xref: /netbsd-src/sys/kern/kern_exec.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /*	$NetBSD: kern_exec.c,v 1.305 2011/01/18 08:21:03 matt 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 1.305 2011/01/18 08:21:03 matt Exp $");
63 
64 #include "opt_ktrace.h"
65 #include "opt_modular.h"
66 #include "opt_syscall_debug.h"
67 #include "veriexec.h"
68 #include "opt_pax.h"
69 #include "opt_sa.h"
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/filedesc.h>
74 #include <sys/kernel.h>
75 #include <sys/proc.h>
76 #include <sys/mount.h>
77 #include <sys/malloc.h>
78 #include <sys/kmem.h>
79 #include <sys/namei.h>
80 #include <sys/vnode.h>
81 #include <sys/file.h>
82 #include <sys/acct.h>
83 #include <sys/exec.h>
84 #include <sys/ktrace.h>
85 #include <sys/uidinfo.h>
86 #include <sys/wait.h>
87 #include <sys/mman.h>
88 #include <sys/ras.h>
89 #include <sys/signalvar.h>
90 #include <sys/stat.h>
91 #include <sys/syscall.h>
92 #include <sys/kauth.h>
93 #include <sys/lwpctl.h>
94 #include <sys/pax.h>
95 #include <sys/cpu.h>
96 #include <sys/module.h>
97 #include <sys/sa.h>
98 #include <sys/savar.h>
99 #include <sys/syscallvar.h>
100 #include <sys/syscallargs.h>
101 #if NVERIEXEC > 0
102 #include <sys/verified_exec.h>
103 #endif /* NVERIEXEC > 0 */
104 #include <sys/sdt.h>
105 
106 #include <uvm/uvm_extern.h>
107 
108 #include <machine/reg.h>
109 
110 #include <compat/common/compat_util.h>
111 
112 static int exec_sigcode_map(struct proc *, const struct emul *);
113 
114 #ifdef DEBUG_EXEC
115 #define DPRINTF(a) printf a
116 #else
117 #define DPRINTF(a)
118 #endif /* DEBUG_EXEC */
119 
120 /*
121  * DTrace SDT provider definitions
122  */
123 SDT_PROBE_DEFINE(proc,,,exec,
124 	    "char *", NULL,
125 	    NULL, NULL, NULL, NULL,
126 	    NULL, NULL, NULL, NULL);
127 SDT_PROBE_DEFINE(proc,,,exec_success,
128 	    "char *", NULL,
129 	    NULL, NULL, NULL, NULL,
130 	    NULL, NULL, NULL, NULL);
131 SDT_PROBE_DEFINE(proc,,,exec_failure,
132 	    "int", NULL,
133 	    NULL, NULL, NULL, NULL,
134 	    NULL, NULL, NULL, NULL);
135 
136 /*
137  * Exec function switch:
138  *
139  * Note that each makecmds function is responsible for loading the
140  * exec package with the necessary functions for any exec-type-specific
141  * handling.
142  *
143  * Functions for specific exec types should be defined in their own
144  * header file.
145  */
146 static const struct execsw	**execsw = NULL;
147 static int			nexecs;
148 
149 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
150 
151 /* list of dynamically loaded execsw entries */
152 static LIST_HEAD(execlist_head, exec_entry) ex_head =
153     LIST_HEAD_INITIALIZER(ex_head);
154 struct exec_entry {
155 	LIST_ENTRY(exec_entry)	ex_list;
156 	SLIST_ENTRY(exec_entry)	ex_slist;
157 	const struct execsw	*ex_sw;
158 };
159 
160 #ifndef __HAVE_SYSCALL_INTERN
161 void	syscall(void);
162 #endif
163 
164 #ifdef KERN_SA
165 static struct sa_emul saemul_netbsd = {
166 	sizeof(ucontext_t),
167 	sizeof(struct sa_t),
168 	sizeof(struct sa_t *),
169 	NULL,
170 	NULL,
171 	cpu_upcall,
172 	(void (*)(struct lwp *, void *))getucontext_sa,
173 	sa_ucsp
174 };
175 #endif /* KERN_SA */
176 
177 /* NetBSD emul struct */
178 struct emul emul_netbsd = {
179 	.e_name =		"netbsd",
180 	.e_path =		NULL,
181 #ifndef __HAVE_MINIMAL_EMUL
182 	.e_flags =		EMUL_HAS_SYS___syscall,
183 	.e_errno =		NULL,
184 	.e_nosys =		SYS_syscall,
185 	.e_nsysent =		SYS_NSYSENT,
186 #endif
187 	.e_sysent =		sysent,
188 #ifdef SYSCALL_DEBUG
189 	.e_syscallnames =	syscallnames,
190 #else
191 	.e_syscallnames =	NULL,
192 #endif
193 	.e_sendsig =		sendsig,
194 	.e_trapsignal =		trapsignal,
195 	.e_tracesig =		NULL,
196 	.e_sigcode =		NULL,
197 	.e_esigcode =		NULL,
198 	.e_sigobject =		NULL,
199 	.e_setregs =		setregs,
200 	.e_proc_exec =		NULL,
201 	.e_proc_fork =		NULL,
202 	.e_proc_exit =		NULL,
203 	.e_lwp_fork =		NULL,
204 	.e_lwp_exit =		NULL,
205 #ifdef __HAVE_SYSCALL_INTERN
206 	.e_syscall_intern =	syscall_intern,
207 #else
208 	.e_syscall =		syscall,
209 #endif
210 	.e_sysctlovly =		NULL,
211 	.e_fault =		NULL,
212 	.e_vm_default_addr =	uvm_default_mapaddr,
213 	.e_usertrap =		NULL,
214 #ifdef KERN_SA
215 	.e_sa =			&saemul_netbsd,
216 #else
217 	.e_sa =			NULL,
218 #endif
219 	.e_ucsize =		sizeof(ucontext_t),
220 	.e_startlwp =		startlwp
221 };
222 
223 /*
224  * Exec lock. Used to control access to execsw[] structures.
225  * This must not be static so that netbsd32 can access it, too.
226  */
227 krwlock_t exec_lock;
228 
229 static kmutex_t sigobject_lock;
230 
231 static void *
232 exec_pool_alloc(struct pool *pp, int flags)
233 {
234 
235 	return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
236 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
237 }
238 
239 static void
240 exec_pool_free(struct pool *pp, void *addr)
241 {
242 
243 	uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
244 }
245 
246 static struct pool exec_pool;
247 
248 static struct pool_allocator exec_palloc = {
249 	.pa_alloc = exec_pool_alloc,
250 	.pa_free = exec_pool_free,
251 	.pa_pagesz = NCARGS
252 };
253 
254 /*
255  * check exec:
256  * given an "executable" described in the exec package's namei info,
257  * see what we can do with it.
