xref: /netbsd-src/sys/kern/kern_exec.c (revision 96230fab84e26a6435963032070e916a951a8b2e)
1 /*	$NetBSD: kern_exec.c,v 1.279 2008/10/15 06:51:20 wrstuden 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 1.279 2008/10/15 06:51:20 wrstuden Exp $");
63 
64 #include "opt_ktrace.h"
65 #include "opt_syscall_debug.h"
66 #include "opt_compat_netbsd.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 
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 extern const struct execsw	execsw_builtin[];
129 extern int			nexecs_builtin;
130 static const struct execsw	**execsw = NULL;
131 static int			nexecs;
132 
133 u_int	exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
134 
135 #ifdef LKM
136 /* list of supported emulations */
137 static
138 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
139 struct emul_entry {
140 	LIST_ENTRY(emul_entry)	el_list;
141 	const struct emul	*el_emul;
142 	int			ro_entry;
143 };
144 
145 /* list of dynamically loaded execsw entries */
146 static
147 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
148 struct exec_entry {
149 	LIST_ENTRY(exec_entry)	ex_list;
150 	const struct execsw	*es;
151 };
152 
153 /* structure used for building execw[] */
154 struct execsw_entry {
155 	struct execsw_entry	*next;
156 	const struct execsw	*es;
157 };
158 #endif /* LKM */
159 
160 #ifdef SYSCALL_DEBUG
161 extern const char * const syscallnames[];
162 #endif
163 
164 #ifdef COMPAT_16
165 extern char	sigcode[], esigcode[];
166 struct uvm_object *emul_netbsd_object;
167 #endif
168 
169 #ifndef __HAVE_SYSCALL_INTERN
170 void	syscall(void);
171 #endif
172 
173 static const struct sa_emul saemul_netbsd = {
174 	sizeof(ucontext_t),
175 	sizeof(struct sa_t),
176 	sizeof(struct sa_t *),
177 	NULL,
178 	NULL,
179 	cpu_upcall,
180 	(void (*)(struct lwp *, void *))getucontext_sa,
181 #ifdef KERN_SA
182 	sa_ucsp
183 #else
184 	NULL
185 #endif
186 };
187 
188 /* NetBSD emul struct */
189 const struct emul emul_netbsd = {
190 	"netbsd",
191 	NULL,		/* emulation path */
192 #ifndef __HAVE_MINIMAL_EMUL
193 	EMUL_HAS_SYS___syscall,
194 	NULL,
195 	SYS_syscall,
196 	SYS_NSYSENT,
197 #endif
198 	sysent,
199 #ifdef SYSCALL_DEBUG
200 	syscallnames,
201 #else
202 	NULL,
203 #endif
204 	sendsig,
205 	trapsignal,
206 	NULL,
207 #ifdef COMPAT_16
208 	sigcode,
209 	esigcode,
210 	&emul_netbsd_object,
211 #else
212 	NULL,
213 	NULL,
214 	NULL,
215 #endif
216 	setregs,
217 	NULL,
218 	NULL,
219 	NULL,
220 	NULL,
221 	NULL,
222 #ifdef __HAVE_SYSCALL_INTERN
223 	syscall_intern,
224 #else
225 	syscall,
226 #endif
227 	NULL,
228 	NULL,
229 
230 	uvm_default_mapaddr,
231 	NULL,
232 	&saemul_netbsd,
233 	sizeof(ucontext_t),
234 	startlwp,
235 };
236 
237 #ifdef LKM
238 /*
239  * Exec lock. Used to control access to execsw[] structures.
240  * This must not be static so that netbsd32 can access it, too.
241  */
242 krwlock_t exec_lock;
243 
244 static void link_es(struct execsw_entry **, const struct execsw *);
245 #endif /* LKM */
246 
247 static kmutex_t sigobject_lock;
248 
249 static void *
250 exec_pool_alloc(struct pool *pp, int flags)
251 {
252 
253 	return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
254 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
255 }
256 
257 static void
258 exec_pool_free(struct pool *pp, void *addr)
259 {
260 
261 	uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
262 }
263 
264 static struct pool exec_pool;
265 
266 static struct pool_allocator exec_palloc = {
267 	.pa_alloc = exec_pool_alloc,
268 	.pa_free = exec_pool_free,
269 	.pa_pagesz = NCARGS
270 };
271 
272 /*
273  * check exec:
274  * given an "executable" described in the exec package's namei info,
275  * see what we can do with it.
276  *
277  * ON ENTRY:
278  *	exec package with appropriate namei info
279  *	lwp pointer of exec'ing lwp
280  *	NO SELF-LOCKED VNODES
281  *
282  * ON EXIT:
283  *	error:	nothing held, etc.  exec header still allocated.
284  *	ok:	filled exec package, executable's vnode (unlocked).
285  *
286  * EXEC SWITCH ENTRY:
287  * 	Locked vnode to check, exec package, proc.
288  *
289  * EXEC SWITCH EXIT:
290  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
291  *	error:	destructive:
292  *			everything deallocated execept exec header.
293  *		non-destructive:
294  *			error code, executable's vnode (unlocked),
295  *			exec header unmodified.
296  */
297 int
298 /*ARGSUSED*/
299 check_exec(struct lwp *l, struct exec_package *epp)
300 {
301 	int		error, i;
302 	struct vnode	*vp;
303 	struct nameidata *ndp;
304 	size_t		resid;
305 
306 	ndp = epp->ep_ndp;
307 	ndp->ni_cnd.cn_nameiop = LOOKUP;
308 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME | TRYEMULROOT;
309 	/* first get the vnode */
310 	if ((error = namei(ndp)) != 0)
311 		return error;
312 	epp->ep_vp = vp = ndp->ni_vp;
313 
314 	/* check access and type */
315 	if (vp->v_type != VREG) {
316 		error = EACCES;
317 		goto bad1;
318 	}
319 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
320 		goto bad1;
321 
322 	/* get attributes */
323 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
324 		goto bad1;
325 
326 	/* Check mount point */
327 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
328 		error = EACCES;
329 		goto bad1;
330 	}
331 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
332 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
333 
334 	/* try to open it */
335 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
336 		goto bad1;
337 
338 	/* unlock vp, since we need it unlocked from here on out. */
339 	VOP_UNLOCK(vp, 0);
340 
341 #if NVERIEXEC > 0
342 	error = veriexec_verify(l, vp, ndp->ni_cnd.cn_pnbuf,
343 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
344 	    NULL);
345 	if (error)
346 		goto bad2;
347 #endif /* NVERIEXEC > 0 */
348 
349 #ifdef PAX_SEGVGUARD
350 	error = pax_segvguard(l, vp, ndp->ni_cnd.cn_pnbuf, false);
351 	if (error)
352 		goto bad2;
353 #endif /* PAX_SEGVGUARD */
354 
355 	/* now we have the file, get the exec header */
356 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
357 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
358 	if (error)
359 		goto bad2;
360 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
361 
362 	/*
363 	 * Set up default address space limits.  Can be overridden
364 	 * by individual exec packages.
