xref: /netbsd-src/sys/kern/kern_exec.c (revision b7ae68fde0d8ef1c03714e8bbb1ee7c6118ea93b)
1 /*	$NetBSD: kern_exec.c,v 1.227 2006/07/30 21:58:11 ad 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 1.227 2006/07/30 21:58:11 ad Exp $");
37 
38 #include "opt_ktrace.h"
39 #include "opt_syscall_debug.h"
40 #include "opt_compat_netbsd.h"
41 #include "veriexec.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 #include <sys/kauth.h>
64 
65 #include <sys/sa.h>
66 #include <sys/savar.h>
67 #include <sys/syscallargs.h>
68 #if NVERIEXEC > 0
69 #include <sys/verified_exec.h>
70 #endif /* NVERIEXEC > 0 */
71 
72 #ifdef SYSTRACE
73 #include <sys/systrace.h>
74 #endif /* SYSTRACE */
75 
76 #include <uvm/uvm_extern.h>
77 
78 #include <machine/cpu.h>
79 #include <machine/reg.h>
80 
81 static int exec_sigcode_map(struct proc *, const struct emul *);
82 
83 #ifdef DEBUG_EXEC
84 #define DPRINTF(a) uprintf a
85 #else
86 #define DPRINTF(a)
87 #endif /* DEBUG_EXEC */
88 
89 MALLOC_DEFINE(M_EXEC, "exec", "argument lists & other mem used by exec");
90 
91 /*
92  * Exec function switch:
93  *
94  * Note that each makecmds function is responsible for loading the
95  * exec package with the necessary functions for any exec-type-specific
96  * handling.
97  *
98  * Functions for specific exec types should be defined in their own
99  * header file.
100  */
101 extern const struct execsw	execsw_builtin[];
102 extern int			nexecs_builtin;
103 static const struct execsw	**execsw = NULL;
104 static int			nexecs;
105 
106 u_int	exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
107 
108 #ifdef LKM
109 /* list of supported emulations */
110 static
111 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
112 struct emul_entry {
113 	LIST_ENTRY(emul_entry)	el_list;
114 	const struct emul	*el_emul;
115 	int			ro_entry;
116 };
117 
118 /* list of dynamically loaded execsw entries */
119 static
120 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
121 struct exec_entry {
122 	LIST_ENTRY(exec_entry)	ex_list;
123 	const struct execsw	*es;
124 };
125 
126 /* structure used for building execw[] */
127 struct execsw_entry {
128 	struct execsw_entry	*next;
129 	const struct execsw	*es;
130 };
131 #endif /* LKM */
132 
133 #ifdef SYSCALL_DEBUG
134 extern const char * const syscallnames[];
135 #endif
136 
137 #ifdef COMPAT_16
138 extern char	sigcode[], esigcode[];
139 struct uvm_object *emul_netbsd_object;
140 #endif
141 
142 #ifndef __HAVE_SYSCALL_INTERN
143 void	syscall(void);
144 #endif
145 
146 static const struct sa_emul saemul_netbsd = {
147 	sizeof(ucontext_t),
148 	sizeof(struct sa_t),
149 	sizeof(struct sa_t *),
150 	NULL,
151 	NULL,
152 	cpu_upcall,
153 	(void (*)(struct lwp *, void *))getucontext,
154 	sa_ucsp
155 };
156 
157 /* NetBSD emul struct */
158 const struct emul emul_netbsd = {
159 	"netbsd",
160 	NULL,		/* emulation path */
161 #ifndef __HAVE_MINIMAL_EMUL
162 	EMUL_HAS_SYS___syscall,
163 	NULL,
164 	SYS_syscall,
165 	SYS_NSYSENT,
166 #endif
167 	sysent,
168 #ifdef SYSCALL_DEBUG
169 	syscallnames,
170 #else
171 	NULL,
172 #endif
173 	sendsig,
174 	trapsignal,
175 	NULL,
176 #ifdef COMPAT_16
177 	sigcode,
178 	esigcode,
179 	&emul_netbsd_object,
180 #else
181 	NULL,
182 	NULL,
183 	NULL,
184 #endif
185 	setregs,
186 	NULL,
187 	NULL,
188 	NULL,
189 	NULL,
190 	NULL,
191 #ifdef __HAVE_SYSCALL_INTERN
192 	syscall_intern,
193 #else
194 	syscall,
195 #endif
196 	NULL,
197 	NULL,
198 
199 	uvm_default_mapaddr,
200 	NULL,
201 	&saemul_netbsd,
202 };
203 
204 #ifdef LKM
205 /*
206  * Exec lock. Used to control access to execsw[] structures.
207  * This must not be static so that netbsd32 can access it, too.
208  */
209 struct lock exec_lock;
210 
211 static void link_es(struct execsw_entry **, const struct execsw *);
212 #endif /* LKM */
213 
214 /*
215  * check exec:
216  * given an "executable" described in the exec package's namei info,
217  * see what we can do with it.
218  *
219  * ON ENTRY:
220  *	exec package with appropriate namei info
221  *	lwp pointer of exec'ing lwp
222  *      if verified exec enabled then flag indicating a direct exec or
223  *        an indirect exec (i.e. for a shell script interpreter)
224  *	NO SELF-LOCKED VNODES
225  *
226  * ON EXIT:
227  *	error:	nothing held, etc.  exec header still allocated.
228  *	ok:	filled exec package, executable's vnode (unlocked).
229  *
230  * EXEC SWITCH ENTRY:
231  * 	Locked vnode to check, exec package, proc.
232  *
233  * EXEC SWITCH EXIT:
234  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
235  *	error:	destructive:
236  *			everything deallocated execept exec header.
