xref: /netbsd-src/sys/kern/kern_exec.c (revision da9817918ec7e88db2912a2882967c7570a83f47)
1 /*	$NetBSD: kern_exec.c,v 1.289 2009/06/02 23:21:38 pooka Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*-
30  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
31  * Copyright (C) 1992 Wolfgang Solfrank.
32  * Copyright (C) 1992 TooLs GmbH.
33  * All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. All advertising materials mentioning features or use of this software
44  *    must display the following acknowledgement:
45  *	This product includes software developed by TooLs GmbH.
46  * 4. The name of TooLs GmbH may not be used to endorse or promote products
47  *    derived from this software without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
50  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
51  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
52  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
53  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
54  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
55  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
56  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
57  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
58  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
59  */
60 
61 #include <sys/cdefs.h>
62 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.289 2009/06/02 23:21:38 pooka Exp $");
63 
64 #include "opt_ktrace.h"
65 #include "opt_modular.h"
66 #include "opt_syscall_debug.h"
67 #include "veriexec.h"
68 #include "opt_pax.h"
69 #include "opt_sa.h"
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/filedesc.h>
74 #include <sys/kernel.h>
75 #include <sys/proc.h>
76 #include <sys/mount.h>
77 #include <sys/malloc.h>
78 #include <sys/kmem.h>
79 #include <sys/namei.h>
80 #include <sys/vnode.h>
81 #include <sys/file.h>
82 #include <sys/acct.h>
83 #include <sys/exec.h>
84 #include <sys/ktrace.h>
85 #include <sys/uidinfo.h>
86 #include <sys/wait.h>
87 #include <sys/mman.h>
88 #include <sys/ras.h>
89 #include <sys/signalvar.h>
90 #include <sys/stat.h>
91 #include <sys/syscall.h>
92 #include <sys/kauth.h>
93 #include <sys/lwpctl.h>
94 #include <sys/pax.h>
95 #include <sys/cpu.h>
96 #include <sys/module.h>
97 #include <sys/sa.h>
98 #include <sys/savar.h>
99 #include <sys/syscallvar.h>
100 #include <sys/syscallargs.h>
101 #if NVERIEXEC > 0
102 #include <sys/verified_exec.h>
103 #endif /* NVERIEXEC > 0 */
104 
105 #include <uvm/uvm_extern.h>
106 
107 #include <machine/reg.h>
108 
109 #include <compat/common/compat_util.h>
110 
111 static int exec_sigcode_map(struct proc *, const struct emul *);
112 
113 #ifdef DEBUG_EXEC
114 #define DPRINTF(a) uprintf a
115 #else
116 #define DPRINTF(a)
117 #endif /* DEBUG_EXEC */
118 
119 /*
120  * Exec function switch:
121  *
122  * Note that each makecmds function is responsible for loading the
123  * exec package with the necessary functions for any exec-type-specific
124  * handling.
125  *
126  * Functions for specific exec types should be defined in their own
127  * header file.
128  */
129 static const struct execsw	**execsw = NULL;
130 static int			nexecs;
131 
132 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
133 
134 /* list of dynamically loaded execsw entries */
135 static LIST_HEAD(execlist_head, exec_entry) ex_head =
136     LIST_HEAD_INITIALIZER(ex_head);
137 struct exec_entry {
138 	LIST_ENTRY(exec_entry)	ex_list;
139 	SLIST_ENTRY(exec_entry)	ex_slist;
140 	const struct execsw	*ex_sw;
141 };
142 
143 #ifndef __HAVE_SYSCALL_INTERN
144 void	syscall(void);
145 #endif
146 
147 #ifdef KERN_SA
148 static struct sa_emul saemul_netbsd = {
149 	sizeof(ucontext_t),
150 	sizeof(struct sa_t),
151 	sizeof(struct sa_t *),
152 	NULL,
153 	NULL,
154 	cpu_upcall,
155 	(void (*)(struct lwp *, void *))getucontext_sa,
156 	sa_ucsp
157 };
158 #endif /* KERN_SA */
159 
160 /* NetBSD emul struct */
161 struct emul emul_netbsd = {
162 	"netbsd",
163 	NULL,		/* emulation path */
164 #ifndef __HAVE_MINIMAL_EMUL
165 	EMUL_HAS_SYS___syscall,
166 	NULL,
167 	SYS_syscall,
168 	SYS_NSYSENT,
169 #endif
170 	sysent,
171 #ifdef SYSCALL_DEBUG
172 	syscallnames,
173 #else
174 	NULL,
175 #endif
176 	sendsig,
177 	trapsignal,
178 	NULL,
179 	NULL,
180 	NULL,
181 	NULL,
182 	setregs,
183 	NULL,
184 	NULL,
185 	NULL,
186 	NULL,
187 	NULL,
188 #ifdef __HAVE_SYSCALL_INTERN
189 	syscall_intern,
190 #else
191 	syscall,
192 #endif
193 	NULL,
194 	NULL,
195 
196 	uvm_default_mapaddr,
197 	NULL,
198 #ifdef KERN_SA
199 	&saemul_netbsd,
200 #else
201 	NULL,
202 #endif
203 	sizeof(ucontext_t),
204 	startlwp,
205 };
206 
207 /*
208  * Exec lock. Used to control access to execsw[] structures.
209  * This must not be static so that netbsd32 can access it, too.
210  */
211 krwlock_t exec_lock;
212 
213 static kmutex_t sigobject_lock;
214 
215 static void *
216 exec_pool_alloc(struct pool *pp, int flags)
217 {
218 
219 	return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
220 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
221 }
222 
223 static void
224 exec_pool_free(struct pool *pp, void *addr)
225 {
226 
227 	uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
228 }
229 
230 static struct pool exec_pool;
231 
232 static struct pool_allocator exec_palloc = {
233 	.pa_alloc = exec_pool_alloc,
234 	.pa_free = exec_pool_free,
235 	.pa_pagesz = NCARGS
236 };
237 
238 /*
239  * check exec:
240  * given an "executable" described in the exec package's namei info,
241  * see what we can do with it.
242  *
243  * ON ENTRY:
244  *	exec package with appropriate namei info
245  *	lwp pointer of exec'ing lwp
246  *	NO SELF-LOCKED VNODES
247  *
248  * ON EXIT:
249  *	error:	nothing held, etc.  exec header still allocated.