365 	 *
366 	 * XXX probably should be all done in the exec packages.
367 	 */
368 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
369 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
370 	/*
371 	 * set up the vmcmds for creation of the process
372 	 * address space
373 	 */
374 	error = ENOEXEC;
375 	for (i = 0; i < nexecs; i++) {
376 		int newerror;
377 
378 		epp->ep_esch = execsw[i];
379 		newerror = (*execsw[i]->es_makecmds)(l, epp);
380 
381 		if (!newerror) {
382 			/* Seems ok: check that entry point is sane */
383 			if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
384 				error = ENOEXEC;
385 				break;
386 			}
387 
388 			/* check limits */
389 			if ((epp->ep_tsize > MAXTSIZ) ||
390 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
391 						    [RLIMIT_DATA].rlim_cur)) {
392 				error = ENOMEM;
393 				break;
394 			}
395 			return 0;
396 		}
397 
398 		if (epp->ep_emul_root != NULL) {
399 			vrele(epp->ep_emul_root);
400 			epp->ep_emul_root = NULL;
401 		}
402 		if (epp->ep_interp != NULL) {
403 			vrele(epp->ep_interp);
404 			epp->ep_interp = NULL;
405 		}
406 
407 		/* make sure the first "interesting" error code is saved. */
408 		if (error == ENOEXEC)
409 			error = newerror;
410 
411 		if (epp->ep_flags & EXEC_DESTR)
412 			/* Error from "#!" code, tidied up by recursive call */
413 			return error;
414 	}
415 
416 	/* not found, error */
417 
418 	/*
419 	 * free any vmspace-creation commands,
420 	 * and release their references
421 	 */
422 	kill_vmcmds(&epp->ep_vmcmds);
423 
424 bad2:
425 	/*
426 	 * close and release the vnode, restore the old one, free the
427 	 * pathname buf, and punt.
428 	 */
429 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
430 	VOP_CLOSE(vp, FREAD, l->l_cred);
431 	vput(vp);
432 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
433 	return error;
434 
435 bad1:
436 	/*
437 	 * free the namei pathname buffer, and put the vnode
438 	 * (which we don't yet have open).
439 	 */
440 	vput(vp);				/* was still locked */
441 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
442 	return error;
443 }
444 
445 #ifdef __MACHINE_STACK_GROWS_UP
446 #define STACK_PTHREADSPACE NBPG
447 #else
448 #define STACK_PTHREADSPACE 0
449 #endif
450 
451 static int
452 execve_fetch_element(char * const *array, size_t index, char **value)
453 {
454 	return copyin(array + index, value, sizeof(*value));
455 }
456 
457 /*
458  * exec system call
459  */
460 /* ARGSUSED */
461 int
462 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
463 {
464 	/* {
465 		syscallarg(const char *)	path;
466 		syscallarg(char * const *)	argp;
467 		syscallarg(char * const *)	envp;
468 	} */
469 
470 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
471 	    SCARG(uap, envp), execve_fetch_element);
472 }
473 
474 int
475 execve1(struct lwp *l, const char *path, char * const *args,
476     char * const *envs, execve_fetch_element_t fetch_element)
477 {
478 	int			error;
479 	struct exec_package	pack;
480 	struct nameidata	nid;
481 	struct vattr		attr;
482 	struct proc		*p;
483 	char			*argp;
484 	char			*dp, *sp;
485 	long			argc, envc;
486 	size_t			i, len;
487 	char			*stack;
488 	struct ps_strings	arginfo;
489 	struct ps_strings	*aip = &arginfo;
490 	struct vmspace		*vm;
491 	struct exec_fakearg	*tmpfap;
492 	int			szsigcode;
493 	struct exec_vmcmd	*base_vcp;
494 	int			oldlwpflags;
495 	ksiginfo_t		ksi;
496 	ksiginfoq_t		kq;
497 	char			*pathbuf;
498 	size_t			pathbuflen;
499 
500 	p = l->l_proc;
501 
502 	/*
503 	 * Check if we have exceeded our number of processes limit.
504 	 * This is so that we handle the case where a root daemon
505 	 * forked, ran setuid to become the desired user and is trying
506 	 * to exec. The obvious place to do the reference counting check
507 	 * is setuid(), but we don't do the reference counting check there
508 	 * like other OS's do because then all the programs that use setuid()
509 	 * must be modified to check the return code of setuid() and exit().
510 	 * It is dangerous to make setuid() fail, because it fails open and
511 	 * the program will continue to run as root. If we make it succeed
512 	 * and return an error code, again we are not enforcing the limit.
513 	 * The best place to enforce the limit is here, when the process tries
514 	 * to execute a new image, because eventually the process will need
515 	 * to call exec in order to do something useful.
516 	 */
517 
518 	if ((p->p_flag & PK_SUGID) &&
519 	    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
520 	    p->p_rlimit[RLIMIT_NPROC].rlim_cur)
521 		return EAGAIN;
522 
523 	oldlwpflags = l->l_flag & (LW_SA | LW_SA_UPCALL);
524 	if (l->l_flag & LW_SA) {
525 		lwp_lock(l);
526 		l->l_flag &= ~(LW_SA | LW_SA_UPCALL);
527 		lwp_unlock(l);
528 	}
529 
530 	/*
531 	 * Drain existing references and forbid new ones.  The process
532 	 * should be left alone until we're done here.  This is necessary
533 	 * to avoid race conditions - e.g. in ptrace() - that might allow
534 	 * a local user to illicitly obtain elevated privileges.
535 	 */
536 	rw_enter(&p->p_reflock, RW_WRITER);
537 
538 	base_vcp = NULL;
539 	/*
540 	 * Init the namei data to point the file user's program name.
541 	 * This is done here rather than in check_exec(), so that it's
542 	 * possible to override this settings if any of makecmd/probe
543 	 * functions call check_exec() recursively - for example,
544 	 * see exec_script_makecmds().
