xref: /netbsd-src/sys/kern/kern_exec.c (revision 69b6d498973bb4d7230c2d3c12bd9a032738ec8e)
1 /*	$NetBSD: kern_exec.c,v 1.207 2005/07/29 22:57:34 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 1.207 2005/07/29 22:57:34 elad Exp $");
37 
38 #include "opt_ktrace.h"
39 #include "opt_syscall_debug.h"
40 #include "opt_compat_netbsd.h"
41 #include "opt_verified_exec.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/filedesc.h>
46 #include <sys/kernel.h>
47 #include <sys/proc.h>
48 #include <sys/mount.h>
49 #include <sys/malloc.h>
50 #include <sys/namei.h>
51 #include <sys/vnode.h>
52 #include <sys/file.h>
53 #include <sys/acct.h>
54 #include <sys/exec.h>
55 #include <sys/ktrace.h>
56 #include <sys/resourcevar.h>
57 #include <sys/wait.h>
58 #include <sys/mman.h>
59 #include <sys/ras.h>
60 #include <sys/signalvar.h>
61 #include <sys/stat.h>
62 #include <sys/syscall.h>
63 
64 #include <sys/sa.h>
65 #include <sys/savar.h>
66 #include <sys/syscallargs.h>
67 #ifdef VERIFIED_EXEC
68 #include <sys/verified_exec.h>
69 #endif
70 
71 #ifdef SYSTRACE
72 #include <sys/systrace.h>
73 #endif /* SYSTRACE */
74 
75 #include <uvm/uvm_extern.h>
76 
77 #include <machine/cpu.h>
78 #include <machine/reg.h>
79 
80 static int exec_sigcode_map(struct proc *, const struct emul *);
81 
82 #ifdef DEBUG_EXEC
83 #define DPRINTF(a) uprintf a
84 #else
85 #define DPRINTF(a)
86 #endif /* DEBUG_EXEC */
87 
88 MALLOC_DEFINE(M_EXEC, "exec", "argument lists & other mem used by exec");
89 
90 /*
91  * Exec function switch:
92  *
93  * Note that each makecmds function is responsible for loading the
94  * exec package with the necessary functions for any exec-type-specific
95  * handling.
96  *
97  * Functions for specific exec types should be defined in their own
98  * header file.
99  */
100 extern const struct execsw	execsw_builtin[];
101 extern int			nexecs_builtin;
102 static const struct execsw	**execsw = NULL;
103 static int			nexecs;
104 
105 u_int	exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
106 
107 #ifdef LKM
108 /* list of supported emulations */
109 static
110 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
111 struct emul_entry {
112 	LIST_ENTRY(emul_entry)	el_list;
113 	const struct emul	*el_emul;
114 	int			ro_entry;
115 };
116 
117 /* list of dynamically loaded execsw entries */
118 static
119 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
120 struct exec_entry {
121 	LIST_ENTRY(exec_entry)	ex_list;
122 	const struct execsw	*es;
123 };
124 
125 /* structure used for building execw[] */
126 struct execsw_entry {
127 	struct execsw_entry	*next;
128 	const struct execsw	*es;
129 };
130 #endif /* LKM */
131 
132 #ifdef SYSCALL_DEBUG
133 extern const char * const syscallnames[];
134 #endif
135 
136 #ifdef COMPAT_16
137 extern char	sigcode[], esigcode[];
138 struct uvm_object *emul_netbsd_object;
139 #endif
140 
141 #ifndef __HAVE_SYSCALL_INTERN
142 void	syscall(void);
143 #endif
144 
145 /* NetBSD emul struct */
146 const struct emul emul_netbsd = {
147 	"netbsd",
148 	NULL,		/* emulation path */
149 #ifndef __HAVE_MINIMAL_EMUL
150 	EMUL_HAS_SYS___syscall,
151 	NULL,
152 	SYS_syscall,
153 	SYS_NSYSENT,
154 #endif
155 	sysent,
156 #ifdef SYSCALL_DEBUG
157 	syscallnames,
158 #else
159 	NULL,
160 #endif
161 	sendsig,
162 	trapsignal,
163 	NULL,
164 #ifdef COMPAT_16
165 	sigcode,
166 	esigcode,
167 	&emul_netbsd_object,
168 #else
169 	NULL,
170 	NULL,
171 	NULL,
172 #endif
173 	setregs,
174 	NULL,
175 	NULL,
176 	NULL,
177 	NULL,
178 	NULL,
179 #ifdef __HAVE_SYSCALL_INTERN
180 	syscall_intern,
181 #else
182 	syscall,
183 #endif
184 	NULL,
185 	NULL,
186 
187 	uvm_default_mapaddr,
188 };
189 
190 #ifdef LKM
191 /*
192  * Exec lock. Used to control access to execsw[] structures.
193  * This must not be static so that netbsd32 can access it, too.
194  */
195 struct lock exec_lock;
196 
197 static void link_es(struct execsw_entry **, const struct execsw *);
198 #endif /* LKM */
199 
200 /*
201  * check exec:
202  * given an "executable" described in the exec package's namei info,
203  * see what we can do with it.
204  *
205  * ON ENTRY:
206  *	exec package with appropriate namei info
207  *	proc pointer of exec'ing proc
208  *      if verified exec enabled then flag indicating a direct exec or
209  *        an indirect exec (i.e. for a shell script interpreter)
210  *	NO SELF-LOCKED VNODES
211  *
212  * ON EXIT:
213  *	error:	nothing held, etc.  exec header still allocated.
