xref: /netbsd-src/sys/kern/kern_exec.c (revision 48fb7bfab72acd4281a53bbee5ccf3f809019e75)
1 /*	$NetBSD: kern_exec.c,v 1.377 2014/02/19 15:23:20 maxv 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 1.377 2014/02/19 15:23:20 maxv Exp $");
63 
64 #include "opt_exec.h"
65 #include "opt_execfmt.h"
66 #include "opt_ktrace.h"
67 #include "opt_modular.h"
68 #include "opt_syscall_debug.h"
69 #include "veriexec.h"
70 #include "opt_pax.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/filedesc.h>
75 #include <sys/kernel.h>
76 #include <sys/proc.h>
77 #include <sys/mount.h>
78 #include <sys/malloc.h>
79 #include <sys/kmem.h>
80 #include <sys/namei.h>
81 #include <sys/vnode.h>
82 #include <sys/file.h>
83 #include <sys/acct.h>
84 #include <sys/atomic.h>
85 #include <sys/exec.h>
86 #include <sys/ktrace.h>
87 #include <sys/uidinfo.h>
88 #include <sys/wait.h>
89 #include <sys/mman.h>
90 #include <sys/ras.h>
91 #include <sys/signalvar.h>
92 #include <sys/stat.h>
93 #include <sys/syscall.h>
94 #include <sys/kauth.h>
95 #include <sys/lwpctl.h>
96 #include <sys/pax.h>
97 #include <sys/cpu.h>
98 #include <sys/module.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 #include <sys/spawn.h>
106 #include <sys/prot.h>
107 #include <sys/cprng.h>
108 
109 #include <uvm/uvm_extern.h>
110 
111 #include <machine/reg.h>
112 
113 #include <compat/common/compat_util.h>
114 
115 #ifndef MD_TOPDOWN_INIT
116 #ifdef __USE_TOPDOWN_VM
117 #define	MD_TOPDOWN_INIT(epp)	(epp)->ep_flags |= EXEC_TOPDOWN_VM
118 #else
119 #define	MD_TOPDOWN_INIT(epp)
120 #endif
121 #endif
122 
123 static int exec_sigcode_map(struct proc *, const struct emul *);
124 
125 #ifdef DEBUG_EXEC
126 #define DPRINTF(a) printf a
127 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
128     __LINE__, (s), (a), (b))
129 #else
130 #define DPRINTF(a)
131 #define COPYPRINTF(s, a, b)
132 #endif /* DEBUG_EXEC */
133 
134 /*
135  * DTrace SDT provider definitions
136  */
137 SDT_PROBE_DEFINE(proc,,,exec,exec,
138 	    "char *", NULL,
139 	    NULL, NULL, NULL, NULL,
140 	    NULL, NULL, NULL, NULL);
141 SDT_PROBE_DEFINE(proc,,,exec_success,exec-success,
142 	    "char *", NULL,
143 	    NULL, NULL, NULL, NULL,
144 	    NULL, NULL, NULL, NULL);
145 SDT_PROBE_DEFINE(proc,,,exec_failure,exec-failure,
146 	    "int", NULL,
147 	    NULL, NULL, NULL, NULL,
148 	    NULL, NULL, NULL, NULL);
149 
150 /*
151  * Exec function switch:
152  *
153  * Note that each makecmds function is responsible for loading the
154  * exec package with the necessary functions for any exec-type-specific
155  * handling.
156  *
157  * Functions for specific exec types should be defined in their own
158  * header file.
159  */
160 static const struct execsw	**execsw = NULL;
161 static int			nexecs;
162 
163 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
164 
165 /* list of dynamically loaded execsw entries */
166 static LIST_HEAD(execlist_head, exec_entry) ex_head =
167     LIST_HEAD_INITIALIZER(ex_head);
168 struct exec_entry {
169 	LIST_ENTRY(exec_entry)	ex_list;
170 	SLIST_ENTRY(exec_entry)	ex_slist;
171 	const struct execsw	*ex_sw;
172 };
173 
174 #ifndef __HAVE_SYSCALL_INTERN
175 void	syscall(void);
176 #endif
177 
178 /* NetBSD emul struct */
179 struct emul emul_netbsd = {
180 	.e_name =		"netbsd",
181 #ifdef EMUL_NATIVEROOT
182 	.e_path =		EMUL_NATIVEROOT,
183 #else
184 	.e_path =		NULL,
185 #endif
186 #ifndef __HAVE_MINIMAL_EMUL
187 	.e_flags =		EMUL_HAS_SYS___syscall,
188 	.e_errno =		NULL,
189 	.e_nosys =		SYS_syscall,
190 	.e_nsysent =		SYS_NSYSENT,
191 #endif
192 	.e_sysent =		sysent,
193 #ifdef SYSCALL_DEBUG
194 	.e_syscallnames =	syscallnames,
195 #else
196 	.e_syscallnames =	NULL,
197 #endif
198 	.e_sendsig =		sendsig,
199 	.e_trapsignal =		trapsignal,
200 	.e_tracesig =		NULL,
201 	.e_sigcode =		NULL,
202 	.e_esigcode =		NULL,
203 	.e_sigobject =		NULL,
204 	.e_setregs =		setregs,
205 	.e_proc_exec =		NULL,
206 	.e_proc_fork =		NULL,
207 	.e_proc_exit =		NULL,
208 	.e_lwp_fork =		NULL,
209 	.e_lwp_exit =		NULL,
210 #ifdef __HAVE_SYSCALL_INTERN
211 	.e_syscall_intern =	syscall_intern,
212 #else
213 	.e_syscall =		syscall,
214 #endif
215 	.e_sysctlovly =		NULL,
216 	.e_fault =		NULL,
217 	.e_vm_default_addr =	uvm_default_mapaddr,
218 	.e_usertrap =		NULL,
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 /*