545 	 */
546 	pathbuf = PNBUF_GET();
547 	error = copyinstr(path, pathbuf, MAXPATHLEN, &pathbuflen);
548 	if (error) {
549 		DPRINTF(("execve: copyinstr path %d", error));
550 		goto clrflg;
551 	}
552 
553 	NDINIT(&nid, LOOKUP, NOFOLLOW | TRYEMULROOT, UIO_SYSSPACE, pathbuf);
554 
555 	/*
556 	 * initialize the fields of the exec package.
557 	 */
558 	pack.ep_name = path;
559 	pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
560 	pack.ep_hdrlen = exec_maxhdrsz;
561 	pack.ep_hdrvalid = 0;
562 	pack.ep_ndp = &nid;
563 	pack.ep_emul_arg = NULL;
564 	pack.ep_vmcmds.evs_cnt = 0;
565 	pack.ep_vmcmds.evs_used = 0;
566 	pack.ep_vap = &attr;
567 	pack.ep_flags = 0;
568 	pack.ep_emul_root = NULL;
569 	pack.ep_interp = NULL;
570 	pack.ep_esch = NULL;
571 	pack.ep_pax_flags = 0;
572 
573 #ifdef LKM
574 	rw_enter(&exec_lock, RW_READER);
575 #endif
576 
577 	/* see if we can run it. */
578 	if ((error = check_exec(l, &pack)) != 0) {
579 		if (error != ENOENT) {
580 			DPRINTF(("execve: check exec failed %d\n", error));
581 		}
582 		goto freehdr;
583 	}
584 
585 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
586 
587 	/* allocate an argument buffer */
588 	argp = pool_get(&exec_pool, PR_WAITOK);
589 	KASSERT(argp != NULL);
590 	dp = argp;
591 	argc = 0;
592 
593 	/* copy the fake args list, if there's one, freeing it as we go */
594 	if (pack.ep_flags & EXEC_HASARGL) {
595 		tmpfap = pack.ep_fa;
596 		while (tmpfap->fa_arg != NULL) {
597 			const char *cp;
598 
599 			cp = tmpfap->fa_arg;
600 			while (*cp)
601 				*dp++ = *cp++;
602 			*dp++ = '\0';
603 
604 			kmem_free(tmpfap->fa_arg, tmpfap->fa_len);
605 			tmpfap++; argc++;
606 		}
607 		kmem_free(pack.ep_fa, pack.ep_fa_len);
608 		pack.ep_flags &= ~EXEC_HASARGL;
609 	}
610 
611 	/* Now get argv & environment */
612 	if (args == NULL) {
613 		DPRINTF(("execve: null args\n"));
614 		error = EINVAL;
615 		goto bad;
616 	}
617 	/* 'i' will index the argp/envp element to be retrieved */
618 	i = 0;
619 	if (pack.ep_flags & EXEC_SKIPARG)
620 		i++;
621 
622 	while (1) {
623 		len = argp + ARG_MAX - dp;
624 		if ((error = (*fetch_element)(args, i, &sp)) != 0) {
625 			DPRINTF(("execve: fetch_element args %d\n", error));
626 			goto bad;
627 		}
628 		if (!sp)
629 			break;
630 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
631 			DPRINTF(("execve: copyinstr args %d\n", error));
632 			if (error == ENAMETOOLONG)
633 				error = E2BIG;
634 			goto bad;
635 		}
636 		ktrexecarg(dp, len - 1);
637 		dp += len;
638 		i++;
639 		argc++;
640 	}
641 
642 	envc = 0;
643 	/* environment need not be there */
644 	if (envs != NULL) {
645 		i = 0;
646 		while (1) {
647 			len = argp + ARG_MAX - dp;
648 			if ((error = (*fetch_element)(envs, i, &sp)) != 0) {
649 				DPRINTF(("execve: fetch_element env %d\n", error));
650 				goto bad;
651 			}
652 			if (!sp)
653 				break;
654 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
655 				DPRINTF(("execve: copyinstr env %d\n", error));
656 				if (error == ENAMETOOLONG)
657 					error = E2BIG;
658 				goto bad;
659 			}
660 			ktrexecenv(dp, len - 1);
661 			dp += len;
662 			i++;
663 			envc++;
664 		}
665 	}
666 
667 	dp = (char *) ALIGN(dp);
668 
669 	szsigcode = pack.ep_esch->es_emul->e_esigcode -
670 	    pack.ep_esch->es_emul->e_sigcode;
671 
672 #ifdef __MACHINE_STACK_GROWS_UP
673 /* See big comment lower down */
674 #define	RTLD_GAP	32
675 #else
676 #define	RTLD_GAP	0
677 #endif
678 
679 	/* Now check if args & environ fit into new stack */
680 	if (pack.ep_flags & EXEC_32)
681 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
682 		    sizeof(int) + sizeof(int) + dp + RTLD_GAP +
683 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
684 		    - argp;
685 	else
686 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
687 		    sizeof(char *) + sizeof(int) + dp + RTLD_GAP +
688 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
689 		    - argp;
690 
691 #ifdef PAX_ASLR
692 	if (pax_aslr_active(l))
693 		len += (arc4random() % PAGE_SIZE);
694 #endif /* PAX_ASLR */
695 
696 #ifdef STACKLALIGN	/* arm, etc. */
697 	len = STACKALIGN(len);	/* make the stack "safely" aligned */
698 #else
699 	len = ALIGN(len);	/* make the stack "safely" aligned */
700 #endif
701 
702 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
703 		DPRINTF(("execve: stack limit exceeded %zu\n", len));
704 		error = ENOMEM;
705 		goto bad;
706 	}
707 
708 	/* Get rid of other LWPs. */
709 	if (p->p_sa || p->p_nlwps > 1) {
710 		mutex_enter(p->p_lock);
711 		exit_lwps(l);
712 		mutex_exit(p->p_lock);
713 	}
714 	KDASSERT(p->p_nlwps == 1);
715 
716 	/* Destroy any lwpctl info. */
717 	if (p->p_lwpctl != NULL)
718 		lwp_ctl_exit();
719 
720 	/* This is now LWP 1 */
721 	l->l_lid = 1;
722 	p->p_nlwpid = 1;
723 
724 #ifdef KERN_SA
725 	/* Release any SA state. */
726 	if (p->p_sa)
727 		sa_release(p);
728 #endif /* KERN_SA */
729 
730 	/* Remove POSIX timers */
731 	timers_free(p, TIMERS_POSIX);
732 
733 	/* adjust "active stack depth" for process VSZ */
734 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
735 
736 	/*
737 	 * Do whatever is necessary to prepare the address space
738 	 * for remapping.  Note that this might replace the current
739 	 * vmspace with another!