214  *	ok:	filled exec package, executable's vnode (unlocked).
215  *
216  * EXEC SWITCH ENTRY:
217  * 	Locked vnode to check, exec package, proc.
218  *
219  * EXEC SWITCH EXIT:
220  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
221  *	error:	destructive:
222  *			everything deallocated execept exec header.
223  *		non-destructive:
224  *			error code, executable's vnode (unlocked),
225  *			exec header unmodified.
226  */
227 int
228 /*ARGSUSED*/
229 check_exec(struct proc *p, struct exec_package *epp, int flag)
230 {
231 	int		error, i;
232 	struct vnode	*vp;
233 	struct nameidata *ndp;
234 	size_t		resid;
235 
236 	ndp = epp->ep_ndp;
237 	ndp->ni_cnd.cn_nameiop = LOOKUP;
238 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
239 	/* first get the vnode */
240 	if ((error = namei(ndp)) != 0)
241 		return error;
242 	epp->ep_vp = vp = ndp->ni_vp;
243 
244 	/* check access and type */
245 	if (vp->v_type != VREG) {
246 		error = EACCES;
247 		goto bad1;
248 	}
249 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
250 		goto bad1;
251 
252 	/* get attributes */
253 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
254 		goto bad1;
255 
256 	/* Check mount point */
257 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
258 		error = EACCES;
259 		goto bad1;
260 	}
261 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
262 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
263 
264 	/* try to open it */
265 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
266 		goto bad1;
267 
268 	/* unlock vp, since we need it unlocked from here on out. */
269 	VOP_UNLOCK(vp, 0);
270 
271 
272 #ifdef VERIFIED_EXEC
273         if ((error = veriexec_verify(p, vp, epp->ep_vap, epp->ep_ndp->ni_dirp,
274 				     flag, NULL)) != 0)
275                 goto bad2;
276 #endif
277 
278 	/* now we have the file, get the exec header */
279 	uvn_attach(vp, VM_PROT_READ);
280 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
281 			UIO_SYSSPACE, 0, p->p_ucred, &resid, NULL);
282 	if (error)
283 		goto bad2;
284 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
285 
286 	/*
287 	 * Set up default address space limits.  Can be overridden
288 	 * by individual exec packages.
289 	 *
290 	 * XXX probably should be all done in the exec pakages.
291 	 */
292 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
293 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
294 	/*
295 	 * set up the vmcmds for creation of the process
296 	 * address space
297 	 */
298 	error = ENOEXEC;
299 	for (i = 0; i < nexecs && error != 0; i++) {
300 		int newerror;
301 
302 		epp->ep_esch = execsw[i];
303 		newerror = (*execsw[i]->es_makecmds)(p, epp);
304 		/* make sure the first "interesting" error code is saved. */
305 		if (!newerror || error == ENOEXEC)
306 			error = newerror;
307 
308 		/* if es_makecmds call was successful, update epp->ep_es */
309 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
310 			epp->ep_es = execsw[i];
311 
312 		if (epp->ep_flags & EXEC_DESTR && error != 0)
313 			return error;
314 	}
315 	if (!error) {
316 		/* check that entry point is sane */
317 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
318 			error = ENOEXEC;
319 
320 		/* check limits */
321 		if ((epp->ep_tsize > MAXTSIZ) ||
322 		    (epp->ep_dsize >
323 		     (u_quad_t)p->p_rlimit[RLIMIT_DATA].rlim_cur))
324 			error = ENOMEM;
325 
326 		if (!error)
327 			return (0);
328 	}
329 
330 	/*
331 	 * free any vmspace-creation commands,
332 	 * and release their references
333 	 */
334 	kill_vmcmds(&epp->ep_vmcmds);
335 
336 bad2:
337 	/*
338 	 * close and release the vnode, restore the old one, free the
339 	 * pathname buf, and punt.
340 	 */
341 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
342 	VOP_CLOSE(vp, FREAD, p->p_ucred, p);
343 	vput(vp);
344 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
345 	return error;
346 
347 bad1:
348 	/*
349 	 * free the namei pathname buffer, and put the vnode
350 	 * (which we don't yet have open).
351 	 */
352 	vput(vp);				/* was still locked */
353 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
354 	return error;
355 }
356 
357 #ifdef __MACHINE_STACK_GROWS_UP
358 #define STACK_PTHREADSPACE NBPG
359 #else
360 #define STACK_PTHREADSPACE 0
361 #endif
362 
363 static int
364 execve_fetch_element(char * const *array, size_t index, char **value)
365 {
366 	return copyin(array + index, value, sizeof(*value));
367 }
368 
369 /*
370  * exec system call
371  */
372 /* ARGSUSED */
373 int
374 sys_execve(struct lwp *l, void *v, register_t *retval)
375 {
376 	struct sys_execve_args /* {
377 		syscallarg(const char *)	path;
378 		syscallarg(char * const *)	argp;
379 		syscallarg(char * const *)	envp;
380 	} */ *uap = v;
381 
382 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
383 	    SCARG(uap, envp), execve_fetch_element);
384 }
385 
386 int
387 execve1(struct lwp *l, const char *path, char * const *args,
388     char * const *envs, execve_fetch_element_t fetch_element)
389 {
390 	int			error;
391 	u_int			i;
392 	struct exec_package	pack;
393 	struct nameidata	nid;
394 	struct vattr		attr;
395 	struct proc		*p;
396 	struct ucred		*cred;
397 	char			*argp;
398 	char			*dp, *sp;
399 	long			argc, envc;
400 	size_t			len;
401 	char			*stack;
402 	struct ps_strings	arginfo;
403 	struct vmspace		*vm;
404 	char			**tmpfap;
405 	int			szsigcode;
406 	struct exec_vmcmd	*base_vcp;
407 	int			oldlwpflags;
408 #ifdef SYSTRACE
409 	int			wassugid = ISSET(p->p_flag, P_SUGID);
410 	char			pathbuf[MAXPATHLEN];
411 	size_t			pathbuflen;
412 #endif /* SYSTRACE */
413 
414 	/* Disable scheduler activation upcalls. */
415 	oldlwpflags = l->l_flag & (L_SA | L_SA_UPCALL);
416 	if (l->l_flag & L_SA)
417 		l->l_flag &= ~(L_SA | L_SA_UPCALL);
418 
419 	p = l->l_proc;
420 	/*
421 	 * Lock the process and set the P_INEXEC flag to indicate that
422 	 * it should be left alone until we're done here.  This is
423 	 * necessary to avoid race conditions - e.g. in ptrace() -
424 	 * that might allow a local user to illicitly obtain elevated
425 	 * privileges.
