xref: /netbsd-src/sys/kern/kern_exec.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /*	$NetBSD: kern_exec.c,v 1.300 2010/08/21 13:19:39 pgoyette 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 1.300 2010/08/21 13:19:39 pgoyette 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) uprintf 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, const char *kpath)
282 {
283 	int		error, i;
284 	struct vnode	*vp;
285 	struct nameidata nd;
286 	size_t		resid;
287 
288 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | SAVENAME | TRYEMULROOT,
289 	       UIO_SYSSPACE, kpath);
290 
291 	/* first get the vnode */
292 	if ((error = namei(&nd)) != 0)
293 		return error;
294 	epp->ep_vp = vp = nd.ni_vp;
295 	/* this cannot overflow as both are size PATH_MAX */
296 	strcpy(epp->ep_resolvedname, nd.ni_cnd.cn_pnbuf);
297 
298 	/* dump this right away */
299 #ifdef DIAGNOSTIC
300 	/* paranoia (take this out once namei stuff stabilizes) */
301 	memset(nd.ni_cnd.cn_pnbuf, '~', PATH_MAX);
302 #endif
303 	PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
304 
305 	/* check access and type */
306 	if (vp->v_type != VREG) {
307 		error = EACCES;
308 		goto bad1;
309 	}
310 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
311 		goto bad1;
312 
313 	/* get attributes */
314 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
315 		goto bad1;
316 
317 	/* Check mount point */
318 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
319 		error = EACCES;
320 		goto bad1;
321 	}
322 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
323 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
324 
325 	/* try to open it */
326 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
327 		goto bad1;
328 
329 	/* unlock vp, since we need it unlocked from here on out. */
330 	VOP_UNLOCK(vp);
331 
332 #if NVERIEXEC > 0
333 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
334 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
335 	    NULL);
336 	if (error)
337 		goto bad2;
338 #endif /* NVERIEXEC > 0 */
339 
340 #ifdef PAX_SEGVGUARD
341 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
342 	if (error)
343 		goto bad2;
344 #endif /* PAX_SEGVGUARD */
345 
346 	/* now we have the file, get the exec header */
347 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
348 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
349 	if (error)
350 		goto bad2;
351 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
352 
353 	/*
354 	 * Set up default address space limits.  Can be overridden
355 	 * by individual exec packages.
356 	 *
357 	 * XXX probably should be all done in the exec packages.
358 	 */
359 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
360 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
361 	/*
362 	 * set up the vmcmds for creation of the process
363 	 * address space
364 	 */
365 	error = ENOEXEC;
366 	for (i = 0; i < nexecs; i++) {
367 		int newerror;
368 
369 		epp->ep_esch = execsw[i];
370 		newerror = (*execsw[i]->es_makecmds)(l, epp);
371 
372 		if (!newerror) {
373 			/* Seems ok: check that entry point is sane */
374 			if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
375 				error = ENOEXEC;
376 				break;
377 			}
378 
379 			/* check limits */
380 			if ((epp->ep_tsize > MAXTSIZ) ||
381 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
382 						    [RLIMIT_DATA].rlim_cur)) {
383 				error = ENOMEM;
384 				break;
385 			}
386 			return 0;
387 		}
388 
389 		if (epp->ep_emul_root != NULL) {
390 			vrele(epp->ep_emul_root);
391 			epp->ep_emul_root = NULL;
392 		}
393 		if (epp->ep_interp != NULL) {
394 			vrele(epp->ep_interp);
395 			epp->ep_interp = NULL;
396 		}
397 
398 		/* make sure the first "interesting" error code is saved. */
399 		if (error == ENOEXEC)
400 			error = newerror;
401 
402 		if (epp->ep_flags & EXEC_DESTR)
403 			/* Error from "#!" code, tidied up by recursive call */
404 			return error;
405 	}
406 
407 	/* not found, error */
408 
409 	/*
410 	 * free any vmspace-creation commands,
411 	 * and release their references
412 	 */
413 	kill_vmcmds(&epp->ep_vmcmds);
414 
415 bad2:
416 	/*
417 	 * close and release the vnode, restore the old one, free the
418 	 * pathname buf, and punt.
419 	 */
420 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
421 	VOP_CLOSE(vp, FREAD, l->l_cred);
422 	vput(vp);
423 	return error;
424 
425 bad1:
426 	/*
427 	 * free the namei pathname buffer, and put the vnode
428 	 * (which we don't yet have open).
429 	 */
430 	vput(vp);				/* was still locked */
431 	return error;
432 }
433 
434 #ifdef __MACHINE_STACK_GROWS_UP
435 #define STACK_PTHREADSPACE NBPG
436 #else
437 #define STACK_PTHREADSPACE 0
438 #endif
439 
440 static int
441 execve_fetch_element(char * const *array, size_t index, char **value)
442 {
443 	return copyin(array + index, value, sizeof(*value));
444 }
445 
446 /*
447  * exec system call
448  */
449 /* ARGSUSED */
450 int
451 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
452 {
453 	/* {
454 		syscallarg(const char *)	path;
455 		syscallarg(char * const *)	argp;
456 		syscallarg(char * const *)	envp;
457 	} */
458 
459 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
460 	    SCARG(uap, envp), execve_fetch_element);
461 }
462 
463 /*
464  * Load modules to try and execute an image that we do not understand.