232  * Data used between a loadvm and execve part of an "exec" operation
233  */
234 struct execve_data {
235 	struct exec_package	ed_pack;
236 	struct pathbuf		*ed_pathbuf;
237 	struct vattr		ed_attr;
238 	struct ps_strings	ed_arginfo;
239 	char			*ed_argp;
240 	const char		*ed_pathstring;
241 	char			*ed_resolvedpathbuf;
242 	size_t			ed_ps_strings_sz;
243 	int			ed_szsigcode;
244 	long			ed_argc;
245 	long			ed_envc;
246 };
247 
248 /*
249  * data passed from parent lwp to child during a posix_spawn()
250  */
251 struct spawn_exec_data {
252 	struct execve_data	sed_exec;
253 	struct posix_spawn_file_actions
254 				*sed_actions;
255 	struct posix_spawnattr	*sed_attrs;
256 	struct proc		*sed_parent;
257 	kcondvar_t		sed_cv_child_ready;
258 	kmutex_t		sed_mtx_child;
259 	int			sed_error;
260 	volatile uint32_t	sed_refcnt;
261 };
262 
263 static void *
264 exec_pool_alloc(struct pool *pp, int flags)
265 {
266 
267 	return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
268 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
269 }
270 
271 static void
272 exec_pool_free(struct pool *pp, void *addr)
273 {
274 
275 	uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
276 }
277 
278 static struct pool exec_pool;
279 
280 static struct pool_allocator exec_palloc = {
281 	.pa_alloc = exec_pool_alloc,
282 	.pa_free = exec_pool_free,
283 	.pa_pagesz = NCARGS
284 };
285 
286 /*
287  * check exec:
288  * given an "executable" described in the exec package's namei info,
289  * see what we can do with it.
290  *
291  * ON ENTRY:
292  *	exec package with appropriate namei info
293  *	lwp pointer of exec'ing lwp
294  *	NO SELF-LOCKED VNODES
295  *
296  * ON EXIT:
297  *	error:	nothing held, etc.  exec header still allocated.
298  *	ok:	filled exec package, executable's vnode (unlocked).
299  *
300  * EXEC SWITCH ENTRY:
301  * 	Locked vnode to check, exec package, proc.
302  *
303  * EXEC SWITCH EXIT:
304  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
305  *	error:	destructive:
306  *			everything deallocated execept exec header.
307  *		non-destructive:
308  *			error code, executable's vnode (unlocked),
309  *			exec header unmodified.
310  */
311 int
312 /*ARGSUSED*/
313 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
314 {
315 	int		error, i;
316 	struct vnode	*vp;
317 	struct nameidata nd;
318 	size_t		resid;
319 
320 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
321 
322 	/* first get the vnode */
323 	if ((error = namei(&nd)) != 0)
324 		return error;
325 	epp->ep_vp = vp = nd.ni_vp;
326 	/* normally this can't fail */
327 	if ((error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL)))
328 		goto bad1;
329 
330 #ifdef DIAGNOSTIC
331 	/* paranoia (take this out once namei stuff stabilizes) */
332 	memset(nd.ni_pnbuf, '~', PATH_MAX);
333 #endif
334 
335 	/* check access and type */
336 	if (vp->v_type != VREG) {
337 		error = EACCES;
338 		goto bad1;
339 	}
340 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
341 		goto bad1;
342 
343 	/* get attributes */
344 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
345 		goto bad1;
346 
347 	/* Check mount point */
348 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
349 		error = EACCES;
350 		goto bad1;
351 	}
352 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
353 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
354 
355 	/* try to open it */
356 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
357 		goto bad1;
358 
359 	/* unlock vp, since we need it unlocked from here on out. */
360 	VOP_UNLOCK(vp);
361 
362 #if NVERIEXEC > 0
363 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
364 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
365 	    NULL);
366 	if (error)
367 		goto bad2;
368 #endif /* NVERIEXEC > 0 */
369 
370 #ifdef PAX_SEGVGUARD
371 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
372 	if (error)
373 		goto bad2;
374 #endif /* PAX_SEGVGUARD */
375 
376 	/* now we have the file, get the exec header */
377 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
378 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
379 	if (error)
380 		goto bad2;
381 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
382 
383 	/*
384 	 * Set up default address space limits.  Can be overridden
385 	 * by individual exec packages.
386 	 *
387 	 * XXX probably should be all done in the exec packages.