740 	 */
741 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
742 
743 	/* record proc's vnode, for use by procfs and others */
744         if (p->p_textvp)
745                 vrele(p->p_textvp);
746 	VREF(pack.ep_vp);
747 	p->p_textvp = pack.ep_vp;
748 
749 	/* Now map address space */
750 	vm = p->p_vmspace;
751 	vm->vm_taddr = (void *)pack.ep_taddr;
752 	vm->vm_tsize = btoc(pack.ep_tsize);
753 	vm->vm_daddr = (void*)pack.ep_daddr;
754 	vm->vm_dsize = btoc(pack.ep_dsize);
755 	vm->vm_ssize = btoc(pack.ep_ssize);
756 	vm->vm_maxsaddr = (void *)pack.ep_maxsaddr;
757 	vm->vm_minsaddr = (void *)pack.ep_minsaddr;
758 
759 #ifdef PAX_ASLR
760 	pax_aslr_init(l, vm);
761 #endif /* PAX_ASLR */
762 
763 	/* create the new process's VM space by running the vmcmds */
764 #ifdef DIAGNOSTIC
765 	if (pack.ep_vmcmds.evs_used == 0)
766 		panic("execve: no vmcmds");
767 #endif
768 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
769 		struct exec_vmcmd *vcp;
770 
771 		vcp = &pack.ep_vmcmds.evs_cmds[i];
772 		if (vcp->ev_flags & VMCMD_RELATIVE) {
773 #ifdef DIAGNOSTIC
774 			if (base_vcp == NULL)
775 				panic("execve: relative vmcmd with no base");
776 			if (vcp->ev_flags & VMCMD_BASE)
777 				panic("execve: illegal base & relative vmcmd");
778 #endif
779 			vcp->ev_addr += base_vcp->ev_addr;
780 		}
781 		error = (*vcp->ev_proc)(l, vcp);
782 #ifdef DEBUG_EXEC
783 		if (error) {
784 			size_t j;
785 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
786 			for (j = 0; j <= i; j++)
787 				uprintf(
788 			"vmcmd[%zu] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
789 				    j, vp[j].ev_addr, vp[j].ev_len,
790 				    vp[j].ev_offset, vp[j].ev_prot,
791 				    vp[j].ev_flags);
792 		}
793 #endif /* DEBUG_EXEC */
794 		if (vcp->ev_flags & VMCMD_BASE)
795 			base_vcp = vcp;
796 	}
797 
798 	/* free the vmspace-creation commands, and release their references */
799 	kill_vmcmds(&pack.ep_vmcmds);
800 
801 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
802 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
803 	vput(pack.ep_vp);
804 
805 	/* if an error happened, deallocate and punt */
806 	if (error) {
807 		DPRINTF(("execve: vmcmd %zu failed: %d\n", i - 1, error));
808 		goto exec_abort;
809 	}
810 
811 	/* remember information about the process */
812 	arginfo.ps_nargvstr = argc;
813 	arginfo.ps_nenvstr = envc;
814 
815 	/* set command name & other accounting info */
816 	i = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
817 	(void)memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, i);
818 	p->p_comm[i] = '\0';
819 
820 	dp = PNBUF_GET();
821 	/*
822 	 * If the path starts with /, we don't need to do any work.
823 	 * This handles the majority of the cases.
824 	 * In the future perhaps we could canonicalize it?
825 	 */
826 	if (pathbuf[0] == '/')
827 		(void)strlcpy(pack.ep_path = dp, pathbuf, MAXPATHLEN);
828 #ifdef notyet
829 	/*
830 	 * Although this works most of the time [since the entry was just
831 	 * entered in the cache] we don't use it because it theoretically
832 	 * can fail and it is not the cleanest interface, because there
833 	 * could be races. When the namei cache is re-written, this can
834 	 * be changed to use the appropriate function.
835 	 */
836 	else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p)))
837 		pack.ep_path = dp;
838 #endif
839 	else {
840 #ifdef notyet
841 		printf("Cannot get path for pid %d [%s] (error %d)",
842 		    (int)p->p_pid, p->p_comm, error);
843 #endif
844 		pack.ep_path = NULL;
845 		PNBUF_PUT(dp);
846 	}
847 
848 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
849 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
850 		len - (sizeof(struct ps_strings) + szsigcode));
851 
852 #ifdef __MACHINE_STACK_GROWS_UP
853 	/*
854 	 * The copyargs call always copies into lower addresses
855 	 * first, moving towards higher addresses, starting with
856 	 * the stack pointer that we give.  When the stack grows
857 	 * down, this puts argc/argv/envp very shallow on the
858 	 * stack, right at the first user stack pointer.
859 	 * When the stack grows up, the situation is reversed.
860 	 *
861 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
862 	 * function expects to be called with a single pointer to
863 	 * a region that has a few words it can stash values into,
864 	 * followed by argc/argv/envp.  When the stack grows down,
865 	 * it's easy to decrement the stack pointer a little bit to
866 	 * allocate the space for these few words and pass the new
867 	 * stack pointer to _rtld.  When the stack grows up, however,
868 	 * a few words before argc is part of the signal trampoline, XXX
869 	 * so we have a problem.
870 	 *
871 	 * Instead of changing how _rtld works, we take the easy way
872 	 * out and steal 32 bytes before we call copyargs.
873 	 * This extra space was allowed for when 'len' was calculated.
874 	 */
875 	stack += RTLD_GAP;
876 #endif /* __MACHINE_STACK_GROWS_UP */
877 
878 	/* Now copy argc, args & environ to new stack */
879 	error = (*pack.ep_esch->es_copyargs)(l, &pack, &arginfo, &stack, argp);
880 	if (pack.ep_path) {
881 		PNBUF_PUT(pack.ep_path);
882 		pack.ep_path = NULL;
883 	}
884 	if (error) {
885 		DPRINTF(("execve: copyargs failed %d\n", error));
886 		goto exec_abort;
887 	}
888 	/* Move the stack back to original point */
889 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
890 
891 	/* fill process ps_strings info */
892 	p->p_psstr = (struct ps_strings *)
893 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
894 	    sizeof(struct ps_strings));
895 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
896 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
897 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
898 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
899 
900 	/* copy out the process's ps_strings structure */
901 	if ((error = copyout(aip, (char *)p->p_psstr,
902 	    sizeof(arginfo))) != 0) {
903 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
904 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
905 		goto exec_abort;
906 	}
907 
908 	fd_closeexec();		/* handle close on exec */
909 	execsigs(p);		/* reset catched signals */
910 
911 	l->l_ctxlink = NULL;	/* reset ucontext link */
912 
913 
914 	p->p_acflag &= ~AFORK;
915 	mutex_enter(p->p_lock);
916 	p->p_flag |= PK_EXEC;
917 	mutex_exit(p->p_lock);
918 
919 	/*
920 	 * Stop profiling.
