xref: /netbsd-src/sys/kern/kern_exec.c (revision eb961d0e02b7a46a9acfa877b02df48df6637278)
1 /*	$NetBSD: kern_exec.c,v 1.214 2006/03/05 07:21:38 christos 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 1.214 2006/03/05 07:21:38 christos 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  *	lwp pointer of exec'ing lwp
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 lwp *l, struct exec_package *epp, int flag)
230 {
231 	int		error, i;
232 	struct vnode	*vp;
233 	struct nameidata *ndp;
234 	size_t		resid;
235 	struct proc	*p;
236 
237 	p = l->l_proc;
238 	ndp = epp->ep_ndp;
239 	ndp->ni_cnd.cn_nameiop = LOOKUP;
240 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
241 	/* first get the vnode */
242 	if ((error = namei(ndp)) != 0)
243 		return error;
244 	epp->ep_vp = vp = ndp->ni_vp;
245 
246 	/* check access and type */
247 	if (vp->v_type != VREG) {
248 		error = EACCES;
249 		goto bad1;
250 	}
251 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, l)) != 0)
252 		goto bad1;
253 
254 	/* get attributes */
255 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, l)) != 0)
256 		goto bad1;
257 
258 	/* Check mount point */
259 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
260 		error = EACCES;
261 		goto bad1;
262 	}
263 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
264 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
265 
266 	/* try to open it */
267 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, l)) != 0)
268 		goto bad1;
269 
270 	/* unlock vp, since we need it unlocked from here on out. */
271 	VOP_UNLOCK(vp, 0);
272 
273 
274 #ifdef VERIFIED_EXEC
275         if ((error = veriexec_verify(l, vp, epp->ep_vap, epp->ep_ndp->ni_dirp,
276 				     flag, NULL)) != 0)
277                 goto bad2;
278 #endif
279 
280 	/* now we have the file, get the exec header */
281 	uvn_attach(vp, VM_PROT_READ);
282 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
283 			UIO_SYSSPACE, 0, p->p_ucred, &resid, NULL);
284 	if (error)
285 		goto bad2;
286 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
287 
288 	/*
289 	 * Set up default address space limits.  Can be overridden
290 	 * by individual exec packages.
291 	 *
292 	 * XXX probably should be all done in the exec pakages.
293 	 */
294 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
295 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
296 	/*
297 	 * set up the vmcmds for creation of the process
298 	 * address space
299 	 */
300 	error = ENOEXEC;
301 	for (i = 0; i < nexecs && error != 0; i++) {
302 		int newerror;
303 
304 		epp->ep_esch = execsw[i];
305 		newerror = (*execsw[i]->es_makecmds)(l, epp);
306 		/* make sure the first "interesting" error code is saved. */
307 		if (!newerror || error == ENOEXEC)
308 			error = newerror;
309 
310 		/* if es_makecmds call was successful, update epp->ep_es */
311 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
312 			epp->ep_es = execsw[i];
313 
314 		if (epp->ep_flags & EXEC_DESTR && error != 0)
315 			return error;
316 	}
317 	if (!error) {
318 		/* check that entry point is sane */
319 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
320 			error = ENOEXEC;
321 
322 		/* check limits */
323 		if ((epp->ep_tsize > MAXTSIZ) ||
324 		    (epp->ep_dsize >
325 		     (u_quad_t)p->p_rlimit[RLIMIT_DATA].rlim_cur))
326 			error = ENOMEM;
327 
328 		if (!error)
329 			return (0);
330 	}
331 
332 	/*
333 	 * free any vmspace-creation commands,
334 	 * and release their references
335 	 */
336 	kill_vmcmds(&epp->ep_vmcmds);
337 
338 bad2:
339 	/*
340 	 * close and release the vnode, restore the old one, free the
341 	 * pathname buf, and punt.
342 	 */
343 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
344 	VOP_CLOSE(vp, FREAD, p->p_ucred, l);
345 	vput(vp);
346 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
347 	return error;
348 
349 bad1:
350 	/*
351 	 * free the namei pathname buffer, and put the vnode
352 	 * (which we don't yet have open).