388 	 */
389 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
390 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
391 	/*
392 	 * set up the vmcmds for creation of the process
393 	 * address space
394 	 */
395 	error = ENOEXEC;
396 	for (i = 0; i < nexecs; i++) {
397 		int newerror;
398 
399 		epp->ep_esch = execsw[i];
400 		newerror = (*execsw[i]->es_makecmds)(l, epp);
401 
402 		if (!newerror) {
403 			/* Seems ok: check that entry point is not too high */
404 			if (epp->ep_entry > epp->ep_vm_maxaddr) {
405 #ifdef DIAGNOSTIC
406 				printf("%s: rejecting %p due to "
407 				    "too high entry address (> %p)\n",
408 					 __func__, (void *)epp->ep_entry,
409 					 (void *)epp->ep_vm_maxaddr);
410 #endif
411 				error = ENOEXEC;
412 				break;
413 			}
414 			/* Seems ok: check that entry point is not too low */
415 			if (epp->ep_entry < epp->ep_vm_minaddr) {
416 #ifdef DIAGNOSTIC
417 				printf("%s: rejecting %p due to "
418 				    "too low entry address (< %p)\n",
419 				     __func__, (void *)epp->ep_entry,
420 				     (void *)epp->ep_vm_minaddr);
421 #endif
422 				error = ENOEXEC;
423 				break;
424 			}
425 
426 			/* check limits */
427 			if ((epp->ep_tsize > MAXTSIZ) ||
428 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
429 						    [RLIMIT_DATA].rlim_cur)) {
430 #ifdef DIAGNOSTIC
431 				printf("%s: rejecting due to "
432 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
433 				    __func__,
434 				    (unsigned long long)epp->ep_tsize,
435 				    (unsigned long long)MAXTSIZ,
436 				    (unsigned long long)epp->ep_dsize,
437 				    (unsigned long long)
438 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
439 #endif
440 				error = ENOMEM;
441 				break;
442 			}
443 			return 0;
444 		}
445 
446 		if (epp->ep_emul_root != NULL) {
447 			vrele(epp->ep_emul_root);
448 			epp->ep_emul_root = NULL;
449 		}
450 		if (epp->ep_interp != NULL) {
451 			vrele(epp->ep_interp);
452 			epp->ep_interp = NULL;
453 		}
454 
455 		/* make sure the first "interesting" error code is saved. */
456 		if (error == ENOEXEC)
457 			error = newerror;
458 
459 		if (epp->ep_flags & EXEC_DESTR)
460 			/* Error from "#!" code, tidied up by recursive call */
461 			return error;
462 	}
463 
464 	/* not found, error */
465 
466 	/*
467 	 * free any vmspace-creation commands,
468 	 * and release their references
469 	 */
470 	kill_vmcmds(&epp->ep_vmcmds);
471 
472 bad2:
473 	/*
474 	 * close and release the vnode, restore the old one, free the
475 	 * pathname buf, and punt.
476 	 */
477 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
478 	VOP_CLOSE(vp, FREAD, l->l_cred);
479 	vput(vp);
480 	return error;
481 
482 bad1:
483 	/*
484 	 * free the namei pathname buffer, and put the vnode
485 	 * (which we don't yet have open).
486 	 */
487 	vput(vp);				/* was still locked */
488 	return error;
489 }
490 
491 #ifdef __MACHINE_STACK_GROWS_UP
492 #define STACK_PTHREADSPACE NBPG
493 #else
494 #define STACK_PTHREADSPACE 0
495 #endif
496 
497 static int
498 execve_fetch_element(char * const *array, size_t index, char **value)
499 {
500 	return copyin(array + index, value, sizeof(*value));
501 }
502 
503 /*
504  * exec system call
505  */
506 int
507 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
508 {
509 	/* {
510 		syscallarg(const char *)	path;
511 		syscallarg(char * const *)	argp;
512 		syscallarg(char * const *)	envp;
513 	} */
514 
515 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
516 	    SCARG(uap, envp), execve_fetch_element);
517 }
518 
519 int
520 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
521     register_t *retval)
522 {
523 	/* {
524 		syscallarg(int)			fd;
525 		syscallarg(char * const *)	argp;
526 		syscallarg(char * const *)	envp;
527 	} */
528 
529 	return ENOSYS;
530 }
531 
532 /*
533  * Load modules to try and execute an image that we do not understand.