237  *		non-destructive:
238  *			error code, executable's vnode (unlocked),
239  *			exec header unmodified.
240  */
241 int
242 /*ARGSUSED*/
243 check_exec(struct lwp *l, struct exec_package *epp, int flag)
244 {
245 	int		error, i;
246 	struct vnode	*vp;
247 	struct nameidata *ndp;
248 	size_t		resid;
249 
250 	ndp = epp->ep_ndp;
251 	ndp->ni_cnd.cn_nameiop = LOOKUP;
252 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
253 	/* first get the vnode */
254 	if ((error = namei(ndp)) != 0)
255 		return error;
256 	epp->ep_vp = vp = ndp->ni_vp;
257 
258 	/* check access and type */
259 	if (vp->v_type != VREG) {
260 		error = EACCES;
261 		goto bad1;
262 	}
263 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred, l)) != 0)
264 		goto bad1;
265 
266 	/* get attributes */
267 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred, l)) != 0)
268 		goto bad1;
269 
270 	/* Check mount point */
271 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
272 		error = EACCES;
273 		goto bad1;
274 	}
275 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
276 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
277 
278 	/* try to open it */
279 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred, l)) != 0)
280 		goto bad1;
281 
282 	/* unlock vp, since we need it unlocked from here on out. */
283 	VOP_UNLOCK(vp, 0);
284 
285 
286 #if NVERIEXEC > 0
287         if ((error = veriexec_verify(l, vp, epp->ep_ndp->ni_dirp, flag,
288 	    NULL)) != 0)
289                 goto bad2;
290 #endif /* NVERIEXEC > 0 */
291 
292 	/* now we have the file, get the exec header */
293 	uvn_attach(vp, VM_PROT_READ);
294 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
295 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
296 	if (error)
297 		goto bad2;
298 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
299 
300 	/*
301 	 * Set up default address space limits.  Can be overridden
302 	 * by individual exec packages.
303 	 *
304 	 * XXX probably should be all done in the exec pakages.
305 	 */
306 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
307 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
308 	/*
309 	 * set up the vmcmds for creation of the process
310 	 * address space
311 	 */
312 	error = ENOEXEC;
313 	for (i = 0; i < nexecs && error != 0; i++) {
314 		int newerror;
315 
316 		epp->ep_esch = execsw[i];
317 		newerror = (*execsw[i]->es_makecmds)(l, epp);
318 		/* make sure the first "interesting" error code is saved. */
319 		if (!newerror || error == ENOEXEC)
320 			error = newerror;
321 
322 		/* if es_makecmds call was successful, update epp->ep_es */
323 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
324 			epp->ep_es = execsw[i];
325 
326 		if (epp->ep_flags & EXEC_DESTR && error != 0)
327 			return error;
328 	}
329 	if (!error) {
330 		/* check that entry point is sane */
331 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
332 			error = ENOEXEC;
333 
334 		/* check limits */
335 		if ((epp->ep_tsize > MAXTSIZ) ||
336 		    (epp->ep_dsize >
337 		     (u_quad_t)l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur))
338 			error = ENOMEM;
339 
340 		if (!error)
341 			return (0);
342 	}
343 
344 	/*
345 	 * free any vmspace-creation commands,
346 	 * and release their references
347 	 */
348 	kill_vmcmds(&epp->ep_vmcmds);
349 
350 bad2:
351 	/*
352 	 * close and release the vnode, restore the old one, free the
353 	 * pathname buf, and punt.
354 	 */
355 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
356 	VOP_CLOSE(vp, FREAD, l->l_cred, l);
357 	vput(vp);
358 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
359 	return error;
360 
361 bad1:
362 	/*
363 	 * free the namei pathname buffer, and put the vnode
364 	 * (which we don't yet have open).
365 	 */
366 	vput(vp);				/* was still locked */
367 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
368 	return error;
369 }
370 
371 #ifdef __MACHINE_STACK_GROWS_UP
372 #define STACK_PTHREADSPACE NBPG
373 #else
374 #define STACK_PTHREADSPACE 0
375 #endif
376 
377 static int
378 execve_fetch_element(char * const *array, size_t index, char **value)
379 {
380 	return copyin(array + index, value, sizeof(*value));
381 }
382 
383 /*
384  * exec system call
385  */
386 /* ARGSUSED */
387 int
388 sys_execve(struct lwp *l, void *v, register_t *retval)
389 {
390 	struct sys_execve_args /* {
391 		syscallarg(const char *)	path;
392 		syscallarg(char * const *)	argp;
393 		syscallarg(char * const *)	envp;
394 	} */ *uap = v;
395 
396 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
397 	    SCARG(uap, envp), execve_fetch_element);
398 }
399 
400 int
401 execve1(struct lwp *l, const char *path, char * const *args,
402     char * const *envs, execve_fetch_element_t fetch_element)
403 {
404 	int			error;
405 	u_int			i;
406 	struct exec_package	pack;
407 	struct nameidata	nid;
408 	struct vattr		attr;
409 	struct proc		*p;
410 	char			*argp;
411 	char			*dp, *sp;
412 	long			argc, envc;
413 	size_t			len;
414 	char			*stack;
415 	struct ps_strings	arginfo;
416 	struct ps_strings	*aip = &arginfo;
417 	struct vmspace		*vm;
418 	char			**tmpfap;
419 	int			szsigcode;
420 	struct exec_vmcmd	*base_vcp;
421 	int			oldlwpflags;
422 #ifdef SYSTRACE
423 	int			wassugid = ISSET(p->p_flag, P_SUGID);
424 	char			pathbuf[MAXPATHLEN];
425 	size_t			pathbuflen;
426 #endif /* SYSTRACE */
427 
428 	/* Disable scheduler activation upcalls. */
429 	oldlwpflags = l->l_flag & (L_SA | L_SA_UPCALL);
430 	if (l->l_flag & L_SA)
431 		l->l_flag &= ~(L_SA | L_SA_UPCALL);
432 
433 	p = l->l_proc;
434 	/*
435 	 * Lock the process and set the P_INEXEC flag to indicate that
436 	 * it should be left alone until we're done here.  This is
437 	 * necessary to avoid race conditions - e.g. in ptrace() -
438 	 * that might allow a local user to illicitly obtain elevated
439 	 * privileges.