353 	 */
354 	vput(vp);				/* was still locked */
355 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
356 	return error;
357 }
358 
359 #ifdef __MACHINE_STACK_GROWS_UP
360 #define STACK_PTHREADSPACE NBPG
361 #else
362 #define STACK_PTHREADSPACE 0
363 #endif
364 
365 static int
366 execve_fetch_element(char * const *array, size_t index, char **value)
367 {
368 	return copyin(array + index, value, sizeof(*value));
369 }
370 
371 /*
372  * exec system call
373  */
374 /* ARGSUSED */
375 int
376 sys_execve(struct lwp *l, void *v, register_t *retval)
377 {
378 	struct sys_execve_args /* {
379 		syscallarg(const char *)	path;
380 		syscallarg(char * const *)	argp;
381 		syscallarg(char * const *)	envp;
382 	} */ *uap = v;
383 
384 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
385 	    SCARG(uap, envp), execve_fetch_element);
386 }
387 
388 int
389 execve1(struct lwp *l, const char *path, char * const *args,
390     char * const *envs, execve_fetch_element_t fetch_element)
391 {
392 	int			error;
393 	u_int			i;
394 	struct exec_package	pack;
395 	struct nameidata	nid;
396 	struct vattr		attr;
397 	struct proc		*p;
398 	struct ucred		*cred;
399 	char			*argp;
400 	char			*dp, *sp;
401 	long			argc, envc;
402 	size_t			len;
403 	char			*stack;
404 	struct ps_strings	arginfo;
405 	struct ps_strings	*aip = &arginfo;
406 	struct vmspace		*vm;
407 	char			**tmpfap;
408 	int			szsigcode;
409 	struct exec_vmcmd	*base_vcp;
410 	int			oldlwpflags;
411 #ifdef SYSTRACE
412 	int			wassugid = ISSET(p->p_flag, P_SUGID);
413 	char			pathbuf[MAXPATHLEN];
414 	size_t			pathbuflen;
415 #endif /* SYSTRACE */
416 
417 	/* Disable scheduler activation upcalls. */
418 	oldlwpflags = l->l_flag & (L_SA | L_SA_UPCALL);
419 	if (l->l_flag & L_SA)
420 		l->l_flag &= ~(L_SA | L_SA_UPCALL);
421 
422 	p = l->l_proc;
423 	/*
424 	 * Lock the process and set the P_INEXEC flag to indicate that
425 	 * it should be left alone until we're done here.  This is
426 	 * necessary to avoid race conditions - e.g. in ptrace() -
427 	 * that might allow a local user to illicitly obtain elevated
428 	 * privileges.
429 	 */
430 	p->p_flag |= P_INEXEC;
431 
432 	cred = p->p_ucred;
433 	base_vcp = NULL;
434 	/*
435 	 * Init the namei data to point the file user's program name.
436 	 * This is done here rather than in check_exec(), so that it's
437 	 * possible to override this settings if any of makecmd/probe
438 	 * functions call check_exec() recursively - for example,
439 	 * see exec_script_makecmds().
440 	 */
441 #ifdef SYSTRACE
442 	if (ISSET(p->p_flag, P_SYSTRACE))
443 		systrace_execve0(p);
444 
445 	error = copyinstr(path, pathbuf, sizeof(pathbuf),
446 			  &pathbuflen);
447 	if (error)
448 		goto clrflg;
449 
450 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_SYSSPACE, pathbuf, l);
451 #else
452 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, path, l);
453 #endif /* SYSTRACE */
454 
455 	/*
456 	 * initialize the fields of the exec package.
457 	 */
458 #ifdef SYSTRACE
459 	pack.ep_name = pathbuf;
460 #else
461 	pack.ep_name = path;
462 #endif /* SYSTRACE */
463 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
464 	pack.ep_hdrlen = exec_maxhdrsz;
465 	pack.ep_hdrvalid = 0;
466 	pack.ep_ndp = &nid;
467 	pack.ep_emul_arg = NULL;
468 	pack.ep_vmcmds.evs_cnt = 0;
469 	pack.ep_vmcmds.evs_used = 0;
470 	pack.ep_vap = &attr;
471 	pack.ep_flags = 0;
472 
473 #ifdef LKM
474 	lockmgr(&exec_lock, LK_SHARED, NULL);
475 #endif
476 
477 	/* see if we can run it. */
478 #ifdef VERIFIED_EXEC
479         if ((error = check_exec(l, &pack, VERIEXEC_DIRECT)) != 0)
480 #else
481         if ((error = check_exec(l, &pack, 0)) != 0)
482 #endif
483 		goto freehdr;
484 
485 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
486 
487 	/* allocate an argument buffer */
488 	argp = (char *) uvm_km_alloc(exec_map, NCARGS, 0,
489 	    UVM_KMF_PAGEABLE|UVM_KMF_WAITVA);
490 #ifdef DIAGNOSTIC
491 	if (argp == NULL)
492 		panic("execve: argp == NULL");
493 #endif
494 	dp = argp;
495 	argc = 0;
496 
497 	/* copy the fake args list, if there's one, freeing it as we go */
498 	if (pack.ep_flags & EXEC_HASARGL) {
499 		tmpfap = pack.ep_fa;
500 		while (*tmpfap != NULL) {
501 			char *cp;
502 
503 			cp = *tmpfap;
504 			while (*cp)
505 				*dp++ = *cp++;
506 			dp++;
507 
508 			FREE(*tmpfap, M_EXEC);
509 			tmpfap++; argc++;
510 		}
511 		FREE(pack.ep_fa, M_EXEC);
512 		pack.ep_flags &= ~EXEC_HASARGL;
513 	}
514 
