xref: /netbsd-src/sys/kern/kern_exec.c (revision c8a35b6227034951e874c2def577388e79ede4a5)
1 /*	$NetBSD: kern_exec.c,v 1.286 2009/03/07 19:23:02 christos 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 1.286 2009/03/07 19:23:02 christos 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/syscallargs.h>
100 #if NVERIEXEC > 0
101 #include <sys/verified_exec.h>
102 #endif /* NVERIEXEC > 0 */
103 
104 #include <uvm/uvm_extern.h>
105 
106 #include <machine/reg.h>
107 
108 #include <compat/common/compat_util.h>
109 
110 static int exec_sigcode_map(struct proc *, const struct emul *);
111 
112 #ifdef DEBUG_EXEC
113 #define DPRINTF(a) uprintf a
114 #else
115 #define DPRINTF(a)
116 #endif /* DEBUG_EXEC */
117 
118 /*
119  * Exec function switch:
120  *
121  * Note that each makecmds function is responsible for loading the
122  * exec package with the necessary functions for any exec-type-specific
123  * handling.
124  *
125  * Functions for specific exec types should be defined in their own
126  * header file.
127  */
128 static const struct execsw	**execsw = NULL;
129 static int			nexecs;
130 
131 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
132 
133 /* list of dynamically loaded execsw entries */
134 static LIST_HEAD(execlist_head, exec_entry) ex_head =
135     LIST_HEAD_INITIALIZER(ex_head);
136 struct exec_entry {
137 	LIST_ENTRY(exec_entry)	ex_list;
138 	SLIST_ENTRY(exec_entry)	ex_slist;
139 	const struct execsw	*ex_sw;
140 };
141 
142 #ifdef SYSCALL_DEBUG
143 extern const char * const syscallnames[];
144 #endif
145 
146 #ifndef __HAVE_SYSCALL_INTERN
147 void	syscall(void);
148 #endif
149 
150 #ifdef KERN_SA
151 static struct sa_emul saemul_netbsd = {
152 	sizeof(ucontext_t),
153 	sizeof(struct sa_t),
154 	sizeof(struct sa_t *),
155 	NULL,
156 	NULL,
157 	cpu_upcall,
158 	(void (*)(struct lwp *, void *))getucontext_sa,
159 	sa_ucsp
160 };
161 #endif /* KERN_SA */
162 
163 /* NetBSD emul struct */
164 struct emul emul_netbsd = {
165 	"netbsd",
166 	NULL,		/* emulation path */
167 #ifndef __HAVE_MINIMAL_EMUL
168 	EMUL_HAS_SYS___syscall,
169 	NULL,
170 	SYS_syscall,
171 	SYS_NSYSENT,
172 #endif
173 	sysent,
174 #ifdef SYSCALL_DEBUG
175 	syscallnames,
176 #else
177 	NULL,
178 #endif
179 	sendsig,
180 	trapsignal,
181 	NULL,
182 	NULL,
183 	NULL,
184 	NULL,
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 #ifdef KERN_SA
202 	&saemul_netbsd,
203 #else
204 	NULL,
205 #endif
206 	sizeof(ucontext_t),
207 	startlwp,
208 };
209 
210 /*
211  * Exec lock. Used to control access to execsw[] structures.
212  * This must not be static so that netbsd32 can access it, too.
213  */
214 krwlock_t exec_lock;
215 
216 static kmutex_t sigobject_lock;
217 
218 static void *
219 exec_pool_alloc(struct pool *pp, int flags)
220 {
221 
222 	return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
223 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
224 }
225 
226 static void
227 exec_pool_free(struct pool *pp, void *addr)
228 {
229 
230 	uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
231 }
232 
233 static struct pool exec_pool;
234 
235 static struct pool_allocator exec_palloc = {
236 	.pa_alloc = exec_pool_alloc,
237 	.pa_free = exec_pool_free,
238 	.pa_pagesz = NCARGS
239 };
240 
241 /*
242  * check exec:
243  * given an "executable" described in the exec package's namei info,
244  * see what we can do with it.
245  *
246  * ON ENTRY:
247  *	exec package with appropriate namei info
248  *	lwp pointer of exec'ing lwp
249  *	NO SELF-LOCKED VNODES
250  *
251  * ON EXIT:
252  *	error:	nothing held, etc.  exec header still allocated.
253  *	ok:	filled exec package, executable's vnode (unlocked).
254  *
255  * EXEC SWITCH ENTRY:
256  * 	Locked vnode to check, exec package, proc.
257  *
258  * EXEC SWITCH EXIT:
259  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
260  *	error:	destructive:
261  *			everything deallocated execept exec header.
262  *		non-destructive:
263  *			error code, executable's vnode (unlocked),
264  *			exec header unmodified.
265  */
266 int
267 /*ARGSUSED*/
268 check_exec(struct lwp *l, struct exec_package *epp)
269 {
270 	int		error, i;
271 	struct vnode	*vp;
272 	struct nameidata *ndp;
273 	size_t		resid;
274 
275 	ndp = epp->ep_ndp;
276 	ndp->ni_cnd.cn_nameiop = LOOKUP;
277 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME | TRYEMULROOT;
278 	/* first get the vnode */
279 	if ((error = namei(ndp)) != 0)
280 		return error;
281 	epp->ep_vp = vp = ndp->ni_vp;
282 
283 	/* check access and type */
284 	if (vp->v_type != VREG) {
285 		error = EACCES;
286 		goto bad1;
287 	}
288 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
289 		goto bad1;
290 
291 	/* get attributes */
292 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
293 		goto bad1;
294 
295 	/* Check mount point */
296 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
297 		error = EACCES;
298 		goto bad1;
299 	}
300 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
301 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
302 
303 	/* try to open it */
304 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
305 		goto bad1;
306 
307 	/* unlock vp, since we need it unlocked from here on out. */
308 	VOP_UNLOCK(vp, 0);
309 
310 #if NVERIEXEC > 0
311 	error = veriexec_verify(l, vp, ndp->ni_cnd.cn_pnbuf,
312 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
313 	    NULL);
314 	if (error)
315 		goto bad2;
316 #endif /* NVERIEXEC > 0 */
317 
318 #ifdef PAX_SEGVGUARD
319 	error = pax_segvguard(l, vp, ndp->ni_cnd.cn_pnbuf, false);
320 	if (error)
321 		goto bad2;
322 #endif /* PAX_SEGVGUARD */
323 
324 	/* now we have the file, get the exec header */
325 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
326 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
327 	if (error)
328 		goto bad2;
329 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
330 
331 	/*
332 	 * Set up default address space limits.  Can be overridden
333 	 * by individual exec packages.