534  * If no execsw entries are present, we load those likely to be needed
535  * in order to run native images only.  Otherwise, we autoload all
536  * possible modules that could let us run the binary.  XXX lame
537  */
538 static void
539 exec_autoload(void)
540 {
541 #ifdef MODULAR
542 	static const char * const native[] = {
543 		"exec_elf32",
544 		"exec_elf64",
545 		"exec_script",
546 		NULL
547 	};
548 	static const char * const compat[] = {
549 		"exec_elf32",
550 		"exec_elf64",
551 		"exec_script",
552 		"exec_aout",
553 		"exec_coff",
554 		"exec_ecoff",
555 		"compat_aoutm68k",
556 		"compat_freebsd",
557 		"compat_ibcs2",
558 		"compat_linux",
559 		"compat_linux32",
560 		"compat_netbsd32",
561 		"compat_sunos",
562 		"compat_sunos32",
563 		"compat_svr4",
564 		"compat_svr4_32",
565 		"compat_ultrix",
566 		NULL
567 	};
568 	char const * const *list;
569 	int i;
570 
571 	list = (nexecs == 0 ? native : compat);
572 	for (i = 0; list[i] != NULL; i++) {
573 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
574 			continue;
575 		}
576 		yield();
577 	}
578 #endif
579 }
580 
581 static int
582 execve_loadvm(struct lwp *l, const char *path, char * const *args,
583 	char * const *envs, execve_fetch_element_t fetch_element,
584 	struct execve_data * restrict data)
585 {
586 	int			error;
587 	struct proc		*p;
588 	char			*dp, *sp;
589 	size_t			i, len;
590 	struct exec_fakearg	*tmpfap;
591 	u_int			modgen;
592 
593 	KASSERT(data != NULL);
594 
595 	p = l->l_proc;
596 	modgen = 0;
597 
598 	SDT_PROBE(proc,,,exec, path, 0, 0, 0, 0);
599 
600 	/*
601 	 * Check if we have exceeded our number of processes limit.
602 	 * This is so that we handle the case where a root daemon
603 	 * forked, ran setuid to become the desired user and is trying
604 	 * to exec. The obvious place to do the reference counting check
605 	 * is setuid(), but we don't do the reference counting check there
606 	 * like other OS's do because then all the programs that use setuid()
607 	 * must be modified to check the return code of setuid() and exit().
608 	 * It is dangerous to make setuid() fail, because it fails open and
609 	 * the program will continue to run as root. If we make it succeed
610 	 * and return an error code, again we are not enforcing the limit.
611 	 * The best place to enforce the limit is here, when the process tries
612 	 * to execute a new image, because eventually the process will need
613 	 * to call exec in order to do something useful.
614 	 */
615  retry:
616 	if (p->p_flag & PK_SUGID) {
617 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
618 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
619 		     &p->p_rlimit[RLIMIT_NPROC],
620 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
621 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
622 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
623 		return EAGAIN;
624 	}
625 
626 	/*
627 	 * Drain existing references and forbid new ones.  The process
628 	 * should be left alone until we're done here.  This is necessary
629 	 * to avoid race conditions - e.g. in ptrace() - that might allow
630 	 * a local user to illicitly obtain elevated privileges.
631 	 */
632 	rw_enter(&p->p_reflock, RW_WRITER);
633 
634 	/*
635 	 * Init the namei data to point the file user's program name.
636 	 * This is done here rather than in check_exec(), so that it's
637 	 * possible to override this settings if any of makecmd/probe
638 	 * functions call check_exec() recursively - for example,
639 	 * see exec_script_makecmds().
640 	 */
641 	error = pathbuf_copyin(path, &data->ed_pathbuf);
642 	if (error) {
643 		DPRINTF(("%s: pathbuf_copyin path @%p %d\n", __func__,
644 		    path, error));
645 		goto clrflg;
646 	}
647 	data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
648 	data->ed_resolvedpathbuf = PNBUF_GET();
649 
650 	/*
651 	 * initialize the fields of the exec package.
652 	 */
653 	data->ed_pack.ep_name = path;
654 	data->ed_pack.ep_kname = data->ed_pathstring;
655 	data->ed_pack.ep_resolvedname = data->ed_resolvedpathbuf;
656 	data->ed_pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
657 	data->ed_pack.ep_hdrlen = exec_maxhdrsz;
658 	data->ed_pack.ep_hdrvalid = 0;
659 	data->ed_pack.ep_emul_arg = NULL;