921 	 */
922 	if ((p->p_stflag & PST_PROFIL) != 0) {
923 		mutex_spin_enter(&p->p_stmutex);
924 		stopprofclock(p);
925 		mutex_spin_exit(&p->p_stmutex);
926 	}
927 
928 	/*
929 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
930 	 * exited and exec()/exit() are the only places it will be cleared.
931 	 */
932 	if ((p->p_lflag & PL_PPWAIT) != 0) {
933 		mutex_enter(proc_lock);
934 		p->p_lflag &= ~PL_PPWAIT;
935 		cv_broadcast(&p->p_pptr->p_waitcv);
936 		mutex_exit(proc_lock);
937 	}
938 
939 	/*
940 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
941 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
942 	 * out additional references on the process for the moment.
943 	 */
944 	if ((p->p_slflag & PSL_TRACED) == 0 &&
945 
946 	    (((attr.va_mode & S_ISUID) != 0 &&
947 	      kauth_cred_geteuid(l->l_cred) != attr.va_uid) ||
948 
949 	     ((attr.va_mode & S_ISGID) != 0 &&
950 	      kauth_cred_getegid(l->l_cred) != attr.va_gid))) {
951 		/*
952 		 * Mark the process as SUGID before we do
953 		 * anything that might block.
954 		 */
955 		proc_crmod_enter();
956 		proc_crmod_leave(NULL, NULL, true);
957 
958 		/* Make sure file descriptors 0..2 are in use. */
959 		if ((error = fd_checkstd()) != 0) {
960 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
961 			goto exec_abort;
962 		}
963 
964 		/*
965 		 * Copy the credential so other references don't see our
966 		 * changes.
967 		 */
968 		l->l_cred = kauth_cred_copy(l->l_cred);
969 #ifdef KTRACE
970 		/*
971 		 * If the persistent trace flag isn't set, turn off.
972 		 */
973 		if (p->p_tracep) {
974 			mutex_enter(&ktrace_lock);
975 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
976 				ktrderef(p);
977 			mutex_exit(&ktrace_lock);
978 		}
979 #endif
980 		if (attr.va_mode & S_ISUID)
981 			kauth_cred_seteuid(l->l_cred, attr.va_uid);
982 		if (attr.va_mode & S_ISGID)
983 			kauth_cred_setegid(l->l_cred, attr.va_gid);
984 	} else {
985 		if (kauth_cred_geteuid(l->l_cred) ==
986 		    kauth_cred_getuid(l->l_cred) &&
987 		    kauth_cred_getegid(l->l_cred) ==
988 		    kauth_cred_getgid(l->l_cred))
989 			p->p_flag &= ~PK_SUGID;
990 	}
991 
992 	/*
993 	 * Copy the credential so other references don't see our changes.
994 	 * Test to see if this is necessary first, since in the common case
995 	 * we won't need a private reference.
996 	 */
997 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
998 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
999 		l->l_cred = kauth_cred_copy(l->l_cred);
1000 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1001 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1002 	}
1003 
1004 	/* Update the master credentials. */
1005 	if (l->l_cred != p->p_cred) {
1006 		kauth_cred_t ocred;
1007 
1008 		kauth_cred_hold(l->l_cred);
1009 		mutex_enter(p->p_lock);
1010 		ocred = p->p_cred;
1011 		p->p_cred = l->l_cred;
1012 		mutex_exit(p->p_lock);
1013 		kauth_cred_free(ocred);
1014 	}
1015 
1016 #if defined(__HAVE_RAS)
1017 	/*
1018 	 * Remove all RASs from the address space.
1019 	 */
1020 	ras_purgeall();
1021 #endif
1022 
1023 	doexechooks(p);
1024 
1025 	/* setup new registers and do misc. setup. */
1026 	(*pack.ep_esch->es_emul->e_setregs)(l, &pack, (u_long) stack);
1027 	if (pack.ep_esch->es_setregs)
1028 		(*pack.ep_esch->es_setregs)(l, &pack, (u_long) stack);
1029 
1030 	/* map the process's signal trampoline code */
1031 	if (exec_sigcode_map(p, pack.ep_esch->es_emul)) {
1032 		DPRINTF(("execve: map sigcode failed %d\n", error));
1033 		goto exec_abort;
1034 	}
1035 
1036 	pool_put(&exec_pool, argp);
1037 
1038 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1039 
1040 	/* notify others that we exec'd */
1041 	KNOTE(&p->p_klist, NOTE_EXEC);
1042 
1043 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1044 
1045 	/* The emulation root will usually have been found when we looked
1046 	 * for the elf interpreter (or similar), if not look now. */
1047 	if (pack.ep_esch->es_emul->e_path != NULL && pack.ep_emul_root == NULL)
1048 		emul_find_root(l, &pack);
1049 
1050 	/* Any old emulation root got removed by fdcloseexec */
1051 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1052 	p->p_cwdi->cwdi_edir = pack.ep_emul_root;
1053 	rw_exit(&p->p_cwdi->cwdi_lock);
1054 	pack.ep_emul_root = NULL;
1055 	if (pack.ep_interp != NULL)
1056 		vrele(pack.ep_interp);
1057 
1058 	/*
1059 	 * Call emulation specific exec hook. This can setup per-process
1060 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1061 	 *
1062 	 * If we are executing process of different emulation than the
1063 	 * original forked process, call e_proc_exit() of the old emulation
1064 	 * first, then e_proc_exec() of new emulation. If the emulation is
1065 	 * same, the exec hook code should deallocate any old emulation
1066 	 * resources held previously by this process.
1067 	 */
1068 	if (p->p_emul && p->p_emul->e_proc_exit
1069 	    && p->p_emul != pack.ep_esch->es_emul)
1070 		(*p->p_emul->e_proc_exit)(p);
1071 
1072 	/*
1073 	 * Call exec hook. Emulation code may NOT store reference to anything
1074 	 * from &pack.