465  * If no execsw entries are present, we load those likely to be needed
466  * in order to run native images only.  Otherwise, we autoload all
467  * possible modules that could let us run the binary.  XXX lame
468  */
469 static void
470 exec_autoload(void)
471 {
472 #ifdef MODULAR
473 	static const char * const native[] = {
474 		"exec_elf32",
475 		"exec_elf64",
476 		"exec_script",
477 		NULL
478 	};
479 	static const char * const compat[] = {
480 		"exec_elf32",
481 		"exec_elf64",
482 		"exec_script",
483 		"exec_aout",
484 		"exec_coff",
485 		"exec_ecoff",
486 		"compat_aoutm68k",
487 		"compat_freebsd",
488 		"compat_ibcs2",
489 		"compat_irix",
490 		"compat_linux",
491 		"compat_linux32",
492 		"compat_netbsd32",
493 		"compat_sunos",
494 		"compat_sunos32",
495 		"compat_svr4",
496 		"compat_svr4_32",
497 		"compat_ultrix",
498 		NULL
499 	};
500 	char const * const *list;
501 	int i;
502 
503 	list = (nexecs == 0 ? native : compat);
504 	for (i = 0; list[i] != NULL; i++) {
505 		if (module_autoload(list[i], MODULE_CLASS_MISC) != 0) {
506 		    	continue;
507 		}
508 	   	yield();
509 	}
510 #endif
511 }
512 
513 int
514 execve1(struct lwp *l, const char *path, char * const *args,
515     char * const *envs, execve_fetch_element_t fetch_element)
516 {
517 	int			error;
518 	struct exec_package	pack;
519 	struct vattr		attr;
520 	struct proc		*p;
521 	char			*argp;
522 	char			*dp, *sp;
523 	long			argc, envc;
524 	size_t			i, len;
525 	char			*stack;
526 	struct ps_strings	arginfo;
527 	struct ps_strings	*aip = &arginfo;
528 	struct vmspace		*vm;
529 	struct exec_fakearg	*tmpfap;
530 	int			szsigcode;
531 	struct exec_vmcmd	*base_vcp;
532 	int			oldlwpflags;
533 	ksiginfo_t		ksi;
534 	ksiginfoq_t		kq;
535 	char			*pathbuf;
536 	char			*resolvedpathbuf;
537 	const char		*commandname;
538 	u_int			modgen;
539 
540 	p = l->l_proc;
541  	modgen = 0;
542 
543 	SDT_PROBE(proc,,,exec, path, 0, 0, 0, 0);
544 
545 	/*
546 	 * Check if we have exceeded our number of processes limit.
547 	 * This is so that we handle the case where a root daemon
548 	 * forked, ran setuid to become the desired user and is trying
549 	 * to exec. The obvious place to do the reference counting check
550 	 * is setuid(), but we don't do the reference counting check there
551 	 * like other OS's do because then all the programs that use setuid()
552 	 * must be modified to check the return code of setuid() and exit().
553 	 * It is dangerous to make setuid() fail, because it fails open and
554 	 * the program will continue to run as root. If we make it succeed
555 	 * and return an error code, again we are not enforcing the limit.
556 	 * The best place to enforce the limit is here, when the process tries
557 	 * to execute a new image, because eventually the process will need
558 	 * to call exec in order to do something useful.
559 	 */
560  retry:
561 	if ((p->p_flag & PK_SUGID) && kauth_authorize_generic(l->l_cred,
562 	    KAUTH_GENERIC_ISSUSER, NULL) != 0 && chgproccnt(kauth_cred_getuid(
563 	    l->l_cred), 0) > p->p_rlimit[RLIMIT_NPROC].rlim_cur)
564 		return EAGAIN;
565 
566 	oldlwpflags = l->l_flag & (LW_SA | LW_SA_UPCALL);
567 	if (l->l_flag & LW_SA) {
568 		lwp_lock(l);
569 		l->l_flag &= ~(LW_SA | LW_SA_UPCALL);
570 		lwp_unlock(l);
571 	}
572 
573 	/*
574 	 * Drain existing references and forbid new ones.  The process
575 	 * should be left alone until we're done here.  This is necessary
576 	 * to avoid race conditions - e.g. in ptrace() - that might allow
577 	 * a local user to illicitly obtain elevated privileges.
578 	 */
579 	rw_enter(&p->p_reflock, RW_WRITER);
580 
581 	base_vcp = NULL;
582 	/*
583 	 * Init the namei data to point the file user's program name.
584 	 * This is done here rather than in check_exec(), so that it's
585 	 * possible to override this settings if any of makecmd/probe
586 	 * functions call check_exec() recursively - for example,
587 	 * see exec_script_makecmds().
588 	 */
589 	pathbuf = PNBUF_GET();
590 	error = copyinstr(path, pathbuf, MAXPATHLEN, NULL);
591 	if (error) {
592 		DPRINTF(("execve: copyinstr path %d", error));
593 		goto clrflg;
594 	}
595 	resolvedpathbuf = PNBUF_GET();
596 #ifdef DIAGNOSTIC
597 	strcpy(resolvedpathbuf, "/wrong");
598 #endif
599 
600 	/*
601 	 * initialize the fields of the exec package.