258  *
259  * ON ENTRY:
260  *	exec package with appropriate namei info
261  *	lwp pointer of exec'ing lwp
262  *	NO SELF-LOCKED VNODES
263  *
264  * ON EXIT:
265  *	error:	nothing held, etc.  exec header still allocated.
266  *	ok:	filled exec package, executable's vnode (unlocked).
267  *
268  * EXEC SWITCH ENTRY:
269  * 	Locked vnode to check, exec package, proc.
270  *
271  * EXEC SWITCH EXIT:
272  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
273  *	error:	destructive:
274  *			everything deallocated execept exec header.
275  *		non-destructive:
276  *			error code, executable's vnode (unlocked),
277  *			exec header unmodified.
278  */
279 int
280 /*ARGSUSED*/
281 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
282 {
283 	int		error, i;
284 	struct vnode	*vp;
285 	struct nameidata nd;
286 	size_t		resid;
287 
288 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
289 
290 	/* first get the vnode */
291 	if ((error = namei(&nd)) != 0)
292 		return error;
293 	epp->ep_vp = vp = nd.ni_vp;
294 	/* this cannot overflow as both are size PATH_MAX */
295 	strcpy(epp->ep_resolvedname, nd.ni_pnbuf);
296 
297 #ifdef DIAGNOSTIC
298 	/* paranoia (take this out once namei stuff stabilizes) */
299 	memset(nd.ni_pnbuf, '~', PATH_MAX);
300 #endif
301 
302 	/* check access and type */
303 	if (vp->v_type != VREG) {
304 		error = EACCES;
305 		goto bad1;
306 	}
307 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
308 		goto bad1;
309 
310 	/* get attributes */
311 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
312 		goto bad1;
313 
314 	/* Check mount point */
315 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
316 		error = EACCES;
317 		goto bad1;
318 	}
319 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
320 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
321 
322 	/* try to open it */
323 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
324 		goto bad1;
325 
326 	/* unlock vp, since we need it unlocked from here on out. */
327 	VOP_UNLOCK(vp);
328 
329 #if NVERIEXEC > 0
330 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
331 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
332 	    NULL);
333 	if (error)
334 		goto bad2;
335 #endif /* NVERIEXEC > 0 */
336 
337 #ifdef PAX_SEGVGUARD
338 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
339 	if (error)
340 		goto bad2;
341 #endif /* PAX_SEGVGUARD */
342 
343 	/* now we have the file, get the exec header */
344 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
345 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
346 	if (error)
347 		goto bad2;
348 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
349 
350 	/*
351 	 * Set up default address space limits.  Can be overridden
352 	 * by individual exec packages.
353 	 *
354 	 * XXX probably should be all done in the exec packages.
355 	 */
356 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
357 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
358 	/*
359 	 * set up the vmcmds for creation of the process
360 	 * address space
361 	 */
362 	error = ENOEXEC;
363 	for (i = 0; i < nexecs; i++) {
364 		int newerror;
365 
366 		epp->ep_esch = execsw[i];
367 		newerror = (*execsw[i]->es_makecmds)(l, epp);
368 
369 		if (!newerror) {
370 			/* Seems ok: check that entry point is sane */
371 			if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
372 				error = ENOEXEC;
373 				break;
374 			}
375 
376 			/* check limits */
377 			if ((epp->ep_tsize > MAXTSIZ) ||
378 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
379 						    [RLIMIT_DATA].rlim_cur)) {
380 				error = ENOMEM;
381 				break;
382 			}
383 			return 0;
384 		}
385 
386 		if (epp->ep_emul_root != NULL) {
387 			vrele(epp->ep_emul_root);
388 			epp->ep_emul_root = NULL;
389 		}
390 		if (epp->ep_interp != NULL) {
391 			vrele(epp->ep_interp);
392 			epp->ep_interp = NULL;
393 		}
394 
395 		/* make sure the first "interesting" error code is saved. */
396 		if (error == ENOEXEC)
397 			error = newerror;
398 
399 		if (epp->ep_flags & EXEC_DESTR)
400 			/* Error from "#!" code, tidied up by recursive call */
401 			return error;
402 	}
403 
404 	/* not found, error */
405 
406 	/*
407 	 * free any vmspace-creation commands,
408 	 * and release their references
409 	 */
410 	kill_vmcmds(&epp->ep_vmcmds);
411 
412 bad2:
413 	/*
414 	 * close and release the vnode, restore the old one, free the
415 	 * pathname buf, and punt.
416 	 */
417 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
418 	VOP_CLOSE(vp, FREAD, l->l_cred);
419 	vput(vp);
420 	return error;
421 
422 bad1:
423 	/*
424 	 * free the namei pathname buffer, and put the vnode
425 	 * (which we don't yet have open).
426 	 */
427 	vput(vp);				/* was still locked */
428 	return error;
429 }
430 
431 #ifdef __MACHINE_STACK_GROWS_UP
432 #define STACK_PTHREADSPACE NBPG
433 #else
434 #define STACK_PTHREADSPACE 0
435 #endif
436 
437 static int
438 execve_fetch_element(char * const *array, size_t index, char **value)
439 {
440 	return copyin(array + index, value, sizeof(*value));
441 }
442 
443 /*
444  * exec system call
445  */
446 /* ARGSUSED */
447 int
448 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
449 {
450 	/* {
451 		syscallarg(const char *)	path;
452 		syscallarg(char * const *)	argp;
453 		syscallarg(char * const *)	envp;
454 	} */
455 
456 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
457 	    SCARG(uap, envp), execve_fetch_element);
458 }
459 
460 /*
461  * Load modules to try and execute an image that we do not understand.