1075 	 */
1076         if (pack.ep_esch->es_emul->e_proc_exec)
1077                 (*pack.ep_esch->es_emul->e_proc_exec)(p, &pack);
1078 
1079 	/* update p_emul, the old value is no longer needed */
1080 	p->p_emul = pack.ep_esch->es_emul;
1081 
1082 	/* ...and the same for p_execsw */
1083 	p->p_execsw = pack.ep_esch;
1084 
1085 #ifdef __HAVE_SYSCALL_INTERN
1086 	(*p->p_emul->e_syscall_intern)(p);
1087 #endif
1088 	ktremul();
1089 
1090 	/* Allow new references from the debugger/procfs. */
1091 	rw_exit(&p->p_reflock);
1092 #ifdef LKM
1093 	rw_exit(&exec_lock);
1094 #endif
1095 
1096 	mutex_enter(proc_lock);
1097 
1098 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1099 		KSI_INIT_EMPTY(&ksi);
1100 		ksi.ksi_signo = SIGTRAP;
1101 		ksi.ksi_lid = l->l_lid;
1102 		kpsignal(p, &ksi, NULL);
1103 	}
1104 
1105 	if (p->p_sflag & PS_STOPEXEC) {
1106 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1107 		p->p_pptr->p_nstopchild++;
1108 		p->p_pptr->p_waited = 0;
1109 		mutex_enter(p->p_lock);
1110 		ksiginfo_queue_init(&kq);
1111 		sigclearall(p, &contsigmask, &kq);
1112 		lwp_lock(l);
1113 		l->l_stat = LSSTOP;
1114 		p->p_stat = SSTOP;
1115 		p->p_nrlwps--;
1116 		mutex_exit(p->p_lock);
1117 		mutex_exit(proc_lock);
1118 		mi_switch(l);
1119 		ksiginfo_queue_drain(&kq);
1120 		KERNEL_LOCK(l->l_biglocks, l);
1121 	} else {
1122 		mutex_exit(proc_lock);
1123 	}
1124 
1125 	PNBUF_PUT(pathbuf);
1126 	return (EJUSTRETURN);
1127 
1128  bad:
1129 	/* free the vmspace-creation commands, and release their references */
1130 	kill_vmcmds(&pack.ep_vmcmds);
1131 	/* kill any opened file descriptor, if necessary */
1132 	if (pack.ep_flags & EXEC_HASFD) {
1133 		pack.ep_flags &= ~EXEC_HASFD;
1134 		fd_close(pack.ep_fd);
1135 	}
1136 	/* close and put the exec'd file */
1137 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
1138 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
1139 	vput(pack.ep_vp);
1140 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1141 	pool_put(&exec_pool, argp);
1142 
1143  freehdr:
1144 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1145 	if (pack.ep_emul_root != NULL)
1146 		vrele(pack.ep_emul_root);
1147 	if (pack.ep_interp != NULL)
1148 		vrele(pack.ep_interp);
1149 
1150 #ifdef LKM
1151 	rw_exit(&exec_lock);
1152 #endif
1153 
1154  clrflg:
1155 	lwp_lock(l);
1156 	l->l_flag |= oldlwpflags;
1157 	lwp_unlock(l);
1158 	PNBUF_PUT(pathbuf);
1159 	rw_exit(&p->p_reflock);
1160 
1161 	return error;
1162 
1163  exec_abort:
1164 	PNBUF_PUT(pathbuf);
1165 	rw_exit(&p->p_reflock);
1166 #ifdef LKM
1167 	rw_exit(&exec_lock);
1168 #endif
1169 
1170 	/*
1171 	 * the old process doesn't exist anymore.  exit gracefully.
1172 	 * get rid of the (new) address space we have created, if any, get rid
1173 	 * of our namei data and vnode, and exit noting failure
1174 	 */
1175 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1176 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1177 	if (pack.ep_emul_arg)
1178 		FREE(pack.ep_emul_arg, M_TEMP);
1179 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1180 	pool_put(&exec_pool, argp);
1181 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1182 	if (pack.ep_emul_root != NULL)
1183 		vrele(pack.ep_emul_root);
1184 	if (pack.ep_interp != NULL)
1185 		vrele(pack.ep_interp);
1186 
1187 	/* Acquire the sched-state mutex (exit1() will release it). */
1188 	mutex_enter(p->p_lock);
1189 	exit1(l, W_EXITCODE(error, SIGABRT));
1190 
1191 	/* NOTREACHED */
1192 	return 0;
1193 }
1194 
1195 
1196 int
1197 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1198     char **stackp, void *argp)
1199 {
1200 	char	**cpp, *dp, *sp;
1201 	size_t	len;
1202 	void	*nullp;
1203 	long	argc, envc;
1204 	int	error;
1205 
1206 	cpp = (char **)*stackp;
1207 	nullp = NULL;
1208 	argc = arginfo->ps_nargvstr;
1209 	envc = arginfo->ps_nenvstr;
1210 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
1211 		return error;
1212 
1213 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen);
1214 	sp = argp;
1215 
1216 	/* XXX don't copy them out, remap them! */
1217 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1218 
1219 	for (; --argc >= 0; sp += len, dp += len)
1220 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1221 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1222 			return error;
1223 
1224 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1225 		return error;
1226 
1227 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1228 
1229 	for (; --envc >= 0; sp += len, dp += len)
1230 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1231 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1232 			return error;
1233 
1234 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1235 		return error;
1236 
1237 	*stackp = (char *)cpp;
1238 	return 0;
1239 }
1240 
1241 #ifdef LKM
1242 /*
1243  * Find an emulation of given name in list of emulations.
1244  * Needs to be called with the exec_lock held.
1245  */
1246 const struct emul *
1247 emul_search(const char *name)
1248 {
1249 	struct emul_entry *it;
1250 
1251 	LIST_FOREACH(it, &el_head, el_list) {
1252 		if (strcmp(name, it->el_emul->e_name) == 0)
1253 			return it->el_emul;
1254 	}
1255 
1256 	return NULL;
1257 }
1258 
1259 /*
1260  * Add an emulation to list, if it's not there already.