250  *	ok:	filled exec package, executable's vnode (unlocked).
251  *
252  * EXEC SWITCH ENTRY:
253  * 	Locked vnode to check, exec package, proc.
254  *
255  * EXEC SWITCH EXIT:
256  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
257  *	error:	destructive:
258  *			everything deallocated execept exec header.
259  *		non-destructive:
260  *			error code, executable's vnode (unlocked),
261  *			exec header unmodified.
262  */
263 int
264 /*ARGSUSED*/
265 check_exec(struct lwp *l, struct exec_package *epp)
266 {
267 	int		error, i;
268 	struct vnode	*vp;
269 	struct nameidata *ndp;
270 	size_t		resid;
271 
272 	ndp = epp->ep_ndp;
273 	ndp->ni_cnd.cn_nameiop = LOOKUP;
274 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME | TRYEMULROOT;
275 	/* first get the vnode */
276 	if ((error = namei(ndp)) != 0)
277 		return error;
278 	epp->ep_vp = vp = ndp->ni_vp;
279 
280 	/* check access and type */
281 	if (vp->v_type != VREG) {
282 		error = EACCES;
283 		goto bad1;
284 	}
285 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
286 		goto bad1;
287 
288 	/* get attributes */
289 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
290 		goto bad1;
291 
292 	/* Check mount point */
293 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
294 		error = EACCES;
295 		goto bad1;
296 	}
297 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
298 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
299 
300 	/* try to open it */
301 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
302 		goto bad1;
303 
304 	/* unlock vp, since we need it unlocked from here on out. */
305 	VOP_UNLOCK(vp, 0);
306 
307 #if NVERIEXEC > 0
308 	error = veriexec_verify(l, vp, ndp->ni_cnd.cn_pnbuf,
309 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
310 	    NULL);
311 	if (error)
312 		goto bad2;
313 #endif /* NVERIEXEC > 0 */
314 
315 #ifdef PAX_SEGVGUARD
316 	error = pax_segvguard(l, vp, ndp->ni_cnd.cn_pnbuf, false);
317 	if (error)
318 		goto bad2;
319 #endif /* PAX_SEGVGUARD */
320 
321 	/* now we have the file, get the exec header */
322 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
323 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
324 	if (error)
325 		goto bad2;
326 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
327 
328 	/*
329 	 * Set up default address space limits.  Can be overridden
330 	 * by individual exec packages.
331 	 *
332 	 * XXX probably should be all done in the exec packages.
333 	 */
334 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
335 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
336 	/*
337 	 * set up the vmcmds for creation of the process
338 	 * address space
339 	 */
340 	error = ENOEXEC;
341 	for (i = 0; i < nexecs; i++) {
342 		int newerror;
343 
344 		epp->ep_esch = execsw[i];
345 		newerror = (*execsw[i]->es_makecmds)(l, epp);
346 
347 		if (!newerror) {
348 			/* Seems ok: check that entry point is sane */
349 			if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
350 				error = ENOEXEC;
351 				break;
352 			}
353 
354 			/* check limits */
355 			if ((epp->ep_tsize > MAXTSIZ) ||
356 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
357 						    [RLIMIT_DATA].rlim_cur)) {
358 				error = ENOMEM;
359 				break;
360 			}
361 			return 0;
362 		}
363 
364 		if (epp->ep_emul_root != NULL) {
365 			vrele(epp->ep_emul_root);
366 			epp->ep_emul_root = NULL;
367 		}
368 		if (epp->ep_interp != NULL) {
369 			vrele(epp->ep_interp);
370 			epp->ep_interp = NULL;
371 		}
372 
373 		/* make sure the first "interesting" error code is saved. */
374 		if (error == ENOEXEC)
375 			error = newerror;
376 
377 		if (epp->ep_flags & EXEC_DESTR)
378 			/* Error from "#!" code, tidied up by recursive call */
379 			return error;
380 	}
381 
382 	/* not found, error */
383 
384 	/*
385 	 * free any vmspace-creation commands,
386 	 * and release their references
387 	 */
388 	kill_vmcmds(&epp->ep_vmcmds);
389 
390 bad2:
391 	/*
392 	 * close and release the vnode, restore the old one, free the
393 	 * pathname buf, and punt.
394 	 */
395 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
396 	VOP_CLOSE(vp, FREAD, l->l_cred);
397 	vput(vp);
398 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
399 	return error;
400 
401 bad1:
402 	/*
403 	 * free the namei pathname buffer, and put the vnode
404 	 * (which we don't yet have open).
405 	 */
406 	vput(vp);				/* was still locked */
407 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
408 	return error;
409 }
410 
411 #ifdef __MACHINE_STACK_GROWS_UP
412 #define STACK_PTHREADSPACE NBPG
413 #else
414 #define STACK_PTHREADSPACE 0
415 #endif
416 
417 static int
418 execve_fetch_element(char * const *array, size_t index, char **value)
419 {
420 	return copyin(array + index, value, sizeof(*value));
421 }
422 
423 /*
424  * exec system call
425  */
426 /* ARGSUSED */
427 int
428 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
429 {
430 	/* {
431 		syscallarg(const char *)	path;
432 		syscallarg(char * const *)	argp;
433 		syscallarg(char * const *)	envp;
434 	} */
435 
436 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
437 	    SCARG(uap, envp), execve_fetch_element);
438 }
439 
440 /*
441  * Load modules to try and execute an image that we do not understand.