462  * If no execsw entries are present, we load those likely to be needed
463  * in order to run native images only.  Otherwise, we autoload all
464  * possible modules that could let us run the binary.  XXX lame
465  */
466 static void
467 exec_autoload(void)
468 {
469 #ifdef MODULAR
470 	static const char * const native[] = {
471 		"exec_elf32",
472 		"exec_elf64",
473 		"exec_script",
474 		NULL
475 	};
476 	static const char * const compat[] = {
477 		"exec_elf32",
478 		"exec_elf64",
479 		"exec_script",
480 		"exec_aout",
481 		"exec_coff",
482 		"exec_ecoff",
483 		"compat_aoutm68k",
484 		"compat_freebsd",
485 		"compat_ibcs2",
486 		"compat_irix",
487 		"compat_linux",
488 		"compat_linux32",
489 		"compat_netbsd32",
490 		"compat_sunos",
491 		"compat_sunos32",
492 		"compat_svr4",
493 		"compat_svr4_32",
494 		"compat_ultrix",
495 		NULL
496 	};
497 	char const * const *list;
498 	int i;
499 
500 	list = (nexecs == 0 ? native : compat);
501 	for (i = 0; list[i] != NULL; i++) {
502 		if (module_autoload(list[i], MODULE_CLASS_MISC) != 0) {
503 		    	continue;
504 		}
505 	   	yield();
506 	}
507 #endif
508 }
509 
510 int
511 execve1(struct lwp *l, const char *path, char * const *args,
512     char * const *envs, execve_fetch_element_t fetch_element)
513 {
514 	int			error;
515 	struct exec_package	pack;
516 	struct pathbuf		*pb;
517 	struct vattr		attr;
518 	struct proc		*p;
519 	char			*argp;
520 	char			*dp, *sp;
521 	long			argc, envc;
522 	size_t			i, len;
523 	char			*stack;
524 	struct ps_strings	arginfo;
525 	struct ps_strings	*aip = &arginfo;
526 	struct vmspace		*vm;
527 	struct exec_fakearg	*tmpfap;
528 	int			szsigcode;
529 	struct exec_vmcmd	*base_vcp;
530 	int			oldlwpflags;
531 	ksiginfo_t		ksi;
532 	ksiginfoq_t		kq;
533 	const char		*pathstring;
534 	char			*resolvedpathbuf;
535 	const char		*commandname;
536 	u_int			modgen;
537 
538 	p = l->l_proc;
539  	modgen = 0;
540 
541 	SDT_PROBE(proc,,,exec, path, 0, 0, 0, 0);
542 
543 	/*
544 	 * Check if we have exceeded our number of processes limit.
545 	 * This is so that we handle the case where a root daemon
546 	 * forked, ran setuid to become the desired user and is trying
547 	 * to exec. The obvious place to do the reference counting check
548 	 * is setuid(), but we don't do the reference counting check there
549 	 * like other OS's do because then all the programs that use setuid()
550 	 * must be modified to check the return code of setuid() and exit().
551 	 * It is dangerous to make setuid() fail, because it fails open and
552 	 * the program will continue to run as root. If we make it succeed
553 	 * and return an error code, again we are not enforcing the limit.
554 	 * The best place to enforce the limit is here, when the process tries
555 	 * to execute a new image, because eventually the process will need
556 	 * to call exec in order to do something useful.
557 	 */
558  retry:
559 	if ((p->p_flag & PK_SUGID) && kauth_authorize_generic(l->l_cred,
560 	    KAUTH_GENERIC_ISSUSER, NULL) != 0 && chgproccnt(kauth_cred_getuid(
561 	    l->l_cred), 0) > p->p_rlimit[RLIMIT_NPROC].rlim_cur)
562 		return EAGAIN;
563 
564 	oldlwpflags = l->l_flag & (LW_SA | LW_SA_UPCALL);
565 	if (l->l_flag & LW_SA) {
566 		lwp_lock(l);
567 		l->l_flag &= ~(LW_SA | LW_SA_UPCALL);
568 		lwp_unlock(l);
569 	}
570 
571 	/*
572 	 * Drain existing references and forbid new ones.  The process
573 	 * should be left alone until we're done here.  This is necessary
574 	 * to avoid race conditions - e.g. in ptrace() - that might allow
575 	 * a local user to illicitly obtain elevated privileges.
576 	 */
577 	rw_enter(&p->p_reflock, RW_WRITER);
578 
579 	base_vcp = NULL;
580 	/*
581 	 * Init the namei data to point the file user's program name.
582 	 * This is done here rather than in check_exec(), so that it's
583 	 * possible to override this settings if any of makecmd/probe
584 	 * functions call check_exec() recursively - for example,
585 	 * see exec_script_makecmds().
586 	 */
587 	error = pathbuf_copyin(path, &pb);
588 	if (error) {
589 		DPRINTF(("execve: pathbuf_copyin path @%p %d\n", path, error));
590 		goto clrflg;
591 	}
592 	pathstring = pathbuf_stringcopy_get(pb);
593 	resolvedpathbuf = PNBUF_GET();
594 #ifdef DIAGNOSTIC
595 	strcpy(resolvedpathbuf, "/wrong");
596 #endif
597 
598 	/*
599 	 * initialize the fields of the exec package.
600 	 */
601 	pack.ep_name = path;
602 	pack.ep_kname = pathstring;
603 	pack.ep_resolvedname = resolvedpathbuf;
604 	pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
605 	pack.ep_hdrlen = exec_maxhdrsz;
606 	pack.ep_hdrvalid = 0;
607 	pack.ep_emul_arg = NULL;
608 	pack.ep_vmcmds.evs_cnt = 0;
609 	pack.ep_vmcmds.evs_used = 0;
610 	pack.ep_vap = &attr;
611 	pack.ep_flags = 0;
612 	pack.ep_emul_root = NULL;
613 	pack.ep_interp = NULL;
614 	pack.ep_esch = NULL;
615 	pack.ep_pax_flags = 0;
616 
617 	rw_enter(&exec_lock, RW_READER);
618 
619 	/* see if we can run it. */
620 	if ((error = check_exec(l, &pack, pb)) != 0) {
621 		if (error != ENOENT) {
622 			DPRINTF(("execve: check exec failed %d\n", error));
623 		}
624 		goto freehdr;
625 	}
626 
627 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
628 
629 	/* allocate an argument buffer */
630 	argp = pool_get(&exec_pool, PR_WAITOK);
631 	KASSERT(argp != NULL);
632 	dp = argp;
633 	argc = 0;
634 
635 	/* copy the fake args list, if there's one, freeing it as we go */
636 	if (pack.ep_flags & EXEC_HASARGL) {
637 		tmpfap = pack.ep_fa;
638 		while (tmpfap->fa_arg != NULL) {
639 			const char *cp;
640 
641 			cp = tmpfap->fa_arg;
642 			while (*cp)
643 				*dp++ = *cp++;
644 			*dp++ = '\0';
645 			ktrexecarg(tmpfap->fa_arg, cp - tmpfap->fa_arg);
646 
647 			kmem_free(tmpfap->fa_arg, tmpfap->fa_len);
648 			tmpfap++; argc++;
649 		}
650 		kmem_free(pack.ep_fa, pack.ep_fa_len);
651 		pack.ep_flags &= ~EXEC_HASARGL;
652 	}
653 
654 	/* Now get argv & environment */
655 	if (args == NULL) {
656 		DPRINTF(("execve: null args\n"));
657 		error = EINVAL;
658 		goto bad;
659 	}
660 	/* 'i' will index the argp/envp element to be retrieved */
