_es;
1346 	nexecs = es_sz;
1347 	if (old_es)
1348 		/*XXXUNCONST*/
1349 		free(__UNCONST(old_es), M_EXEC);
1350 
1351 	/*
1352 	 * Figure out the maximum size of an exec header.
1353 	 */
1354 	exec_maxhdrsz = 0;
1355 	for (i = 0; i < nexecs; i++) {
1356 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1357 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1358 	}
1359 
1360 	return 0;
1361 }
1362 #endif
1363 
1364 #ifndef LKM
1365 /*
1366  * Simplified exec_init() for kernels without LKMs. Only initialize
1367  * exec_maxhdrsz and execsw[].
1368  */
1369 int
1370 exec_init(int init_boot)
1371 {
1372 	int i;
1373 
1374 #ifdef DIAGNOSTIC
1375 	if (!init_boot)
1376 		panic("exec_init(): called with init_boot == 0");
1377 #endif
1378 
1379 	/* do one-time initializations */
1380 	nexecs = nexecs_builtin;
1381 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
1382 
1383 	/*
1384 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1385 	 */
1386 	exec_maxhdrsz = 0;
1387 	for(i=0; i < nexecs; i++) {
1388 		execsw[i] = &execsw_builtin[i];
1389 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1390 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1391 	}
1392 
1393 	return 0;
1394 
1395 }
1396 #endif /* !LKM */
1397 
1398 static int
1399 exec_sigcode_map(struct proc *p, const struct emul *e)
1400 {
1401 	vaddr_t va;
1402 	vsize_t sz;
1403 	int error;
1404 	struct uvm_object *uobj;
1405 
1406 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1407 
1408 	if (e->e_sigobject == NULL || sz == 0) {
1409 		return 0;
1410 	}
1411 
1412 	/*
1413 	 * If we don't have a sigobject for this emulation, create one.
1414 	 *
1415 	 * sigobject is an anonymous memory object (just like SYSV shared
1416 	 * memory) that we keep a permanent reference to and that we map
1417 	 * in all processes that need this sigcode. The creation is simple,
1418 	 * we create an object, add a permanent reference to it, map it in
1419 	 * kernel space, copy out the sigcode to it and unmap it.
1420 	 * We map it with PROT_READ|PROT_EXEC into the process just
1421 	 * the way sys_mmap() would map it.
1422 	 */
1423 
1424 	uobj = *e->e_sigobject;
1425 	if (uobj == NULL) {
1426 		uobj = uao_create(sz, 0);
1427 		(*uobj->pgops->pgo_reference)(uobj);
1428 		va = vm_map_min(kernel_map);
1429 		if ((error = uvm_map(kernel_map, &va, round_page(sz),
1430 		    uobj, 0, 0,
1431 		    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1432 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1433 			printf("kernel mapping failed %d\n", error);
1434 			(*uobj->pgops->pgo_detach)(uobj);
1435 			return (error);
1436 		}
1437 		memcpy((void *)va, e->e_sigcode, sz);
1438 #ifdef PMAP_NEED_PROCWR
1439 		pmap_procwr(&proc0, va, sz);
1440 #endif
1441 		uvm_unmap(kernel_map, va, va + round_page(sz));
1442 		*e->e_sigobject = uobj;
1443 	}
1444 
1445 	/* Just a hint to uvm_map where to put it. */
1446 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1447 	    round_page(sz));
1448 
1449 #ifdef __alpha__
1450 	/*
1451 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1452 	 * which causes the above calculation to put the sigcode at
1453 	 * an invalid address.  Put it just below the text instead.
1454 	 */
1455 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1456 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1457 	}
1458 #endif
1459 
1460 	(*uobj->pgops->pgo_reference)(uobj);
1461 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1462 			uobj, 0, 0,
1463 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1464 				    UVM_ADV_RANDOM, 0));
1465 	if (error) {
1466 		(*uobj->pgops->pgo_detach)(uobj);
1467 		return (error);
1468 	}
1469 	p->p_sigctx.ps_sigcode = (void *)va;
1470 	return (0);
1471 }
1472