602 	 */
603 	pack.ep_name = path;
604 	pack.ep_kname = pathbuf;
605 	pack.ep_resolvedname = resolvedpathbuf;
606 	pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
607 	pack.ep_hdrlen = exec_maxhdrsz;
608 	pack.ep_hdrvalid = 0;
609 	pack.ep_emul_arg = NULL;
610 	pack.ep_vmcmds.evs_cnt = 0;
611 	pack.ep_vmcmds.evs_used = 0;
612 	pack.ep_vap = &attr;
613 	pack.ep_flags = 0;
614 	pack.ep_emul_root = NULL;
615 	pack.ep_interp = NULL;
616 	pack.ep_esch = NULL;
617 	pack.ep_pax_flags = 0;
618 
619 	rw_enter(&exec_lock, RW_READER);
620 
621 	/* see if we can run it. */
622 	if ((error = check_exec(l, &pack, pathbuf)) != 0) {
623 		if (error != ENOENT) {
624 			DPRINTF(("execve: check exec failed %d\n", error));
625 		}
626 		goto freehdr;
627 	}
628 
629 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
630 
631 	/* allocate an argument buffer */
632 	argp = pool_get(&exec_pool, PR_WAITOK);
633 	KASSERT(argp != NULL);
634 	dp = argp;
635 	argc = 0;
636 
637 	/* copy the fake args list, if there's one, freeing it as we go */
638 	if (pack.ep_flags & EXEC_HASARGL) {
639 		tmpfap = pack.ep_fa;
640 		while (tmpfap->fa_arg != NULL) {
641 			const char *cp;
642 
643 			cp = tmpfap->fa_arg;
644 			while (*cp)
645 				*dp++ = *cp++;
646 			*dp++ = '\0';
647 			ktrexecarg(tmpfap->fa_arg, cp - tmpfap->fa_arg);
648 
649 			kmem_free(tmpfap->fa_arg, tmpfap->fa_len);
650 			tmpfap++; argc++;
651 		}
652 		kmem_free(pack.ep_fa, pack.ep_fa_len);
653 		pack.ep_flags &= ~EXEC_HASARGL;
654 	}
655 
656 	/* Now get argv & environment */
657 	if (args == NULL) {
658 		DPRINTF(("execve: null args\n"));
659 		error = EINVAL;
660 		goto bad;
661 	}
662 	/* 'i' will index the argp/envp element to be retrieved */
663 	i = 0;
664 	if (pack.ep_flags & EXEC_SKIPARG)
665 		i++;
666 
667 	while (1) {
668 		len = argp + ARG_MAX - dp;
669 		if ((error = (*fetch_element)(args, i, &sp)) != 0) {
670 			DPRINTF(("execve: fetch_element args %d\n", error));
671 			goto bad;
672 		}
673 		if (!sp)
674 			break;
675 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
676 			DPRINTF(("execve: copyinstr args %d\n", error));
677 			if (error == ENAMETOOLONG)
678 				error = E2BIG;
679 			goto bad;
680 		}
681 		ktrexecarg(dp, len - 1);
682 		dp += len;
683 		i++;
684 		argc++;
685 	}
686 
687 	envc = 0;
688 	/* environment need not be there */
689 	if (envs != NULL) {
690 		i = 0;
691 		while (1) {
692 			len = argp + ARG_MAX - dp;
693 			if ((error = (*fetch_element)(envs, i, &sp)) != 0) {
694 				DPRINTF(("execve: fetch_element env %d\n", error));
695 				goto bad;
696 			}
697 			if (!sp)
698 				break;
699 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
700 				DPRINTF(("execve: copyinstr env %d\n", error));
701 				if (error == ENAMETOOLONG)
702 					error = E2BIG;
703 				goto bad;
704 			}
705 			ktrexecenv(dp, len - 1);
706 			dp += len;
707 			i++;
708 			envc++;
709 		}
710 	}
711 
712 	dp = (char *) ALIGN(dp);
713 
714 	szsigcode = pack.ep_esch->es_emul->e_esigcode -
715 	    pack.ep_esch->es_emul->e_sigcode;
716 
717 #ifdef __MACHINE_STACK_GROWS_UP
718 /* See big comment lower down */
719 #define	RTLD_GAP	32
720 #else
721 #define	RTLD_GAP	0
722 #endif
723 
724 	/* Now check if args & environ fit into new stack */
725 	if (pack.ep_flags & EXEC_32)
726 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
727 		    sizeof(int) + sizeof(int) + dp + RTLD_GAP +
728 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
729 		    - argp;
730 	else
731 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
732 		    sizeof(char *) + sizeof(int) + dp + RTLD_GAP +
733 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
734 		    - argp;
735 
736 #ifdef PAX_ASLR
737 	if (pax_aslr_active(l))
738 		len += (arc4random() % PAGE_SIZE);
739 #endif /* PAX_ASLR */
740 
741 #ifdef STACKLALIGN	/* arm, etc. */
742 	len = STACKALIGN(len);	/* make the stack "safely" aligned */
743 #else
744 	len = ALIGN(len);	/* make the stack "safely" aligned */
745 #endif
746 
747 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
748 		DPRINTF(("execve: stack limit exceeded %zu\n", len));
749 		error = ENOMEM;
750 		goto bad;
751 	}
752 
753 	/* Get rid of other LWPs. */
754 	if (p->p_sa || p->p_nlwps > 1) {
755 		mutex_enter(p->p_lock);
756 		exit_lwps(l);
757 		mutex_exit(p->p_lock);
758 	}
759 	KDASSERT(p->p_nlwps == 1);
760 
761 	/* Destroy any lwpctl info. */
762 	if (p->p_lwpctl != NULL)
763 		lwp_ctl_exit();
764 
765 #ifdef KERN_SA
766 	/* Release any SA state. */
767 	if (p->p_sa)
768 		sa_release(p);
769 #endif /* KERN_SA */
770 
771 	/* Remove POSIX timers */
772 	timers_free(p, TIMERS_POSIX);
773 
774 	/* adjust "active stack depth" for process VSZ */
775 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
776 
777 	/*
778 	 * Do whatever is necessary to prepare the address space
779 	 * for remapping.  Note that this might replace the current
780 	 * vmspace with another!