426 	 */
427 	p->p_flag |= P_INEXEC;
428 
429 	cred = p->p_ucred;
430 	base_vcp = NULL;
431 	/*
432 	 * Init the namei data to point the file user's program name.
433 	 * This is done here rather than in check_exec(), so that it's
434 	 * possible to override this settings if any of makecmd/probe
435 	 * functions call check_exec() recursively - for example,
436 	 * see exec_script_makecmds().
437 	 */
438 #ifdef SYSTRACE
439 	if (ISSET(p->p_flag, P_SYSTRACE))
440 		systrace_execve0(p);
441 
442 	error = copyinstr(path, pathbuf, sizeof(pathbuf),
443 			  &pathbuflen);
444 	if (error)
445 		goto clrflg;
446 
447 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_SYSSPACE, pathbuf, p);
448 #else
449 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, path, p);
450 #endif /* SYSTRACE */
451 
452 	/*
453 	 * initialize the fields of the exec package.
454 	 */
455 #ifdef SYSTRACE
456 	pack.ep_name = pathbuf;
457 #else
458 	pack.ep_name = path;
459 #endif /* SYSTRACE */
460 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
461 	pack.ep_hdrlen = exec_maxhdrsz;
462 	pack.ep_hdrvalid = 0;
463 	pack.ep_ndp = &nid;
464 	pack.ep_emul_arg = NULL;
465 	pack.ep_vmcmds.evs_cnt = 0;
466 	pack.ep_vmcmds.evs_used = 0;
467 	pack.ep_vap = &attr;
468 	pack.ep_flags = 0;
469 
470 #ifdef LKM
471 	lockmgr(&exec_lock, LK_SHARED, NULL);
472 #endif
473 
474 	/* see if we can run it. */
475 #ifdef VERIFIED_EXEC
476         if ((error = check_exec(p, &pack, VERIEXEC_DIRECT)) != 0)
477 #else
478         if ((error = check_exec(p, &pack, 0)) != 0)
479 #endif
480 		goto freehdr;
481 
482 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
483 
484 	/* allocate an argument buffer */
485 	argp = (char *) uvm_km_alloc(exec_map, NCARGS, 0,
486 	    UVM_KMF_PAGEABLE|UVM_KMF_WAITVA);
487 #ifdef DIAGNOSTIC
488 	if (argp == (vaddr_t) 0)
489 		panic("execve: argp == NULL");
490 #endif
491 	dp = argp;
492 	argc = 0;
493 
494 	/* copy the fake args list, if there's one, freeing it as we go */
495 	if (pack.ep_flags & EXEC_HASARGL) {
496 		tmpfap = pack.ep_fa;
497 		while (*tmpfap != NULL) {
498 			char *cp;
499 
500 			cp = *tmpfap;
501 			while (*cp)
502 				*dp++ = *cp++;
503 			dp++;
504 
505 			FREE(*tmpfap, M_EXEC);
506 			tmpfap++; argc++;
507 		}
508 		FREE(pack.ep_fa, M_EXEC);
509 		pack.ep_flags &= ~EXEC_HASARGL;
510 	}
511 
512 	/* Now get argv & environment */
513 	if (args == NULL) {
514 		error = EINVAL;
515 		goto bad;
516 	}
517 	/* 'i' will index the argp/envp element to be retrieved */
518 	i = 0;
519 	if (pack.ep_flags & EXEC_SKIPARG)
520 		i++;
521 
522 	while (1) {
523 		len = argp + ARG_MAX - dp;
524 		if ((error = (*fetch_element)(args, i, &sp)) != 0)
525 			goto bad;
526 		if (!sp)
527 			break;
528 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
529 			if (error == ENAMETOOLONG)
530 				error = E2BIG;
531 			goto bad;
532 		}
533 #ifdef KTRACE
534 		if (KTRPOINT(p, KTR_EXEC_ARG))
535 			ktrkmem(p, KTR_EXEC_ARG, dp, len - 1);
536 #endif
537 		dp += len;
538 		i++;
539 		argc++;
540 	}
541 
542 	envc = 0;
543 	/* environment need not be there */
544 	if (envs != NULL) {
545 		i = 0;
546 		while (1) {
547 			len = argp + ARG_MAX - dp;
548 			if ((error = (*fetch_element)(envs, i, &sp)) != 0)
549 				goto bad;
550 			if (!sp)
551 				break;
552 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
553 				if (error == ENAMETOOLONG)
554 					error = E2BIG;
555 				goto bad;
556 			}
557 #ifdef KTRACE
558 			if (KTRPOINT(p, KTR_EXEC_ENV))
559 				ktrkmem(p, KTR_EXEC_ENV, dp, len - 1);
560 #endif
561 			dp += len;
562 			i++;
563 			envc++;
564 		}
565 	}
566 
567 	dp = (char *) ALIGN(dp);
568 
569 	szsigcode = pack.ep_es->es_emul->e_esigcode -
570 	    pack.ep_es->es_emul->e_sigcode;
571 
572 	/* Now check if args & environ fit into new stack */
573 	if (pack.ep_flags & EXEC_32)
574 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
575 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
576 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
577 		    - argp;
578 	else
579 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
580 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
581 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
582 		    - argp;
583 
584 	len = ALIGN(len);	/* make the stack "safely" aligned */
585 