440 	 */
441 	p->p_flag |= P_INEXEC;
442 
443 	base_vcp = NULL;
444 	/*
445 	 * Init the namei data to point the file user's program name.
446 	 * This is done here rather than in check_exec(), so that it's
447 	 * possible to override this settings if any of makecmd/probe
448 	 * functions call check_exec() recursively - for example,
449 	 * see exec_script_makecmds().
450 	 */
451 #ifdef SYSTRACE
452 	if (ISSET(p->p_flag, P_SYSTRACE))
453 		systrace_execve0(p);
454 
455 	error = copyinstr(path, pathbuf, sizeof(pathbuf),
456 			  &pathbuflen);
457 	if (error)
458 		goto clrflg;
459 
460 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_SYSSPACE, pathbuf, l);
461 #else
462 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, path, l);
463 #endif /* SYSTRACE */
464 
465 	/*
466 	 * initialize the fields of the exec package.
467 	 */
468 #ifdef SYSTRACE
469 	pack.ep_name = pathbuf;
470 #else
471 	pack.ep_name = path;
472 #endif /* SYSTRACE */
473 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
474 	pack.ep_hdrlen = exec_maxhdrsz;
475 	pack.ep_hdrvalid = 0;
476 	pack.ep_ndp = &nid;
477 	pack.ep_emul_arg = NULL;
478 	pack.ep_vmcmds.evs_cnt = 0;
479 	pack.ep_vmcmds.evs_used = 0;
480 	pack.ep_vap = &attr;
481 	pack.ep_flags = 0;
482 
483 #ifdef LKM
484 	lockmgr(&exec_lock, LK_SHARED, NULL);
485 #endif
486 
487 	/* see if we can run it. */
488 #if NVERIEXEC > 0
489         if ((error = check_exec(l, &pack, VERIEXEC_DIRECT)) != 0)
490 #else
491         if ((error = check_exec(l, &pack, 0)) != 0)
492 #endif /* NVERIEXEC > 0 */
493 		goto freehdr;
494 
495 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
496 
497 	/* allocate an argument buffer */
498 	argp = (char *) uvm_km_alloc(exec_map, NCARGS, 0,
499 	    UVM_KMF_PAGEABLE|UVM_KMF_WAITVA);
500 #ifdef DIAGNOSTIC
501 	if (argp == NULL)
502 		panic("execve: argp == NULL");
503 #endif
504 	dp = argp;
505 	argc = 0;
506 
507 	/* copy the fake args list, if there's one, freeing it as we go */
508 	if (pack.ep_flags & EXEC_HASARGL) {
509 		tmpfap = pack.ep_fa;
510 		while (*tmpfap != NULL) {
511 			char *cp;
512 
513 			cp = *tmpfap;
514 			while (*cp)
515 				*dp++ = *cp++;
516 			dp++;
517 
518 			FREE(*tmpfap, M_EXEC);
519 			tmpfap++; argc++;
520 		}
521 		FREE(pack.ep_fa, M_EXEC);
522 		pack.ep_flags &= ~EXEC_HASARGL;
523 	}
524 
525 	/* Now get argv & environment */
526 	if (args == NULL) {
527 		error = EINVAL;
528 		goto bad;
529 	}
530 	/* 'i' will index the argp/envp element to be retrieved */
531 	i = 0;
532 	if (pack.ep_flags & EXEC_SKIPARG)
533 		i++;
534 
535 	while (1) {
536 		len = argp + ARG_MAX - dp;
537 		if ((error = (*fetch_element)(args, i, &sp)) != 0)
538 			goto bad;
539 		if (!sp)
540 			break;
541 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
542 			if (error == ENAMETOOLONG)
543 				error = E2BIG;
544 			goto bad;
545 		}
546 #ifdef KTRACE
547 		if (KTRPOINT(p, KTR_EXEC_ARG))
548 			ktrkmem(l, KTR_EXEC_ARG, dp, len - 1);
549 #endif
550 		dp += len;
551 		i++;
552 		argc++;
553 	}
554 
555 	envc = 0;
556 	/* environment need not be there */
557 	if (envs != NULL) {
558 		i = 0;
559 		while (1) {
560 			len = argp + ARG_MAX - dp;
561 			if ((error = (*fetch_element)(envs, i, &sp)) != 0)
562 				goto bad;
563 			if (!sp)
564 				break;
565 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
566 				if (error == ENAMETOOLONG)
567 					error = E2BIG;
568 				goto bad;
569 			}
570 #ifdef KTRACE
571 			if (KTRPOINT(p, KTR_EXEC_ENV))
572 				ktrkmem(l, KTR_EXEC_ENV, dp, len - 1);
573 #endif
574 			dp += len;
575 			i++;
576 			envc++;
577 		}
578 	}
579 
580 	dp = (char *) ALIGN(dp);
581 
582 	szsigcode = pack.ep_es->es_emul->e_esigcode -
583 	    pack.ep_es->es_emul->e_sigcode;
584 
585 	/* Now check if args & environ fit into new stack */
586 	if (pack.ep_flags & EXEC_32)
587 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
588 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
589 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
590 		    - argp;
591 	else
592 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
593 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
594 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
595 		    - argp;
596 
597 	len = ALIGN(len);	/* make the stack "safely" aligned */
598 
599 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
600 		error = ENOMEM;
601 		goto bad;
602 	}
603 
604 	/* Get rid of other LWPs/ */
605 	p->p_flag |= P_WEXIT; /* XXX hack. lwp-exit stuff wants to see it. */
606 	exit_lwps(l);
607 	p->p_flag &= ~P_WEXIT;
608 	KDASSERT(p->p_nlwps == 1);
609 
610 	/* This is now LWP 1 */
611 	l->l_lid = 1;
612 	p->p_nlwpid = 1;
613 
614 	/* Release any SA state. */
615 	if (p->p_sa)
616 		sa_release(p);
617 
618 	/* Remove POSIX timers */
619 	timers_free(p, TIMERS_POSIX);
620 
621 	/* adjust "active stack depth" for process VSZ */
622 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
623 
624 	/*
625 	 * Do whatever is necessary to prepare the address space
626 	 * for remapping.  Note that this might replace the current
627 	 * vmspace with another!