442  * If no execsw entries are present, we load those likely to be needed
443  * in order to run native images only.  Otherwise, we autoload all
444  * possible modules that could let us run the binary.  XXX lame
445  */
446 static void
447 exec_autoload(void)
448 {
449 #ifdef MODULAR
450 	static const char * const native[] = {
451 		"exec_elf32",
452 		"exec_elf64",
453 		"exec_script",
454 		NULL
455 	};
456 	static const char * const compat[] = {
457 		"exec_elf32",
458 		"exec_elf64",
459 		"exec_script",
460 		"exec_aout",
461 		"exec_coff",
462 		"exec_ecoff",
463 		"compat_aoutm68k",
464 		"compat_freebsd",
465 		"compat_ibcs2",
466 		"compat_irix",
467 		"compat_linux",
468 		"compat_linux32",
469 		"compat_netbsd32",
470 		"compat_sunos",
471 		"compat_sunos32",
472 		"compat_svr4",
473 		"compat_svr4_32",
474 		"compat_ultrix",
475 		NULL
476 	};
477 	char const * const *list;
478 	int i;
479 
480 	mutex_enter(&module_lock);
481 	list = (nexecs == 0 ? native : compat);
482 	for (i = 0; list[i] != NULL; i++) {
483 		if (module_autoload(list[i], MODULE_CLASS_MISC) != 0) {
484 		    	continue;
485 		}
486 		mutex_exit(&module_lock);
487 	   	yield();
488 		mutex_enter(&module_lock);
489 	}
490 	mutex_exit(&module_lock);
491 #endif
492 }
493 
494 int
495 execve1(struct lwp *l, const char *path, char * const *args,
496     char * const *envs, execve_fetch_element_t fetch_element)
497 {
498 	int			error;
499 	struct exec_package	pack;
500 	struct nameidata	nid;
501 	struct vattr		attr;
502 	struct proc		*p;
503 	char			*argp;
504 	char			*dp, *sp;
505 	long			argc, envc;
506 	size_t			i, len;
507 	char			*stack;
508 	struct ps_strings	arginfo;
509 	struct ps_strings	*aip = &arginfo;
510 	struct vmspace		*vm;
511 	struct exec_fakearg	*tmpfap;
512 	int			szsigcode;
513 	struct exec_vmcmd	*base_vcp;
514 	int			oldlwpflags;
515 	ksiginfo_t		ksi;
516 	ksiginfoq_t		kq;
517 	char			*pathbuf;
518 	size_t			pathbuflen;
519 	u_int			modgen;
520 
521 	p = l->l_proc;
522  	modgen = 0;
523 
524 	/*
525 	 * Check if we have exceeded our number of processes limit.
526 	 * This is so that we handle the case where a root daemon
527 	 * forked, ran setuid to become the desired user and is trying
528 	 * to exec. The obvious place to do the reference counting check
529 	 * is setuid(), but we don't do the reference counting check there
530 	 * like other OS's do because then all the programs that use setuid()
531 	 * must be modified to check the return code of setuid() and exit().
532 	 * It is dangerous to make setuid() fail, because it fails open and
533 	 * the program will continue to run as root. If we make it succeed
534 	 * and return an error code, again we are not enforcing the limit.
535 	 * The best place to enforce the limit is here, when the process tries
536 	 * to execute a new image, because eventually the process will need
537 	 * to call exec in order to do something useful.
538 	 */
539  retry:
540 	if ((p->p_flag & PK_SUGID) && kauth_authorize_generic(l->l_cred,
541 	    KAUTH_GENERIC_ISSUSER, NULL) != 0 && chgproccnt(kauth_cred_getuid(
542 	    l->l_cred), 0) > p->p_rlimit[RLIMIT_NPROC].rlim_cur)
543 		return EAGAIN;
544 
545 	oldlwpflags = l->l_flag & (LW_SA | LW_SA_UPCALL);
546 	if (l->l_flag & LW_SA) {
547 		lwp_lock(l);
548 		l->l_flag &= ~(LW_SA | LW_SA_UPCALL);
549 		lwp_unlock(l);
550 	}
551 
552 	/*
553 	 * Drain existing references and forbid new ones.  The process
554 	 * should be left alone until we're done here.  This is necessary
555 	 * to avoid race conditions - e.g. in ptrace() - that might allow
556 	 * a local user to illicitly obtain elevated privileges.
557 	 */
558 	rw_enter(&p->p_reflock, RW_WRITER);
559 
560 	base_vcp = NULL;
561 	/*
562 	 * Init the namei data to point the file user's program name.
563 	 * This is done here rather than in check_exec(), so that it's
564 	 * possible to override this settings if any of makecmd/probe
565 	 * functions call check_exec() recursively - for example,
566 	 * see exec_script_makecmds().
567 	 */
568 	pathbuf = PNBUF_GET();
569 	error = copyinstr(path, pathbuf, MAXPATHLEN, &pathbuflen);
570 	if (error) {
571 		DPRINTF(("execve: copyinstr path %d", error));
572 		goto clrflg;
573 	}
574 
575 	NDINIT(&nid, LOOKUP, NOFOLLOW | TRYEMULROOT, UIO_SYSSPACE, pathbuf);
576 
577 	/*
578 	 * initialize the fields of the exec package.