586 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
587 		error = ENOMEM;
588 		goto bad;
589 	}
590 
591 	/* Get rid of other LWPs/ */
592 	p->p_flag |= P_WEXIT; /* XXX hack. lwp-exit stuff wants to see it. */
593 	exit_lwps(l);
594 	p->p_flag &= ~P_WEXIT;
595 	KDASSERT(p->p_nlwps == 1);
596 
597 	/* This is now LWP 1 */
598 	l->l_lid = 1;
599 	p->p_nlwpid = 1;
600 
601 	/* Release any SA state. */
602 	if (p->p_sa)
603 		sa_release(p);
604 
605 	/* Remove POSIX timers */
606 	timers_free(p, TIMERS_POSIX);
607 
608 	/* adjust "active stack depth" for process VSZ */
609 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
610 
611 	/*
612 	 * Do whatever is necessary to prepare the address space
613 	 * for remapping.  Note that this might replace the current
614 	 * vmspace with another!
615 	 */
616 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
617 
618 	/* record proc's vnode, for use by procfs and others */
619         if (p->p_textvp)
620                 vrele(p->p_textvp);
621 	VREF(pack.ep_vp);
622 	p->p_textvp = pack.ep_vp;
623 
624 	/* Now map address space */
625 	vm = p->p_vmspace;
626 	vm->vm_taddr = (caddr_t) pack.ep_taddr;
627 	vm->vm_tsize = btoc(pack.ep_tsize);
628 	vm->vm_daddr = (caddr_t) pack.ep_daddr;
629 	vm->vm_dsize = btoc(pack.ep_dsize);
630 	vm->vm_ssize = btoc(pack.ep_ssize);
631 	vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr;
632 	vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr;
633 
634 	/* create the new process's VM space by running the vmcmds */
635 #ifdef DIAGNOSTIC
636 	if (pack.ep_vmcmds.evs_used == 0)
637 		panic("execve: no vmcmds");
638 #endif
639 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
640 		struct exec_vmcmd *vcp;
641 
642 		vcp = &pack.ep_vmcmds.evs_cmds[i];
643 		if (vcp->ev_flags & VMCMD_RELATIVE) {
644 #ifdef DIAGNOSTIC
645 			if (base_vcp == NULL)
646 				panic("execve: relative vmcmd with no base");
647 			if (vcp->ev_flags & VMCMD_BASE)
648 				panic("execve: illegal base & relative vmcmd");
649 #endif
650 			vcp->ev_addr += base_vcp->ev_addr;
651 		}
652 		error = (*vcp->ev_proc)(p, vcp);
653 #ifdef DEBUG_EXEC
654 		if (error) {
655 			int j;
656 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
657 			for (j = 0; j <= i; j++)
658 				uprintf(
659 			    "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
660 				    j, vp[j].ev_addr, vp[j].ev_len,
661 				    vp[j].ev_offset, vp[j].ev_prot,
662 				    vp[j].ev_flags);
663 		}
664 #endif /* DEBUG_EXEC */
665 		if (vcp->ev_flags & VMCMD_BASE)
666 			base_vcp = vcp;
667 	}
668 
669 	/* free the vmspace-creation commands, and release their references */
670 	kill_vmcmds(&pack.ep_vmcmds);
671 
672 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
673 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
674 	vput(pack.ep_vp);
675 
676 	/* if an error happened, deallocate and punt */
677 	if (error) {
678 		DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error));
679 		goto exec_abort;
680 	}
681 
682 	/* remember information about the process */
683 	arginfo.ps_nargvstr = argc;
684 	arginfo.ps_nenvstr = envc;
685 
686 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
687 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
688 		len - (sizeof(struct ps_strings) + szsigcode));
689 #ifdef __MACHINE_STACK_GROWS_UP
690 	/*
691 	 * The copyargs call always copies into lower addresses
692 	 * first, moving towards higher addresses, starting with
693 	 * the stack pointer that we give.  When the stack grows
694 	 * down, this puts argc/argv/envp very shallow on the
695 	 * stack, right at the first user stack pointer, and puts
696 	 * STACKGAPLEN very deep in the stack.  When the stack
697 	 * grows up, the situation is reversed.
698 	 *
699 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
700 	 * function expects to be called with a single pointer to
701 	 * a region that has a few words it can stash values into,
702 	 * followed by argc/argv/envp.  When the stack grows down,
703 	 * it's easy to decrement the stack pointer a little bit to
704 	 * allocate the space for these few words and pass the new
705 	 * stack pointer to _rtld.  When the stack grows up, however,
706 	 * a few words before argc is part of the signal trampoline, XXX
707 	 * so we have a problem.