628 	 */
629 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
630 
631 	/* record proc's vnode, for use by procfs and others */
632         if (p->p_textvp)
633                 vrele(p->p_textvp);
634 	VREF(pack.ep_vp);
635 	p->p_textvp = pack.ep_vp;
636 
637 	/* Now map address space */
638 	vm = p->p_vmspace;
639 	vm->vm_taddr = (caddr_t) pack.ep_taddr;
640 	vm->vm_tsize = btoc(pack.ep_tsize);
641 	vm->vm_daddr = (caddr_t) pack.ep_daddr;
642 	vm->vm_dsize = btoc(pack.ep_dsize);
643 	vm->vm_ssize = btoc(pack.ep_ssize);
644 	vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr;
645 	vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr;
646 
647 	/* create the new process's VM space by running the vmcmds */
648 #ifdef DIAGNOSTIC
649 	if (pack.ep_vmcmds.evs_used == 0)
650 		panic("execve: no vmcmds");
651 #endif
652 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
653 		struct exec_vmcmd *vcp;
654 
655 		vcp = &pack.ep_vmcmds.evs_cmds[i];
656 		if (vcp->ev_flags & VMCMD_RELATIVE) {
657 #ifdef DIAGNOSTIC
658 			if (base_vcp == NULL)
659 				panic("execve: relative vmcmd with no base");
660 			if (vcp->ev_flags & VMCMD_BASE)
661 				panic("execve: illegal base & relative vmcmd");
662 #endif
663 			vcp->ev_addr += base_vcp->ev_addr;
664 		}
665 		error = (*vcp->ev_proc)(l, vcp);
666 #ifdef DEBUG_EXEC
667 		if (error) {
668 			int j;
669 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
670 			for (j = 0; j <= i; j++)
671 				uprintf(
672 			    "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
673 				    j, vp[j].ev_addr, vp[j].ev_len,
674 				    vp[j].ev_offset, vp[j].ev_prot,
675 				    vp[j].ev_flags);
676 		}
677 #endif /* DEBUG_EXEC */
678 		if (vcp->ev_flags & VMCMD_BASE)
679 			base_vcp = vcp;
680 	}
681 
682 	/* free the vmspace-creation commands, and release their references */
683 	kill_vmcmds(&pack.ep_vmcmds);
684 
685 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
686 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred, l);
687 	vput(pack.ep_vp);
688 
689 	/* if an error happened, deallocate and punt */
690 	if (error) {
691 		DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error));
692 		goto exec_abort;
693 	}
694 
695 	/* remember information about the process */
696 	arginfo.ps_nargvstr = argc;
697 	arginfo.ps_nenvstr = envc;
698 
699 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
700 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
701 		len - (sizeof(struct ps_strings) + szsigcode));
702 #ifdef __MACHINE_STACK_GROWS_UP
703 	/*
704 	 * The copyargs call always copies into lower addresses
705 	 * first, moving towards higher addresses, starting with
706 	 * the stack pointer that we give.  When the stack grows
707 	 * down, this puts argc/argv/envp very shallow on the
708 	 * stack, right at the first user stack pointer, and puts
709 	 * STACKGAPLEN very deep in the stack.  When the stack
710 	 * grows up, the situation is reversed.
711 	 *
712 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
713 	 * function expects to be called with a single pointer to
714 	 * a region that has a few words it can stash values into,
715 	 * followed by argc/argv/envp.  When the stack grows down,
716 	 * it's easy to decrement the stack pointer a little bit to
717 	 * allocate the space for these few words and pass the new
718 	 * stack pointer to _rtld.  When the stack grows up, however,
719 	 * a few words before argc is part of the signal trampoline, XXX
720 	 * so we have a problem.
721 	 *
722 	 * Instead of changing how _rtld works, we take the easy way
723 	 * out and steal 32 bytes before we call copyargs.  This
724 	 * space is effectively stolen from STACKGAPLEN.