334 	 *
335 	 * XXX probably should be all done in the exec packages.
336 	 */
337 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
338 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
339 	/*
340 	 * set up the vmcmds for creation of the process
341 	 * address space
342 	 */
343 	error = ENOEXEC;
344 	for (i = 0; i < nexecs; i++) {
345 		int newerror;
346 
347 		epp->ep_esch = execsw[i];
348 		newerror = (*execsw[i]->es_makecmds)(l, epp);
349 
350 		if (!newerror) {
351 			/* Seems ok: check that entry point is sane */
352 			if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
353 				error = ENOEXEC;
354 				break;
355 			}
356 
357 			/* check limits */
358 			if ((epp->ep_tsize > MAXTSIZ) ||
359 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
360 						    [RLIMIT_DATA].rlim_cur)) {
361 				error = ENOMEM;
362 				break;
363 			}
364 			return 0;
365 		}
366 
367 		if (epp->ep_emul_root != NULL) {
368 			vrele(epp->ep_emul_root);
369 			epp->ep_emul_root = NULL;
370 		}
371 		if (epp->ep_interp != NULL) {
372 			vrele(epp->ep_interp);
373 			epp->ep_interp = NULL;
374 		}
375 
376 		/* make sure the first "interesting" error code is saved. */
377 		if (error == ENOEXEC)
378 			error = newerror;
379 
380 		if (epp->ep_flags & EXEC_DESTR)
381 			/* Error from "#!" code, tidied up by recursive call */
382 			return error;
383 	}
384 
385 	/* not found, error */
386 
387 	/*
388 	 * free any vmspace-creation commands,
389 	 * and release their references
390 	 */
391 	kill_vmcmds(&epp->ep_vmcmds);
392 
393 bad2:
394 	/*
395 	 * close and release the vnode, restore the old one, free the
396 	 * pathname buf, and punt.
397 	 */
398 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
399 	VOP_CLOSE(vp, FREAD, l->l_cred);
400 	vput(vp);
401 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
402 	return error;
403 
404 bad1:
405 	/*
406 	 * free the namei pathname buffer, and put the vnode
407 	 * (which we don't yet have open).
408 	 */
409 	vput(vp);				/* was still locked */
410 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
411 	return error;
412 }
413 
414 #ifdef __MACHINE_STACK_GROWS_UP
415 #define STACK_PTHREADSPACE NBPG
416 #else
417 #define STACK_PTHREADSPACE 0
418 #endif
419 
420 static int
421 execve_fetch_element(char * const *array, size_t index, char **value)
422 {
423 	return copyin(array + index, value, sizeof(*value));
424 }
425 
426 /*
427  * exec system call
428  */
429 /* ARGSUSED */
430 int
431 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
432 {
433 	/* {
434 		syscallarg(const char *)	path;
435 		syscallarg(char * const *)	argp;
436 		syscallarg(char * const *)	envp;
437 	} */
438 
439 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
440 	    SCARG(uap, envp), execve_fetch_element);
441 }
442 
443 /*
444  * Load modules to try and execute an image that we do not understand.
445  * If no execsw entries are present, we load those likely to be needed
446  * in order to run native images only.  Otherwise, we autoload all
447  * possible modules that could let us run the binary.  XXX lame
448  */
449 static void
450 exec_autoload(void)
451 {
452 #ifdef MODULAR
453 	static const char * const native[] = {
454 		"exec_elf32",
455 		"exec_elf64",
456 		"exec_script",
457 		NULL
458 	};
459 	static const char * const compat[] = {
460 		"exec_elf32",
461 		"exec_elf64",
462 		"exec_script",
463 		"exec_aout",
464 		"exec_coff",
465 		"exec_ecoff",
466 		"compat_aoutm68k",
467 		"compat_freebsd",
468 		"compat_ibcs2",
469 		"compat_irix",
470 		"compat_linux",
471 		"compat_linux32",
472 		"compat_netbsd32",
473 		"compat_sunos",
474 		"compat_sunos32",
475 		"compat_svr4",
476 		"compat_svr4_32",
477 		"compat_ultrix",
478 		NULL
479 	};
480 	char const * const *list;
481 	int i;
482 
483 	mutex_enter(&module_lock);
484 	list = (nexecs == 0 ? native : compat);
485 	for (i = 0; list[i] != NULL; i++) {
486 		if (module_autoload(list[i], MODULE_CLASS_MISC) != 0) {
487 		    	continue;
488 		}
489 		mutex_exit(&module_lock);
490 	   	yield();
491 		mutex_enter(&module_lock);
492 	}
493 	mutex_exit(&module_lock);
494 #endif
495 }
496 
497 int
498 execve1(struct lwp *l, const char *path, char * const *args,
499     char * const *envs, execve_fetch_element_t fetch_element)
500 {
501 	int			error;
502 	struct exec_package	pack;
503 	struct nameidata	nid;
504 	struct vattr		attr;
505 	struct proc		*p;
506 	char			*argp;
507 	char			*dp, *sp;
508 	long			argc, envc;
509 	size_t			i, len;
510 	char			*stack;
511 	struct ps_strings	arginfo;
512 	struct ps_strings	*aip = &arginfo;
513 	struct vmspace		*vm;
514 	struct exec_fakearg	*tmpfap;
515 	int			szsigcode;
516 	struct exec_vmcmd	*base_vcp;
517 	int			oldlwpflags;
518 	ksiginfo_t		ksi;
519 	ksiginfoq_t		kq;
520 	char			*pathbuf;
521 	size_t			pathbuflen;
522 	u_int			modgen;
523 	uid_t			uid;
524 
525 	p = l->l_proc;
526  	modgen = 0;
527 
528 	/*
529 	 * Check if we have exceeded our number of processes limit.