515 	/* Now get argv & environment */
516 	if (args == NULL) {
517 		error = EINVAL;
518 		goto bad;
519 	}
520 	/* 'i' will index the argp/envp element to be retrieved */
521 	i = 0;
522 	if (pack.ep_flags & EXEC_SKIPARG)
523 		i++;
524 
525 	while (1) {
526 		len = argp + ARG_MAX - dp;
527 		if ((error = (*fetch_element)(args, i, &sp)) != 0)
528 			goto bad;
529 		if (!sp)
530 			break;
531 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
532 			if (error == ENAMETOOLONG)
533 				error = E2BIG;
534 			goto bad;
535 		}
536 #ifdef KTRACE
537 		if (KTRPOINT(p, KTR_EXEC_ARG))
538 			ktrkmem(l, KTR_EXEC_ARG, dp, len - 1);
539 #endif
540 		dp += len;
541 		i++;
542 		argc++;
543 	}
544 
545 	envc = 0;
546 	/* environment need not be there */
547 	if (envs != NULL) {
548 		i = 0;
549 		while (1) {
550 			len = argp + ARG_MAX - dp;
551 			if ((error = (*fetch_element)(envs, i, &sp)) != 0)
552 				goto bad;
553 			if (!sp)
554 				break;
555 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
556 				if (error == ENAMETOOLONG)
557 					error = E2BIG;
558 				goto bad;
559 			}
560 #ifdef KTRACE
561 			if (KTRPOINT(p, KTR_EXEC_ENV))
562 				ktrkmem(l, KTR_EXEC_ENV, dp, len - 1);
563 #endif
564 			dp += len;
565 			i++;
566 			envc++;
567 		}
568 	}
569 
570 	dp = (char *) ALIGN(dp);
571 
572 	szsigcode = pack.ep_es->es_emul->e_esigcode -
573 	    pack.ep_es->es_emul->e_sigcode;
574 
575 	/* Now check if args & environ fit into new stack */
576 	if (pack.ep_flags & EXEC_32)
577 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
578 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
579 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
580 		    - argp;
581 	else
582 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
583 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
584 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
585 		    - argp;
586 
587 	len = ALIGN(len);	/* make the stack "safely" aligned */
588 
589 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
590 		error = ENOMEM;
591 		goto bad;
592 	}
593 
594 	/* Get rid of other LWPs/ */
595 	p->p_flag |= P_WEXIT; /* XXX hack. lwp-exit stuff wants to see it. */
596 	exit_lwps(l);
597 	p->p_flag &= ~P_WEXIT;
598 	KDASSERT(p->p_nlwps == 1);
599 
600 	/* This is now LWP 1 */
601 	l->l_lid = 1;
602 	p->p_nlwpid = 1;
603 
604 	/* Release any SA state. */
605 	if (p->p_sa)
606 		sa_release(p);
607 
608 	/* Remove POSIX timers */
609 	timers_free(p, TIMERS_POSIX);
610 
611 	/* adjust "active stack depth" for process VSZ */
612 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
613 
614 	/*
615 	 * Do whatever is necessary to prepare the address space
616 	 * for remapping.  Note that this might replace the current
617 	 * vmspace with another!
618 	 */
619 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
620 
621 	/* record proc's vnode, for use by procfs and others */
622         if (p->p_textvp)
623                 vrele(p->p_textvp);
624 	VREF(pack.ep_vp);
625 	p->p_textvp = pack.ep_vp;
626 
627 	/* Now map address space */
628 	vm = p->p_vmspace;
629 	vm->vm_taddr = (caddr_t) pack.ep_taddr;
630 	vm->vm_tsize = btoc(pack.ep_tsize);
631 	vm->vm_daddr = (caddr_t) pack.ep_daddr;
632 	vm->vm_dsize = btoc(pack.ep_dsize);
633 	vm->vm_ssize = btoc(pack.ep_ssize);
634 	vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr;
635 	vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr;
636 
637 	/* create the new process's VM space by running the vmcmds */
638 #ifdef DIAGNOSTIC
639 	if (pack.ep_vmcmds.evs_used == 0)
640 		panic("execve: no vmcmds");
641 #endif
642 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
643 		struct exec_vmcmd *vcp;
644 
645 		vcp = &pack.ep_vmcmds.evs_cmds[i];
646 		if (vcp->ev_flags & VMCMD_RELATIVE) {
647 #ifdef DIAGNOSTIC
648 			if (base_vcp == NULL)
649 				panic("execve: relative vmcmd with no base");
650 			if (vcp->ev_flags & VMCMD_BASE)
651 				panic("execve: illegal base & relative vmcmd");
652 #endif
653 			vcp->ev_addr += base_vcp->ev_addr;
654 		}
655 		error = (*vcp->ev_proc)(l, vcp);
656 #ifdef DEBUG_EXEC
657 		if (error) {
658 			int j;
659 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
660 			for (j = 0; j <= i; j++)
661 				uprintf(
662 			    "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
663 				    j, vp[j].ev_addr, vp[j].ev_len,
664 				    vp[j].ev_offset, vp[j].ev_prot,
665 				    vp[j].ev_flags);
666 		}
667 #endif /* DEBUG_EXEC */