781 	 */
782 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
783 
784 	/* record proc's vnode, for use by procfs and others */
785         if (p->p_textvp)
786                 vrele(p->p_textvp);
787 	vref(pack.ep_vp);
788 	p->p_textvp = pack.ep_vp;
789 
790 	/* Now map address space */
791 	vm = p->p_vmspace;
792 	vm->vm_taddr = (void *)pack.ep_taddr;
793 	vm->vm_tsize = btoc(pack.ep_tsize);
794 	vm->vm_daddr = (void*)pack.ep_daddr;
795 	vm->vm_dsize = btoc(pack.ep_dsize);
796 	vm->vm_ssize = btoc(pack.ep_ssize);
797 	vm->vm_issize = 0;
798 	vm->vm_maxsaddr = (void *)pack.ep_maxsaddr;
799 	vm->vm_minsaddr = (void *)pack.ep_minsaddr;
800 
801 #ifdef PAX_ASLR
802 	pax_aslr_init(l, vm);
803 #endif /* PAX_ASLR */
804 
805 	/* create the new process's VM space by running the vmcmds */
806 #ifdef DIAGNOSTIC
807 	if (pack.ep_vmcmds.evs_used == 0)
808 		panic("execve: no vmcmds");
809 #endif
810 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
811 		struct exec_vmcmd *vcp;
812 
813 		vcp = &pack.ep_vmcmds.evs_cmds[i];
814 		if (vcp->ev_flags & VMCMD_RELATIVE) {
815 #ifdef DIAGNOSTIC
816 			if (base_vcp == NULL)
817 				panic("execve: relative vmcmd with no base");
818 			if (vcp->ev_flags & VMCMD_BASE)
819 				panic("execve: illegal base & relative vmcmd");
820 #endif
821 			vcp->ev_addr += base_vcp->ev_addr;
822 		}
823 		error = (*vcp->ev_proc)(l, vcp);
824 #ifdef DEBUG_EXEC
825 		if (error) {
826 			size_t j;
827 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
828 			for (j = 0; j <= i; j++)
829 				uprintf(
830 			"vmcmd[%zu] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
831 				    j, vp[j].ev_addr, vp[j].ev_len,
832 				    vp[j].ev_offset, vp[j].ev_prot,
833 				    vp[j].ev_flags);
834 		}
835 #endif /* DEBUG_EXEC */
836 		if (vcp->ev_flags & VMCMD_BASE)
837 			base_vcp = vcp;
838 	}
839 
840 	/* free the vmspace-creation commands, and release their references */
841 	kill_vmcmds(&pack.ep_vmcmds);
842 
843 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
844 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
845 	vput(pack.ep_vp);
846 
847 	/* if an error happened, deallocate and punt */
848 	if (error) {
849 		DPRINTF(("execve: vmcmd %zu failed: %d\n", i - 1, error));
850 		goto exec_abort;
851 	}
852 
853 	/* remember information about the process */
854 	arginfo.ps_nargvstr = argc;
855 	arginfo.ps_nenvstr = envc;
856 
857 	/* set command name & other accounting info */
858 	commandname = strrchr(pack.ep_resolvedname, '/');
859 	if (commandname != NULL) {
860 		commandname++;
861 	} else {
862 		commandname = pack.ep_resolvedname;
863 	}
864 	i = min(strlen(commandname), MAXCOMLEN);
865 	(void)memcpy(p->p_comm, commandname, i);
866 	p->p_comm[i] = '\0';
867 
868 	dp = PNBUF_GET();
869 	/*
870 	 * If the path starts with /, we don't need to do any work.
871 	 * This handles the majority of the cases.
872 	 * In the future perhaps we could canonicalize it?
873 	 */
874 	if (pathbuf[0] == '/')
875 		(void)strlcpy(pack.ep_path = dp, pathbuf, MAXPATHLEN);
876 #ifdef notyet
877 	/*
878 	 * Although this works most of the time [since the entry was just
879 	 * entered in the cache] we don't use it because it theoretically
880 	 * can fail and it is not the cleanest interface, because there
881 	 * could be races. When the namei cache is re-written, this can
882 	 * be changed to use the appropriate function.
883 	 */
884 	else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p)))
885 		pack.ep_path = dp;
886 #endif
887 	else {
888 #ifdef notyet
889 		printf("Cannot get path for pid %d [%s] (error %d)",
890 		    (int)p->p_pid, p->p_comm, error);
891 #endif
892 		pack.ep_path = NULL;
893 		PNBUF_PUT(dp);
894 	}
895 
896 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
897 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
898 		len - (sizeof(struct ps_strings) + szsigcode));
899 
900 #ifdef __MACHINE_STACK_GROWS_UP
901 	/*
902 	 * The copyargs call always copies into lower addresses
903 	 * first, moving towards higher addresses, starting with
904 	 * the stack pointer that we give.  When the stack grows
905 	 * down, this puts argc/argv/envp very shallow on the
906 	 * stack, right at the first user stack pointer.
907 	 * When the stack grows up, the situation is reversed.
908 	 *
909 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
910 	 * function expects to be called with a single pointer to
911 	 * a region that has a few words it can stash values into,
912 	 * followed by argc/argv/envp.  When the stack grows down,
913 	 * it's easy to decrement the stack pointer a little bit to
914 	 * allocate the space for these few words and pass the new
915 	 * stack pointer to _rtld.  When the stack grows up, however,
916 	 * a few words before argc is part of the signal trampoline, XXX
917 	 * so we have a problem.