725 	 */
726 	stack += 32;
727 #endif /* __MACHINE_STACK_GROWS_UP */
728 
729 	/* Now copy argc, args & environ to new stack */
730 	error = (*pack.ep_es->es_copyargs)(l, &pack, &arginfo, &stack, argp);
731 	if (error) {
732 		DPRINTF(("execve: copyargs failed %d\n", error));
733 		goto exec_abort;
734 	}
735 	/* Move the stack back to original point */
736 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
737 
738 	/* fill process ps_strings info */
739 	p->p_psstr = (struct ps_strings *)
740 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
741 	    sizeof(struct ps_strings));
742 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
743 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
744 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
745 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
746 
747 	/* copy out the process's ps_strings structure */
748 	if ((error = copyout(aip, (char *)p->p_psstr,
749 	    sizeof(arginfo))) != 0) {
750 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
751 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
752 		goto exec_abort;
753 	}
754 
755 	stopprofclock(p);	/* stop profiling */
756 	fdcloseexec(l);		/* handle close on exec */
757 	execsigs(p);		/* reset catched signals */
758 
759 	l->l_ctxlink = NULL;	/* reset ucontext link */
760 
761 	/* set command name & other accounting info */
762 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
763 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
764 	p->p_comm[len] = 0;
765 	p->p_acflag &= ~AFORK;
766 
767 	p->p_flag |= P_EXEC;
768 	if (p->p_flag & P_PPWAIT) {
769 		p->p_flag &= ~P_PPWAIT;
770 		wakeup((caddr_t) p->p_pptr);
771 	}
772 
773 	/*
774 	 * deal with set[ug]id.
775 	 * MNT_NOSUID has already been used to disable s[ug]id.
776 	 */
777 	if ((p->p_flag & P_TRACED) == 0 &&
778 
779 	    (((attr.va_mode & S_ISUID) != 0 &&
780 	      kauth_cred_geteuid(l->l_cred) != attr.va_uid) ||
781 
782 	     ((attr.va_mode & S_ISGID) != 0 &&
783 	      kauth_cred_getegid(l->l_cred) != attr.va_gid))) {
784 		/*
785 		 * Mark the process as SUGID before we do
786 		 * anything that might block.
787 		 */
788 		p_sugid(p);
789 
790 		/* Make sure file descriptors 0..2 are in use. */
791 		if ((error = fdcheckstd(l)) != 0) {
792 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
793 			goto exec_abort;
794 		}
795 
796 		/*
797 		 * Copy the credential so other references don't see our
798 		 * changes.
799 		 */
800 		l->l_cred = kauth_cred_copy(l->l_cred);
801 #ifdef KTRACE
802 		/*
803 		 * If process is being ktraced, turn off - unless
804 		 * root set it.
805 		 */
806 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
807 			ktrderef(p);
808 #endif
809 		if (attr.va_mode & S_ISUID)
810 			kauth_cred_seteuid(l->l_cred, attr.va_uid);
811 		if (attr.va_mode & S_ISGID)
812 			kauth_cred_setegid(l->l_cred, attr.va_gid);
813 	} else {
814 		if (kauth_cred_geteuid(l->l_cred) ==
815 		    kauth_cred_getuid(l->l_cred) &&
816 		    kauth_cred_getegid(l->l_cred) ==
817 		    kauth_cred_getgid(l->l_cred))
818 			p->p_flag &= ~P_SUGID;
819 	}
820 
821 	/*
822 	 * Copy the credential so other references don't see our changes.
823 	 * Test to see if this is necessary first, since in the common case
824 	 * we won't need a private reference.
825 	 */
826 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
827 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
828 		l->l_cred = kauth_cred_copy(l->l_cred);
829 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
830 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
831 	}
832 
833 	/* Update the master credentials. */
834 	if (l->l_cred != p->p_cred) {
835 		kauth_cred_t ocred;
836 
837 		kauth_cred_hold(l->l_cred);
838 		simple_lock(&p->p_lock);
839 		ocred = p->p_cred;
840 		p->p_cred = l->l_cred;
841 		simple_unlock(&p->p_lock);
842 		kauth_cred_free(ocred);
843 	}
844 
845 #if defined(__HAVE_RAS)
846 	/*
847 	 * Remove all RASs from the address space.
848 	 */
849 	ras_purgeall(p);
850 #endif
851 
852 	doexechooks(p);
853 
854 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
855 
856 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
857 
858 	/* notify others that we exec'd */
859 	KNOTE(&p->p_klist, NOTE_EXEC);
860 
861 	/* setup new registers and do misc. setup. */
862 	(*pack.ep_es->es_emul->e_setregs)(l, &pack, (u_long) stack);
863 	if (pack.ep_es->es_setregs)
864 		(*pack.ep_es->es_setregs)(l, &pack, (u_long) stack);
865 
866 	/* map the process's signal trampoline code */
867 	if (exec_sigcode_map(p, pack.ep_es->es_emul)) {
868 		DPRINTF(("execve: map sigcode failed %d\n", error));
869 		goto exec_abort;
870 	}
871 
872 	if ((p->p_flag & (P_TRACED|P_SYSCALL)) == P_TRACED)
873 		psignal(p, SIGTRAP);
874 
875 	free(pack.ep_hdr, M_EXEC);
876 
877 	/*
878 	 * Call emulation specific exec hook. This can setup per-process
879 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
880 	 *
881 	 * If we are executing process of different emulation than the
882 	 * original forked process, call e_proc_exit() of the old emulation
883 	 * first, then e_proc_exec() of new emulation. If the emulation is
884 	 * same, the exec hook code should deallocate any old emulation
885 	 * resources held previously by this process.
886 	 */
887 	if (p->p_emul && p->p_emul->e_proc_exit
888 	    && p->p_emul != pack.ep_es->es_emul)
889 		(*p->p_emul->e_proc_exit)(p);
890 
891 	/*
892 	 * Call exec hook. Emulation code may NOT store reference to anything
893 	 * from &pack.