530 	 * This is so that we handle the case where a root daemon
531 	 * forked, ran setuid to become the desired user and is trying
532 	 * to exec. The obvious place to do the reference counting check
533 	 * is setuid(), but we don't do the reference counting check there
534 	 * like other OS's do because then all the programs that use setuid()
535 	 * must be modified to check the return code of setuid() and exit().
536 	 * It is dangerous to make setuid() fail, because it fails open and
537 	 * the program will continue to run as root. If we make it succeed
538 	 * and return an error code, again we are not enforcing the limit.
539 	 * The best place to enforce the limit is here, when the process tries
540 	 * to execute a new image, because eventually the process will need
541 	 * to call exec in order to do something useful.
542 	 */
543  retry:
544 	if ((p->p_flag & PK_SUGID) && (uid = kauth_cred_getuid(l->l_cred)) != 0
545 	    && chgproccnt(uid, 0) > p->p_rlimit[RLIMIT_NPROC].rlim_cur)
546 		return EAGAIN;
547 
548 	oldlwpflags = l->l_flag & (LW_SA | LW_SA_UPCALL);
549 	if (l->l_flag & LW_SA) {
550 		lwp_lock(l);
551 		l->l_flag &= ~(LW_SA | LW_SA_UPCALL);
552 		lwp_unlock(l);
553 	}
554 
555 	/*
556 	 * Drain existing references and forbid new ones.  The process
557 	 * should be left alone until we're done here.  This is necessary
558 	 * to avoid race conditions - e.g. in ptrace() - that might allow
559 	 * a local user to illicitly obtain elevated privileges.
560 	 */
561 	rw_enter(&p->p_reflock, RW_WRITER);
562 
563 	base_vcp = NULL;
564 	/*
565 	 * Init the namei data to point the file user's program name.
566 	 * This is done here rather than in check_exec(), so that it's
567 	 * possible to override this settings if any of makecmd/probe
568 	 * functions call check_exec() recursively - for example,
569 	 * see exec_script_makecmds().
570 	 */
571 	pathbuf = PNBUF_GET();
572 	error = copyinstr(path, pathbuf, MAXPATHLEN, &pathbuflen);
573 	if (error) {
574 		DPRINTF(("execve: copyinstr path %d", error));
575 		goto clrflg;
576 	}
577 
578 	NDINIT(&nid, LOOKUP, NOFOLLOW | TRYEMULROOT, UIO_SYSSPACE, pathbuf);
579 
580 	/*
581 	 * initialize the fields of the exec package.
582 	 */
583 	pack.ep_name = path;
584 	pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
585 	pack.ep_hdrlen = exec_maxhdrsz;
586 	pack.ep_hdrvalid = 0;
587 	pack.ep_ndp = &nid;
588 	pack.ep_emul_arg = NULL;
589 	pack.ep_vmcmds.evs_cnt = 0;
590 	pack.ep_vmcmds.evs_used = 0;
591 	pack.ep_vap = &attr;
592 	pack.ep_flags = 0;
593 	pack.ep_emul_root = NULL;
594 	pack.ep_interp = NULL;
595 	pack.ep_esch = NULL;
596 	pack.ep_pax_flags = 0;
597 
598 	rw_enter(&exec_lock, RW_READER);
599 
600 	/* see if we can run it. */
601 	if ((error = check_exec(l, &pack)) != 0) {
602 		if (error != ENOENT) {
603 			DPRINTF(("execve: check exec failed %d\n", error));
604 		}
605 		goto freehdr;
606 	}
607 
608 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
609 
610 	/* allocate an argument buffer */
611 	argp = pool_get(&exec_pool, PR_WAITOK);
612 	KASSERT(argp != NULL);
613 	dp = argp;
614 	argc = 0;
615 
616 	/* copy the fake args list, if there's one, freeing it as we go */
617 	if (pack.ep_flags & EXEC_HASARGL) {
618 		tmpfap = pack.ep_fa;
619 		while (tmpfap->fa_arg != NULL) {
620 			const char *cp;
621 
622 			cp = tmpfap->fa_arg;
623 			while (*cp)
624 				*dp++ = *cp++;
625 			*dp++ = '\0';
626 
627 			kmem_free(tmpfap->fa_arg, tmpfap->fa_len);
628 			tmpfap++; argc++;
629 		}
630 		kmem_free(pack.ep_fa, pack.ep_fa_len);
631 		pack.ep_flags &= ~EXEC_HASARGL;
632 	}
633 
634 	/* Now get argv & environment */
635 	if (args == NULL) {
636 		DPRINTF(("execve: null args\n"));
637 		error = EINVAL;
638 		goto bad;
639 	}
640 	/* 'i' will index the argp/envp element to be retrieved */
641 	i = 0;
642 	if (pack.ep_flags & EXEC_SKIPARG)
643 		i++;
644 
645 	while (1) {
646 		len = argp + ARG_MAX - dp;
647 		if ((error = (*fetch_element)(args, i, &sp)) != 0) {
648 			DPRINTF(("execve: fetch_element args %d\n", error));
649 			goto bad;
650 		}
651 		if (!sp)
652 			break;
653 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
654 			DPRINTF(("execve: copyinstr args %d\n", error));
655 			if (error == ENAMETOOLONG)
656 				error = E2BIG;
657 			goto bad;
658 		}
659 		ktrexecarg(dp, len - 1);
660 		dp += len;
661 		i++;
662 		argc++;
663 	}
664 
665 	envc = 0;
666 	/* environment need not be there */
667 	if (envs != NULL) {
668 		i = 0;
669 		while (1) {
670 			len = argp + ARG_MAX - dp;
671 			if ((error = (*fetch_element)(envs, i, &sp)) != 0) {
672 				DPRINTF(("execve: fetch_element env %d\n", error));
673 				goto bad;
674 			}
675 			if (!sp)
676 				break;
677 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
678 				DPRINTF(("execve: copyinstr env %d\n", error));
679 				if (error == ENAMETOOLONG)
680 					error = E2BIG;