661 	i = 0;
662 	if (pack.ep_flags & EXEC_SKIPARG)
663 		i++;
664 
665 	while (1) {
666 		len = argp + ARG_MAX - dp;
667 		if ((error = (*fetch_element)(args, i, &sp)) != 0) {
668 			DPRINTF(("execve: fetch_element args %d\n", error));
669 			goto bad;
670 		}
671 		if (!sp)
672 			break;
673 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
674 			DPRINTF(("execve: copyinstr args %d\n", error));
675 			if (error == ENAMETOOLONG)
676 				error = E2BIG;
677 			goto bad;
678 		}
679 		ktrexecarg(dp, len - 1);
680 		dp += len;
681 		i++;
682 		argc++;
683 	}
684 
685 	envc = 0;
686 	/* environment need not be there */
687 	if (envs != NULL) {
688 		i = 0;
689 		while (1) {
690 			len = argp + ARG_MAX - dp;
691 			if ((error = (*fetch_element)(envs, i, &sp)) != 0) {
692 				DPRINTF(("execve: fetch_element env %d\n", error));
693 				goto bad;
694 			}
695 			if (!sp)
696 				break;
697 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
698 				DPRINTF(("execve: copyinstr env %d\n", error));
699 				if (error == ENAMETOOLONG)
700 					error = E2BIG;
701 				goto bad;
702 			}
703 			ktrexecenv(dp, len - 1);
704 			dp += len;
705 			i++;
706 			envc++;
707 		}
708 	}
709 
710 	dp = (char *) ALIGN(dp);
711 
712 	szsigcode = pack.ep_esch->es_emul->e_esigcode -
713 	    pack.ep_esch->es_emul->e_sigcode;
714 
715 #ifdef __MACHINE_STACK_GROWS_UP
716 /* See big comment lower down */
717 #define	RTLD_GAP	32
718 #else
719 #define	RTLD_GAP	0
720 #endif
721 
722 	/* Now check if args & environ fit into new stack */
723 	if (pack.ep_flags & EXEC_32)
724 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
725 		    sizeof(int) + sizeof(int) + dp + RTLD_GAP +
726 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
727 		    - argp;
728 	else
729 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
730 		    sizeof(char *) + sizeof(int) + dp + RTLD_GAP +
731 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
732 		    - argp;
733 
734 #ifdef PAX_ASLR
735 	if (pax_aslr_active(l))
736 		len += (arc4random() % PAGE_SIZE);
737 #endif /* PAX_ASLR */
738 
739 #ifdef STACKLALIGN	/* arm, etc. */
740 	len = STACKALIGN(len);	/* make the stack "safely" aligned */
741 #else
742 	len = ALIGN(len);	/* make the stack "safely" aligned */
743 #endif
744 
745 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
746 		DPRINTF(("execve: stack limit exceeded %zu\n", len));
747 		error = ENOMEM;
748 		goto bad;
749 	}
750 
751 	/* Get rid of other LWPs. */
752 	if (p->p_sa || p->p_nlwps > 1) {
753 		mutex_enter(p->p_lock);
754 		exit_lwps(l);
755 		mutex_exit(p->p_lock);
756 	}
757 	KDASSERT(p->p_nlwps == 1);
758 
759 	/* Destroy any lwpctl info. */
760 	if (p->p_lwpctl != NULL)
761 		lwp_ctl_exit();
762 
763 #ifdef KERN_SA
764 	/* Release any SA state. */
765 	if (p->p_sa)
766 		sa_release(p);
767 #endif /* KERN_SA */
768 
769 	/* Remove POSIX timers */
770 	timers_free(p, TIMERS_POSIX);
771 
772 	/* adjust "active stack depth" for process VSZ */
773 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
774 
775 	/*
776 	 * Do whatever is necessary to prepare the address space
777 	 * for remapping.  Note that this might replace the current
778 	 * vmspace with another!
779 	 */
780 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
781 
782 	/* record proc's vnode, for use by procfs and others */
783         if (p->p_textvp)
784                 vrele(p->p_textvp);
785 	vref(pack.ep_vp);
786 	p->p_textvp = pack.ep_vp;
787 
788 	/* Now map address space */
789 	vm = p->p_vmspace;
790 	vm->vm_taddr = (void *)pack.ep_taddr;
791 	vm->vm_tsize = btoc(pack.ep_tsize);
792 	vm->vm_daddr = (void*)pack.ep_daddr;
793 	vm->vm_dsize = btoc(pack.ep_dsize);
794 	vm->vm_ssize = btoc(pack.ep_ssize);
795 	vm->vm_issize = 0;
796 	vm->vm_maxsaddr = (void *)pack.ep_maxsaddr;
797 	vm->vm_minsaddr = (void *)pack.ep_minsaddr;
798 
799 #ifdef PAX_ASLR
800 	pax_aslr_init(l, vm);
801 #endif /* PAX_ASLR */
802 
803 	/* create the new process's VM space by running the vmcmds */
804 #ifdef DIAGNOSTIC
805 	if (pack.ep_vmcmds.evs_used == 0)
806 		panic("execve: no vmcmds");
807 #endif
808 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
809 		struct exec_vmcmd *vcp;
810 
811 		vcp = &pack.ep_vmcmds.evs_cmds[i];
812 		if (vcp->ev_flags & VMCMD_RELATIVE) {
813 #ifdef DIAGNOSTIC
814 			if (base_vcp == NULL)
815 				panic("execve: relative vmcmd with no base");
816 			if (vcp->ev_flags & VMCMD_BASE)
817 				panic("execve: illegal base & relative vmcmd");
818 #endif
819 			vcp->ev_addr += base_vcp->ev_addr;
820 		}
821 		error = (*vcp->ev_proc)(l, vcp);
822 #ifdef DEBUG_EXEC
823 		if (error) {
824 			size_t j;
825 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
826 			for (j = 0; j <= i; j++)
827 				uprintf(
828 			"vmcmd[%zu] = %#"PRIxVADDR"/%#"PRIxVSIZE" fd@%#"PRIxVSIZE" prot=0%o flags=%d\n",
829 				    j, vp[j].ev_addr, vp[j].ev_len,
830 				    vp[j].ev_offset, vp[j].ev_prot,
831 				    vp[j].ev_flags);
832 		}
833 #endif /* DEBUG_EXEC */
834 		if (vcp->ev_flags & VMCMD_BASE)
835 			base_vcp = vcp;
836 	}
837 
838 	/* free the vmspace-creation commands, and release their references */
839 	kill_vmcmds(&pack.ep_vmcmds);
840 
841 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
842 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
843 	vput(pack.ep_vp);
844 
845 	/* if an error happened, deallocate and punt */
846 	if (error) {
847 		DPRINTF(("execve: vmcmd %zu failed: %d\n", i - 1, error));
848 		goto exec_abort;
849 	}
850 
851 	/* remember information about the process */
852 	arginfo.ps_nargvstr = argc;
853 	arginfo.ps_nenvstr = envc;
854 
855 	/* set command name & other accounting info */
856 	commandname = strrchr(pack.ep_resolvedname, '/');
857 	if (commandname != NULL) {
858 		commandname++;
859 	} else {
860 		commandname = pack.ep_resolvedname;
861 	}
862 	i = min(strlen(commandname), MAXCOMLEN);
863 	(void)memcpy(p->p_comm, commandname, i);
864 	p->p_comm[i] = '\0';
865 
866 	dp = PNBUF_GET();
867 	/*
868 	 * If the path starts with /, we don't need to do any work.