1261  */
1262 int
1263 emul_register(const struct emul *emul, int ro_entry)
1264 {
1265 	struct emul_entry	*ee;
1266 	int			error;
1267 
1268 	error = 0;
1269 	rw_enter(&exec_lock, RW_WRITER);
1270 
1271 	if (emul_search(emul->e_name)) {
1272 		error = EEXIST;
1273 		goto out;
1274 	}
1275 
1276 	ee = kmem_alloc(sizeof(*ee), KM_SLEEP);
1277 	ee->el_emul = emul;
1278 	ee->ro_entry = ro_entry;
1279 	LIST_INSERT_HEAD(&el_head, ee, el_list);
1280 
1281  out:
1282 	rw_exit(&exec_lock);
1283 	return error;
1284 }
1285 
1286 /*
1287  * Remove emulation with name 'name' from list of supported emulations.
1288  */
1289 int
1290 emul_unregister(const char *name)
1291 {
1292 	const struct proclist_desc *pd;
1293 	struct emul_entry	*it;
1294 	int			i, error;
1295 	struct proc		*ptmp;
1296 
1297 	error = 0;
1298 	rw_enter(&exec_lock, RW_WRITER);
1299 
1300 	LIST_FOREACH(it, &el_head, el_list) {
1301 		if (strcmp(it->el_emul->e_name, name) == 0)
1302 			break;
1303 	}
1304 
1305 	if (!it) {
1306 		error = ENOENT;
1307 		goto out;
1308 	}
1309 
1310 	if (it->ro_entry) {
1311 		error = EBUSY;
1312 		goto out;
1313 	}
1314 
1315 	/* test if any execw[] entry is still using this */
1316 	for(i=0; i < nexecs; i++) {
1317 		if (execsw[i]->es_emul == it->el_emul) {
1318 			error = EBUSY;
1319 			goto out;
1320 		}
1321 	}
1322 
1323 	/*
1324 	 * Test if any process is running under this emulation - since
1325 	 * emul_unregister() is running quite sendomly, it's better
1326 	 * to do expensive check here than to use any locking.
1327 	 */
1328 	mutex_enter(proc_lock);
1329 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
1330 		PROCLIST_FOREACH(ptmp, pd->pd_list) {
1331 			if (ptmp->p_emul == it->el_emul) {
1332 				error = EBUSY;
1333 				break;
1334 			}
1335 		}
1336 	}
1337 	mutex_exit(proc_lock);
1338 
1339 	if (error)
1340 		goto out;
1341 
1342 
1343 	/* entry is not used, remove it */
1344 	LIST_REMOVE(it, el_list);
1345 	kmem_free(it, sizeof(*it));
1346 
1347  out:
1348 	rw_exit(&exec_lock);
1349 	return error;
1350 }
1351 
1352 /*
1353  * Add execsw[] entry.
1354  */
1355 int
1356 exec_add(struct execsw *esp, const char *e_name)
1357 {
1358 	struct exec_entry	*it;
1359 	int			error;
1360 
1361 	error = 0;
1362 	rw_enter(&exec_lock, RW_WRITER);
1363 
1364 	if (!esp->es_emul) {
1365 		esp->es_emul = emul_search(e_name);
1366 		if (!esp->es_emul) {
1367 			error = ENOENT;
1368 			goto out;
1369 		}
1370 	}
1371 
1372 	LIST_FOREACH(it, &ex_head, ex_list) {
1373 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1374 		if (it->es->es_makecmds == esp->es_makecmds
1375 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1376 		    && it->es->es_emul == esp->es_emul) {
1377 			error = EEXIST;
1378 			goto out;
1379 		}
1380 	}
1381 
1382 	/* if we got here, the entry doesn't exist yet */
1383 	it = kmem_alloc(sizeof(*it), KM_SLEEP);
1384 	it->es = esp;
1385 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
1386 
1387 	/* update execsw[] */
1388 	exec_init(0);
1389 
1390  out:
1391 	rw_exit(&exec_lock);
1392 	return error;
1393 }
1394 
1395 /*
1396  * Remove execsw[] entry.
1397  */
1398 int
1399 exec_remove(const struct execsw *esp)
1400 {
1401 	struct exec_entry	*it;
1402 	int			error;
1403 
1404 	error = 0;
1405 	rw_enter(&exec_lock, RW_WRITER);
1406 
1407 	LIST_FOREACH(it, &ex_head, ex_list) {
1408 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1409 		if (it->es->es_makecmds == esp->es_makecmds
1410 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1411 		    && it->es->es_emul == esp->es_emul)
1412 			break;
1413 	}
1414 	if (!it) {
1415 		error = ENOENT;
1416 		goto out;
1417 	}
1418 
1419 	/* remove item from list and free resources */
1420 	LIST_REMOVE(it, ex_list);
1421 	kmem_free(it, sizeof(*it));
1422 
1423 	/* update execsw[] */
1424 	exec_init(0);
1425 
1426  out:
1427 	rw_exit(&exec_lock);
1428 	return error;
1429 }
1430 
1431 static void
1432 link_es(struct execsw_entry **listp, const struct execsw *esp)
1433 {
1434 	struct execsw_entry *et, *e1;
1435 
1436 	et = (struct execsw_entry *) malloc(sizeof(struct execsw_entry),
1437 			M_TEMP, M_WAITOK);
1438 	et->next = NULL;
1439 	et->es = esp;
1440 	if (*listp == NULL) {
1441 		*listp = et;
1442 		return;
1443 	}
1444 
1445 	switch(et->es->es_prio) {
1446 	case EXECSW_PRIO_FIRST:
1447 		/* put new entry as the first */
1448 		et->next = *listp;
1449 		*listp = et;
1450 		break;
1451 	case EXECSW_PRIO_ANY:
1452 		/* put new entry after all *_FIRST and *_ANY entries */
1453 		for(e1 = *listp; e1->next
1454 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
1455 			e1 = e1->next);
1456 		et->next = e1->next;
1457 		e1->next = et;
1458 		break;
1459 	case EXECSW_PRIO_LAST:
1460 		/* put new entry as the last one */
1461 		for(e1 = *listp; e1->next; e1 = e1->next);
1462 		e1->next = et;
1463 		break;
1464 	default:
1465 #ifdef DIAGNOSTIC
1466 		panic("execw[] entry with unknown priority %d found",
1467 			et->es->es_prio);
1468 #else
1469 		free(et, M_TEMP);
1470 #endif
1471 		break;
1472 	}
1473 }
1474 
1475 /*
1476  * Initialize exec structures. If init_boot is true, also does necessary
1477  * one-time initialization (it's called from main() that way).
1478  * Once system is multiuser, this should be called with exec_lock held,
1479  * i.e. via exec_{add|remove}().