681 				goto bad;
682 			}
683 			ktrexecenv(dp, len - 1);
684 			dp += len;
685 			i++;
686 			envc++;
687 		}
688 	}
689 
690 	dp = (char *) ALIGN(dp);
691 
692 	szsigcode = pack.ep_esch->es_emul->e_esigcode -
693 	    pack.ep_esch->es_emul->e_sigcode;
694 
695 #ifdef __MACHINE_STACK_GROWS_UP
696 /* See big comment lower down */
697 #define	RTLD_GAP	32
698 #else
699 #define	RTLD_GAP	0
700 #endif
701 
702 	/* Now check if args & environ fit into new stack */
703 	if (pack.ep_flags & EXEC_32)
704 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
705 		    sizeof(int) + sizeof(int) + dp + RTLD_GAP +
706 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
707 		    - argp;
708 	else
709 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
710 		    sizeof(char *) + sizeof(int) + dp + RTLD_GAP +
711 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
712 		    - argp;
713 
714 #ifdef PAX_ASLR
715 	if (pax_aslr_active(l))
716 		len += (arc4random() % PAGE_SIZE);
717 #endif /* PAX_ASLR */
718 
719 #ifdef STACKLALIGN	/* arm, etc. */
720 	len = STACKALIGN(len);	/* make the stack "safely" aligned */
721 #else
722 	len = ALIGN(len);	/* make the stack "safely" aligned */
723 #endif
724 
725 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
726 		DPRINTF(("execve: stack limit exceeded %zu\n", len));
727 		error = ENOMEM;
728 		goto bad;
729 	}
730 
731 	/* Get rid of other LWPs. */
732 	if (p->p_sa || p->p_nlwps > 1) {
733 		mutex_enter(p->p_lock);
734 		exit_lwps(l);
735 		mutex_exit(p->p_lock);
736 	}
737 	KDASSERT(p->p_nlwps == 1);
738 
739 	/* Destroy any lwpctl info. */
740 	if (p->p_lwpctl != NULL)
741 		lwp_ctl_exit();
742 
743 	/* This is now LWP 1 */
744 	l->l_lid = 1;
745 	p->p_nlwpid = 1;
746 
747 #ifdef KERN_SA
748 	/* Release any SA state. */
749 	if (p->p_sa)
750 		sa_release(p);
751 #endif /* KERN_SA */
752 
753 	/* Remove POSIX timers */
754 	timers_free(p, TIMERS_POSIX);
755 
756 	/* adjust "active stack depth" for process VSZ */
757 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
758 
759 	/*
760 	 * Do whatever is necessary to prepare the address space
761 	 * for remapping.  Note that this might replace the current
762 	 * vmspace with another!
763 	 */
764 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
765 
766 	/* record proc's vnode, for use by procfs and others */
767         if (p->p_textvp)
768                 vrele(p->p_textvp);
769 	VREF(pack.ep_vp);
770 	p->p_textvp = pack.ep_vp;
771 
772 	/* Now map address space */
773 	vm = p->p_vmspace;
774 	vm->vm_taddr = (void *)pack.ep_taddr;
775 	vm->vm_tsize = btoc(pack.ep_tsize);
776 	vm->vm_daddr = (void*)pack.ep_daddr;
777 	vm->vm_dsize = btoc(pack.ep_dsize);
778 	vm->vm_ssize = btoc(pack.ep_ssize);
779 	vm->vm_maxsaddr = (void *)pack.ep_maxsaddr;
780 	vm->vm_minsaddr = (void *)pack.ep_minsaddr;
781 
782 #ifdef PAX_ASLR
783 	pax_aslr_init(l, vm);
784 #endif /* PAX_ASLR */
785 
786 	/* create the new process's VM space by running the vmcmds */
787 #ifdef DIAGNOSTIC
788 	if (pack.ep_vmcmds.evs_used == 0)
789 		panic("execve: no vmcmds");
790 #endif
791 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
792 		struct exec_vmcmd *vcp;
793 
794 		vcp = &pack.ep_vmcmds.evs_cmds[i];
795 		if (vcp->ev_flags & VMCMD_RELATIVE) {
796 #ifdef DIAGNOSTIC
797 			if (base_vcp == NULL)
798 				panic("execve: relative vmcmd with no base");
799 			if (vcp->ev_flags & VMCMD_BASE)
800 				panic("execve: illegal base & relative vmcmd");
801 #endif
802 			vcp->ev_addr += base_vcp->ev_addr;
803 		}
804 		error = (*vcp->ev_proc)(l, vcp);
805 #ifdef DEBUG_EXEC
806 		if (error) {
807 			size_t j;
808 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
809 			for (j = 0; j <= i; j++)
810 				uprintf(
811 			"vmcmd[%zu] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
812 				    j, vp[j].ev_addr, vp[j].ev_len,
813 				    vp[j].ev_offset, vp[j].ev_prot,
814 				    vp[j].ev_flags);
815 		}
816 #endif /* DEBUG_EXEC */
817 		if (vcp->ev_flags & VMCMD_BASE)
818 			base_vcp = vcp;
819 	}
820 
821 	/* free the vmspace-creation commands, and release their references */
822 	kill_vmcmds(&pack.ep_vmcmds);
823 
824 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
825 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
826 	vput(pack.ep_vp);
827 
828 	/* if an error happened, deallocate and punt */
829 	if (error) {
830 		DPRINTF(("execve: vmcmd %zu failed: %d\n", i - 1, error));
831 		goto exec_abort;
832 	}
833 
834 	/* remember information about the process */
835 	arginfo.ps_nargvstr = argc;
836 	arginfo.ps_nenvstr = envc;
837 
838 	/* set command name & other accounting info */
839 	i = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
840 	(void)memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, i);
841 	p->p_comm[i] = '\0';
842 
843 	dp = PNBUF_GET();
844 	/*
845 	 * If the path starts with /, we don't need to do any work.