894 	 */
895         if (pack.ep_es->es_emul->e_proc_exec)
896                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
897 
898 	/* update p_emul, the old value is no longer needed */
899 	p->p_emul = pack.ep_es->es_emul;
900 
901 	/* ...and the same for p_execsw */
902 	p->p_execsw = pack.ep_es;
903 
904 #ifdef __HAVE_SYSCALL_INTERN
905 	(*p->p_emul->e_syscall_intern)(p);
906 #endif
907 #ifdef KTRACE
908 	if (KTRPOINT(p, KTR_EMUL))
909 		ktremul(l);
910 #endif
911 
912 #ifdef LKM
913 	lockmgr(&exec_lock, LK_RELEASE, NULL);
914 #endif
915 	p->p_flag &= ~P_INEXEC;
916 
917 	if (p->p_flag & P_STOPEXEC) {
918 		int s;
919 
920 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
921 		SCHED_LOCK(s);
922 		p->p_pptr->p_nstopchild++;
923 		p->p_stat = SSTOP;
924 		l->l_stat = LSSTOP;
925 		p->p_nrlwps--;
926 		mi_switch(l, NULL);
927 		SCHED_ASSERT_UNLOCKED();
928 		splx(s);
929 	}
930 
931 #ifdef SYSTRACE
932 	if (ISSET(p->p_flag, P_SYSTRACE) &&
933 	    wassugid && !ISSET(p->p_flag, P_SUGID))
934 		systrace_execve1(pathbuf, p);
935 #endif /* SYSTRACE */
936 
937 	return (EJUSTRETURN);
938 
939  bad:
940 	p->p_flag &= ~P_INEXEC;
941 	/* free the vmspace-creation commands, and release their references */
942 	kill_vmcmds(&pack.ep_vmcmds);
943 	/* kill any opened file descriptor, if necessary */
944 	if (pack.ep_flags & EXEC_HASFD) {
945 		pack.ep_flags &= ~EXEC_HASFD;
946 		(void) fdrelease(l, pack.ep_fd);
947 	}
948 	/* close and put the exec'd file */
949 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
950 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred, l);
951 	vput(pack.ep_vp);
952 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
953 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
954 
955  freehdr:
956 	free(pack.ep_hdr, M_EXEC);
957 
958 #ifdef SYSTRACE
959  clrflg:
960 #endif /* SYSTRACE */
961 	l->l_flag |= oldlwpflags;
962 	p->p_flag &= ~P_INEXEC;
963 #ifdef LKM
964 	lockmgr(&exec_lock, LK_RELEASE, NULL);
965 #endif
966 
967 	return error;
968 
969  exec_abort:
970 	p->p_flag &= ~P_INEXEC;
971 #ifdef LKM
972 	lockmgr(&exec_lock, LK_RELEASE, NULL);
973 #endif
974 
975 	/*
976 	 * the old process doesn't exist anymore.  exit gracefully.
977 	 * get rid of the (new) address space we have created, if any, get rid
978 	 * of our namei data and vnode, and exit noting failure
979 	 */
980 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
981 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
982 	if (pack.ep_emul_arg)
983 		FREE(pack.ep_emul_arg, M_TEMP);
984 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
985 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
986 	free(pack.ep_hdr, M_EXEC);
987 	exit1(l, W_EXITCODE(error, SIGABRT));
988 
989 	/* NOTREACHED */
990 	return 0;
991 }
992 
993 
994 int
995 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
996     char **stackp, void *argp)
997 {
998 	char	**cpp, *dp, *sp;
999 	size_t	len;
1000 	void	*nullp;
1001 	long	argc, envc;
1002 	int	error;
1003 
1004 	cpp = (char **)*stackp;
1005 	nullp = NULL;
1006 	argc = arginfo->ps_nargvstr;
1007 	envc = arginfo->ps_nenvstr;
1008 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
1009 		return error;
1010 
1011 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
1012 	sp = argp;
1013 
1014 	/* XXX don't copy them out, remap them! */
1015 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1016 
1017 	for (; --argc >= 0; sp += len, dp += len)
1018 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1019 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1020 			return error;
1021 
1022 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1023 		return error;
1024 
1025 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1026 
1027 	for (; --envc >= 0; sp += len, dp += len)
1028 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1029 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1030 			return error;
1031 
1032 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1033 		return error;
1034 
1035 	*stackp = (char *)cpp;
1036 	return 0;
1037 }
1038 
1039 #ifdef LKM
1040 /*
1041  * Find an emulation of given name in list of emulations.
1042  * Needs to be called with the exec_lock held.
1043  */
1044 const struct emul *
1045 emul_search(const char *name)
1046 {
1047 	struct emul_entry *it;
1048 
1049 	LIST_FOREACH(it, &el_head, el_list) {
1050 		if (strcmp(name, it->el_emul->e_name) == 0)
1051 			return it->el_emul;
1052 	}
1053 
1054 	return NULL;
1055 }
1056 
1057 /*
1058  * Add an emulation to list, if it's not there already.
1059  */
1060 int
1061 emul_register(const struct emul *emul, int ro_entry)
1062 {
1063 	struct emul_entry	*ee;
1064 	int			error;
1065 
1066 	error = 0;
1067 	lockmgr(&exec_lock, LK_SHARED, NULL);
1068 
1069 	if (emul_search(emul->e_name)) {
1070 		error = EEXIST;
1071 		goto out;
1072 	}
1073 
1074 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
1075 		M_EXEC, M_WAITOK);
1076 	ee->el_emul = emul;
1077 	ee->ro_entry = ro_entry;
1078 	LIST_INSERT_HEAD(&el_head, ee, el_list);
1079 
1080  out:
1081 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1082 	return error;
1083 }
1084 
1085 /*
1086  * Remove emulation with name 'name' from list of supported emulations.