579 	 */
580 	pack.ep_name = path;
581 	pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
582 	pack.ep_hdrlen = exec_maxhdrsz;
583 	pack.ep_hdrvalid = 0;
584 	pack.ep_ndp = &nid;
585 	pack.ep_emul_arg = NULL;
586 	pack.ep_vmcmds.evs_cnt = 0;
587 	pack.ep_vmcmds.evs_used = 0;
588 	pack.ep_vap = &attr;
589 	pack.ep_flags = 0;
590 	pack.ep_emul_root = NULL;
591 	pack.ep_interp = NULL;
592 	pack.ep_esch = NULL;
593 	pack.ep_pax_flags = 0;
594 
595 	rw_enter(&exec_lock, RW_READER);
596 
597 	/* see if we can run it. */
598 	if ((error = check_exec(l, &pack)) != 0) {
599 		if (error != ENOENT) {
600 			DPRINTF(("execve: check exec failed %d\n", error));
601 		}
602 		goto freehdr;
603 	}
604 
605 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
606 
607 	/* allocate an argument buffer */
608 	argp = pool_get(&exec_pool, PR_WAITOK);
609 	KASSERT(argp != NULL);
610 	dp = argp;
611 	argc = 0;
612 
613 	/* copy the fake args list, if there's one, freeing it as we go */
614 	if (pack.ep_flags & EXEC_HASARGL) {
615 		tmpfap = pack.ep_fa;
616 		while (tmpfap->fa_arg != NULL) {
617 			const char *cp;
618 
619 			cp = tmpfap->fa_arg;
620 			while (*cp)
621 				*dp++ = *cp++;
622 			*dp++ = '\0';
623 
624 			kmem_free(tmpfap->fa_arg, tmpfap->fa_len);
625 			tmpfap++; argc++;
626 		}
627 		kmem_free(pack.ep_fa, pack.ep_fa_len);
628 		pack.ep_flags &= ~EXEC_HASARGL;
629 	}
630 
631 	/* Now get argv & environment */
632 	if (args == NULL) {
633 		DPRINTF(("execve: null args\n"));
634 		error = EINVAL;
635 		goto bad;
636 	}
637 	/* 'i' will index the argp/envp element to be retrieved */
638 	i = 0;
639 	if (pack.ep_flags & EXEC_SKIPARG)
640 		i++;
641 
642 	while (1) {
643 		len = argp + ARG_MAX - dp;
644 		if ((error = (*fetch_element)(args, i, &sp)) != 0) {
645 			DPRINTF(("execve: fetch_element args %d\n", error));
646 			goto bad;
647 		}
648 		if (!sp)
649 			break;
650 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
651 			DPRINTF(("execve: copyinstr args %d\n", error));
652 			if (error == ENAMETOOLONG)
653 				error = E2BIG;
654 			goto bad;
655 		}
656 		ktrexecarg(dp, len - 1);
657 		dp += len;
658 		i++;
659 		argc++;
660 	}
661 
662 	envc = 0;
663 	/* environment need not be there */
664 	if (envs != NULL) {
665 		i = 0;
666 		while (1) {
667 			len = argp + ARG_MAX - dp;
668 			if ((error = (*fetch_element)(envs, i, &sp)) != 0) {
669 				DPRINTF(("execve: fetch_element env %d\n", error));
670 				goto bad;
671 			}
672 			if (!sp)
673 				break;
674 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
675 				DPRINTF(("execve: copyinstr env %d\n", error));
676 				if (error == ENAMETOOLONG)
677 					error = E2BIG;
678 				goto bad;
679 			}
680 			ktrexecenv(dp, len - 1);
681 			dp += len;
682 			i++;
683 			envc++;
684 		}
685 	}
686 
687 	dp = (char *) ALIGN(dp);
688 
689 	szsigcode = pack.ep_esch->es_emul->e_esigcode -
690 	    pack.ep_esch->es_emul->e_sigcode;
691 
692 #ifdef __MACHINE_STACK_GROWS_UP
693 /* See big comment lower down */
694 #define	RTLD_GAP	32
695 #else
696 #define	RTLD_GAP	0
697 #endif
698 
699 	/* Now check if args & environ fit into new stack */
700 	if (pack.ep_flags & EXEC_32)
701 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
702 		    sizeof(int) + sizeof(int) + dp + RTLD_GAP +
703 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
704 		    - argp;
705 	else
706 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
707 		    sizeof(char *) + sizeof(int) + dp + RTLD_GAP +
708 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
709 		    - argp;
710 
711 #ifdef PAX_ASLR
712 	if (pax_aslr_active(l))
713 		len += (arc4random() % PAGE_SIZE);
714 #endif /* PAX_ASLR */
715 
716 #ifdef STACKLALIGN	/* arm, etc. */
717 	len = STACKALIGN(len);	/* make the stack "safely" aligned */
718 #else
719 	len = ALIGN(len);	/* make the stack "safely" aligned */
720 #endif
721 
722 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
723 		DPRINTF(("execve: stack limit exceeded %zu\n", len));
724 		error = ENOMEM;
725 		goto bad;
726 	}
727 
728 	/* Get rid of other LWPs. */
729 	if (p->p_sa || p->p_nlwps > 1) {
730 		mutex_enter(p->p_lock);
731 		exit_lwps(l);
732 		mutex_exit(p->p_lock);
733 	}
734 	KDASSERT(p->p_nlwps == 1);
735 
736 	/* Destroy any lwpctl info. */
737 	if (p->p_lwpctl != NULL)
738 		lwp_ctl_exit();
739 
740 	/* This is now LWP 1 */
741 	l->l_lid = 1;
742 	p->p_nlwpid = 1;
743 
744 #ifdef KERN_SA
745 	/* Release any SA state. */
746 	if (p->p_sa)
747 		sa_release(p);
748 #endif /* KERN_SA */
749 
750 	/* Remove POSIX timers */
751 	timers_free(p, TIMERS_POSIX);
752 
753 	/* adjust "active stack depth" for process VSZ */
754 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
755 
756 	/*
757 	 * Do whatever is necessary to prepare the address space
758 	 * for remapping.  Note that this might replace the current
759 	 * vmspace with another!
760 	 */
761 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
762 
763 	/* record proc's vnode, for use by procfs and others */
764         if (p->p_textvp)
765                 vrele(p->p_textvp);
766 	VREF(pack.ep_vp);
767 	p->p_textvp = pack.ep_vp;
768 
769 	/* Now map address space */
770 	vm = p->p_vmspace;
771 	vm->vm_taddr = (void *)pack.ep_taddr;
772 	vm->vm_tsize = btoc(pack.ep_tsize);
773 	vm->vm_daddr = (void*)pack.ep_daddr;
774 	vm->vm_dsize = btoc(pack.ep_dsize);
775 	vm->vm_ssize = btoc(pack.ep_ssize);
776 	vm->vm_issize = 0;
777 	vm->vm_maxsaddr = (void *)pack.ep_maxsaddr;
778 	vm->vm_minsaddr = (void *)pack.ep_minsaddr;
779 
780 #ifdef PAX_ASLR
781 	pax_aslr_init(l, vm);
782 #endif /* PAX_ASLR */
783 
784 	/* create the new process's VM space by running the vmcmds */
785 #ifdef DIAGNOSTIC
786 	if (pack.ep_vmcmds.evs_used == 0)
787 		panic("execve: no vmcmds");
788 #endif
789 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
790 		struct exec_vmcmd *vcp;
791 
792 		vcp = &pack.ep_vmcmds.evs_cmds[i];
793 		if (vcp->ev_flags & VMCMD_RELATIVE) {
794 #ifdef DIAGNOSTIC
795 			if (base_vcp == NULL)
796 				panic("execve: relative vmcmd with no base");
797 			if (vcp->ev_flags & VMCMD_BASE)
798 				panic("execve: illegal base & relative vmcmd");
799 #endif
800 			vcp->ev_addr += base_vcp->ev_addr;
801 		}
802 		error = (*vcp->ev_proc)(l, vcp);
803 #ifdef DEBUG_EXEC
804 		if (error) {
805 			size_t j;
806 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
807 			for (j = 0; j <= i; j++)
808 				uprintf(
809 			"vmcmd[%zu] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
810 				    j, vp[j].ev_addr, vp[j].ev_len,
811 				    vp[j].ev_offset, vp[j].ev_prot,
812 				    vp[j].ev_flags);
813 		}
814 #endif /* DEBUG_EXEC */
815 		if (vcp->ev_flags & VMCMD_BASE)
816 			base_vcp = vcp;
817 	}
818 
819 	/* free the vmspace-creation commands, and release their references */
820 	kill_vmcmds(&pack.ep_vmcmds);
821 
822 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
823 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
824 	vput(pack.ep_vp);
825 
826 	/* if an error happened, deallocate and punt */
827 	if (error) {
828 		DPRINTF(("execve: vmcmd %zu failed: %d\n", i - 1, error));
829 		goto exec_abort;
830 	}
831 
832 	/* remember information about the process */
833 	arginfo.ps_nargvstr = argc;
834 	arginfo.ps_nenvstr = envc;
835 
836 	/* set command name & other accounting info */
837 	i = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
838 	(void)memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, i);
839 	p->p_comm[i] = '\0';
840 
841 	dp = PNBUF_GET();
842 	/*
843 	 * If the path starts with /, we don't need to do any work.