668 		if (vcp->ev_flags & VMCMD_BASE)
669 			base_vcp = vcp;
670 	}
671 
672 	/* free the vmspace-creation commands, and release their references */
673 	kill_vmcmds(&pack.ep_vmcmds);
674 
675 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
676 	VOP_CLOSE(pack.ep_vp, FREAD, cred, l);
677 	vput(pack.ep_vp);
678 
679 	/* if an error happened, deallocate and punt */
680 	if (error) {
681 		DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error));
682 		goto exec_abort;
683 	}
684 
685 	/* remember information about the process */
686 	arginfo.ps_nargvstr = argc;
687 	arginfo.ps_nenvstr = envc;
688 
689 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
690 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
691 		len - (sizeof(struct ps_strings) + szsigcode));
692 #ifdef __MACHINE_STACK_GROWS_UP
693 	/*
694 	 * The copyargs call always copies into lower addresses
695 	 * first, moving towards higher addresses, starting with
696 	 * the stack pointer that we give.  When the stack grows
697 	 * down, this puts argc/argv/envp very shallow on the
698 	 * stack, right at the first user stack pointer, and puts
699 	 * STACKGAPLEN very deep in the stack.  When the stack
700 	 * grows up, the situation is reversed.
701 	 *
702 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
703 	 * function expects to be called with a single pointer to
704 	 * a region that has a few words it can stash values into,
705 	 * followed by argc/argv/envp.  When the stack grows down,
706 	 * it's easy to decrement the stack pointer a little bit to
707 	 * allocate the space for these few words and pass the new
708 	 * stack pointer to _rtld.  When the stack grows up, however,
709 	 * a few words before argc is part of the signal trampoline, XXX
710 	 * so we have a problem.
711 	 *
712 	 * Instead of changing how _rtld works, we take the easy way
713 	 * out and steal 32 bytes before we call copyargs.  This
714 	 * space is effectively stolen from STACKGAPLEN.
715 	 */
716 	stack += 32;
717 #endif /* __MACHINE_STACK_GROWS_UP */
718 
719 	/* Now copy argc, args & environ to new stack */
720 	error = (*pack.ep_es->es_copyargs)(l, &pack, &arginfo, &stack, argp);
721 	if (error) {
722 		DPRINTF(("execve: copyargs failed %d\n", error));
723 		goto exec_abort;
724 	}
725 	/* Move the stack back to original point */
726 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
727 
728 	/* fill process ps_strings info */
729 	p->p_psstr = (struct ps_strings *)
730 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
731 	    sizeof(struct ps_strings));
732 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
733 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
734 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
735 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
736 
737 	/* copy out the process's ps_strings structure */
738 	if ((error = copyout(aip, (char *)p->p_psstr,
739 	    sizeof(arginfo))) != 0) {
740 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
741 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
742 		goto exec_abort;
743 	}
744 
745 	stopprofclock(p);	/* stop profiling */
746 	fdcloseexec(l);		/* handle close on exec */
747 	execsigs(p);		/* reset catched signals */
748 
749 	l->l_ctxlink = NULL;	/* reset ucontext link */
750 
751 	/* set command name & other accounting info */
752 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
753 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
754 	p->p_comm[len] = 0;
755 	p->p_acflag &= ~AFORK;
756 
757 	p->p_flag |= P_EXEC;
758 	if (p->p_flag & P_PPWAIT) {
759 		p->p_flag &= ~P_PPWAIT;
760 		wakeup((caddr_t) p->p_pptr);
761 	}
762 
763 	/*
764 	 * deal with set[ug]id.
765 	 * MNT_NOSUID has already been used to disable s[ug]id.
766 	 */
767 	if ((p->p_flag & P_TRACED) == 0 &&
768 
769 	    (((attr.va_mode & S_ISUID) != 0 &&
770 	      p->p_ucred->cr_uid != attr.va_uid) ||
771 
772 	     ((attr.va_mode & S_ISGID) != 0 &&
773 	      p->p_ucred->cr_gid != attr.va_gid))) {
774 		/*
775 		 * Mark the process as SUGID before we do
776 		 * anything that might block.
777 		 */
778 		p_sugid(p);
779 
780 		/* Make sure file descriptors 0..2 are in use. */
781 		if ((error = fdcheckstd(l)) != 0) {
782 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
783 			goto exec_abort;
784 		}
785 
786 		p->p_ucred = crcopy(cred);
787 #ifdef KTRACE
788 		/*
789 		 * If process is being ktraced, turn off - unless
790 		 * root set it.