708 	 *
709 	 * Instead of changing how _rtld works, we take the easy way
710 	 * out and steal 32 bytes before we call copyargs.  This
711 	 * space is effectively stolen from STACKGAPLEN.
712 	 */
713 	stack += 32;
714 #endif /* __MACHINE_STACK_GROWS_UP */
715 
716 	/* Now copy argc, args & environ to new stack */
717 	error = (*pack.ep_es->es_copyargs)(p, &pack, &arginfo, &stack, argp);
718 	if (error) {
719 		DPRINTF(("execve: copyargs failed %d\n", error));
720 		goto exec_abort;
721 	}
722 	/* Move the stack back to original point */
723 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
724 
725 	/* fill process ps_strings info */
726 	p->p_psstr = (struct ps_strings *)
727 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
728 	    sizeof(struct ps_strings));
729 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
730 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
731 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
732 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
733 
734 	/* copy out the process's ps_strings structure */
735 	if ((error = copyout(&arginfo, (char *)p->p_psstr,
736 	    sizeof(arginfo))) != 0) {
737 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
738 		       &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo)));
739 		goto exec_abort;
740 	}
741 
742 	stopprofclock(p);	/* stop profiling */
743 	fdcloseexec(p);		/* handle close on exec */
744 	execsigs(p);		/* reset catched signals */
745 
746 	l->l_ctxlink = NULL;	/* reset ucontext link */
747 
748 	/* set command name & other accounting info */
749 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
750 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
751 	p->p_comm[len] = 0;
752 	p->p_acflag &= ~AFORK;
753 
754 	p->p_flag |= P_EXEC;
755 	if (p->p_flag & P_PPWAIT) {
756 		p->p_flag &= ~P_PPWAIT;
757 		wakeup((caddr_t) p->p_pptr);
758 	}
759 
760 	/*
761 	 * deal with set[ug]id.
762 	 * MNT_NOSUID has already been used to disable s[ug]id.
763 	 */
764 	if ((p->p_flag & P_TRACED) == 0 &&
765 
766 	    (((attr.va_mode & S_ISUID) != 0 &&
767 	      p->p_ucred->cr_uid != attr.va_uid) ||
768 
769 	     ((attr.va_mode & S_ISGID) != 0 &&
770 	      p->p_ucred->cr_gid != attr.va_gid))) {
771 		/*
772 		 * Mark the process as SUGID before we do
773 		 * anything that might block.
774 		 */
775 		p_sugid(p);
776 
777 		/* Make sure file descriptors 0..2 are in use. */
778 		if ((error = fdcheckstd(p)) != 0)
779 			goto exec_abort;
780 
781 		p->p_ucred = crcopy(cred);
782 #ifdef KTRACE
783 		/*
784 		 * If process is being ktraced, turn off - unless
785 		 * root set it.
786 		 */
787 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
788 			ktrderef(p);
789 #endif
790 		if (attr.va_mode & S_ISUID)
791 			p->p_ucred->cr_uid = attr.va_uid;
792 		if (attr.va_mode & S_ISGID)
793 			p->p_ucred->cr_gid = attr.va_gid;
794 	} else
795 		p->p_flag &= ~P_SUGID;
796 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
797 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
798 
799 #if defined(__HAVE_RAS)
800 	/*
801 	 * Remove all RASs from the address space.
802 	 */
803 	ras_purgeall(p);
804 #endif
805 
806 	doexechooks(p);
807 
808 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
809 
810 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
811 
812 	/* notify others that we exec'd */
813 	KNOTE(&p->p_klist, NOTE_EXEC);
814 
815 	/* setup new registers and do misc. setup. */
816 	(*pack.ep_es->es_emul->e_setregs)(l, &pack, (u_long) stack);
817 	if (pack.ep_es->es_setregs)
818 		(*pack.ep_es->es_setregs)(l, &pack, (u_long) stack);
819 
820 	/* map the process's signal trampoline code */
821 	if (exec_sigcode_map(p, pack.ep_es->es_emul))
822 		goto exec_abort;
823 
824 	if (p->p_flag & P_TRACED)
825 		psignal(p, SIGTRAP);
826 
827 	free(pack.ep_hdr, M_EXEC);
828 
829 	/*
830 	 * Call emulation specific exec hook. This can setup per-process
831 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
832 	 *
833 	 * If we are executing process of different emulation than the
834 	 * original forked process, call e_proc_exit() of the old emulation
835 	 * first, then e_proc_exec() of new emulation. If the emulation is
836 	 * same, the exec hook code should deallocate any old emulation
837 	 * resources held previously by this process.
838 	 */
839 	if (p->p_emul && p->p_emul->e_proc_exit
840 	    && p->p_emul != pack.ep_es->es_emul)
841 		(*p->p_emul->e_proc_exit)(p);
842 
843 	/*
844 	 * Call exec hook. Emulation code may NOT store reference to anything
845 	 * from &pack.