844 	 * This handles the majority of the cases.
845 	 * In the future perhaps we could canonicalize it?
846 	 */
847 	if (pathbuf[0] == '/')
848 		(void)strlcpy(pack.ep_path = dp, pathbuf, MAXPATHLEN);
849 #ifdef notyet
850 	/*
851 	 * Although this works most of the time [since the entry was just
852 	 * entered in the cache] we don't use it because it theoretically
853 	 * can fail and it is not the cleanest interface, because there
854 	 * could be races. When the namei cache is re-written, this can
855 	 * be changed to use the appropriate function.
856 	 */
857 	else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p)))
858 		pack.ep_path = dp;
859 #endif
860 	else {
861 #ifdef notyet
862 		printf("Cannot get path for pid %d [%s] (error %d)",
863 		    (int)p->p_pid, p->p_comm, error);
864 #endif
865 		pack.ep_path = NULL;
866 		PNBUF_PUT(dp);
867 	}
868 
869 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
870 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
871 		len - (sizeof(struct ps_strings) + szsigcode));
872 
873 #ifdef __MACHINE_STACK_GROWS_UP
874 	/*
875 	 * The copyargs call always copies into lower addresses
876 	 * first, moving towards higher addresses, starting with
877 	 * the stack pointer that we give.  When the stack grows
878 	 * down, this puts argc/argv/envp very shallow on the
879 	 * stack, right at the first user stack pointer.
880 	 * When the stack grows up, the situation is reversed.
881 	 *
882 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
883 	 * function expects to be called with a single pointer to
884 	 * a region that has a few words it can stash values into,
885 	 * followed by argc/argv/envp.  When the stack grows down,
886 	 * it's easy to decrement the stack pointer a little bit to
887 	 * allocate the space for these few words and pass the new
888 	 * stack pointer to _rtld.  When the stack grows up, however,
889 	 * a few words before argc is part of the signal trampoline, XXX
890 	 * so we have a problem.
891 	 *
892 	 * Instead of changing how _rtld works, we take the easy way
893 	 * out and steal 32 bytes before we call copyargs.
894 	 * This extra space was allowed for when 'len' was calculated.
895 	 */
896 	stack += RTLD_GAP;
897 #endif /* __MACHINE_STACK_GROWS_UP */
898 
899 	/* Now copy argc, args & environ to new stack */
900 	error = (*pack.ep_esch->es_copyargs)(l, &pack, &arginfo, &stack, argp);
901 	if (pack.ep_path) {
902 		PNBUF_PUT(pack.ep_path);
903 		pack.ep_path = NULL;
904 	}
905 	if (error) {
906 		DPRINTF(("execve: copyargs failed %d\n", error));
907 		goto exec_abort;
908 	}
909 	/* Move the stack back to original point */
910 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
911 
912 	/* fill process ps_strings info */
913 	p->p_psstr = (struct ps_strings *)
914 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
915 	    sizeof(struct ps_strings));
916 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
917 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
918 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
919 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
920 
921 	/* copy out the process's ps_strings structure */
922 	if ((error = copyout(aip, (char *)p->p_psstr,
923 	    sizeof(arginfo))) != 0) {
924 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
925 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
926 		goto exec_abort;
927 	}
928 
929 	fd_closeexec();		/* handle close on exec */
930 	execsigs(p);		/* reset catched signals */
931 
932 	l->l_ctxlink = NULL;	/* reset ucontext link */
933 
934 
935 	p->p_acflag &= ~AFORK;
936 	mutex_enter(p->p_lock);
937 	p->p_flag |= PK_EXEC;
938 	mutex_exit(p->p_lock);
939 
940 	/*
941 	 * Stop profiling.
942 	 */
943 	if ((p->p_stflag & PST_PROFIL) != 0) {
944 		mutex_spin_enter(&p->p_stmutex);
945 		stopprofclock(p);
946 		mutex_spin_exit(&p->p_stmutex);
947 	}
948 
949 	/*
950 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
951 	 * exited and exec()/exit() are the only places it will be cleared.
952 	 */
953 	if ((p->p_lflag & PL_PPWAIT) != 0) {
954 		mutex_enter(proc_lock);
955 		p->p_lflag &= ~PL_PPWAIT;
956 		cv_broadcast(&p->p_pptr->p_waitcv);
957 		mutex_exit(proc_lock);
958 	}
959 
960 	/*
961 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
962 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
963 	 * out additional references on the process for the moment.
964 	 */
965 	if ((p->p_slflag & PSL_TRACED) == 0 &&
966 
967 	    (((attr.va_mode & S_ISUID) != 0 &&
968 	      kauth_cred_geteuid(l->l_cred) != attr.va_uid) ||
969 
970 	     ((attr.va_mode & S_ISGID) != 0 &&
971 	      kauth_cred_getegid(l->l_cred) != attr.va_gid))) {
972 		/*
973 		 * Mark the process as SUGID before we do
974 		 * anything that might block.