918 	 *
919 	 * Instead of changing how _rtld works, we take the easy way
920 	 * out and steal 32 bytes before we call copyargs.
921 	 * This extra space was allowed for when 'len' was calculated.
922 	 */
923 	stack += RTLD_GAP;
924 #endif /* __MACHINE_STACK_GROWS_UP */
925 
926 	/* Now copy argc, args & environ to new stack */
927 	error = (*pack.ep_esch->es_copyargs)(l, &pack, &arginfo, &stack, argp);
928 	if (pack.ep_path) {
929 		PNBUF_PUT(pack.ep_path);
930 		pack.ep_path = NULL;
931 	}
932 	if (error) {
933 		DPRINTF(("execve: copyargs failed %d\n", error));
934 		goto exec_abort;
935 	}
936 	/* Move the stack back to original point */
937 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
938 
939 	/* fill process ps_strings info */
940 	p->p_psstr = (struct ps_strings *)
941 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
942 	    sizeof(struct ps_strings));
943 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
944 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
945 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
946 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
947 
948 	/* copy out the process's ps_strings structure */
949 	if ((error = copyout(aip, (char *)p->p_psstr,
950 	    sizeof(arginfo))) != 0) {
951 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
952 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
953 		goto exec_abort;
954 	}
955 
956 	fd_closeexec();		/* handle close on exec */
957 	execsigs(p);		/* reset catched signals */
958 
959 	l->l_ctxlink = NULL;	/* reset ucontext link */
960 
961 
962 	p->p_acflag &= ~AFORK;
963 	mutex_enter(p->p_lock);
964 	p->p_flag |= PK_EXEC;
965 	mutex_exit(p->p_lock);
966 
967 	/*
968 	 * Stop profiling.
969 	 */
970 	if ((p->p_stflag & PST_PROFIL) != 0) {
971 		mutex_spin_enter(&p->p_stmutex);
972 		stopprofclock(p);
973 		mutex_spin_exit(&p->p_stmutex);
974 	}
975 
976 	/*
977 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
978 	 * exited and exec()/exit() are the only places it will be cleared.
979 	 */
980 	if ((p->p_lflag & PL_PPWAIT) != 0) {
981 		mutex_enter(proc_lock);
982 		p->p_lflag &= ~PL_PPWAIT;
983 		cv_broadcast(&p->p_pptr->p_waitcv);
984 		mutex_exit(proc_lock);
985 	}
986 
987 	/*
988 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
989 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
990 	 * out additional references on the process for the moment.
991 	 */
992 	if ((p->p_slflag & PSL_TRACED) == 0 &&
993 
994 	    (((attr.va_mode & S_ISUID) != 0 &&
995 	      kauth_cred_geteuid(l->l_cred) != attr.va_uid) ||
996 
997 	     ((attr.va_mode & S_ISGID) != 0 &&
998 	      kauth_cred_getegid(l->l_cred) != attr.va_gid))) {
999 		/*
1000 		 * Mark the process as SUGID before we do
1001 		 * anything that might block.
1002 		 */
1003 		proc_crmod_enter();
1004 		proc_crmod_leave(NULL, NULL, true);
1005 
1006 		/* Make sure file descriptors 0..2 are in use. */
1007 		if ((error = fd_checkstd()) != 0) {
1008 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
1009 			goto exec_abort;
1010 		}
1011 
1012 		/*
1013 		 * Copy the credential so other references don't see our
1014 		 * changes.
1015 		 */
1016 		l->l_cred = kauth_cred_copy(l->l_cred);
1017 #ifdef KTRACE
1018 		/*
1019 		 * If the persistent trace flag isn't set, turn off.
1020 		 */
1021 		if (p->p_tracep) {
1022 			mutex_enter(&ktrace_lock);
1023 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1024 				ktrderef(p);
1025 			mutex_exit(&ktrace_lock);
1026 		}
1027 #endif
1028 		if (attr.va_mode & S_ISUID)
1029 			kauth_cred_seteuid(l->l_cred, attr.va_uid);
1030 		if (attr.va_mode & S_ISGID)
1031 			kauth_cred_setegid(l->l_cred, attr.va_gid);
1032 	} else {
1033 		if (kauth_cred_geteuid(l->l_cred) ==
1034 		    kauth_cred_getuid(l->l_cred) &&
1035 		    kauth_cred_getegid(l->l_cred) ==
1036 		    kauth_cred_getgid(l->l_cred))
1037 			p->p_flag &= ~PK_SUGID;
1038 	}
1039 
1040 	/*
1041 	 * Copy the credential so other references don't see our changes.
1042 	 * Test to see if this is necessary first, since in the common case
1043 	 * we won't need a private reference.
1044 	 */
1045 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1046 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1047 		l->l_cred = kauth_cred_copy(l->l_cred);
1048 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1049 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1050 	}
1051 
1052 	/* Update the master credentials. */
1053 	if (l->l_cred != p->p_cred) {
1054 		kauth_cred_t ocred;
1055 
1056 		kauth_cred_hold(l->l_cred);
1057 		mutex_enter(p->p_lock);
1058 		ocred = p->p_cred;
1059 		p->p_cred = l->l_cred;
1060 		mutex_exit(p->p_lock);
1061 		kauth_cred_free(ocred);
1062 	}
1063 
1064 #if defined(__HAVE_RAS)
1065 	/*
1066 	 * Remove all RASs from the address space.