846 	 */
847         if (pack.ep_es->es_emul->e_proc_exec)
848                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
849 
850 	/* update p_emul, the old value is no longer needed */
851 	p->p_emul = pack.ep_es->es_emul;
852 
853 	/* ...and the same for p_execsw */
854 	p->p_execsw = pack.ep_es;
855 
856 #ifdef __HAVE_SYSCALL_INTERN
857 	(*p->p_emul->e_syscall_intern)(p);
858 #endif
859 #ifdef KTRACE
860 	if (KTRPOINT(p, KTR_EMUL))
861 		ktremul(p);
862 #endif
863 
864 #ifdef LKM
865 	lockmgr(&exec_lock, LK_RELEASE, NULL);
866 #endif
867 	p->p_flag &= ~P_INEXEC;
868 
869 	if (p->p_flag & P_STOPEXEC) {
870 		int s;
871 
872 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
873 		SCHED_LOCK(s);
874 		p->p_pptr->p_nstopchild++;
875 		p->p_stat = SSTOP;
876 		l->l_stat = LSSTOP;
877 		p->p_nrlwps--;
878 		mi_switch(l, NULL);
879 		SCHED_ASSERT_UNLOCKED();
880 		splx(s);
881 	}
882 
883 #ifdef SYSTRACE
884 	if (ISSET(p->p_flag, P_SYSTRACE) &&
885 	    wassugid && !ISSET(p->p_flag, P_SUGID))
886 		systrace_execve1(pathbuf, p);
887 #endif /* SYSTRACE */
888 
889 	return (EJUSTRETURN);
890 
891  bad:
892 	p->p_flag &= ~P_INEXEC;
893 	/* free the vmspace-creation commands, and release their references */
894 	kill_vmcmds(&pack.ep_vmcmds);
895 	/* kill any opened file descriptor, if necessary */
896 	if (pack.ep_flags & EXEC_HASFD) {
897 		pack.ep_flags &= ~EXEC_HASFD;
898 		(void) fdrelease(p, pack.ep_fd);
899 	}
900 	/* close and put the exec'd file */
901 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
902 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
903 	vput(pack.ep_vp);
904 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
905 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
906 
907  freehdr:
908 	free(pack.ep_hdr, M_EXEC);
909 
910 #ifdef SYSTRACE
911  clrflg:
912 #endif /* SYSTRACE */
913 	l->l_flag |= oldlwpflags;
914 	p->p_flag &= ~P_INEXEC;
915 #ifdef LKM
916 	lockmgr(&exec_lock, LK_RELEASE, NULL);
917 #endif
918 
919 	return error;
920 
921  exec_abort:
922 	p->p_flag &= ~P_INEXEC;
923 #ifdef LKM
924 	lockmgr(&exec_lock, LK_RELEASE, NULL);
925 #endif
926 
927 	/*
928 	 * the old process doesn't exist anymore.  exit gracefully.
929 	 * get rid of the (new) address space we have created, if any, get rid
930 	 * of our namei data and vnode, and exit noting failure
931 	 */
932 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
933 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
934 	if (pack.ep_emul_arg)
935 		FREE(pack.ep_emul_arg, M_TEMP);
936 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
937 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
938 	free(pack.ep_hdr, M_EXEC);
939 	exit1(l, W_EXITCODE(error, SIGABRT));
940 
941 	/* NOTREACHED */
942 	return 0;
943 }
944 
945 
946 int
947 copyargs(struct proc *p, struct exec_package *pack, struct ps_strings *arginfo,
948     char **stackp, void *argp)
949 {
950 	char	**cpp, *dp, *sp;
951 	size_t	len;
952 	void	*nullp;
953 	long	argc, envc;
954 	int	error;
955 
956 	cpp = (char **)*stackp;
957 	nullp = NULL;
958 	argc = arginfo->ps_nargvstr;
959 	envc = arginfo->ps_nenvstr;
960 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
961 		return error;
962 
963 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
964 	sp = argp;
965 
966 	/* XXX don't copy them out, remap them! */
967 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
968 
969 	for (; --argc >= 0; sp += len, dp += len)
970 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
971 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
972 			return error;
973 
974 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
975 		return error;
976 
977 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
978 
979 	for (; --envc >= 0; sp += len, dp += len)
980 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
981 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
982 			return error;
983 
984 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
985 		return error;
986 
987 	*stackp = (char *)cpp;
988 	return 0;
989 }
990 
991 #ifdef LKM
992 /*
993  * Find an emulation of given name in list of emulations.
994  * Needs to be called with the exec_lock held.
995  */
996 const struct emul *
997 emul_search(const char *name)
998 {
999 	struct emul_entry *it;
1000 
1001 	LIST_FOREACH(it, &el_head, el_list) {
1002 		if (strcmp(name, it->el_emul->e_name) == 0)
1003 			return it->el_emul;
1004 	}
1005 
1006 	return NULL;
1007 }
1008 
1009 /*
1010  * Add an emulation to list, if it's not there already.
1011  */
1012 int
1013 emul_register(const struct emul *emul, int ro_entry)
1014 {
1015 	struct emul_entry	*ee;
1016 	int			error;
1017 
1018 	error = 0;
1019 	lockmgr(&exec_lock, LK_SHARED, NULL);
1020 
1021 	if (emul_search(emul->e_name)) {
1022 		error = EEXIST;
1023 		goto out;
1024 	}
1025 
1026 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
1027 		M_EXEC, M_WAITOK);
1028 	ee->el_emul = emul;
1029 	ee->ro_entry = ro_entry;
1030 	LIST_INSERT_HEAD(&el_head, ee, el_list);
1031 
1032  out:
1033 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1034 	return error;
1035 }
1036 
1037 /*
1038  * Remove emulation with name 'name' from list of supported emulations.