869 	 * This handles the majority of the cases.
870 	 * In the future perhaps we could canonicalize it?
871 	 */
872 	if (pathstring[0] == '/')
873 		(void)strlcpy(pack.ep_path = dp, pathstring, MAXPATHLEN);
874 #ifdef notyet
875 	/*
876 	 * Although this works most of the time [since the entry was just
877 	 * entered in the cache] we don't use it because it theoretically
878 	 * can fail and it is not the cleanest interface, because there
879 	 * could be races. When the namei cache is re-written, this can
880 	 * be changed to use the appropriate function.
881 	 */
882 	else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p)))
883 		pack.ep_path = dp;
884 #endif
885 	else {
886 #ifdef notyet
887 		printf("Cannot get path for pid %d [%s] (error %d)",
888 		    (int)p->p_pid, p->p_comm, error);
889 #endif
890 		pack.ep_path = NULL;
891 		PNBUF_PUT(dp);
892 	}
893 
894 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
895 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
896 		len - (sizeof(struct ps_strings) + szsigcode));
897 
898 #ifdef __MACHINE_STACK_GROWS_UP
899 	/*
900 	 * The copyargs call always copies into lower addresses
901 	 * first, moving towards higher addresses, starting with
902 	 * the stack pointer that we give.  When the stack grows
903 	 * down, this puts argc/argv/envp very shallow on the
904 	 * stack, right at the first user stack pointer.
905 	 * When the stack grows up, the situation is reversed.
906 	 *
907 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
908 	 * function expects to be called with a single pointer to
909 	 * a region that has a few words it can stash values into,
910 	 * followed by argc/argv/envp.  When the stack grows down,
911 	 * it's easy to decrement the stack pointer a little bit to
912 	 * allocate the space for these few words and pass the new
913 	 * stack pointer to _rtld.  When the stack grows up, however,
914 	 * a few words before argc is part of the signal trampoline, XXX
915 	 * so we have a problem.
916 	 *
917 	 * Instead of changing how _rtld works, we take the easy way
918 	 * out and steal 32 bytes before we call copyargs.
919 	 * This extra space was allowed for when 'len' was calculated.
920 	 */
921 	stack += RTLD_GAP;
922 #endif /* __MACHINE_STACK_GROWS_UP */
923 
924 	/* Now copy argc, args & environ to new stack */
925 	error = (*pack.ep_esch->es_copyargs)(l, &pack, &arginfo, &stack, argp);
926 	if (pack.ep_path) {
927 		PNBUF_PUT(pack.ep_path);
928 		pack.ep_path = NULL;
929 	}
930 	if (error) {
931 		DPRINTF(("execve: copyargs failed %d\n", error));
932 		goto exec_abort;
933 	}
934 	/* Move the stack back to original point */
935 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
936 
937 	/* fill process ps_strings info */
938 	p->p_psstr = (struct ps_strings *)
939 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
940 	    sizeof(struct ps_strings));
941 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
942 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
943 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
944 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
945 
946 	/* copy out the process's ps_strings structure */
947 	if ((error = copyout(aip, (char *)p->p_psstr,
948 	    sizeof(arginfo))) != 0) {
949 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
950 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
951 		goto exec_abort;
952 	}
953 
954 	fd_closeexec();		/* handle close on exec */
955 	execsigs(p);		/* reset catched signals */
956 
957 	l->l_ctxlink = NULL;	/* reset ucontext link */
958 
959 
960 	p->p_acflag &= ~AFORK;
961 	mutex_enter(p->p_lock);
962 	p->p_flag |= PK_EXEC;
963 	mutex_exit(p->p_lock);
964 
965 	/*
966 	 * Stop profiling.
967 	 */
968 	if ((p->p_stflag & PST_PROFIL) != 0) {
969 		mutex_spin_enter(&p->p_stmutex);
970 		stopprofclock(p);
971 		mutex_spin_exit(&p->p_stmutex);
972 	}
973 
974 	/*
975 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
976 	 * exited and exec()/exit() are the only places it will be cleared.
977 	 */
978 	if ((p->p_lflag & PL_PPWAIT) != 0) {
979 		mutex_enter(proc_lock);
980 		p->p_lflag &= ~PL_PPWAIT;
981 		cv_broadcast(&p->p_pptr->p_waitcv);
982 		mutex_exit(proc_lock);
983 	}
984 
985 	/*
986 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
987 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
988 	 * out additional references on the process for the moment.
989 	 */
990 	if ((p->p_slflag & PSL_TRACED) == 0 &&
991 
992 	    (((attr.va_mode & S_ISUID) != 0 &&
993 	      kauth_cred_geteuid(l->l_cred) != attr.va_uid) ||
994 
995 	     ((attr.va_mode & S_ISGID) != 0 &&
996 	      kauth_cred_getegid(l->l_cred) != attr.va_gid))) {
997 		/*
998 		 * Mark the process as SUGID before we do
999 		 * anything that might block.
1000 		 */
1001 		proc_crmod_enter();
1002 		proc_crmod_leave(NULL, NULL, true);
1003 
1004 		/* Make sure file descriptors 0..2 are in use. */
1005 		if ((error = fd_checkstd()) != 0) {
1006 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
1007 			goto exec_abort;
1008 		}
1009 
1010 		/*
1011 		 * Copy the credential so other references don't see our
1012 		 * changes.
1013 		 */
1014 		l->l_cred = kauth_cred_copy(l->l_cred);
1015 #ifdef KTRACE
1016 		/*
1017 		 * If the persistent trace flag isn't set, turn off.