791 		 */
792 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
793 			ktrderef(p);
794 #endif
795 		if (attr.va_mode & S_ISUID)
796 			p->p_ucred->cr_uid = attr.va_uid;
797 		if (attr.va_mode & S_ISGID)
798 			p->p_ucred->cr_gid = attr.va_gid;
799 	} else {
800 		if (p->p_ucred->cr_uid == p->p_cred->p_ruid &&
801 		    p->p_ucred->cr_gid == p->p_cred->p_rgid)
802 			p->p_flag &= ~P_SUGID;
803 	}
804 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
805 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
806 
807 #if defined(__HAVE_RAS)
808 	/*
809 	 * Remove all RASs from the address space.
810 	 */
811 	ras_purgeall(p);
812 #endif
813 
814 	doexechooks(p);
815 
816 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
817 
818 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
819 
820 	/* notify others that we exec'd */
821 	KNOTE(&p->p_klist, NOTE_EXEC);
822 
823 	/* setup new registers and do misc. setup. */
824 	(*pack.ep_es->es_emul->e_setregs)(l, &pack, (u_long) stack);
825 	if (pack.ep_es->es_setregs)
826 		(*pack.ep_es->es_setregs)(l, &pack, (u_long) stack);
827 
828 	/* map the process's signal trampoline code */
829 	if (exec_sigcode_map(p, pack.ep_es->es_emul)) {
830 		DPRINTF(("execve: map sigcode failed %d\n", error));
831 		goto exec_abort;
832 	}
833 
834 	if ((p->p_flag & (P_TRACED|P_SYSCALL)) == P_TRACED)
835 		psignal(p, SIGTRAP);
836 
837 	free(pack.ep_hdr, M_EXEC);
838 
839 	/*
840 	 * Call emulation specific exec hook. This can setup per-process
841 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
842 	 *
843 	 * If we are executing process of different emulation than the
844 	 * original forked process, call e_proc_exit() of the old emulation
845 	 * first, then e_proc_exec() of new emulation. If the emulation is
846 	 * same, the exec hook code should deallocate any old emulation
847 	 * resources held previously by this process.
848 	 */
849 	if (p->p_emul && p->p_emul->e_proc_exit
850 	    && p->p_emul != pack.ep_es->es_emul)
851 		(*p->p_emul->e_proc_exit)(p);
852 
853 	/*
854 	 * Call exec hook. Emulation code may NOT store reference to anything
855 	 * from &pack.
856 	 */
857         if (pack.ep_es->es_emul->e_proc_exec)
858                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
859 
860 	/* update p_emul, the old value is no longer needed */
861 	p->p_emul = pack.ep_es->es_emul;
862 
863 	/* ...and the same for p_execsw */
864 	p->p_execsw = pack.ep_es;
865 
866 #ifdef __HAVE_SYSCALL_INTERN
867 	(*p->p_emul->e_syscall_intern)(p);
868 #endif
869 #ifdef KTRACE
870 	if (KTRPOINT(p, KTR_EMUL))
871 		ktremul(l);
872 #endif
873 
874 #ifdef LKM
875 	lockmgr(&exec_lock, LK_RELEASE, NULL);
876 #endif
877 	p->p_flag &= ~P_INEXEC;
878 
879 	if (p->p_flag & P_STOPEXEC) {
880 		int s;
881 
882 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
883 		SCHED_LOCK(s);
884 		p->p_pptr->p_nstopchild++;
885 		p->p_stat = SSTOP;
886 		l->l_stat = LSSTOP;
887 		p->p_nrlwps--;
888 		mi_switch(l, NULL);
889 		SCHED_ASSERT_UNLOCKED();
890 		splx(s);
891 	}
892 
893 #ifdef SYSTRACE
894 	if (ISSET(p->p_flag, P_SYSTRACE) &&
895 	    wassugid && !ISSET(p->p_flag, P_SUGID))
896 		systrace_execve1(pathbuf, p);
897 #endif /* SYSTRACE */
898 
899 	return (EJUSTRETURN);
900 
901  bad:
902 	p->p_flag &= ~P_INEXEC;
903 	/* free the vmspace-creation commands, and release their references */
904 	kill_vmcmds(&pack.ep_vmcmds);
905 	/* kill any opened file descriptor, if necessary */
906 	if (pack.ep_flags & EXEC_HASFD) {
907 		pack.ep_flags &= ~EXEC_HASFD;
908 		(void) fdrelease(l, pack.ep_fd);
909 	}
910 	/* close and put the exec'd file */
911 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
912 	VOP_CLOSE(pack.ep_vp, FREAD, cred, l);
913 	vput(pack.ep_vp);
914 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
915 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
916 
917  freehdr:
918 	free(pack.ep_hdr, M_EXEC);
919 
920 #ifdef SYSTRACE
921  clrflg:
922 #endif /* SYSTRACE */
923 	l->l_flag |= oldlwpflags;
924 	p->p_flag &= ~P_INEXEC;
925 #ifdef LKM
926 	lockmgr(&exec_lock, LK_RELEASE, NULL);
927 #endif
928 
929 	return error;
930 
931  exec_abort:
932 	p->p_flag &= ~P_INEXEC;
933 #ifdef LKM
934 	lockmgr(&exec_lock, LK_RELEASE, NULL);
935 #endif
936 
937 	/*
938 	 * the old process doesn't exist anymore.  exit gracefully.