1039  */
1040 int
1041 emul_unregister(const char *name)
1042 {
1043 	const struct proclist_desc *pd;
1044 	struct emul_entry	*it;
1045 	int			i, error;
1046 	struct proc		*ptmp;
1047 
1048 	error = 0;
1049 	lockmgr(&exec_lock, LK_SHARED, NULL);
1050 
1051 	LIST_FOREACH(it, &el_head, el_list) {
1052 		if (strcmp(it->el_emul->e_name, name) == 0)
1053 			break;
1054 	}
1055 
1056 	if (!it) {
1057 		error = ENOENT;
1058 		goto out;
1059 	}
1060 
1061 	if (it->ro_entry) {
1062 		error = EBUSY;
1063 		goto out;
1064 	}
1065 
1066 	/* test if any execw[] entry is still using this */
1067 	for(i=0; i < nexecs; i++) {
1068 		if (execsw[i]->es_emul == it->el_emul) {
1069 			error = EBUSY;
1070 			goto out;
1071 		}
1072 	}
1073 
1074 	/*
1075 	 * Test if any process is running under this emulation - since
1076 	 * emul_unregister() is running quite sendomly, it's better
1077 	 * to do expensive check here than to use any locking.
1078 	 */
1079 	proclist_lock_read();
1080 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
1081 		PROCLIST_FOREACH(ptmp, pd->pd_list) {
1082 			if (ptmp->p_emul == it->el_emul) {
1083 				error = EBUSY;
1084 				break;
1085 			}
1086 		}
1087 	}
1088 	proclist_unlock_read();
1089 
1090 	if (error)
1091 		goto out;
1092 
1093 
1094 	/* entry is not used, remove it */
1095 	LIST_REMOVE(it, el_list);
1096 	FREE(it, M_EXEC);
1097 
1098  out:
1099 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1100 	return error;
1101 }
1102 
1103 /*
1104  * Add execsw[] entry.
1105  */
1106 int
1107 exec_add(struct execsw *esp, const char *e_name)
1108 {
1109 	struct exec_entry	*it;
1110 	int			error;
1111 
1112 	error = 0;
1113 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
1114 
1115 	if (!esp->es_emul) {
1116 		esp->es_emul = emul_search(e_name);
1117 		if (!esp->es_emul) {
1118 			error = ENOENT;
1119 			goto out;
1120 		}
1121 	}
1122 
1123 	LIST_FOREACH(it, &ex_head, ex_list) {
1124 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1125 		if (it->es->es_makecmds == esp->es_makecmds
1126 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1127 		    && it->es->es_emul == esp->es_emul) {
1128 			error = EEXIST;
1129 			goto out;
1130 		}
1131 	}
1132 
1133 	/* if we got here, the entry doesn't exist yet */
1134 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
1135 		M_EXEC, M_WAITOK);
1136 	it->es = esp;
1137 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
1138 
1139 	/* update execsw[] */
1140 	exec_init(0);
1141 
1142  out:
1143 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1144 	return error;
1145 }
1146 
1147 /*
1148  * Remove execsw[] entry.
1149  */
1150 int
1151 exec_remove(const struct execsw *esp)
1152 {
1153 	struct exec_entry	*it;
1154 	int			error;
1155 
1156 	error = 0;
1157 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
1158 
1159 	LIST_FOREACH(it, &ex_head, ex_list) {
1160 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1161 		if (it->es->es_makecmds == esp->es_makecmds
1162 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1163 		    && it->es->es_emul == esp->es_emul)
1164 			break;
1165 	}
1166 	if (!it) {
1167 		error = ENOENT;
1168 		goto out;
1169 	}
1170 
1171 	/* remove item from list and free resources */
1172 	LIST_REMOVE(it, ex_list);
1173 	FREE(it, M_EXEC);
1174 
1175 	/* update execsw[] */
1176 	exec_init(0);
1177 
1178  out:
1179 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1180 	return error;
1181 }
1182 
1183 static void
1184 link_es(struct execsw_entry **listp, const struct execsw *esp)
1185 {
1186 	struct execsw_entry *et, *e1;
1187 
1188 	MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry),
1189 			M_TEMP, M_WAITOK);
1190 	et->next = NULL;
1191 	et->es = esp;
1192 	if (*listp == NULL) {
1193 		*listp = et;
1194 		return;
1195 	}
1196 
1197 	switch(et->es->es_prio) {
1198 	case EXECSW_PRIO_FIRST:
1199 		/* put new entry as the first */
1200 		et->next = *listp;
1201 		*listp = et;
1202 		break;
1203 	case EXECSW_PRIO_ANY:
1204 		/* put new entry after all *_FIRST and *_ANY entries */
1205 		for(e1 = *listp; e1->next
1206 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
1207 			e1 = e1->next);
1208 		et->next = e1->next;
1209 		e1->next = et;
1210 		break;
1211 	case EXECSW_PRIO_LAST:
1212 		/* put new entry as the last one */
1213 		for(e1 = *listp; e1->next; e1 = e1->next);
1214 		e1->next = et;
1215 		break;
1216 	default:
1217 #ifdef DIAGNOSTIC
1218 		panic("execw[] entry with unknown priority %d found",
1219 			et->es->es_prio);
1220 #endif
1221 		break;
1222 	}
1223 }
1224 
1225 /*
1226  * Initialize exec structures. If init_boot is true, also does necessary
1227  * one-time initialization (it's called from main() that way).
1228  * Once system is multiuser, this should be called with exec_lock held,
1229  * i.e. via exec_{add|remove}().