1067 	 */
1068 	ras_purgeall();
1069 #endif
1070 
1071 	doexechooks(p);
1072 
1073 	/* setup new registers and do misc. setup. */
1074 	(*pack.ep_esch->es_emul->e_setregs)(l, &pack, (vaddr_t)stack);
1075 	if (pack.ep_esch->es_setregs)
1076 		(*pack.ep_esch->es_setregs)(l, &pack, (vaddr_t)stack);
1077 
1078 	/* map the process's signal trampoline code */
1079 	if (exec_sigcode_map(p, pack.ep_esch->es_emul)) {
1080 		DPRINTF(("execve: map sigcode failed %d\n", error));
1081 		goto exec_abort;
1082 	}
1083 
1084 	pool_put(&exec_pool, argp);
1085 
1086 	/* notify others that we exec'd */
1087 	KNOTE(&p->p_klist, NOTE_EXEC);
1088 
1089 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1090 
1091 	SDT_PROBE(proc,,,exec_success, path, 0, 0, 0, 0);
1092 
1093 	/* The emulation root will usually have been found when we looked
1094 	 * for the elf interpreter (or similar), if not look now. */
1095 	if (pack.ep_esch->es_emul->e_path != NULL && pack.ep_emul_root == NULL)
1096 		emul_find_root(l, &pack);
1097 
1098 	/* Any old emulation root got removed by fdcloseexec */
1099 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1100 	p->p_cwdi->cwdi_edir = pack.ep_emul_root;
1101 	rw_exit(&p->p_cwdi->cwdi_lock);
1102 	pack.ep_emul_root = NULL;
1103 	if (pack.ep_interp != NULL)
1104 		vrele(pack.ep_interp);
1105 
1106 	/*
1107 	 * Call emulation specific exec hook. This can setup per-process
1108 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1109 	 *
1110 	 * If we are executing process of different emulation than the
1111 	 * original forked process, call e_proc_exit() of the old emulation
1112 	 * first, then e_proc_exec() of new emulation. If the emulation is
1113 	 * same, the exec hook code should deallocate any old emulation
1114 	 * resources held previously by this process.
1115 	 */
1116 	if (p->p_emul && p->p_emul->e_proc_exit
1117 	    && p->p_emul != pack.ep_esch->es_emul)
1118 		(*p->p_emul->e_proc_exit)(p);
1119 
1120 	/*
1121 	 * This is now LWP 1.
1122 	 */
1123 	mutex_enter(p->p_lock);
1124 	p->p_nlwpid = 1;
1125 	l->l_lid = 1;
1126 	mutex_exit(p->p_lock);
1127 
1128 	/*
1129 	 * Call exec hook. Emulation code may NOT store reference to anything
1130 	 * from &pack.
1131 	 */
1132         if (pack.ep_esch->es_emul->e_proc_exec)
1133                 (*pack.ep_esch->es_emul->e_proc_exec)(p, &pack);
1134 
1135 	/* update p_emul, the old value is no longer needed */
1136 	p->p_emul = pack.ep_esch->es_emul;
1137 
1138 	/* ...and the same for p_execsw */
1139 	p->p_execsw = pack.ep_esch;
1140 
1141 #ifdef __HAVE_SYSCALL_INTERN
1142 	(*p->p_emul->e_syscall_intern)(p);
1143 #endif
1144 	ktremul();
1145 
1146 	/* Allow new references from the debugger/procfs. */
1147 	rw_exit(&p->p_reflock);
1148 	rw_exit(&exec_lock);
1149 
1150 	mutex_enter(proc_lock);
1151 
1152 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1153 		KSI_INIT_EMPTY(&ksi);
1154 		ksi.ksi_signo = SIGTRAP;
1155 		ksi.ksi_lid = l->l_lid;
1156 		kpsignal(p, &ksi, NULL);
1157 	}
1158 
1159 	if (p->p_sflag & PS_STOPEXEC) {
1160 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1161 		p->p_pptr->p_nstopchild++;
1162 		p->p_pptr->p_waited = 0;
1163 		mutex_enter(p->p_lock);
1164 		ksiginfo_queue_init(&kq);
1165 		sigclearall(p, &contsigmask, &kq);
1166 		lwp_lock(l);
1167 		l->l_stat = LSSTOP;
1168 		p->p_stat = SSTOP;
1169 		p->p_nrlwps--;
1170 		mutex_exit(p->p_lock);
1171 		mutex_exit(proc_lock);
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 	PNBUF_PUT(pathbuf);
1180 	PNBUF_PUT(resolvedpathbuf);
1181 	return (EJUSTRETURN);
1182 
1183  bad:
1184 	/* free the vmspace-creation commands, and release their references */
1185 	kill_vmcmds(&pack.ep_vmcmds);
1186 	/* kill any opened file descriptor, if necessary */
1187 	if (pack.ep_flags & EXEC_HASFD) {
1188 		pack.ep_flags &= ~EXEC_HASFD;
1189 		fd_close(pack.ep_fd);
1190 	}
1191 	/* close and put the exec'd file */
1192 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
1193 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
1194 	vput(pack.ep_vp);
1195 	pool_put(&exec_pool, argp);
1196 
1197  freehdr:
1198 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1199 	if (pack.ep_emul_root != NULL)
1200 		vrele(pack.ep_emul_root);
1201 	if (pack.ep_interp != NULL)
1202 		vrele(pack.ep_interp);
1203 
1204 	rw_exit(&exec_lock);
1205 
1206 	PNBUF_PUT(resolvedpathbuf);
1207 
1208  clrflg:
1209 	lwp_lock(l);
1210 	l->l_flag |= oldlwpflags;
1211 	lwp_unlock(l);
1212 	rw_exit(&p->p_reflock);
1213 
1214 	PNBUF_PUT(pathbuf);
1215 
1216 	if (modgen != module_gen && error == ENOEXEC) {
1217 		modgen = module_gen;
1218 		exec_autoload();
1219 		goto retry;
1220 	}
1221 
1222 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
1223 	return error;
1224 
1225  exec_abort:
1226 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
1227 	rw_exit(&p->p_reflock);
1228 	rw_exit(&exec_lock);
1229 
1230 	PNBUF_PUT(pathbuf);
1231 	PNBUF_PUT(resolvedpathbuf);
1232 
1233 	/*
1234 	 * the old process doesn't exist anymore.  exit gracefully.