1087  */
1088 int
1089 emul_unregister(const char *name)
1090 {
1091 	const struct proclist_desc *pd;
1092 	struct emul_entry	*it;
1093 	int			i, error;
1094 	struct proc		*ptmp;
1095 
1096 	error = 0;
1097 	lockmgr(&exec_lock, LK_SHARED, NULL);
1098 
1099 	LIST_FOREACH(it, &el_head, el_list) {
1100 		if (strcmp(it->el_emul->e_name, name) == 0)
1101 			break;
1102 	}
1103 
1104 	if (!it) {
1105 		error = ENOENT;
1106 		goto out;
1107 	}
1108 
1109 	if (it->ro_entry) {
1110 		error = EBUSY;
1111 		goto out;
1112 	}
1113 
1114 	/* test if any execw[] entry is still using this */
1115 	for(i=0; i < nexecs; i++) {
1116 		if (execsw[i]->es_emul == it->el_emul) {
1117 			error = EBUSY;
1118 			goto out;
1119 		}
1120 	}
1121 
1122 	/*
1123 	 * Test if any process is running under this emulation - since
1124 	 * emul_unregister() is running quite sendomly, it's better
1125 	 * to do expensive check here than to use any locking.
1126 	 */
1127 	proclist_lock_read();
1128 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
1129 		PROCLIST_FOREACH(ptmp, pd->pd_list) {
1130 			if (ptmp->p_emul == it->el_emul) {
1131 				error = EBUSY;
1132 				break;
1133 			}
1134 		}
1135 	}
1136 	proclist_unlock_read();
1137 
1138 	if (error)
1139 		goto out;
1140 
1141 
1142 	/* entry is not used, remove it */
1143 	LIST_REMOVE(it, el_list);
1144 	FREE(it, M_EXEC);
1145 
1146  out:
1147 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1148 	return error;
1149 }
1150 
1151 /*
1152  * Add execsw[] entry.
1153  */
1154 int
1155 exec_add(struct execsw *esp, const char *e_name)
1156 {
1157 	struct exec_entry	*it;
1158 	int			error;
1159 
1160 	error = 0;
1161 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
1162 
1163 	if (!esp->es_emul) {
1164 		esp->es_emul = emul_search(e_name);
1165 		if (!esp->es_emul) {
1166 			error = ENOENT;
1167 			goto out;
1168 		}
1169 	}
1170 
1171 	LIST_FOREACH(it, &ex_head, ex_list) {
1172 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1173 		if (it->es->es_makecmds == esp->es_makecmds
1174 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1175 		    && it->es->es_emul == esp->es_emul) {
1176 			error = EEXIST;
1177 			goto out;
1178 		}
1179 	}
1180 
1181 	/* if we got here, the entry doesn't exist yet */
1182 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
1183 		M_EXEC, M_WAITOK);
1184 	it->es = esp;
1185 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
1186 
1187 	/* update execsw[] */
1188 	exec_init(0);
1189 
1190  out:
1191 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1192 	return error;
1193 }
1194 
1195 /*
1196  * Remove execsw[] entry.
1197  */
1198 int
1199 exec_remove(const struct execsw *esp)
1200 {
1201 	struct exec_entry	*it;
1202 	int			error;
1203 
1204 	error = 0;
1205 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
1206 
1207 	LIST_FOREACH(it, &ex_head, ex_list) {
1208 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1209 		if (it->es->es_makecmds == esp->es_makecmds
1210 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1211 		    && it->es->es_emul == esp->es_emul)
1212 			break;
1213 	}
1214 	if (!it) {
1215 		error = ENOENT;
1216 		goto out;
1217 	}
1218 
1219 	/* remove item from list and free resources */
1220 	LIST_REMOVE(it, ex_list);
1221 	FREE(it, M_EXEC);
1222 
1223 	/* update execsw[] */
1224 	exec_init(0);
1225 
1226  out:
1227 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1228 	return error;
1229 }
1230 
1231 static void
1232 link_es(struct execsw_entry **listp, const struct execsw *esp)
1233 {
1234 	struct execsw_entry *et, *e1;
1235 
1236 	et = (struct execsw_entry *) malloc(sizeof(struct execsw_entry),
1237 			M_TEMP, M_WAITOK);
1238 	et->next = NULL;
1239 	et->es = esp;
1240 	if (*listp == NULL) {
1241 		*listp = et;
1242 		return;
1243 	}
1244 
1245 	switch(et->es->es_prio) {
1246 	case EXECSW_PRIO_FIRST:
1247 		/* put new entry as the first */
1248 		et->next = *listp;
1249 		*listp = et;
1250 		break;
1251 	case EXECSW_PRIO_ANY:
1252 		/* put new entry after all *_FIRST and *_ANY entries */
1253 		for(e1 = *listp; e1->next
1254 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
1255 			e1 = e1->next);
1256 		et->next = e1->next;
1257 		e1->next = et;
1258 		break;
1259 	case EXECSW_PRIO_LAST:
1260 		/* put new entry as the last one */
1261 		for(e1 = *listp; e1->next; e1 = e1->next);
1262 		e1->next = et;
1263 		break;
1264 	default:
1265 #ifdef DIAGNOSTIC
1266 		panic("execw[] entry with unknown priority %d found",
1267 			et->es->es_prio);
1268 #else
1269 		free(et, M_TEMP);
1270 #endif
1271 		break;
1272 	}
1273 }
1274 
1275 /*
1276  * Initialize exec structures. If init_boot is true, also does necessary
1277  * one-time initialization (it's called from main() that way).
1278  * Once system is multiuser, this should be called with exec_lock held,
1279  * i.e. via exec_{add|remove}().