846 	 * This handles the majority of the cases.
847 	 * In the future perhaps we could canonicalize it?
848 	 */
849 	if (pathbuf[0] == '/')
850 		(void)strlcpy(pack.ep_path = dp, pathbuf, MAXPATHLEN);
851 #ifdef notyet
852 	/*
853 	 * Although this works most of the time [since the entry was just
854 	 * entered in the cache] we don't use it because it theoretically
855 	 * can fail and it is not the cleanest interface, because there
856 	 * could be races. When the namei cache is re-written, this can
857 	 * be changed to use the appropriate function.
858 	 */
859 	else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p)))
860 		pack.ep_path = dp;
861 #endif
862 	else {
863 #ifdef notyet
864 		printf("Cannot get path for pid %d [%s] (error %d)",
865 		    (int)p->p_pid, p->p_comm, error);
866 #endif
867 		pack.ep_path = NULL;
868 		PNBUF_PUT(dp);
869 	}
870 
871 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
872 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
873 		len - (sizeof(struct ps_strings) + szsigcode));
874 
875 #ifdef __MACHINE_STACK_GROWS_UP
876 	/*
877 	 * The copyargs call always copies into lower addresses
878 	 * first, moving towards higher addresses, starting with
879 	 * the stack pointer that we give.  When the stack grows
880 	 * down, this puts argc/argv/envp very shallow on the
881 	 * stack, right at the first user stack pointer.
882 	 * When the stack grows up, the situation is reversed.
883 	 *
884 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
885 	 * function expects to be called with a single pointer to
886 	 * a region that has a few words it can stash values into,
887 	 * followed by argc/argv/envp.  When the stack grows down,
888 	 * it's easy to decrement the stack pointer a little bit to
889 	 * allocate the space for these few words and pass the new
890 	 * stack pointer to _rtld.  When the stack grows up, however,
891 	 * a few words before argc is part of the signal trampoline, XXX
892 	 * so we have a problem.
893 	 *
894 	 * Instead of changing how _rtld works, we take the easy way
895 	 * out and steal 32 bytes before we call copyargs.
896 	 * This extra space was allowed for when 'len' was calculated.
897 	 */
898 	stack += RTLD_GAP;
899 #endif /* __MACHINE_STACK_GROWS_UP */
900 
901 	/* Now copy argc, args & environ to new stack */
902 	error = (*pack.ep_esch->es_copyargs)(l, &pack, &arginfo, &stack, argp);
903 	if (pack.ep_path) {
904 		PNBUF_PUT(pack.ep_path);
905 		pack.ep_path = NULL;
906 	}
907 	if (error) {
908 		DPRINTF(("execve: copyargs failed %d\n", error));
909 		goto exec_abort;
910 	}
911 	/* Move the stack back to original point */
912 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
913 
914 	/* fill process ps_strings info */
915 	p->p_psstr = (struct ps_strings *)
916 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
917 	    sizeof(struct ps_strings));
918 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
919 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
920 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
921 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
922 
923 	/* copy out the process's ps_strings structure */
924 	if ((error = copyout(aip, (char *)p->p_psstr,
925 	    sizeof(arginfo))) != 0) {
926 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
927 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
928 		goto exec_abort;
929 	}
930 
931 	fd_closeexec();		/* handle close on exec */
932 	execsigs(p);		/* reset catched signals */
933 
934 	l->l_ctxlink = NULL;	/* reset ucontext link */
935 
936 
937 	p->p_acflag &= ~AFORK;
938 	mutex_enter(p->p_lock);
939 	p->p_flag |= PK_EXEC;
940 	mutex_exit(p->p_lock);
941 
942 	/*
943 	 * Stop profiling.
944 	 */
945 	if ((p->p_stflag & PST_PROFIL) != 0) {
946 		mutex_spin_enter(&p->p_stmutex);
947 		stopprofclock(p);
948 		mutex_spin_exit(&p->p_stmutex);
949 	}
950 
951 	/*
952 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
953 	 * exited and exec()/exit() are the only places it will be cleared.
954 	 */
955 	if ((p->p_lflag & PL_PPWAIT) != 0) {
956 		mutex_enter(proc_lock);
957 		p->p_lflag &= ~PL_PPWAIT;
958 		cv_broadcast(&p->p_pptr->p_waitcv);
959 		mutex_exit(proc_lock);
960 	}
961 
962 	/*
963 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
964 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
965 	 * out additional references on the process for the moment.
966 	 */
967 	if ((p->p_slflag & PSL_TRACED) == 0 &&
968 
969 	    (((attr.va_mode & S_ISUID) != 0 &&
970 	      kauth_cred_geteuid(l->l_cred) != attr.va_uid) ||
971 
972 	     ((attr.va_mode & S_ISGID) != 0 &&
973 	      kauth_cred_getegid(l->l_cred) != attr.va_gid))) {
974 		/*
975 		 * Mark the process as SUGID before we do
976 		 * anything that might block.
977 		 */
978 		proc_crmod_enter();
979 		proc_crmod_leave(NULL, NULL, true);
980 
981 		/* Make sure file descriptors 0..2 are in use. */
982 		if ((error = fd_checkstd()) != 0) {
983 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
984 			goto exec_abort;
985 		}
986 
987 		/*
988 		 * Copy the credential so other references don't see our
989 		 * changes.
990 		 */
991 		l->l_cred = kauth_cred_copy(l->l_cred);
992 #ifdef KTRACE
993 		/*
994 		 * If the persistent trace flag isn't set, turn off.