660 	data->ed_pack.ep_emul_arg_free = NULL;
661 	memset(&data->ed_pack.ep_vmcmds, 0, sizeof(data->ed_pack.ep_vmcmds));
662 	data->ed_pack.ep_vap = &data->ed_attr;
663 	data->ed_pack.ep_flags = 0;
664 	MD_TOPDOWN_INIT(&data->ed_pack);
665 	data->ed_pack.ep_emul_root = NULL;
666 	data->ed_pack.ep_interp = NULL;
667 	data->ed_pack.ep_esch = NULL;
668 	data->ed_pack.ep_pax_flags = 0;
669 	memset(data->ed_pack.ep_machine_arch, 0,
670 	    sizeof(data->ed_pack.ep_machine_arch));
671 
672 	rw_enter(&exec_lock, RW_READER);
673 
674 	/* see if we can run it. */
675 	if ((error = check_exec(l, &data->ed_pack, data->ed_pathbuf)) != 0) {
676 		if (error != ENOENT) {
677 			DPRINTF(("%s: check exec failed %d\n",
678 			    __func__, error));
679 		}
680 		goto freehdr;
681 	}
682 
683 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
684 
685 	/* allocate an argument buffer */
686 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
687 	KASSERT(data->ed_argp != NULL);
688 	dp = data->ed_argp;
689 	data->ed_argc = 0;
690 
691 	/* copy the fake args list, if there's one, freeing it as we go */
692 	if (data->ed_pack.ep_flags & EXEC_HASARGL) {
693 		tmpfap = data->ed_pack.ep_fa;
694 		while (tmpfap->fa_arg != NULL) {
695 			const char *cp;
696 
697 			cp = tmpfap->fa_arg;
698 			while (*cp)
699 				*dp++ = *cp++;
700 			*dp++ = '\0';
701 			ktrexecarg(tmpfap->fa_arg, cp - tmpfap->fa_arg);
702 
703 			kmem_free(tmpfap->fa_arg, tmpfap->fa_len);
704 			tmpfap++; data->ed_argc++;
705 		}
706 		kmem_free(data->ed_pack.ep_fa, data->ed_pack.ep_fa_len);
707 		data->ed_pack.ep_flags &= ~EXEC_HASARGL;
708 	}
709 
710 	/* Now get argv & environment */
711 	if (args == NULL) {
712 		DPRINTF(("%s: null args\n", __func__));
713 		error = EINVAL;
714 		goto bad;
715 	}
716 	/* 'i' will index the argp/envp element to be retrieved */
717 	i = 0;
718 	if (data->ed_pack.ep_flags & EXEC_SKIPARG)
719 		i++;
720 
721 	while (1) {
722 		len = data->ed_argp + ARG_MAX - dp;
723 		if ((error = (*fetch_element)(args, i, &sp)) != 0) {
724 			DPRINTF(("%s: fetch_element args %d\n",
725 			    __func__, error));
726 			goto bad;
727 		}
728 		if (!sp)
729 			break;
730 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
731 			DPRINTF(("%s: copyinstr args %d\n", __func__, error));
732 			if (error == ENAMETOOLONG)
733 				error = E2BIG;
734 			goto bad;
735 		}
736 		ktrexecarg(dp, len - 1);
737 		dp += len;
738 		i++;
739 		data->ed_argc++;
740 	}
741 
742 	data->ed_envc = 0;
743 	/* environment need not be there */
744 	if (envs != NULL) {
745 		i = 0;
746 		while (1) {
747 			len = data->ed_argp + ARG_MAX - dp;
748 			if ((error = (*fetch_element)(envs, i, &sp)) != 0) {
749 				DPRINTF(("%s: fetch_element env %d\n",
750 				    __func__, error));
751 				goto bad;
752 			}
753 			if (!sp)
754 				break;
755 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
756 				DPRINTF(("%s: copyinstr env %d\n",
757 				    __func__, error));
758 				if (error == ENAMETOOLONG)
759 					error = E2BIG;
760 				goto bad;
761 			}
762 
763 			ktrexecenv(dp, len - 1);
764 			dp += len;
765 			i++;
766 			data->ed_envc++;
767 		}
768 	}
769 
770 	dp = (char *) ALIGN(dp);
771 
772 	data->ed_szsigcode = data->ed_pack.ep_esch->es_emul->e_esigcode -
773 	    data->ed_pack.ep_esch->es_emul->e_sigcode;
774 
775 #ifdef __MACHINE_STACK_GROWS_UP
776 /* See big comment lower down */
777 #define	RTLD_GAP	32
778 #else
779 #define	RTLD_GAP	0
780 #endif
781 
782 	/* Now check if args & environ fit into new stack */
783 	if (data->ed_pack.ep_flags & EXEC_32) {
784 		data->ed_ps_strings_sz = sizeof(struct ps_strings32);
785 		len = ((data->ed_argc + data->ed_envc + 2 +
786 		    data->ed_pack.ep_esch->es_arglen) *
787 		    sizeof(int) + sizeof(int) + dp + RTLD_GAP +
788 		    data->ed_szsigcode + data->ed_ps_strings_sz + STACK_PTHREADSPACE)
789 		    - data->ed_argp;
790 	} else {
791 		data->ed_ps_strings_sz = sizeof(struct ps_strings);
792 		len = ((data->ed_argc + data->ed_envc + 2 +
793 		    data->ed_pack.ep_esch->es_arglen) *
794 		    sizeof(char *) + sizeof(int) + dp + RTLD_GAP +
795 		    data->ed_szsigcode + data->ed_ps_strings_sz + STACK_PTHREADSPACE)
796 		    - data->ed_argp;
797 	}
798 
799 #ifdef PAX_ASLR