1018 		 */
1019 		if (p->p_tracep) {
1020 			mutex_enter(&ktrace_lock);
1021 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1022 				ktrderef(p);
1023 			mutex_exit(&ktrace_lock);
1024 		}
1025 #endif
1026 		if (attr.va_mode & S_ISUID)
1027 			kauth_cred_seteuid(l->l_cred, attr.va_uid);
1028 		if (attr.va_mode & S_ISGID)
1029 			kauth_cred_setegid(l->l_cred, attr.va_gid);
1030 	} else {
1031 		if (kauth_cred_geteuid(l->l_cred) ==
1032 		    kauth_cred_getuid(l->l_cred) &&
1033 		    kauth_cred_getegid(l->l_cred) ==
1034 		    kauth_cred_getgid(l->l_cred))
1035 			p->p_flag &= ~PK_SUGID;
1036 	}
1037 
1038 	/*
1039 	 * Copy the credential so other references don't see our changes.
1040 	 * Test to see if this is necessary first, since in the common case
1041 	 * we won't need a private reference.
1042 	 */
1043 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1044 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1045 		l->l_cred = kauth_cred_copy(l->l_cred);
1046 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1047 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1048 	}
1049 
1050 	/* Update the master credentials. */
1051 	if (l->l_cred != p->p_cred) {
1052 		kauth_cred_t ocred;
1053 
1054 		kauth_cred_hold(l->l_cred);
1055 		mutex_enter(p->p_lock);
1056 		ocred = p->p_cred;
1057 		p->p_cred = l->l_cred;
1058 		mutex_exit(p->p_lock);
1059 		kauth_cred_free(ocred);
1060 	}
1061 
1062 #if defined(__HAVE_RAS)
1063 	/*
1064 	 * Remove all RASs from the address space.
1065 	 */
1066 	ras_purgeall();
1067 #endif
1068 
1069 	doexechooks(p);
1070 
1071 	/* setup new registers and do misc. setup. */
1072 	(*pack.ep_esch->es_emul->e_setregs)(l, &pack, (vaddr_t)stack);
1073 	if (pack.ep_esch->es_setregs)
1074 		(*pack.ep_esch->es_setregs)(l, &pack, (vaddr_t)stack);
1075 
1076 	/* map the process's signal trampoline code */
1077 	if ((error = exec_sigcode_map(p, pack.ep_esch->es_emul)) != 0) {
1078 		DPRINTF(("execve: map sigcode failed %d\n", error));
1079 		goto exec_abort;
1080 	}
1081 
1082 	pool_put(&exec_pool, argp);
1083 
1084 	/* notify others that we exec'd */
1085 	KNOTE(&p->p_klist, NOTE_EXEC);
1086 
1087 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1088 
1089 	SDT_PROBE(proc,,,exec_success, path, 0, 0, 0, 0);
1090 
1091 	/* The emulation root will usually have been found when we looked
1092 	 * for the elf interpreter (or similar), if not look now. */
1093 	if (pack.ep_esch->es_emul->e_path != NULL && pack.ep_emul_root == NULL)
1094 		emul_find_root(l, &pack);
1095 
1096 	/* Any old emulation root got removed by fdcloseexec */
1097 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1098 	p->p_cwdi->cwdi_edir = pack.ep_emul_root;
1099 	rw_exit(&p->p_cwdi->cwdi_lock);
1100 	pack.ep_emul_root = NULL;
1101 	if (pack.ep_interp != NULL)
1102 		vrele(pack.ep_interp);
1103 
1104 	/*
1105 	 * Call emulation specific exec hook. This can setup per-process
1106 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1107 	 *
1108 	 * If we are executing process of different emulation than the
1109 	 * original forked process, call e_proc_exit() of the old emulation
1110 	 * first, then e_proc_exec() of new emulation. If the emulation is
1111 	 * same, the exec hook code should deallocate any old emulation
1112 	 * resources held previously by this process.
1113 	 */
1114 	if (p->p_emul && p->p_emul->e_proc_exit
1115 	    && p->p_emul != pack.ep_esch->es_emul)
1116 		(*p->p_emul->e_proc_exit)(p);
1117 
1118 	/*
1119 	 * This is now LWP 1.
1120 	 */
1121 	mutex_enter(p->p_lock);
1122 	p->p_nlwpid = 1;
1123 	l->l_lid = 1;
1124 	mutex_exit(p->p_lock);
1125 
1126 	/*
1127 	 * Call exec hook. Emulation code may NOT store reference to anything
1128 	 * from &pack.
1129 	 */
1130         if (pack.ep_esch->es_emul->e_proc_exec)
1131                 (*pack.ep_esch->es_emul->e_proc_exec)(p, &pack);
1132 
1133 	/* update p_emul, the old value is no longer needed */
1134 	p->p_emul = pack.ep_esch->es_emul;
1135 
1136 	/* ...and the same for p_execsw */
1137 	p->p_execsw = pack.ep_esch;
1138 
1139 #ifdef __HAVE_SYSCALL_INTERN
1140 	(*p->p_emul->e_syscall_intern)(p);
1141 #endif
1142 	ktremul();
1143 
1144 	/* Allow new references from the debugger/procfs. */
1145 	rw_exit(&p->p_reflock);
1146 	rw_exit(&exec_lock);
1147 
1148 	mutex_enter(proc_lock);
1149 
1150 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1151 		KSI_INIT_EMPTY(&ksi);
1152 		ksi.ksi_signo = SIGTRAP;
1153 		ksi.ksi_lid = l->l_lid;
1154 		kpsignal(p, &ksi, NULL);
1155 	}
1156 
1157 	if (p->p_sflag & PS_STOPEXEC) {
1158 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1159 		p->p_pptr->p_nstopchild++;
1160 		p->p_pptr->p_waited = 0;
1161 		mutex_enter(p->p_lock);
1162 		ksiginfo_queue_init(&kq);
1163 		sigclearall(p, &contsigmask, &kq);
1164 		lwp_lock(l);
1165 		l->l_stat = LSSTOP;
1166 		p->p_stat = SSTOP;
1167 		p->p_nrlwps--;
1168 		lwp_unlock(l);
1169 		mutex_exit(p->p_lock);
1170 		mutex_exit(proc_lock);
1171 		lwp_lock(l);
1172 		mi_switch(l);
1173 		ksiginfo_queue_drain(&kq);
1174 		KERNEL_LOCK(l->l_biglocks, l);
1175 	} else {
1176 		mutex_exit(proc_lock);
1177 	}
1178 
1179 	pathbuf_stringcopy_put(pb, pathstring);
1180 	pathbuf_destroy(pb);
1181 	PNBUF_PUT(resolvedpathbuf);
1182 	return (EJUSTRETURN);
1183 
1184  bad:
1185 	/* free the vmspace-creation commands, and release their references */
1186 	kill_vmcmds(&pack.ep_vmcmds);
1187 	/* kill any opened file descriptor, if necessary */
1188 	if (pack.ep_flags & EXEC_HASFD) {
1189 		pack.ep_flags &= ~EXEC_HASFD;
1190 		fd_close(pack.ep_fd);
1191 	}
1192 	/* close and put the exec'd file */
1193 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
1194 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
1195 	vput(pack.ep_vp);
1196 	pool_put(&exec_pool, argp);
1197 
1198  freehdr:
1199 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1200 	if (pack.ep_emul_root != NULL)
1201 		vrele(pack.ep_emul_root);
1202 	if (pack.ep_interp != NULL)
1203 		vrele(pack.ep_interp);
1204 
1205 	rw_exit(&exec_lock);
1206 
1207 	pathbuf_stringcopy_put(pb, pathstring);
1208 	pathbuf_destroy(pb);
1209 	PNBUF_PUT(resolvedpathbuf);
1210 
1211  clrflg:
1212 	lwp_lock(l);
1213 	l->l_flag |= oldlwpflags;
1214 	lwp_unlock(l);
1215 	rw_exit(&p->p_reflock);
1216 
1217 	if (modgen != module_gen && error == ENOEXEC) {
1218 		modgen = module_gen;
1219 		exec_autoload();
1220 		goto retry;
1221 	}
1222 
1223 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
1224 	return error;
1225 
1226  exec_abort:
1227 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
1228 	rw_exit(&p->p_reflock);
1229 	rw_exit(&exec_lock);
1230 
1231 	pathbuf_stringcopy_put(pb, pathstring);
1232 	pathbuf_destroy(pb);
1233 	PNBUF_PUT(resolvedpathbuf);
1234 
1235 	/*
1236 	 * the old process doesn't exist anymore.  exit gracefully.