939 	 * get rid of the (new) address space we have created, if any, get rid
940 	 * of our namei data and vnode, and exit noting failure
941 	 */
942 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
943 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
944 	if (pack.ep_emul_arg)
945 		FREE(pack.ep_emul_arg, M_TEMP);
946 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
947 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
948 	free(pack.ep_hdr, M_EXEC);
949 	exit1(l, W_EXITCODE(error, SIGABRT));
950 
951 	/* NOTREACHED */
952 	return 0;
953 }
954 
955 
956 int
957 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
958     char **stackp, void *argp)
959 {
960 	char	**cpp, *dp, *sp;
961 	size_t	len;
962 	void	*nullp;
963 	long	argc, envc;
964 	int	error;
965 
966 	cpp = (char **)*stackp;
967 	nullp = NULL;
968 	argc = arginfo->ps_nargvstr;
969 	envc = arginfo->ps_nenvstr;
970 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
971 		return error;
972 
973 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
974 	sp = argp;
975 
976 	/* XXX don't copy them out, remap them! */
977 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
978 
979 	for (; --argc >= 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 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
988 
989 	for (; --envc >= 0; sp += len, dp += len)
990 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
991 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
992 			return error;
993 
994 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
995 		return error;
996 
997 	*stackp = (char *)cpp;
998 	return 0;
999 }
1000 
1001 #ifdef LKM
1002 /*
1003  * Find an emulation of given name in list of emulations.
1004  * Needs to be called with the exec_lock held.
1005  */
1006 const struct emul *
1007 emul_search(const char *name)
1008 {
1009 	struct emul_entry *it;
1010 
1011 	LIST_FOREACH(it, &el_head, el_list) {
1012 		if (strcmp(name, it->el_emul->e_name) == 0)
1013 			return it->el_emul;
1014 	}
1015 
1016 	return NULL;
1017 }
1018 
1019 /*
1020  * Add an emulation to list, if it's not there already.
1021  */
1022 int
1023 emul_register(const struct emul *emul, int ro_entry)
1024 {
1025 	struct emul_entry	*ee;
1026 	int			error;
1027 
1028 	error = 0;
1029 	lockmgr(&exec_lock, LK_SHARED, NULL);
1030 
1031 	if (emul_search(emul->e_name)) {
1032 		error = EEXIST;
1033 		goto out;
1034 	}
1035 
1036 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
1037 		M_EXEC, M_WAITOK);
1038 	ee->el_emul = emul;
1039 	ee->ro_entry = ro_entry;
1040 	LIST_INSERT_HEAD(&el_head, ee, el_list);
1041 
1042  out:
1043 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1044 	return error;
1045 }
1046 
1047 /*
1048  * Remove emulation with name 'name' from list of supported emulations.
1049  */
1050 int
1051 emul_unregister(const char *name)
1052 {
1053 	const struct proclist_desc *pd;
1054 	struct emul_entry	*it;
1055 	int			i, error;
1056 	struct proc		*ptmp;
1057 
1058 	error = 0;
1059 	lockmgr(&exec_lock, LK_SHARED, NULL);
1060 
1061 	LIST_FOREACH(it, &el_head, el_list) {
1062 		if (strcmp(it->el_emul->e_name, name) == 0)
1063 			break;
1064 	}
1065 
1066 	if (!it) {
1067 		error = ENOENT;
1068 		goto out;
1069 	}
1070 
1071 	if (it->ro_entry) {
1072 		error = EBUSY;
1073 		goto out;
1074 	}
1075 
1076 	/* test if any execw[] entry is still using this */
1077 	for(i=0; i < nexecs; i++) {
1078 		if (execsw[i]->es_emul == it->el_emul) {
1079 			error = EBUSY;
1080 			goto out;
1081 		}
1082 	}
1083 
1084 	/*
1085 	 * Test if any process is running under this emulation - since
1086 	 * emul_unregister() is running quite sendomly, it's better
1087 	 * to do expensive check here than to use any locking.
1088 	 */
1089 	proclist_lock_read();
1090 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
1091 		PROCLIST_FOREACH(ptmp, pd->pd_list) {
1092 			if (ptmp->p_emul == it->el_emul) {
1093 				error = EBUSY;
1094 				break;
1095 			}
1096 		}
1097 	}
1098 	proclist_unlock_read();
1099 
1100 	if (error)
1101 		goto out;
1102 
1103 
1104 	/* entry is not used, remove it */
1105 	LIST_REMOVE(it, el_list);
1106 	FREE(it, M_EXEC);
1107 
1108  out:
1109 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1110 	return error;
1111 }
1112 
1113 /*
1114  * Add execsw[] entry.