1235 	 * get rid of the (new) address space we have created, if any, get rid
1236 	 * of our namei data and vnode, and exit noting failure
1237 	 */
1238 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1239 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1240 	if (pack.ep_emul_arg)
1241 		free(pack.ep_emul_arg, M_TEMP);
1242 	pool_put(&exec_pool, argp);
1243 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1244 	if (pack.ep_emul_root != NULL)
1245 		vrele(pack.ep_emul_root);
1246 	if (pack.ep_interp != NULL)
1247 		vrele(pack.ep_interp);
1248 
1249 	/* Acquire the sched-state mutex (exit1() will release it). */
1250 	mutex_enter(p->p_lock);
1251 	exit1(l, W_EXITCODE(error, SIGABRT));
1252 
1253 	/* NOTREACHED */
1254 	return 0;
1255 }
1256 
1257 
1258 int
1259 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1260     char **stackp, void *argp)
1261 {
1262 	char	**cpp, *dp, *sp;
1263 	size_t	len;
1264 	void	*nullp;
1265 	long	argc, envc;
1266 	int	error;
1267 
1268 	cpp = (char **)*stackp;
1269 	nullp = NULL;
1270 	argc = arginfo->ps_nargvstr;
1271 	envc = arginfo->ps_nenvstr;
1272 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
1273 		return error;
1274 
1275 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen);
1276 	sp = argp;
1277 
1278 	/* XXX don't copy them out, remap them! */
1279 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1280 
1281 	for (; --argc >= 0; sp += len, dp += len)
1282 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1283 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1284 			return error;
1285 
1286 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1287 		return error;
1288 
1289 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1290 
1291 	for (; --envc >= 0; sp += len, dp += len)
1292 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1293 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1294 			return error;
1295 
1296 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1297 		return error;
1298 
1299 	*stackp = (char *)cpp;
1300 	return 0;
1301 }
1302 
1303 
1304 /*
1305  * Add execsw[] entries.
1306  */
1307 int
1308 exec_add(struct execsw *esp, int count)
1309 {
1310 	struct exec_entry	*it;
1311 	int			i;
1312 
1313 	if (count == 0) {
1314 		return 0;
1315 	}
1316 
1317 	/* Check for duplicates. */
1318 	rw_enter(&exec_lock, RW_WRITER);
1319 	for (i = 0; i < count; i++) {
1320 		LIST_FOREACH(it, &ex_head, ex_list) {
1321 			/* assume unique (makecmds, probe_func, emulation) */
1322 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1323 			    it->ex_sw->u.elf_probe_func ==
1324 			    esp[i].u.elf_probe_func &&
1325 			    it->ex_sw->es_emul == esp[i].es_emul) {
1326 				rw_exit(&exec_lock);
1327 				return EEXIST;
1328 			}
1329 		}
1330 	}
1331 
1332 	/* Allocate new entries. */
1333 	for (i = 0; i < count; i++) {
1334 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1335 		it->ex_sw = &esp[i];
1336 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1337 	}
1338 
1339 	/* update execsw[] */
1340 	exec_init(0);
1341 	rw_exit(&exec_lock);
1342 	return 0;
1343 }
1344 
1345 /*
1346  * Remove execsw[] entry.
1347  */
1348 int
1349 exec_remove(struct execsw *esp, int count)
1350 {
1351 	struct exec_entry	*it, *next;
1352 	int			i;
1353 	const struct proclist_desc *pd;
1354 	proc_t			*p;
1355 
1356 	if (count == 0) {
1357 		return 0;
1358 	}
1359 
1360 	/* Abort if any are busy. */
1361 	rw_enter(&exec_lock, RW_WRITER);
1362 	for (i = 0; i < count; i++) {
1363 		mutex_enter(proc_lock);
1364 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1365 			PROCLIST_FOREACH(p, pd->pd_list) {
1366 				if (p->p_execsw == &esp[i]) {
1367 					mutex_exit(proc_lock);
1368 					rw_exit(&exec_lock);
1369 					return EBUSY;
1370 				}
1371 			}
1372 		}
1373 		mutex_exit(proc_lock);
1374 	}
1375 
1376 	/* None are busy, so remove them all. */
1377 	for (i = 0; i < count; i++) {
1378 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1379 			next = LIST_NEXT(it, ex_list);
1380 			if (it->ex_sw == &esp[i]) {
1381 				LIST_REMOVE(it, ex_list);
1382 				kmem_free(it, sizeof(*it));
1383 				break;
1384 			}
1385 		}
1386 	}
1387 
1388 	/* update execsw[] */
1389 	exec_init(0);
1390 	rw_exit(&exec_lock);
1391 	return 0;
1392 }
1393 
1394 /*
1395  * Initialize exec structures. If init_boot is true, also does necessary
1396  * one-time initialization (it's called from main() that way).