975 		 */
976 		proc_crmod_enter();
977 		proc_crmod_leave(NULL, NULL, true);
978 
979 		/* Make sure file descriptors 0..2 are in use. */
980 		if ((error = fd_checkstd()) != 0) {
981 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
982 			goto exec_abort;
983 		}
984 
985 		/*
986 		 * Copy the credential so other references don't see our
987 		 * changes.
988 		 */
989 		l->l_cred = kauth_cred_copy(l->l_cred);
990 #ifdef KTRACE
991 		/*
992 		 * If the persistent trace flag isn't set, turn off.
993 		 */
994 		if (p->p_tracep) {
995 			mutex_enter(&ktrace_lock);
996 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
997 				ktrderef(p);
998 			mutex_exit(&ktrace_lock);
999 		}
1000 #endif
1001 		if (attr.va_mode & S_ISUID)
1002 			kauth_cred_seteuid(l->l_cred, attr.va_uid);
1003 		if (attr.va_mode & S_ISGID)
1004 			kauth_cred_setegid(l->l_cred, attr.va_gid);
1005 	} else {
1006 		if (kauth_cred_geteuid(l->l_cred) ==
1007 		    kauth_cred_getuid(l->l_cred) &&
1008 		    kauth_cred_getegid(l->l_cred) ==
1009 		    kauth_cred_getgid(l->l_cred))
1010 			p->p_flag &= ~PK_SUGID;
1011 	}
1012 
1013 	/*
1014 	 * Copy the credential so other references don't see our changes.
1015 	 * Test to see if this is necessary first, since in the common case
1016 	 * we won't need a private reference.
1017 	 */
1018 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1019 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1020 		l->l_cred = kauth_cred_copy(l->l_cred);
1021 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1022 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1023 	}
1024 
1025 	/* Update the master credentials. */
1026 	if (l->l_cred != p->p_cred) {
1027 		kauth_cred_t ocred;
1028 
1029 		kauth_cred_hold(l->l_cred);
1030 		mutex_enter(p->p_lock);
1031 		ocred = p->p_cred;
1032 		p->p_cred = l->l_cred;
1033 		mutex_exit(p->p_lock);
1034 		kauth_cred_free(ocred);
1035 	}
1036 
1037 #if defined(__HAVE_RAS)
1038 	/*
1039 	 * Remove all RASs from the address space.
1040 	 */
1041 	ras_purgeall();
1042 #endif
1043 
1044 	doexechooks(p);
1045 
1046 	/* setup new registers and do misc. setup. */
1047 	(*pack.ep_esch->es_emul->e_setregs)(l, &pack, (u_long) stack);
1048 	if (pack.ep_esch->es_setregs)
1049 		(*pack.ep_esch->es_setregs)(l, &pack, (u_long) stack);
1050 
1051 	/* map the process's signal trampoline code */
1052 	if (exec_sigcode_map(p, pack.ep_esch->es_emul)) {
1053 		DPRINTF(("execve: map sigcode failed %d\n", error));
1054 		goto exec_abort;
1055 	}
1056 
1057 	pool_put(&exec_pool, argp);
1058 
1059 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1060 
1061 	/* notify others that we exec'd */
1062 	KNOTE(&p->p_klist, NOTE_EXEC);
1063 
1064 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1065 
1066 	/* The emulation root will usually have been found when we looked
1067 	 * for the elf interpreter (or similar), if not look now. */
1068 	if (pack.ep_esch->es_emul->e_path != NULL && pack.ep_emul_root == NULL)
1069 		emul_find_root(l, &pack);
1070 
1071 	/* Any old emulation root got removed by fdcloseexec */
1072 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1073 	p->p_cwdi->cwdi_edir = pack.ep_emul_root;
1074 	rw_exit(&p->p_cwdi->cwdi_lock);
1075 	pack.ep_emul_root = NULL;
1076 	if (pack.ep_interp != NULL)
1077 		vrele(pack.ep_interp);
1078 
1079 	/*
1080 	 * Call emulation specific exec hook. This can setup per-process
1081 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1082 	 *
1083 	 * If we are executing process of different emulation than the
1084 	 * original forked process, call e_proc_exit() of the old emulation
1085 	 * first, then e_proc_exec() of new emulation. If the emulation is
1086 	 * same, the exec hook code should deallocate any old emulation
1087 	 * resources held previously by this process.
1088 	 */
1089 	if (p->p_emul && p->p_emul->e_proc_exit
1090 	    && p->p_emul != pack.ep_esch->es_emul)
1091 		(*p->p_emul->e_proc_exit)(p);
1092 
1093 	/*
1094 	 * Call exec hook. Emulation code may NOT store reference to anything
1095 	 * from &pack.