995 		 */
996 		if (p->p_tracep) {
997 			mutex_enter(&ktrace_lock);
998 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
999 				ktrderef(p);
1000 			mutex_exit(&ktrace_lock);
1001 		}
1002 #endif
1003 		if (attr.va_mode & S_ISUID)
1004 			kauth_cred_seteuid(l->l_cred, attr.va_uid);
1005 		if (attr.va_mode & S_ISGID)
1006 			kauth_cred_setegid(l->l_cred, attr.va_gid);
1007 	} else {
1008 		if (kauth_cred_geteuid(l->l_cred) ==
1009 		    kauth_cred_getuid(l->l_cred) &&
1010 		    kauth_cred_getegid(l->l_cred) ==
1011 		    kauth_cred_getgid(l->l_cred))
1012 			p->p_flag &= ~PK_SUGID;
1013 	}
1014 
1015 	/*
1016 	 * Copy the credential so other references don't see our changes.
1017 	 * Test to see if this is necessary first, since in the common case
1018 	 * we won't need a private reference.
1019 	 */
1020 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1021 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1022 		l->l_cred = kauth_cred_copy(l->l_cred);
1023 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1024 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1025 	}
1026 
1027 	/* Update the master credentials. */
1028 	if (l->l_cred != p->p_cred) {
1029 		kauth_cred_t ocred;
1030 
1031 		kauth_cred_hold(l->l_cred);
1032 		mutex_enter(p->p_lock);
1033 		ocred = p->p_cred;
1034 		p->p_cred = l->l_cred;
1035 		mutex_exit(p->p_lock);
1036 		kauth_cred_free(ocred);
1037 	}
1038 
1039 #if defined(__HAVE_RAS)
1040 	/*
1041 	 * Remove all RASs from the address space.
1042 	 */
1043 	ras_purgeall();
1044 #endif
1045 
1046 	doexechooks(p);
1047 
1048 	/* setup new registers and do misc. setup. */
1049 	(*pack.ep_esch->es_emul->e_setregs)(l, &pack, (u_long) stack);
1050 	if (pack.ep_esch->es_setregs)
1051 		(*pack.ep_esch->es_setregs)(l, &pack, (u_long) stack);
1052 
1053 	/* map the process's signal trampoline code */
1054 	if (exec_sigcode_map(p, pack.ep_esch->es_emul)) {
1055 		DPRINTF(("execve: map sigcode failed %d\n", error));
1056 		goto exec_abort;
1057 	}
1058 
1059 	pool_put(&exec_pool, argp);
1060 
1061 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1062 
1063 	/* notify others that we exec'd */
1064 	KNOTE(&p->p_klist, NOTE_EXEC);
1065 
1066 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1067 
1068 	/* The emulation root will usually have been found when we looked
1069 	 * for the elf interpreter (or similar), if not look now. */
1070 	if (pack.ep_esch->es_emul->e_path != NULL && pack.ep_emul_root == NULL)
1071 		emul_find_root(l, &pack);
1072 
1073 	/* Any old emulation root got removed by fdcloseexec */
1074 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1075 	p->p_cwdi->cwdi_edir = pack.ep_emul_root;
1076 	rw_exit(&p->p_cwdi->cwdi_lock);
1077 	pack.ep_emul_root = NULL;
1078 	if (pack.ep_interp != NULL)
1079 		vrele(pack.ep_interp);
1080 
1081 	/*
1082 	 * Call emulation specific exec hook. This can setup per-process
1083 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1084 	 *
1085 	 * If we are executing process of different emulation than the
1086 	 * original forked process, call e_proc_exit() of the old emulation
1087 	 * first, then e_proc_exec() of new emulation. If the emulation is
1088 	 * same, the exec hook code should deallocate any old emulation
1089 	 * resources held previously by this process.
1090 	 */
1091 	if (p->p_emul && p->p_emul->e_proc_exit
1092 	    && p->p_emul != pack.ep_esch->es_emul)
1093 		(*p->p_emul->e_proc_exit)(p);
1094 
1095 	/*
1096 	 * Call exec hook. Emulation code may NOT store reference to anything
1097 	 * from &pack.
1098 	 */
1099         if (pack.ep_esch->es_emul->e_proc_exec)
1100                 (*pack.ep_esch->es_emul->e_proc_exec)(p, &pack);
1101 
1102 	/* update p_emul, the old value is no longer needed */
1103 	p->p_emul = pack.ep_esch->es_emul;
1104 
1105 	/* ...and the same for p_execsw */
1106 	p->p_execsw = pack.ep_esch;
1107 
1108 #ifdef __HAVE_SYSCALL_INTERN
1109 	(*p->p_emul->e_syscall_intern)(p);
1110 #endif
1111 	ktremul();
1112 
1113 	/* Allow new references from the debugger/procfs. */
1114 	rw_exit(&p->p_reflock);
1115 	rw_exit(&exec_lock);
1116 
1117 	mutex_enter(proc_lock);
1118 
1119 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1120 		KSI_INIT_EMPTY(&ksi);
1121 		ksi.ksi_signo = SIGTRAP;
1122 		ksi.ksi_lid = l->l_lid;
1123 		kpsignal(p, &ksi, NULL);
1124 	}
1125 
1126 	if (p->p_sflag & PS_STOPEXEC) {
1127 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1128 		p->p_pptr->p_nstopchild++;
1129 		p->p_pptr->p_waited = 0;
1130 		mutex_enter(p->p_lock);
1131 		ksiginfo_queue_init(&kq);
1132 		sigclearall(p, &contsigmask, &kq);
1133 		lwp_lock(l);
1134 		l->l_stat = LSSTOP;
1135 		p->p_stat = SSTOP;
1136 		p->p_nrlwps--;
1137 		mutex_exit(p->p_lock);
1138 		mutex_exit(proc_lock);
1139 		mi_switch(l);
1140 		ksiginfo_queue_drain(&kq);
1141 		KERNEL_LOCK(l->l_biglocks, l);
1142 	} else {
1143 		mutex_exit(proc_lock);
1144 	}
1145 
1146 	PNBUF_PUT(pathbuf);
1147 	return (EJUSTRETURN);
1148 
1149  bad:
1150 	/* free the vmspace-creation commands, and release their references */
1151 	kill_vmcmds(&pack.ep_vmcmds);
1152 	/* kill any opened file descriptor, if necessary */
1153 	if (pack.ep_flags & EXEC_HASFD) {
1154 		pack.ep_flags &= ~EXEC_HASFD;
1155 		fd_close(pack.ep_fd);
1156 	}
1157 	/* close and put the exec'd file */
1158 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
1159 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
1160 	vput(pack.ep_vp);
1161 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1162 	pool_put(&exec_pool, argp);
1163 
1164  freehdr:
1165 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1166 	if (pack.ep_emul_root != NULL)
1167 		vrele(pack.ep_emul_root);
1168 	if (pack.ep_interp != NULL)
1169 		vrele(pack.ep_interp);
1170 
1171 	rw_exit(&exec_lock);
1172 
1173  clrflg:
1174 	lwp_lock(l);
1175 	l->l_flag |= oldlwpflags;
1176 	lwp_unlock(l);
1177 	PNBUF_PUT(pathbuf);
1178 	rw_exit(&p->p_reflock);
1179 
1180 	if (modgen != module_gen && error == ENOEXEC) {
1181 		modgen = module_gen;
1182 		exec_autoload();
1183 		goto retry;
1184 	}
1185 
1186 	return error;
1187 
1188  exec_abort:
1189 	PNBUF_PUT(pathbuf);
1190 	rw_exit(&p->p_reflock);
1191 	rw_exit(&exec_lock);
1192 
1193 	/*
1194 	 * the old process doesn't exist anymore.  exit gracefully.