800 	if (pax_aslr_active(l))
801 		len += (cprng_fast32() % PAGE_SIZE);
802 #endif /* PAX_ASLR */
803 
804 	/* make the stack "safely" aligned */
805 	len = STACK_LEN_ALIGN(len, STACK_ALIGNBYTES);
806 
807 	if (len > data->ed_pack.ep_ssize) {
808 		/* in effect, compare to initial limit */
809 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
810 		goto bad;
811 	}
812 	/* adjust "active stack depth" for process VSZ */
813 	data->ed_pack.ep_ssize = len;
814 
815 	return 0;
816 
817  bad:
818 	/* free the vmspace-creation commands, and release their references */
819 	kill_vmcmds(&data->ed_pack.ep_vmcmds);
820 	/* kill any opened file descriptor, if necessary */
821 	if (data->ed_pack.ep_flags & EXEC_HASFD) {
822 		data->ed_pack.ep_flags &= ~EXEC_HASFD;
823 		fd_close(data->ed_pack.ep_fd);
824 	}
825 	/* close and put the exec'd file */
826 	vn_lock(data->ed_pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
827 	VOP_CLOSE(data->ed_pack.ep_vp, FREAD, l->l_cred);
828 	vput(data->ed_pack.ep_vp);
829 	pool_put(&exec_pool, data->ed_argp);
830 
831  freehdr:
832 	kmem_free(data->ed_pack.ep_hdr, data->ed_pack.ep_hdrlen);
833 	if (data->ed_pack.ep_emul_root != NULL)
834 		vrele(data->ed_pack.ep_emul_root);
835 	if (data->ed_pack.ep_interp != NULL)
836 		vrele(data->ed_pack.ep_interp);
837 
838 	rw_exit(&exec_lock);
839 
840 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
841 	pathbuf_destroy(data->ed_pathbuf);
842 	PNBUF_PUT(data->ed_resolvedpathbuf);
843 
844  clrflg:
845 	rw_exit(&p->p_reflock);
846 
847 	if (modgen != module_gen && error == ENOEXEC) {
848 		modgen = module_gen;
849 		exec_autoload();
850 		goto retry;
851 	}
852 
853 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
854 	return error;
855 }
856 
857 static void
858 execve_free_data(struct execve_data *data)
859 {
860 
861 	/* free the vmspace-creation commands, and release their references */
862 	kill_vmcmds(&data->ed_pack.ep_vmcmds);
863 	/* kill any opened file descriptor, if necessary */
864 	if (data->ed_pack.ep_flags & EXEC_HASFD) {
865 		data->ed_pack.ep_flags &= ~EXEC_HASFD;
866 		fd_close(data->ed_pack.ep_fd);
867 	}
868 
869 	/* close and put the exec'd file */
870 	vn_lock(data->ed_pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
871 	VOP_CLOSE(data->ed_pack.ep_vp, FREAD, curlwp->l_cred);
872 	vput(data->ed_pack.ep_vp);
873 	pool_put(&exec_pool, data->ed_argp);
874 
875 	kmem_free(data->ed_pack.ep_hdr, data->ed_pack.ep_hdrlen);
876 	if (data->ed_pack.ep_emul_root != NULL)
877 		vrele(data->ed_pack.ep_emul_root);
878 	if (data->ed_pack.ep_interp != NULL)
879 		vrele(data->ed_pack.ep_interp);
880 
881 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
882 	pathbuf_destroy(data->ed_pathbuf);
883 	PNBUF_PUT(data->ed_resolvedpathbuf);
884 }
885 
886 static int
887 execve_runproc(struct lwp *l, struct execve_data * restrict data,
888 	bool no_local_exec_lock, bool is_spawn)
889 {
890 	int error = 0;
891 	struct proc		*p;
892 	size_t			i;
893 	char			*stack, *dp;
894 	const char		*commandname;
895 	struct ps_strings32	arginfo32;
896 	struct exec_vmcmd	*base_vcp;
897 	void			*aip;
898 	struct vmspace		*vm;
899 	ksiginfo_t		ksi;
900 	ksiginfoq_t		kq;
901 
902 	/*
903 	 * In case of a posix_spawn operation, the child doing the exec
904 	 * might not hold the reader lock on exec_lock, but the parent
905 	 * will do this instead.
906 	 */
907 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
908 	KASSERT(data != NULL);
909 	if (data == NULL)
910 		return EINVAL;
911 
912 	p = l->l_proc;
913 	if (no_local_exec_lock)
914 		KASSERT(is_spawn);
915 
916 	base_vcp = NULL;
917 
918 	if (data->ed_pack.ep_flags & EXEC_32)
919 		aip = &arginfo32;
920 	else
921 		aip = &data->ed_arginfo;
922 
923 	/* Get rid of other LWPs. */
924 	if (p->p_nlwps > 1) {
925 		mutex_enter(p->p_lock);
926 		exit_lwps(l);
927 		mutex_exit(p->p_lock);
928 	}
929 	KDASSERT(p->p_nlwps == 1);
930 
931 	/* Destroy any lwpctl info. */
932 	if (p->p_lwpctl != NULL)
933 		lwp_ctl_exit();
934 
935 	/* Remove POSIX timers */
936 	timers_free(p, TIMERS_POSIX);
937 
938 	/*
939 	 * Do whatever is necessary to prepare the address space
940 	 * for remapping.  Note that this might replace the current
941 	 * vmspace with another!