1096 	 */
1097         if (pack.ep_esch->es_emul->e_proc_exec)
1098                 (*pack.ep_esch->es_emul->e_proc_exec)(p, &pack);
1099 
1100 	/* update p_emul, the old value is no longer needed */
1101 	p->p_emul = pack.ep_esch->es_emul;
1102 
1103 	/* ...and the same for p_execsw */
1104 	p->p_execsw = pack.ep_esch;
1105 
1106 #ifdef __HAVE_SYSCALL_INTERN
1107 	(*p->p_emul->e_syscall_intern)(p);
1108 #endif
1109 	ktremul();
1110 
1111 	/* Allow new references from the debugger/procfs. */
1112 	rw_exit(&p->p_reflock);
1113 	rw_exit(&exec_lock);
1114 
1115 	mutex_enter(proc_lock);
1116 
1117 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1118 		KSI_INIT_EMPTY(&ksi);
1119 		ksi.ksi_signo = SIGTRAP;
1120 		ksi.ksi_lid = l->l_lid;
1121 		kpsignal(p, &ksi, NULL);
1122 	}
1123 
1124 	if (p->p_sflag & PS_STOPEXEC) {
1125 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1126 		p->p_pptr->p_nstopchild++;
1127 		p->p_pptr->p_waited = 0;
1128 		mutex_enter(p->p_lock);
1129 		ksiginfo_queue_init(&kq);
1130 		sigclearall(p, &contsigmask, &kq);
1131 		lwp_lock(l);
1132 		l->l_stat = LSSTOP;
1133 		p->p_stat = SSTOP;
1134 		p->p_nrlwps--;
1135 		mutex_exit(p->p_lock);
1136 		mutex_exit(proc_lock);
1137 		mi_switch(l);
1138 		ksiginfo_queue_drain(&kq);
1139 		KERNEL_LOCK(l->l_biglocks, l);
1140 	} else {
1141 		mutex_exit(proc_lock);
1142 	}
1143 
1144 	PNBUF_PUT(pathbuf);
1145 	return (EJUSTRETURN);
1146 
1147  bad:
1148 	/* free the vmspace-creation commands, and release their references */
1149 	kill_vmcmds(&pack.ep_vmcmds);
1150 	/* kill any opened file descriptor, if necessary */
1151 	if (pack.ep_flags & EXEC_HASFD) {
1152 		pack.ep_flags &= ~EXEC_HASFD;
1153 		fd_close(pack.ep_fd);
1154 	}
1155 	/* close and put the exec'd file */
1156 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
1157 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
1158 	vput(pack.ep_vp);
1159 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1160 	pool_put(&exec_pool, argp);
1161 
1162  freehdr:
1163 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1164 	if (pack.ep_emul_root != NULL)
1165 		vrele(pack.ep_emul_root);
1166 	if (pack.ep_interp != NULL)
1167 		vrele(pack.ep_interp);
1168 
1169 	rw_exit(&exec_lock);
1170 
1171  clrflg:
1172 	lwp_lock(l);
1173 	l->l_flag |= oldlwpflags;
1174 	lwp_unlock(l);
1175 	PNBUF_PUT(pathbuf);
1176 	rw_exit(&p->p_reflock);
1177 
1178 	if (modgen != module_gen && error == ENOEXEC) {
1179 		modgen = module_gen;
1180 		exec_autoload();
1181 		goto retry;
1182 	}
1183 
1184 	return error;
1185 
1186  exec_abort:
1187 	PNBUF_PUT(pathbuf);
1188 	rw_exit(&p->p_reflock);
1189 	rw_exit(&exec_lock);
1190 
1191 	/*
1192 	 * the old process doesn't exist anymore.  exit gracefully.
1193 	 * get rid of the (new) address space we have created, if any, get rid
1194 	 * of our namei data and vnode, and exit noting failure
1195 	 */
1196 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1197 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1198 	if (pack.ep_emul_arg)
1199 		free(pack.ep_emul_arg, M_TEMP);
1200 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1201 	pool_put(&exec_pool, argp);
1202 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1203 	if (pack.ep_emul_root != NULL)
1204 		vrele(pack.ep_emul_root);
1205 	if (pack.ep_interp != NULL)
1206 		vrele(pack.ep_interp);
1207 
1208 	/* Acquire the sched-state mutex (exit1() will release it). */
1209 	mutex_enter(p->p_lock);
1210 	exit1(l, W_EXITCODE(error, SIGABRT));
1211 
1212 	/* NOTREACHED */
1213 	return 0;
1214 }
1215 
1216 
1217 int
1218 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1219     char **stackp, void *argp)
1220 {
1221 	char	**cpp, *dp, *sp;
1222 	size_t	len;
1223 	void	*nullp;
1224 	long	argc, envc;
1225 	int	error;
1226 
1227 	cpp = (char **)*stackp;
1228 	nullp = NULL;
1229 	argc = arginfo->ps_nargvstr;
1230 	envc = arginfo->ps_nenvstr;
1231 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
1232 		return error;
1233 
1234 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen);
1235 	sp = argp;
1236 
1237 	/* XXX don't copy them out, remap them! */
1238 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1239 
1240 	for (; --argc >= 0; sp += len, dp += len)
1241 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1242 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1243 			return error;
1244 
1245 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1246 		return error;
1247 
1248 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1249 
1250 	for (; --envc >= 0; sp += len, dp += len)
1251 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1252 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1253 			return error;
1254 
1255 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1256 		return error;
1257 
1258 	*stackp = (char *)cpp;
1259 	return 0;
1260 }
1261 
1262 
1263 /*
1264  * Add execsw[] entries.
1265  */
1266 int
1267 exec_add(struct execsw *esp, int count)
1268 {
1269 	struct exec_entry	*it;
1270 	int			i;
1271 
1272 	if (count == 0) {
1273 		return 0;
1274 	}
1275 
1276 	/* Check for duplicates. */
1277 	rw_enter(&exec_lock, RW_WRITER);
1278 	for (i = 0; i < count; i++) {
1279 		LIST_FOREACH(it, &ex_head, ex_list) {
1280 			/* assume unique (makecmds, probe_func, emulation) */
1281 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1282 			    it->ex_sw->u.elf_probe_func ==
1283 			    esp[i].u.elf_probe_func &&
1284 			    it->ex_sw->es_emul == esp[i].es_emul) {
1285 				rw_exit(&exec_lock);
1286 				return EEXIST;
1287 			}
1288 		}
1289 	}
1290 
1291 	/* Allocate new entries. */
1292 	for (i = 0; i < count; i++) {
1293 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1294 		it->ex_sw = &esp[i];
1295 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1296 	}
1297 
1298 	/* update execsw[] */
1299 	exec_init(0);
1300 	rw_exit(&exec_lock);
1301 	return 0;
1302 }
1303 
1304 /*
1305  * Remove execsw[] entry.
1306  */
1307 int
1308 exec_remove(struct execsw *esp, int count)
1309 {
1310 	struct exec_entry	*it, *next;
1311 	int			i;
1312 	const struct proclist_desc *pd;
1313 	proc_t			*p;
1314 
1315 	if (count == 0) {
1316 		return 0;
1317 	}
1318 
1319 	/* Abort if any are busy. */
1320 	rw_enter(&exec_lock, RW_WRITER);
1321 	for (i = 0; i < count; i++) {
1322 		mutex_enter(proc_lock);
1323 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1324 			PROCLIST_FOREACH(p, pd->pd_list) {
1325 				if (p->p_execsw == &esp[i]) {
1326 					mutex_exit(proc_lock);
1327 					rw_exit(&exec_lock);
1328 					return EBUSY;
1329 				}
1330 			}
1331 		}
1332 		mutex_exit(proc_lock);
1333 	}
1334 
1335 	/* None are busy, so remove them all. */
1336 	for (i = 0; i < count; i++) {
1337 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1338 			next = LIST_NEXT(it, ex_list);
1339 			if (it->ex_sw == &esp[i]) {
1340 				LIST_REMOVE(it, ex_list);
1341 				kmem_free(it, sizeof(*it));
1342 				break;
1343 			}
1344 		}
1345 	}
1346 
1347 	/* update execsw[] */
1348 	exec_init(0);
1349 	rw_exit(&exec_lock);
1350 	return 0;
1351 }
1352 
1353 /*
1354  * Initialize exec structures. If init_boot is true, also does necessary
1355  * one-time initialization (it's called from main() that way).