1230  */
1231 int
1232 exec_init(int init_boot)
1233 {
1234 	const struct execsw	**new_es, * const *old_es;
1235 	struct execsw_entry	*list, *e1;
1236 	struct exec_entry	*e2;
1237 	int			i, es_sz;
1238 
1239 	if (init_boot) {
1240 		/* do one-time initializations */
1241 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1242 
1243 		/* register compiled-in emulations */
1244 		for(i=0; i < nexecs_builtin; i++) {
1245 			if (execsw_builtin[i].es_emul)
1246 				emul_register(execsw_builtin[i].es_emul, 1);
1247 		}
1248 #ifdef DIAGNOSTIC
1249 		if (i == 0)
1250 			panic("no emulations found in execsw_builtin[]");
1251 #endif
1252 	}
1253 
1254 	/*
1255 	 * Build execsw[] array from builtin entries and entries added
1256 	 * at runtime.
1257 	 */
1258 	list = NULL;
1259 	for(i=0; i < nexecs_builtin; i++)
1260 		link_es(&list, &execsw_builtin[i]);
1261 
1262 	/* Add dynamically loaded entries */
1263 	es_sz = nexecs_builtin;
1264 	LIST_FOREACH(e2, &ex_head, ex_list) {
1265 		link_es(&list, e2->es);
1266 		es_sz++;
1267 	}
1268 
1269 	/*
1270 	 * Now that we have sorted all execw entries, create new execsw[]
1271 	 * and free no longer needed memory in the process.
1272 	 */
1273 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
1274 	for(i=0; list; i++) {
1275 		new_es[i] = list->es;
1276 		e1 = list->next;
1277 		FREE(list, M_TEMP);
1278 		list = e1;
1279 	}
1280 
1281 	/*
1282 	 * New execsw[] array built, now replace old execsw[] and free
1283 	 * used memory.
1284 	 */
1285 	old_es = execsw;
1286 	execsw = new_es;
1287 	nexecs = es_sz;
1288 	if (old_es)
1289 		/*XXXUNCONST*/
1290 		free(__UNCONST(old_es), M_EXEC);
1291 
1292 	/*
1293 	 * Figure out the maximum size of an exec header.
1294 	 */
1295 	exec_maxhdrsz = 0;
1296 	for (i = 0; i < nexecs; i++) {
1297 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1298 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1299 	}
1300 
1301 	return 0;
1302 }
1303 #endif
1304 
1305 #ifndef LKM
1306 /*
1307  * Simplified exec_init() for kernels without LKMs. Only initialize
1308  * exec_maxhdrsz and execsw[].
1309  */
1310 int
1311 exec_init(int init_boot)
1312 {
1313 	int i;
1314 
1315 #ifdef DIAGNOSTIC
1316 	if (!init_boot)
1317 		panic("exec_init(): called with init_boot == 0");
1318 #endif
1319 
1320 	/* do one-time initializations */
1321 	nexecs = nexecs_builtin;
1322 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
1323 
1324 	/*
1325 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1326 	 */
1327 	exec_maxhdrsz = 0;
1328 	for(i=0; i < nexecs; i++) {
1329 		execsw[i] = &execsw_builtin[i];
1330 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1331 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1332 	}
1333 
1334 	return 0;
1335 
1336 }
1337 #endif /* !LKM */
1338 
1339 static int
1340 exec_sigcode_map(struct proc *p, const struct emul *e)
1341 {
1342 	vaddr_t va;
1343 	vsize_t sz;
1344 	int error;
1345 	struct uvm_object *uobj;
1346 
1347 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1348 
1349 	if (e->e_sigobject == NULL || sz == 0) {
1350 		return 0;
1351 	}
1352 
1353 	/*
1354 	 * If we don't have a sigobject for this emulation, create one.
1355 	 *
1356 	 * sigobject is an anonymous memory object (just like SYSV shared
1357 	 * memory) that we keep a permanent reference to and that we map
1358 	 * in all processes that need this sigcode. The creation is simple,
1359 	 * we create an object, add a permanent reference to it, map it in
1360 	 * kernel space, copy out the sigcode to it and unmap it.
1361 	 * We map it with PROT_READ|PROT_EXEC into the process just
1362 	 * the way sys_mmap() would map it.
1363 	 */
1364 
1365 	uobj = *e->e_sigobject;
1366 	if (uobj == NULL) {
1367 		uobj = uao_create(sz, 0);
1368 		(*uobj->pgops->pgo_reference)(uobj);
1369 		va = vm_map_min(kernel_map);
1370 		if ((error = uvm_map(kernel_map, &va, round_page(sz),
1371 		    uobj, 0, 0,
1372 		    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1373 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1374 			printf("kernel mapping failed %d\n", error);
1375 			(*uobj->pgops->pgo_detach)(uobj);
1376 			return (error);
1377 		}
1378 		memcpy((void *)va, e->e_sigcode, sz);
1379 #ifdef PMAP_NEED_PROCWR
1380 		pmap_procwr(&proc0, va, sz);
1381 #endif
1382 		uvm_unmap(kernel_map, va, va + round_page(sz));
1383 		*e->e_sigobject = uobj;
1384 	}
1385 
1386 	/* Just a hint to uvm_map where to put it. */
1387 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1388 	    round_page(sz));
1389 
1390 #ifdef __alpha__
1391 	/*
1392 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1393 	 * which causes the above calculation to put the sigcode at
1394 	 * an invalid address.  Put it just below the text instead.
1395 	 */
1396 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1397 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1398 	}
1399 #endif
1400 
1401 	(*uobj->pgops->pgo_reference)(uobj);
1402 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1403 			uobj, 0, 0,
1404 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1405 				    UVM_ADV_RANDOM, 0));
1406 	if (error) {
1407 		(*uobj->pgops->pgo_detach)(uobj);
1408 		return (error);
1409 	}
1410 	p->p_sigctx.ps_sigcode = (void *)va;
1411 	return (0);
1412 }
1413