1115  */
1116 int
1117 exec_add(struct execsw *esp, const char *e_name)
1118 {
1119 	struct exec_entry	*it;
1120 	int			error;
1121 
1122 	error = 0;
1123 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
1124 
1125 	if (!esp->es_emul) {
1126 		esp->es_emul = emul_search(e_name);
1127 		if (!esp->es_emul) {
1128 			error = ENOENT;
1129 			goto out;
1130 		}
1131 	}
1132 
1133 	LIST_FOREACH(it, &ex_head, ex_list) {
1134 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1135 		if (it->es->es_makecmds == esp->es_makecmds
1136 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1137 		    && it->es->es_emul == esp->es_emul) {
1138 			error = EEXIST;
1139 			goto out;
1140 		}
1141 	}
1142 
1143 	/* if we got here, the entry doesn't exist yet */
1144 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
1145 		M_EXEC, M_WAITOK);
1146 	it->es = esp;
1147 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
1148 
1149 	/* update execsw[] */
1150 	exec_init(0);
1151 
1152  out:
1153 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1154 	return error;
1155 }
1156 
1157 /*
1158  * Remove execsw[] entry.
1159  */
1160 int
1161 exec_remove(const struct execsw *esp)
1162 {
1163 	struct exec_entry	*it;
1164 	int			error;
1165 
1166 	error = 0;
1167 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
1168 
1169 	LIST_FOREACH(it, &ex_head, ex_list) {
1170 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1171 		if (it->es->es_makecmds == esp->es_makecmds
1172 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1173 		    && it->es->es_emul == esp->es_emul)
1174 			break;
1175 	}
1176 	if (!it) {
1177 		error = ENOENT;
1178 		goto out;
1179 	}
1180 
1181 	/* remove item from list and free resources */
1182 	LIST_REMOVE(it, ex_list);
1183 	FREE(it, M_EXEC);
1184 
1185 	/* update execsw[] */
1186 	exec_init(0);
1187 
1188  out:
1189 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1190 	return error;
1191 }
1192 
1193 static void
1194 link_es(struct execsw_entry **listp, const struct execsw *esp)
1195 {
1196 	struct execsw_entry *et, *e1;
1197 
1198 	MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry),
1199 			M_TEMP, M_WAITOK);
1200 	et->next = NULL;
1201 	et->es = esp;
1202 	if (*listp == NULL) {
1203 		*listp = et;
1204 		return;
1205 	}
1206 
1207 	switch(et->es->es_prio) {
1208 	case EXECSW_PRIO_FIRST:
1209 		/* put new entry as the first */
1210 		et->next = *listp;
1211 		*listp = et;
1212 		break;
1213 	case EXECSW_PRIO_ANY:
1214 		/* put new entry after all *_FIRST and *_ANY entries */
1215 		for(e1 = *listp; e1->next
1216 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
1217 			e1 = e1->next);
1218 		et->next = e1->next;
1219 		e1->next = et;
1220 		break;
1221 	case EXECSW_PRIO_LAST:
1222 		/* put new entry as the last one */
1223 		for(e1 = *listp; e1->next; e1 = e1->next);
1224 		e1->next = et;
1225 		break;
1226 	default:
1227 #ifdef DIAGNOSTIC
1228 		panic("execw[] entry with unknown priority %d found",
1229 			et->es->es_prio);
1230 #endif
1231 		break;
1232 	}
1233 }
1234 
1235 /*
1236  * Initialize exec structures. If init_boot is true, also does necessary
1237  * one-time initialization (it's called from main() that way).
1238  * Once system is multiuser, this should be called with exec_lock held,
1239  * i.e. via exec_{add|remove}().
1240  */
1241 int
1242 exec_init(int init_boot)
1243 {
1244 	const struct execsw	**new_es, * const *old_es;
1245 	struct execsw_entry	*list, *e1;
1246 	struct exec_entry	*e2;
1247 	int			i, es_sz;
1248 
1249 	if (init_boot) {
1250 		/* do one-time initializations */
1251 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1252 
1253 		/* register compiled-in emulations */
1254 		for(i=0; i < nexecs_builtin; i++) {
1255 			if (execsw_builtin[i].es_emul)
1256 				emul_register(execsw_builtin[i].es_emul, 1);
1257 		}
1258 #ifdef DIAGNOSTIC
1259 		if (i == 0)
1260 			panic("no emulations found in execsw_builtin[]");
1261 #endif
1262 	}
1263 
1264 	/*
1265 	 * Build execsw[] array from builtin entries and entries added
1266 	 * at runtime.