1356  * Once system is multiuser, this should be called with exec_lock held,
1357  * i.e. via exec_{add|remove}().
1358  */
1359 int
1360 exec_init(int init_boot)
1361 {
1362 	const struct execsw 	**sw;
1363 	struct exec_entry	*ex;
1364 	SLIST_HEAD(,exec_entry)	first;
1365 	SLIST_HEAD(,exec_entry)	any;
1366 	SLIST_HEAD(,exec_entry)	last;
1367 	int			i, sz;
1368 
1369 	if (init_boot) {
1370 		/* do one-time initializations */
1371 		rw_init(&exec_lock);
1372 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1373 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1374 		    "execargs", &exec_palloc, IPL_NONE);
1375 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1376 	} else {
1377 		KASSERT(rw_write_held(&exec_lock));
1378 	}
1379 
1380 	/* Sort each entry onto the appropriate queue. */
1381 	SLIST_INIT(&first);
1382 	SLIST_INIT(&any);
1383 	SLIST_INIT(&last);
1384 	sz = 0;
1385 	LIST_FOREACH(ex, &ex_head, ex_list) {
1386 		switch(ex->ex_sw->es_prio) {
1387 		case EXECSW_PRIO_FIRST:
1388 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1389 			break;
1390 		case EXECSW_PRIO_ANY:
1391 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1392 			break;
1393 		case EXECSW_PRIO_LAST:
1394 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1395 			break;
1396 		default:
1397 			panic("exec_init");
1398 			break;
1399 		}
1400 		sz++;
1401 	}
1402 
1403 	/*
1404 	 * Create new execsw[].  Ensure we do not try a zero-sized
1405 	 * allocation.
1406 	 */
1407 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1408 	i = 0;
1409 	SLIST_FOREACH(ex, &first, ex_slist) {
1410 		sw[i++] = ex->ex_sw;
1411 	}
1412 	SLIST_FOREACH(ex, &any, ex_slist) {
1413 		sw[i++] = ex->ex_sw;
1414 	}
1415 	SLIST_FOREACH(ex, &last, ex_slist) {
1416 		sw[i++] = ex->ex_sw;
1417 	}
1418 
1419 	/* Replace old execsw[] and free used memory. */
1420 	if (execsw != NULL) {
1421 		kmem_free(__UNCONST(execsw),
1422 		    nexecs * sizeof(struct execsw *) + 1);
1423 	}
1424 	execsw = sw;
1425 	nexecs = sz;
1426 
1427 	/* Figure out the maximum size of an exec header. */
1428 	exec_maxhdrsz = sizeof(int);
1429 	for (i = 0; i < nexecs; i++) {
1430 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1431 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1432 	}
1433 
1434 	return 0;
1435 }
1436 
1437 static int
1438 exec_sigcode_map(struct proc *p, const struct emul *e)
1439 {
1440 	vaddr_t va;
1441 	vsize_t sz;
1442 	int error;
1443 	struct uvm_object *uobj;
1444 
1445 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1446 
1447 	if (e->e_sigobject == NULL || sz == 0) {
1448 		return 0;
1449 	}
1450 
1451 	/*
1452 	 * If we don't have a sigobject for this emulation, create one.
1453 	 *
1454 	 * sigobject is an anonymous memory object (just like SYSV shared
1455 	 * memory) that we keep a permanent reference to and that we map
1456 	 * in all processes that need this sigcode. The creation is simple,
1457 	 * we create an object, add a permanent reference to it, map it in
1458 	 * kernel space, copy out the sigcode to it and unmap it.
1459 	 * We map it with PROT_READ|PROT_EXEC into the process just
1460 	 * the way sys_mmap() would map it.
1461 	 */
1462 
1463 	uobj = *e->e_sigobject;
1464 	if (uobj == NULL) {
1465 		mutex_enter(&sigobject_lock);
1466 		if ((uobj = *e->e_sigobject) == NULL) {
1467 			uobj = uao_create(sz, 0);
1468 			(*uobj->pgops->pgo_reference)(uobj);
1469 			va = vm_map_min(kernel_map);
1470 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1471 			    uobj, 0, 0,
1472 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1473 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1474 				printf("kernel mapping failed %d\n", error);
1475 				(*uobj->pgops->pgo_detach)(uobj);
1476 				mutex_exit(&sigobject_lock);
1477 				return (error);
1478 			}
1479 			memcpy((void *)va, e->e_sigcode, sz);
1480 #ifdef PMAP_NEED_PROCWR
1481 			pmap_procwr(&proc0, va, sz);
1482 #endif
1483 			uvm_unmap(kernel_map, va, va + round_page(sz));
1484 			*e->e_sigobject = uobj;
1485 		}
1486 		mutex_exit(&sigobject_lock);
1487 	}
1488 
1489 	/* Just a hint to uvm_map where to put it. */
1490 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1491 	    round_page(sz));
1492 
1493 #ifdef __alpha__
1494 	/*
1495 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1496 	 * which causes the above calculation to put the sigcode at
1497 	 * an invalid address.  Put it just below the text instead.
1498 	 */
1499 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1500 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1501 	}
1502 #endif
1503 
1504 	(*uobj->pgops->pgo_reference)(uobj);
1505 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1506 			uobj, 0, 0,
1507 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1508 				    UVM_ADV_RANDOM, 0));
1509 	if (error) {
1510 		(*uobj->pgops->pgo_detach)(uobj);
1511 		return (error);
1512 	}
1513 	p->p_sigctx.ps_sigcode = (void *)va;
1514 	return (0);
1515 }
1516