942 	 */
943 	if (is_spawn)
944 		uvmspace_spawn(l, data->ed_pack.ep_vm_minaddr,
945 		    data->ed_pack.ep_vm_maxaddr,
946 		    data->ed_pack.ep_flags & EXEC_TOPDOWN_VM);
947 	else
948 		uvmspace_exec(l, data->ed_pack.ep_vm_minaddr,
949 		    data->ed_pack.ep_vm_maxaddr,
950 		    data->ed_pack.ep_flags & EXEC_TOPDOWN_VM);
951 
952 	/* record proc's vnode, for use by procfs and others */
953 	if (p->p_textvp)
954 		vrele(p->p_textvp);
955 	vref(data->ed_pack.ep_vp);
956 	p->p_textvp = data->ed_pack.ep_vp;
957 
958 	/* Now map address space */
959 	vm = p->p_vmspace;
960 	vm->vm_taddr = (void *)data->ed_pack.ep_taddr;
961 	vm->vm_tsize = btoc(data->ed_pack.ep_tsize);
962 	vm->vm_daddr = (void*)data->ed_pack.ep_daddr;
963 	vm->vm_dsize = btoc(data->ed_pack.ep_dsize);
964 	vm->vm_ssize = btoc(data->ed_pack.ep_ssize);
965 	vm->vm_issize = 0;
966 	vm->vm_maxsaddr = (void *)data->ed_pack.ep_maxsaddr;
967 	vm->vm_minsaddr = (void *)data->ed_pack.ep_minsaddr;
968 
969 #ifdef PAX_ASLR
970 	pax_aslr_init(l, vm);
971 #endif /* PAX_ASLR */
972 
973 	/* create the new process's VM space by running the vmcmds */
974 #ifdef DIAGNOSTIC
975 	if (data->ed_pack.ep_vmcmds.evs_used == 0)
976 		panic("%s: no vmcmds", __func__);
977 #endif
978 
979 #ifdef DEBUG_EXEC
980 	{
981 		size_t j;
982 		struct exec_vmcmd *vp = &data->ed_pack.ep_vmcmds.evs_cmds[0];
983 		DPRINTF(("vmcmds %u\n", data->ed_pack.ep_vmcmds.evs_used));
984 		for (j = 0; j < data->ed_pack.ep_vmcmds.evs_used; j++) {
985 			DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
986 			    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
987 			    PRIxVSIZE" prot=0%o flags=%d\n", j,
988 			    vp[j].ev_proc == vmcmd_map_pagedvn ?
989 			    "pagedvn" :
990 			    vp[j].ev_proc == vmcmd_map_readvn ?
991 			    "readvn" :
992 			    vp[j].ev_proc == vmcmd_map_zero ?
993 			    "zero" : "*unknown*",
994 			    vp[j].ev_addr, vp[j].ev_len,
995 			    vp[j].ev_offset, vp[j].ev_prot,
996 			    vp[j].ev_flags));
997 		}
998 	}
999 #endif	/* DEBUG_EXEC */
1000 
1001 	for (i = 0; i < data->ed_pack.ep_vmcmds.evs_used && !error; i++) {
1002 		struct exec_vmcmd *vcp;
1003 
1004 		vcp = &data->ed_pack.ep_vmcmds.evs_cmds[i];
1005 		if (vcp->ev_flags & VMCMD_RELATIVE) {
1006 #ifdef DIAGNOSTIC
1007 			if (base_vcp == NULL)
1008 				panic("%s: relative vmcmd with no base",
1009 				    __func__);
1010 			if (vcp->ev_flags & VMCMD_BASE)
1011 				panic("%s: illegal base & relative vmcmd",
1012 				    __func__);
1013 #endif
1014 			vcp->ev_addr += base_vcp->ev_addr;
1015 		}
1016 		error = (*vcp->ev_proc)(l, vcp);
1017 #ifdef DEBUG_EXEC
1018 		if (error) {
1019 			size_t j;
1020 			struct exec_vmcmd *vp =
1021 			    &data->ed_pack.ep_vmcmds.evs_cmds[0];
1022 			DPRINTF(("vmcmds %zu/%u, error %d\n", i,
1023 			    data->ed_pack.ep_vmcmds.evs_used, error));
1024 			for (j = 0; j < data->ed_pack.ep_vmcmds.evs_used; j++) {
1025 				DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
1026 				    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
1027 				    PRIxVSIZE" prot=0%o flags=%d\n", j,
1028 				    vp[j].ev_proc == vmcmd_map_pagedvn ?
1029 				    "pagedvn" :
1030 				    vp[j].ev_proc == vmcmd_map_readvn ?
1031 				    "readvn" :
1032 				    vp[j].ev_proc == vmcmd_map_zero ?
1033 				    "zero" : "*unknown*",
1034 				    vp[j].ev_addr, vp[j].ev_len,
1035 				    vp[j].ev_offset, vp[j].ev_prot,
1036 				    vp[j].ev_flags));
1037 				if (j == i)
1038 					DPRINTF(("     ^--- failed\n"));
1039 			}
1040 		}
1041 #endif /* DEBUG_EXEC */
1042 		if (vcp->ev_flags & VMCMD_BASE)
1043 			base_vcp = vcp;
1044 	}
1045 
1046 	/* free the vmspace-creation commands, and release their references */
1047 	kill_vmcmds(&data->ed_pack.ep_vmcmds);
1048 
1049 	vn_lock(data->ed_pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
1050 	VOP_CLOSE(data->ed_pack.ep_vp, FREAD, l->l_cred);
1051 	vput(data->ed_pack.ep_vp);
1052 
1053 	/* if an error happened, deallocate and punt */
1054 	if (error) {
1055 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
1056 		goto exec_abort;
1057 	}
1058 
1059 	/* remember information about the process */
1060 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1061 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1062 
1063 	/* set command name & other accounting info */
1064 	commandname = strrchr(data->ed_pack.ep_resolvedname, '/');
1065 	if (commandname != NULL) {
1066 		commandname++;
1067 	} else {
1068 		commandname = data->ed_pack.ep_resolvedname;
1069 	}
1070 	i = min(strlen(commandname), MAXCOMLEN);
1071 	(void)memcpy(p->p_comm, commandname, i);
1072 	p->p_comm[i] = '\0';
1073 
1074 	dp = PNBUF_GET();
1075 	/*
1076 	 * If the path starts with /, we don't need to do any work.