1195 	 * get rid of the (new) address space we have created, if any, get rid
1196 	 * of our namei data and vnode, and exit noting failure
1197 	 */
1198 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1199 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1200 	if (pack.ep_emul_arg)
1201 		free(pack.ep_emul_arg, M_TEMP);
1202 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1203 	pool_put(&exec_pool, argp);
1204 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1205 	if (pack.ep_emul_root != NULL)
1206 		vrele(pack.ep_emul_root);
1207 	if (pack.ep_interp != NULL)
1208 		vrele(pack.ep_interp);
1209 
1210 	/* Acquire the sched-state mutex (exit1() will release it). */
1211 	mutex_enter(p->p_lock);
1212 	exit1(l, W_EXITCODE(error, SIGABRT));
1213 
1214 	/* NOTREACHED */
1215 	return 0;
1216 }
1217 
1218 
1219 int
1220 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1221     char **stackp, void *argp)
1222 {
1223 	char	**cpp, *dp, *sp;
1224 	size_t	len;
1225 	void	*nullp;
1226 	long	argc, envc;
1227 	int	error;
1228 
1229 	cpp = (char **)*stackp;
1230 	nullp = NULL;
1231 	argc = arginfo->ps_nargvstr;
1232 	envc = arginfo->ps_nenvstr;
1233 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
1234 		return error;
1235 
1236 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen);
1237 	sp = argp;
1238 
1239 	/* XXX don't copy them out, remap them! */
1240 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1241 
1242 	for (; --argc >= 0; sp += len, dp += len)
1243 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1244 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1245 			return error;
1246 
1247 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1248 		return error;
1249 
1250 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1251 
1252 	for (; --envc >= 0; sp += len, dp += len)
1253 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1254 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1255 			return error;
1256 
1257 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1258 		return error;
1259 
1260 	*stackp = (char *)cpp;
1261 	return 0;
1262 }
1263 
1264 
1265 /*
1266  * Add execsw[] entries.
1267  */
1268 int
1269 exec_add(struct execsw *esp, int count)
1270 {
1271 	struct exec_entry	*it;
1272 	int			i;
1273 
1274 	if (count == 0) {
1275 		return 0;
1276 	}
1277 
1278 	/* Check for duplicates. */
1279 	rw_enter(&exec_lock, RW_WRITER);
1280 	for (i = 0; i < count; i++) {
1281 		LIST_FOREACH(it, &ex_head, ex_list) {
1282 			/* assume unique (makecmds, probe_func, emulation) */
1283 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1284 			    it->ex_sw->u.elf_probe_func ==
1285 			    esp[i].u.elf_probe_func &&
1286 			    it->ex_sw->es_emul == esp[i].es_emul) {
1287 				rw_exit(&exec_lock);
1288 				return EEXIST;
1289 			}
1290 		}
1291 	}
1292 
1293 	/* Allocate new entries. */
1294 	for (i = 0; i < count; i++) {
1295 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1296 		it->ex_sw = &esp[i];
1297 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1298 	}
1299 
1300 	/* update execsw[] */
1301 	exec_init(0);
1302 	rw_exit(&exec_lock);
1303 	return 0;
1304 }
1305 
1306 /*
1307  * Remove execsw[] entry.
1308  */
1309 int
1310 exec_remove(struct execsw *esp, int count)
1311 {
1312 	struct exec_entry	*it, *next;
1313 	int			i;
1314 	const struct proclist_desc *pd;
1315 	proc_t			*p;
1316 
1317 	if (count == 0) {
1318 		return 0;
1319 	}
1320 
1321 	/* Abort if any are busy. */
1322 	rw_enter(&exec_lock, RW_WRITER);
1323 	for (i = 0; i < count; i++) {
1324 		mutex_enter(proc_lock);
1325 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1326 			PROCLIST_FOREACH(p, pd->pd_list) {
1327 				if (p->p_execsw == &esp[i]) {
1328 					mutex_exit(proc_lock);
1329 					rw_exit(&exec_lock);
1330 					return EBUSY;
1331 				}
1332 			}
1333 		}
1334 		mutex_exit(proc_lock);
1335 	}
1336 
1337 	/* None are busy, so remove them all. */
1338 	for (i = 0; i < count; i++) {
1339 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1340 			next = LIST_NEXT(it, ex_list);
1341 			if (it->ex_sw == &esp[i]) {
1342 				LIST_REMOVE(it, ex_list);
1343 				kmem_free(it, sizeof(*it));
1344 				break;
1345 			}
1346 		}
1347 	}
1348 
1349 	/* update execsw[] */
1350 	exec_init(0);
1351 	rw_exit(&exec_lock);
1352 	return 0;
1353 }
1354 
1355 /*
1356  * Initialize exec structures. If init_boot is true, also does necessary
1357  * one-time initialization (it's called from main() that way).