1280  */
1281 int
1282 exec_init(int init_boot)
1283 {
1284 	const struct execsw	**new_es, * const *old_es;
1285 	struct execsw_entry	*list, *e1;
1286 	struct exec_entry	*e2;
1287 	int			i, es_sz;
1288 
1289 	if (init_boot) {
1290 		/* do one-time initializations */
1291 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1292 
1293 		/* register compiled-in emulations */
1294 		for(i=0; i < nexecs_builtin; i++) {
1295 			if (execsw_builtin[i].es_emul)
1296 				emul_register(execsw_builtin[i].es_emul, 1);
1297 		}
1298 #ifdef DIAGNOSTIC
1299 		if (i == 0)
1300 			panic("no emulations found in execsw_builtin[]");
1301 #endif
1302 	}
1303 
1304 	/*
1305 	 * Build execsw[] array from builtin entries and entries added
1306 	 * at runtime.
1307 	 */
1308 	list = NULL;
1309 	for(i=0; i < nexecs_builtin; i++)
1310 		link_es(&list, &execsw_builtin[i]);
1311 
1312 	/* Add dynamically loaded entries */
1313 	es_sz = nexecs_builtin;
1314 	LIST_FOREACH(e2, &ex_head, ex_list) {
1315 		link_es(&list, e2->es);
1316 		es_sz++;
1317 	}
1318 
1319 	/*
1320 	 * Now that we have sorted all execw entries, create new execsw[]
1321 	 * and free no longer needed memory in the process.
1322 	 */
1323 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
1324 	for(i=0; list; i++) {
1325 		new_es[i] = list->es;
1326 		e1 = list->next;
1327 		free(list, M_TEMP);
1328 		list = e1;
1329 	}
1330 
1331 	/*
1332 	 * New execsw[] array built, now replace old execsw[] and free
1333 	 * used memory.
1334 	 */
1335 	old_es = execsw;
1336 	execsw = new_es;
1337 	nexecs = es_sz;
1338 	if (old_es)
1339 		/*XXXUNCONST*/
1340 		free(__UNCONST(old_es), M_EXEC);
1341 
1342 	/*
1343 	 * Figure out the maximum size of an exec header.
1344 	 */
1345 	exec_maxhdrsz = 0;
1346 	for (i = 0; i < nexecs; i++) {
1347 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1348 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1349 	}
1350 
1351 	return 0;
1352 }
1353 #endif
1354 
1355 #ifndef LKM
1356 /*
1357  * Simplified exec_init() for kernels without LKMs. Only initialize
1358  * exec_maxhdrsz and execsw[].
1359  */
1360 int
1361 exec_init(int init_boot)
1362 {
1363 	int i;
1364 
1365 #ifdef DIAGNOSTIC
1366 	if (!init_boot)
1367 		panic("exec_init(): called with init_boot == 0");
1368 #endif
1369 
1370 	/* do one-time initializations */
1371 	nexecs = nexecs_builtin;
1372 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
1373 
1374 	/*
1375 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1376 	 */
1377 	exec_maxhdrsz = 0;
1378 	for(i=0; i < nexecs; i++) {
1379 		execsw[i] = &execsw_builtin[i];
1380 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1381 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1382 	}
1383 
1384 	return 0;
1385 
1386 }
1387 #endif /* !LKM */
1388 
1389 static int
1390 exec_sigcode_map(struct proc *p, const struct emul *e)
1391 {
1392 	vaddr_t va;
1393 	vsize_t sz;
1394 	int error;
1395 	struct uvm_object *uobj;
1396 
1397 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1398 
1399 	if (e->e_sigobject == NULL || sz == 0) {
1400 		return 0;
1401 	}
1402 
1403 	/*
1404 	 * If we don't have a sigobject for this emulation, create one.
1405 	 *
1406 	 * sigobject is an anonymous memory object (just like SYSV shared
1407 	 * memory) that we keep a permanent reference to and that we map
1408 	 * in all processes that need this sigcode. The creation is simple,
1409 	 * we create an object, add a permanent reference to it, map it in
1410 	 * kernel space, copy out the sigcode to it and unmap it.
1411 	 * We map it with PROT_READ|PROT_EXEC into the process just
1412 	 * the way sys_mmap() would map it.
1413 	 */
1414 
1415 	uobj = *e->e_sigobject;
1416 	if (uobj == NULL) {
1417 		uobj = uao_create(sz, 0);
1418 		(*uobj->pgops->pgo_reference)(uobj);
1419 		va = vm_map_min(kernel_map);
1420 		if ((error = uvm_map(kernel_map, &va, round_page(sz),
1421 		    uobj, 0, 0,
1422 		    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1423 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1424 			printf("kernel mapping failed %d\n", error);
1425 			(*uobj->pgops->pgo_detach)(uobj);
1426 			return (error);
1427 		}
1428 		memcpy((void *)va, e->e_sigcode, sz);
1429 #ifdef PMAP_NEED_PROCWR
1430 		pmap_procwr(&proc0, va, sz);
1431 #endif
1432 		uvm_unmap(kernel_map, va, va + round_page(sz));
1433 		*e->e_sigobject = uobj;
1434 	}
1435 
1436 	/* Just a hint to uvm_map where to put it. */
1437 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1438 	    round_page(sz));
1439 
1440 #ifdef __alpha__
1441 	/*
1442 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1443 	 * which causes the above calculation to put the sigcode at
1444 	 * an invalid address.  Put it just below the text instead.
1445 	 */
1446 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1447 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1448 	}
1449 #endif
1450 
1451 	(*uobj->pgops->pgo_reference)(uobj);
1452 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1453 			uobj, 0, 0,
1454 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1455 				    UVM_ADV_RANDOM, 0));
1456 	if (error) {
1457 		(*uobj->pgops->pgo_detach)(uobj);
1458 		return (error);
1459 	}
1460 	p->p_sigctx.ps_sigcode = (void *)va;
1461 	return (0);
1462 }
1463