1237 	 * get rid of the (new) address space we have created, if any, get rid
1238 	 * of our namei data and vnode, and exit noting failure
1239 	 */
1240 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1241 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1242 	if (pack.ep_emul_arg)
1243 		free(pack.ep_emul_arg, M_TEMP);
1244 	pool_put(&exec_pool, argp);
1245 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1246 	if (pack.ep_emul_root != NULL)
1247 		vrele(pack.ep_emul_root);
1248 	if (pack.ep_interp != NULL)
1249 		vrele(pack.ep_interp);
1250 
1251 	/* Acquire the sched-state mutex (exit1() will release it). */
1252 	mutex_enter(p->p_lock);
1253 	exit1(l, W_EXITCODE(error, SIGABRT));
1254 
1255 	/* NOTREACHED */
1256 	return 0;
1257 }
1258 
1259 
1260 int
1261 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1262     char **stackp, void *argp)
1263 {
1264 	char	**cpp, *dp, *sp;
1265 	size_t	len;
1266 	void	*nullp;
1267 	long	argc, envc;
1268 	int	error;
1269 
1270 	cpp = (char **)*stackp;
1271 	nullp = NULL;
1272 	argc = arginfo->ps_nargvstr;
1273 	envc = arginfo->ps_nenvstr;
1274 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1275 		DPRINTF(("copyargs:%d copyout @%p %zu\n", __LINE__, cpp-1, sizeof(argc)));
1276 		return error;
1277 	}
1278 
1279 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen);
1280 	sp = argp;
1281 
1282 	/* XXX don't copy them out, remap them! */
1283 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1284 
1285 	for (; --argc >= 0; sp += len, dp += len) {
1286 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1287 			DPRINTF(("copyargs:%d copyout @%p %zu\n", __LINE__, cpp-1, sizeof(dp)));
1288 			return error;
1289 		}
1290 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1291 			DPRINTF(("copyargs:%d copyoutstr @%p %u\n", __LINE__, dp, ARG_MAX));
1292 			return error;
1293 		}
1294 	}
1295 
1296 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1297 		DPRINTF(("copyargs:%d copyout @%p %zu\n", __LINE__, cpp-1, sizeof(nullp)));
1298 		return error;
1299 	}
1300 
1301 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1302 
1303 	for (; --envc >= 0; sp += len, dp += len) {
1304 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1305 			DPRINTF(("copyargs:%d copyout @%p %zu\n", __LINE__, cpp-1, sizeof(dp)));
1306 			return error;
1307 		}
1308 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1309 			DPRINTF(("copyargs:%d copyoutstr @%p %u\n", __LINE__, dp, ARG_MAX));
1310 			return error;
1311 		}
1312 	}
1313 
1314 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1315 		DPRINTF(("copyargs:%d copyout @%p %zu\n", __LINE__, cpp-1, sizeof(nullp)));
1316 		return error;
1317 	}
1318 
1319 	*stackp = (char *)cpp;
1320 	return 0;
1321 }
1322 
1323 
1324 /*
1325  * Add execsw[] entries.
1326  */
1327 int
1328 exec_add(struct execsw *esp, int count)
1329 {
1330 	struct exec_entry	*it;
1331 	int			i;
1332 
1333 	if (count == 0) {
1334 		return 0;
1335 	}
1336 
1337 	/* Check for duplicates. */
1338 	rw_enter(&exec_lock, RW_WRITER);
1339 	for (i = 0; i < count; i++) {
1340 		LIST_FOREACH(it, &ex_head, ex_list) {
1341 			/* assume unique (makecmds, probe_func, emulation) */
1342 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1343 			    it->ex_sw->u.elf_probe_func ==
1344 			    esp[i].u.elf_probe_func &&
1345 			    it->ex_sw->es_emul == esp[i].es_emul) {
1346 				rw_exit(&exec_lock);
1347 				return EEXIST;
1348 			}
1349 		}
1350 	}
1351 
1352 	/* Allocate new entries. */
1353 	for (i = 0; i < count; i++) {
1354 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1355 		it->ex_sw = &esp[i];
1356 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1357 	}
1358 
1359 	/* update execsw[] */
1360 	exec_init(0);
1361 	rw_exit(&exec_lock);
1362 	return 0;
1363 }
1364 
1365 /*
1366  * Remove execsw[] entry.