1358  * Once system is multiuser, this should be called with exec_lock held,
1359  * i.e. via exec_{add|remove}().
1360  */
1361 int
1362 exec_init(int init_boot)
1363 {
1364 	const struct execsw 	**sw;
1365 	struct exec_entry	*ex;
1366 	SLIST_HEAD(,exec_entry)	first;
1367 	SLIST_HEAD(,exec_entry)	any;
1368 	SLIST_HEAD(,exec_entry)	last;
1369 	int			i, sz;
1370 
1371 	if (init_boot) {
1372 		/* do one-time initializations */
1373 		rw_init(&exec_lock);
1374 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1375 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1376 		    "execargs", &exec_palloc, IPL_NONE);
1377 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1378 	} else {
1379 		KASSERT(rw_write_held(&exec_lock));
1380 	}
1381 
1382 	/* Sort each entry onto the appropriate queue. */
1383 	SLIST_INIT(&first);
1384 	SLIST_INIT(&any);
1385 	SLIST_INIT(&last);
1386 	sz = 0;
1387 	LIST_FOREACH(ex, &ex_head, ex_list) {
1388 		switch(ex->ex_sw->es_prio) {
1389 		case EXECSW_PRIO_FIRST:
1390 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1391 			break;
1392 		case EXECSW_PRIO_ANY:
1393 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1394 			break;
1395 		case EXECSW_PRIO_LAST:
1396 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1397 			break;
1398 		default:
1399 			panic("exec_init");
1400 			break;
1401 		}
1402 		sz++;
1403 	}
1404 
1405 	/*
1406 	 * Create new execsw[].  Ensure we do not try a zero-sized
1407 	 * allocation.
1408 	 */
1409 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1410 	i = 0;
1411 	SLIST_FOREACH(ex, &first, ex_slist) {
1412 		sw[i++] = ex->ex_sw;
1413 	}
1414 	SLIST_FOREACH(ex, &any, ex_slist) {
1415 		sw[i++] = ex->ex_sw;
1416 	}
1417 	SLIST_FOREACH(ex, &last, ex_slist) {
1418 		sw[i++] = ex->ex_sw;
1419 	}
1420 
1421 	/* Replace old execsw[] and free used memory. */
1422 	if (execsw != NULL) {
1423 		kmem_free(__UNCONST(execsw),
1424 		    nexecs * sizeof(struct execsw *) + 1);
1425 	}
1426 	execsw = sw;
1427 	nexecs = sz;
1428 
1429 	/* Figure out the maximum size of an exec header. */
1430 	exec_maxhdrsz = sizeof(int);
1431 	for (i = 0; i < nexecs; i++) {
1432 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1433 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1434 	}
1435 
1436 	return 0;
1437 }
1438 
1439 static int
1440 exec_sigcode_map(struct proc *p, const struct emul *e)
1441 {
1442 	vaddr_t va;
1443 	vsize_t sz;
1444 	int error;
1445 	struct uvm_object *uobj;
1446 
1447 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1448 
1449 	if (e->e_sigobject == NULL || sz == 0) {
1450 		return 0;
1451 	}
1452 
1453 	/*
1454 	 * If we don't have a sigobject for this emulation, create one.
1455 	 *
1456 	 * sigobject is an anonymous memory object (just like SYSV shared
1457 	 * memory) that we keep a permanent reference to and that we map
1458 	 * in all processes that need this sigcode. The creation is simple,
1459 	 * we create an object, add a permanent reference to it, map it in
1460 	 * kernel space, copy out the sigcode to it and unmap it.
1461 	 * We map it with PROT_READ|PROT_EXEC into the process just
1462 	 * the way sys_mmap() would map it.
1463 	 */
1464 
1465 	uobj = *e->e_sigobject;
1466 	if (uobj == NULL) {
1467 		mutex_enter(&sigobject_lock);
1468 		if ((uobj = *e->e_sigobject) == NULL) {
1469 			uobj = uao_create(sz, 0);
1470 			(*uobj->pgops->pgo_reference)(uobj);
1471 			va = vm_map_min(kernel_map);
1472 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1473 			    uobj, 0, 0,
1474 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1475 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1476 				printf("kernel mapping failed %d\n", error);
1477 				(*uobj->pgops->pgo_detach)(uobj);
1478 				mutex_exit(&sigobject_lock);
1479 				return (error);
1480 			}
1481 			memcpy((void *)va, e->e_sigcode, sz);
1482 #ifdef PMAP_NEED_PROCWR
1483 			pmap_procwr(&proc0, va, sz);
1484 #endif
1485 			uvm_unmap(kernel_map, va, va + round_page(sz));
1486 			*e->e_sigobject = uobj;
1487 		}
1488 		mutex_exit(&sigobject_lock);
1489 	}
1490 
1491 	/* Just a hint to uvm_map where to put it. */
1492 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1493 	    round_page(sz));
1494 
1495 #ifdef __alpha__
1496 	/*
1497 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1498 	 * which causes the above calculation to put the sigcode at
1499 	 * an invalid address.  Put it just below the text instead.
1500 	 */
1501 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1502 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1503 	}
1504 #endif
1505 
1506 	(*uobj->pgops->pgo_reference)(uobj);
1507 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1508 			uobj, 0, 0,
1509 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1510 				    UVM_ADV_RANDOM, 0));
1511 	if (error) {
1512 		(*uobj->pgops->pgo_detach)(uobj);
1513 		return (error);
1514 	}
1515 	p->p_sigctx.ps_sigcode = (void *)va;
1516 	return (0);
1517 }
1518