1397  * Once system is multiuser, this should be called with exec_lock held,
1398  * i.e. via exec_{add|remove}().
1399  */
1400 int
1401 exec_init(int init_boot)
1402 {
1403 	const struct execsw 	**sw;
1404 	struct exec_entry	*ex;
1405 	SLIST_HEAD(,exec_entry)	first;
1406 	SLIST_HEAD(,exec_entry)	any;
1407 	SLIST_HEAD(,exec_entry)	last;
1408 	int			i, sz;
1409 
1410 	if (init_boot) {
1411 		/* do one-time initializations */
1412 		rw_init(&exec_lock);
1413 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1414 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1415 		    "execargs", &exec_palloc, IPL_NONE);
1416 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1417 	} else {
1418 		KASSERT(rw_write_held(&exec_lock));
1419 	}
1420 
1421 	/* Sort each entry onto the appropriate queue. */
1422 	SLIST_INIT(&first);
1423 	SLIST_INIT(&any);
1424 	SLIST_INIT(&last);
1425 	sz = 0;
1426 	LIST_FOREACH(ex, &ex_head, ex_list) {
1427 		switch(ex->ex_sw->es_prio) {
1428 		case EXECSW_PRIO_FIRST:
1429 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1430 			break;
1431 		case EXECSW_PRIO_ANY:
1432 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1433 			break;
1434 		case EXECSW_PRIO_LAST:
1435 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1436 			break;
1437 		default:
1438 			panic("exec_init");
1439 			break;
1440 		}
1441 		sz++;
1442 	}
1443 
1444 	/*
1445 	 * Create new execsw[].  Ensure we do not try a zero-sized
1446 	 * allocation.
1447 	 */
1448 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1449 	i = 0;
1450 	SLIST_FOREACH(ex, &first, ex_slist) {
1451 		sw[i++] = ex->ex_sw;
1452 	}
1453 	SLIST_FOREACH(ex, &any, ex_slist) {
1454 		sw[i++] = ex->ex_sw;
1455 	}
1456 	SLIST_FOREACH(ex, &last, ex_slist) {
1457 		sw[i++] = ex->ex_sw;
1458 	}
1459 
1460 	/* Replace old execsw[] and free used memory. */
1461 	if (execsw != NULL) {
1462 		kmem_free(__UNCONST(execsw),
1463 		    nexecs * sizeof(struct execsw *) + 1);
1464 	}
1465 	execsw = sw;
1466 	nexecs = sz;
1467 
1468 	/* Figure out the maximum size of an exec header. */
1469 	exec_maxhdrsz = sizeof(int);
1470 	for (i = 0; i < nexecs; i++) {
1471 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1472 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1473 	}
1474 
1475 	return 0;
1476 }
1477 
1478 static int
1479 exec_sigcode_map(struct proc *p, const struct emul *e)
1480 {
1481 	vaddr_t va;
1482 	vsize_t sz;
1483 	int error;
1484 	struct uvm_object *uobj;
1485 
1486 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1487 
1488 	if (e->e_sigobject == NULL || sz == 0) {
1489 		return 0;
1490 	}
1491 
1492 	/*
1493 	 * If we don't have a sigobject for this emulation, create one.
1494 	 *
1495 	 * sigobject is an anonymous memory object (just like SYSV shared
1496 	 * memory) that we keep a permanent reference to and that we map
1497 	 * in all processes that need this sigcode. The creation is simple,
1498 	 * we create an object, add a permanent reference to it, map it in
1499 	 * kernel space, copy out the sigcode to it and unmap it.
1500 	 * We map it with PROT_READ|PROT_EXEC into the process just
1501 	 * the way sys_mmap() would map it.
1502 	 */
1503 
1504 	uobj = *e->e_sigobject;
1505 	if (uobj == NULL) {
1506 		mutex_enter(&sigobject_lock);
1507 		if ((uobj = *e->e_sigobject) == NULL) {
1508 			uobj = uao_create(sz, 0);
1509 			(*uobj->pgops->pgo_reference)(uobj);
1510 			va = vm_map_min(kernel_map);
1511 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1512 			    uobj, 0, 0,
1513 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1514 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1515 				printf("kernel mapping failed %d\n", error);
1516 				(*uobj->pgops->pgo_detach)(uobj);
1517 				mutex_exit(&sigobject_lock);
1518 				return (error);
1519 			}
1520 			memcpy((void *)va, e->e_sigcode, sz);
1521 #ifdef PMAP_NEED_PROCWR
1522 			pmap_procwr(&proc0, va, sz);
1523 #endif
1524 			uvm_unmap(kernel_map, va, va + round_page(sz));
1525 			*e->e_sigobject = uobj;
1526 		}
1527 		mutex_exit(&sigobject_lock);
1528 	}
1529 
1530 	/* Just a hint to uvm_map where to put it. */
1531 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1532 	    round_page(sz));
1533 
1534 #ifdef __alpha__
1535 	/*
1536 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1537 	 * which causes the above calculation to put the sigcode at
1538 	 * an invalid address.  Put it just below the text instead.
1539 	 */
1540 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1541 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1542 	}
1543 #endif
1544 
1545 	(*uobj->pgops->pgo_reference)(uobj);
1546 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1547 			uobj, 0, 0,
1548 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1549 				    UVM_ADV_RANDOM, 0));
1550 	if (error) {
1551 		(*uobj->pgops->pgo_detach)(uobj);
1552 		return (error);
1553 	}
1554 	p->p_sigctx.ps_sigcode = (void *)va;
1555 	return (0);
1556 }
1557