1480  */
1481 int
1482 exec_init(int init_boot)
1483 {
1484 	const struct execsw	**new_es, * const *old_es;
1485 	struct execsw_entry	*list, *e1;
1486 	struct exec_entry	*e2;
1487 	int			i, es_sz;
1488 
1489 	if (init_boot) {
1490 		/* do one-time initializations */
1491 		rw_init(&exec_lock);
1492 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1493 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1494 		    "execargs", &exec_palloc, IPL_NONE);
1495 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1496 
1497 		/* register compiled-in emulations */
1498 		for(i=0; i < nexecs_builtin; i++) {
1499 			if (execsw_builtin[i].es_emul)
1500 				emul_register(execsw_builtin[i].es_emul, 1);
1501 		}
1502 #ifdef DIAGNOSTIC
1503 		if (i == 0)
1504 			panic("no emulations found in execsw_builtin[]");
1505 #endif
1506 	}
1507 
1508 	/*
1509 	 * Build execsw[] array from builtin entries and entries added
1510 	 * at runtime.
1511 	 */
1512 	list = NULL;
1513 	for(i=0; i < nexecs_builtin; i++)
1514 		link_es(&list, &execsw_builtin[i]);
1515 
1516 	/* Add dynamically loaded entries */
1517 	es_sz = nexecs_builtin;
1518 	LIST_FOREACH(e2, &ex_head, ex_list) {
1519 		link_es(&list, e2->es);
1520 		es_sz++;
1521 	}
1522 
1523 	/*
1524 	 * Now that we have sorted all execw entries, create new execsw[]
1525 	 * and free no longer needed memory in the process.
1526 	 */
1527 	new_es = kmem_alloc(es_sz * sizeof(struct execsw *), KM_SLEEP);
1528 	for(i=0; list; i++) {
1529 		new_es[i] = list->es;
1530 		e1 = list->next;
1531 		free(list, M_TEMP);
1532 		list = e1;
1533 	}
1534 
1535 	/*
1536 	 * New execsw[] array built, now replace old execsw[] and free
1537 	 * used memory.
1538 	 */
1539 	old_es = execsw;
1540 	if (old_es)
1541 		/*XXXUNCONST*/
1542 		kmem_free(__UNCONST(old_es), nexecs * sizeof(struct execsw *));
1543 	execsw = new_es;
1544 	nexecs = es_sz;
1545 
1546 	/*
1547 	 * Figure out the maximum size of an exec header.
1548 	 */
1549 	exec_maxhdrsz = 0;
1550 	for (i = 0; i < nexecs; i++) {
1551 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1552 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1553 	}
1554 
1555 	return 0;
1556 }
1557 #endif
1558 
1559 #ifndef LKM
1560 /*
1561  * Simplified exec_init() for kernels without LKMs. Only initialize
1562  * exec_maxhdrsz and execsw[].
1563  */
1564 int
1565 exec_init(int init_boot)
1566 {
1567 	int i;
1568 
1569 #ifdef DIAGNOSTIC
1570 	if (!init_boot)
1571 		panic("exec_init(): called with init_boot == 0");
1572 #endif
1573 
1574 	/* do one-time initializations */
1575 	nexecs = nexecs_builtin;
1576 	execsw = kmem_alloc(nexecs * sizeof(struct execsw *), KM_SLEEP);
1577 
1578 	pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1579 	    "execargs", &exec_palloc, IPL_NONE);
1580 	pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1581 
1582 	/*
1583 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1584 	 */
1585 	exec_maxhdrsz = 0;
1586 	for(i=0; i < nexecs; i++) {
1587 		execsw[i] = &execsw_builtin[i];
1588 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1589 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1590 	}
1591 
1592 	return 0;
1593 
1594 }
1595 #endif /* !LKM */
1596 
1597 static int
1598 exec_sigcode_map(struct proc *p, const struct emul *e)
1599 {
1600 	vaddr_t va;
1601 	vsize_t sz;
1602 	int error;
1603 	struct uvm_object *uobj;
1604 
1605 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1606 
1607 	if (e->e_sigobject == NULL || sz == 0) {
1608 		return 0;
1609 	}
1610 
1611 	/*
1612 	 * If we don't have a sigobject for this emulation, create one.
1613 	 *
1614 	 * sigobject is an anonymous memory object (just like SYSV shared
1615 	 * memory) that we keep a permanent reference to and that we map
1616 	 * in all processes that need this sigcode. The creation is simple,
1617 	 * we create an object, add a permanent reference to it, map it in
1618 	 * kernel space, copy out the sigcode to it and unmap it.
1619 	 * We map it with PROT_READ|PROT_EXEC into the process just
1620 	 * the way sys_mmap() would map it.
1621 	 */
1622 
1623 	uobj = *e->e_sigobject;
1624 	if (uobj == NULL) {
1625 		mutex_enter(&sigobject_lock);
1626 		if ((uobj = *e->e_sigobject) == NULL) {
1627 			uobj = uao_create(sz, 0);
1628 			(*uobj->pgops->pgo_reference)(uobj);
1629 			va = vm_map_min(kernel_map);
1630 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1631 			    uobj, 0, 0,
1632 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1633 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1634 				printf("kernel mapping failed %d\n", error);
1635 				(*uobj->pgops->pgo_detach)(uobj);
1636 				mutex_exit(&sigobject_lock);
1637 				return (error);
1638 			}
1639 			memcpy((void *)va, e->e_sigcode, sz);
1640 #ifdef PMAP_NEED_PROCWR
1641 			pmap_procwr(&proc0, va, sz);
1642 #endif
1643 			uvm_unmap(kernel_map, va, va + round_page(sz));
1644 			*e->e_sigobject = uobj;
1645 		}
1646 		mutex_exit(&sigobject_lock);
1647 	}
1648 
1649 	/* Just a hint to uvm_map where to put it. */
1650 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1651 	    round_page(sz));
1652 
1653 #ifdef __alpha__
1654 	/*
1655 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1656 	 * which causes the above calculation to put the sigcode at
1657 	 * an invalid address.  Put it just below the text instead.
1658 	 */
1659 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1660 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1661 	}
1662 #endif
1663 
1664 	(*uobj->pgops->pgo_reference)(uobj);
1665 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1666 			uobj, 0, 0,
1667 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1668 				    UVM_ADV_RANDOM, 0));
1669 	if (error) {
1670 		(*uobj->pgops->pgo_detach)(uobj);
1671 		return (error);
1672 	}
1673 	p->p_sigctx.ps_sigcode = (void *)va;
1674 	return (0);
1675 }
1676