1267 	 */
1268 	list = NULL;
1269 	for(i=0; i < nexecs_builtin; i++)
1270 		link_es(&list, &execsw_builtin[i]);
1271 
1272 	/* Add dynamically loaded entries */
1273 	es_sz = nexecs_builtin;
1274 	LIST_FOREACH(e2, &ex_head, ex_list) {
1275 		link_es(&list, e2->es);
1276 		es_sz++;
1277 	}
1278 
1279 	/*
1280 	 * Now that we have sorted all execw entries, create new execsw[]
1281 	 * and free no longer needed memory in the process.
1282 	 */
1283 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
1284 	for(i=0; list; i++) {
1285 		new_es[i] = list->es;
1286 		e1 = list->next;
1287 		FREE(list, M_TEMP);
1288 		list = e1;
1289 	}
1290 
1291 	/*
1292 	 * New execsw[] array built, now replace old execsw[] and free
1293 	 * used memory.
1294 	 */
1295 	old_es = execsw;
1296 	execsw = new_es;
1297 	nexecs = es_sz;
1298 	if (old_es)
1299 		/*XXXUNCONST*/
1300 		free(__UNCONST(old_es), M_EXEC);
1301 
1302 	/*
1303 	 * Figure out the maximum size of an exec header.
1304 	 */
1305 	exec_maxhdrsz = 0;
1306 	for (i = 0; i < nexecs; i++) {
1307 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1308 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1309 	}
1310 
1311 	return 0;
1312 }
1313 #endif
1314 
1315 #ifndef LKM
1316 /*
1317  * Simplified exec_init() for kernels without LKMs. Only initialize
1318  * exec_maxhdrsz and execsw[].
1319  */
1320 int
1321 exec_init(int init_boot)
1322 {
1323 	int i;
1324 
1325 #ifdef DIAGNOSTIC
1326 	if (!init_boot)
1327 		panic("exec_init(): called with init_boot == 0");
1328 #endif
1329 
1330 	/* do one-time initializations */
1331 	nexecs = nexecs_builtin;
1332 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
1333 
1334 	/*
1335 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1336 	 */
1337 	exec_maxhdrsz = 0;
1338 	for(i=0; i < nexecs; i++) {
1339 		execsw[i] = &execsw_builtin[i];
1340 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1341 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1342 	}
1343 
1344 	return 0;
1345 
1346 }
1347 #endif /* !LKM */
1348 
1349 static int
1350 exec_sigcode_map(struct proc *p, const struct emul *e)
1351 {
1352 	vaddr_t va;
1353 	vsize_t sz;
1354 	int error;
1355 	struct uvm_object *uobj;
1356 
1357 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1358 
1359 	if (e->e_sigobject == NULL || sz == 0) {
1360 		return 0;
1361 	}
1362 
1363 	/*
1364 	 * If we don't have a sigobject for this emulation, create one.
1365 	 *
1366 	 * sigobject is an anonymous memory object (just like SYSV shared
1367 	 * memory) that we keep a permanent reference to and that we map
1368 	 * in all processes that need this sigcode. The creation is simple,
1369 	 * we create an object, add a permanent reference to it, map it in
1370 	 * kernel space, copy out the sigcode to it and unmap it.
1371 	 * We map it with PROT_READ|PROT_EXEC into the process just
1372 	 * the way sys_mmap() would map it.
1373 	 */
1374 
1375 	uobj = *e->e_sigobject;
1376 	if (uobj == NULL) {
1377 		uobj = uao_create(sz, 0);
1378 		(*uobj->pgops->pgo_reference)(uobj);
1379 		va = vm_map_min(kernel_map);
1380 		if ((error = uvm_map(kernel_map, &va, round_page(sz),
1381 		    uobj, 0, 0,
1382 		    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1383 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1384 			printf("kernel mapping failed %d\n", error);
1385 			(*uobj->pgops->pgo_detach)(uobj);
1386 			return (error);
1387 		}
1388 		memcpy((void *)va, e->e_sigcode, sz);
1389 #ifdef PMAP_NEED_PROCWR
1390 		pmap_procwr(&proc0, va, sz);
1391 #endif
1392 		uvm_unmap(kernel_map, va, va + round_page(sz));
1393 		*e->e_sigobject = uobj;
1394 	}
1395 
1396 	/* Just a hint to uvm_map where to put it. */
1397 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1398 	    round_page(sz));
1399 
1400 #ifdef __alpha__
1401 	/*
1402 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1403 	 * which causes the above calculation to put the sigcode at
1404 	 * an invalid address.  Put it just below the text instead.
1405 	 */
1406 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1407 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1408 	}
1409 #endif
1410 
1411 	(*uobj->pgops->pgo_reference)(uobj);
1412 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1413 			uobj, 0, 0,
1414 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1415 				    UVM_ADV_RANDOM, 0));
1416 	if (error) {
1417 		(*uobj->pgops->pgo_detach)(uobj);
1418 		return (error);
1419 	}
1420 	p->p_sigctx.ps_sigcode = (void *)va;
1421 	return (0);
1422 }
1423