1367  */
1368 int
1369 exec_remove(struct execsw *esp, int count)
1370 {
1371 	struct exec_entry	*it, *next;
1372 	int			i;
1373 	const struct proclist_desc *pd;
1374 	proc_t			*p;
1375 
1376 	if (count == 0) {
1377 		return 0;
1378 	}
1379 
1380 	/* Abort if any are busy. */
1381 	rw_enter(&exec_lock, RW_WRITER);
1382 	for (i = 0; i < count; i++) {
1383 		mutex_enter(proc_lock);
1384 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1385 			PROCLIST_FOREACH(p, pd->pd_list) {
1386 				if (p->p_execsw == &esp[i]) {
1387 					mutex_exit(proc_lock);
1388 					rw_exit(&exec_lock);
1389 					return EBUSY;
1390 				}
1391 			}
1392 		}
1393 		mutex_exit(proc_lock);
1394 	}
1395 
1396 	/* None are busy, so remove them all. */
1397 	for (i = 0; i < count; i++) {
1398 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1399 			next = LIST_NEXT(it, ex_list);
1400 			if (it->ex_sw == &esp[i]) {
1401 				LIST_REMOVE(it, ex_list);
1402 				kmem_free(it, sizeof(*it));
1403 				break;
1404 			}
1405 		}
1406 	}
1407 
1408 	/* update execsw[] */
1409 	exec_init(0);
1410 	rw_exit(&exec_lock);
1411 	return 0;
1412 }
1413 
1414 /*
1415  * Initialize exec structures. If init_boot is true, also does necessary
1416  * one-time initialization (it's called from main() that way).
1417  * Once system is multiuser, this should be called with exec_lock held,
1418  * i.e. via exec_{add|remove}().
1419  */
1420 int
1421 exec_init(int init_boot)
1422 {
1423 	const struct execsw 	**sw;
1424 	struct exec_entry	*ex;
1425 	SLIST_HEAD(,exec_entry)	first;
1426 	SLIST_HEAD(,exec_entry)	any;
1427 	SLIST_HEAD(,exec_entry)	last;
1428 	int			i, sz;
1429 
1430 	if (init_boot) {
1431 		/* do one-time initializations */
1432 		rw_init(&exec_lock);
1433 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1434 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1435 		    "execargs", &exec_palloc, IPL_NONE);
1436 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1437 	} else {
1438 		KASSERT(rw_write_held(&exec_lock));
1439 	}
1440 
1441 	/* Sort each entry onto the appropriate queue. */
1442 	SLIST_INIT(&first);
1443 	SLIST_INIT(&any);
1444 	SLIST_INIT(&last);
1445 	sz = 0;
1446 	LIST_FOREACH(ex, &ex_head, ex_list) {
1447 		switch(ex->ex_sw->es_prio) {
1448 		case EXECSW_PRIO_FIRST:
1449 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1450 			break;
1451 		case EXECSW_PRIO_ANY:
1452 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1453 			break;
1454 		case EXECSW_PRIO_LAST:
1455 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1456 			break;
1457 		default:
1458 			panic("exec_init");
1459 			break;
1460 		}
1461 		sz++;
1462 	}
1463 
1464 	/*
1465 	 * Create new execsw[].  Ensure we do not try a zero-sized
1466 	 * allocation.
1467 	 */
1468 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1469 	i = 0;
1470 	SLIST_FOREACH(ex, &first, ex_slist) {
1471 		sw[i++] = ex->ex_sw;
1472 	}
1473 	SLIST_FOREACH(ex, &any, ex_slist) {
1474 		sw[i++] = ex->ex_sw;
1475 	}
1476 	SLIST_FOREACH(ex, &last, ex_slist) {
1477 		sw[i++] = ex->ex_sw;
1478 	}
1479 
1480 	/* Replace old execsw[] and free used memory. */
1481 	if (execsw != NULL) {
1482 		kmem_free(__UNCONST(execsw),
1483 		    nexecs * sizeof(struct execsw *) + 1);
1484 	}
1485 	execsw = sw;
1486 	nexecs = sz;
1487 
1488 	/* Figure out the maximum size of an exec header. */
1489 	exec_maxhdrsz = sizeof(int);
1490 	for (i = 0; i < nexecs; i++) {
1491 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1492 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1493 	}
1494 
1495 	return 0;
1496 }
1497 
1498 static int
1499 exec_sigcode_map(struct proc *p, const struct emul *e)
1500 {
1501 	vaddr_t va;
1502 	vsize_t sz;
1503 	int error;
1504 	struct uvm_object *uobj;
1505 
1506 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1507 
1508 	if (e->e_sigobject == NULL || sz == 0) {
1509 		return 0;
1510 	}
1511 
1512 	/*
1513 	 * If we don't have a sigobject for this emulation, create one.
1514 	 *
1515 	 * sigobject is an anonymous memory object (just like SYSV shared
1516 	 * memory) that we keep a permanent reference to and that we map
1517 	 * in all processes that need this sigcode. The creation is simple,
1518 	 * we create an object, add a permanent reference to it, map it in
1519 	 * kernel space, copy out the sigcode to it and unmap it.
1520 	 * We map it with PROT_READ|PROT_EXEC into the process just
1521 	 * the way sys_mmap() would map it.
1522 	 */
1523 
1524 	uobj = *e->e_sigobject;
1525 	if (uobj == NULL) {
1526 		mutex_enter(&sigobject_lock);
1527 		if ((uobj = *e->e_sigobject) == NULL) {
1528 			uobj = uao_create(sz, 0);
1529 			(*uobj->pgops->pgo_reference)(uobj);
1530 			va = vm_map_min(kernel_map);
1531 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1532 			    uobj, 0, 0,
1533 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1534 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1535 				printf("kernel mapping failed %d\n", error);
1536 				(*uobj->pgops->pgo_detach)(uobj);
1537 				mutex_exit(&sigobject_lock);
1538 				return (error);
1539 			}
1540 			memcpy((void *)va, e->e_sigcode, sz);
1541 #ifdef PMAP_NEED_PROCWR
1542 			pmap_procwr(&proc0, va, sz);
1543 #endif
1544 			uvm_unmap(kernel_map, va, va + round_page(sz));
1545 			*e->e_sigobject = uobj;
1546 		}
1547 		mutex_exit(&sigobject_lock);
1548 	}
1549 
1550 	/* Just a hint to uvm_map where to put it. */
1551 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1552 	    round_page(sz));
1553 
1554 #ifdef __alpha__
1555 	/*
1556 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1557 	 * which causes the above calculation to put the sigcode at
1558 	 * an invalid address.  Put it just below the text instead.
1559 	 */
1560 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1561 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1562 	}
1563 #endif
1564 
1565 	(*uobj->pgops->pgo_reference)(uobj);
1566 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1567 			uobj, 0, 0,
1568 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1569 				    UVM_ADV_RANDOM, 0));
1570 	if (error) {
1571 		DPRINTF(("exec_sigcode_map:%d map %p "
1572 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1573 		    __LINE__, &p->p_vmspace->vm_map, round_page(sz), va,
1574 		    error));
1575 		(*uobj->pgops->pgo_detach)(uobj);
1576 		return (error);
1577 	}
1578 	p->p_sigctx.ps_sigcode = (void *)va;
1579 	return (0);
1580 }
1581