1077 	 * This handles the majority of the cases.
1078 	 * In the future perhaps we could canonicalize it?
1079 	 */
1080 	if (data->ed_pathstring[0] == '/')
1081 		(void)strlcpy(data->ed_pack.ep_path = dp, data->ed_pathstring,
1082 		    MAXPATHLEN);
1083 #ifdef notyet
1084 	/*
1085 	 * Although this works most of the time [since the entry was just
1086 	 * entered in the cache] we don't use it because it will fail for
1087 	 * entries that are not placed in the cache because their name is
1088 	 * longer than NCHNAMLEN and it is not the cleanest interface,
1089 	 * because there could be races. When the namei cache is re-written,
1090 	 * this can be changed to use the appropriate function.
1091 	 */
1092 	else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p)))
1093 		data->ed_pack.ep_path = dp;
1094 #endif
1095 	else {
1096 #ifdef notyet
1097 		printf("Cannot get path for pid %d [%s] (error %d)\n",
1098 		    (int)p->p_pid, p->p_comm, error);
1099 #endif
1100 		data->ed_pack.ep_path = NULL;
1101 		PNBUF_PUT(dp);
1102 	}
1103 
1104 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
1105 		STACK_PTHREADSPACE + data->ed_ps_strings_sz + data->ed_szsigcode),
1106 		data->ed_pack.ep_ssize - (data->ed_ps_strings_sz + data->ed_szsigcode));
1107 
1108 #ifdef __MACHINE_STACK_GROWS_UP
1109 	/*
1110 	 * The copyargs call always copies into lower addresses
1111 	 * first, moving towards higher addresses, starting with
1112 	 * the stack pointer that we give.  When the stack grows
1113 	 * down, this puts argc/argv/envp very shallow on the
1114 	 * stack, right at the first user stack pointer.
1115 	 * When the stack grows up, the situation is reversed.
1116 	 *
1117 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
1118 	 * function expects to be called with a single pointer to
1119 	 * a region that has a few words it can stash values into,
1120 	 * followed by argc/argv/envp.  When the stack grows down,
1121 	 * it's easy to decrement the stack pointer a little bit to
1122 	 * allocate the space for these few words and pass the new
1123 	 * stack pointer to _rtld.  When the stack grows up, however,
1124 	 * a few words before argc is part of the signal trampoline, XXX
1125 	 * so we have a problem.
1126 	 *
1127 	 * Instead of changing how _rtld works, we take the easy way
1128 	 * out and steal 32 bytes before we call copyargs.
1129 	 * This extra space was allowed for when 'pack.ep_ssize' was calculated.
1130 	 */
1131 	stack += RTLD_GAP;
1132 #endif /* __MACHINE_STACK_GROWS_UP */
1133 
1134 	/* Now copy argc, args & environ to new stack */
1135 	error = (*data->ed_pack.ep_esch->es_copyargs)(l, &data->ed_pack,
1136 	    &data->ed_arginfo, &stack, data->ed_argp);
1137 
1138 	if (data->ed_pack.ep_path) {
1139 		PNBUF_PUT(data->ed_pack.ep_path);
1140 		data->ed_pack.ep_path = NULL;
1141 	}
1142 	if (error) {
1143 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1144 		goto exec_abort;
1145 	}
1146 	/* Move the stack back to original point */
1147 	stack = (char *)STACK_GROW(vm->vm_minsaddr, data->ed_pack.ep_ssize);
1148 
1149 	/* fill process ps_strings info */
1150 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
1151 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1152 
1153 	if (data->ed_pack.ep_flags & EXEC_32) {
1154 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1155 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1156 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1157 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1158 	}
1159 
1160 	/* copy out the process's ps_strings structure */
1161 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1162 	    != 0) {
1163 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1164 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1165 		goto exec_abort;
1166 	}
1167 
1168 	cwdexec(p);
1169 	fd_closeexec();		/* handle close on exec */
1170 
1171 	if (__predict_false(ktrace_on))
1172 		fd_ktrexecfd();
1173 
1174 	execsigs(p);		/* reset catched signals */
1175 
1176 	l->l_ctxlink = NULL;	/* reset ucontext link */
1177 
1178 
1179 	p->p_acflag &= ~AFORK;
1180 	mutex_enter(p->p_lock);
1181 	p->p_flag |= PK_EXEC;
1182 	mutex_exit(p->p_lock);
1183 
1184 	/*
1185 	 * Stop profiling.
1186 	 */
1187 	if ((p->p_stflag & PST_PROFIL) != 0) {
1188 		mutex_spin_enter(&p->p_stmutex);
1189 		stopprofclock(p);
1190 		mutex_spin_exit(&p->p_stmutex);
1191 	}
1192 
1193 	/*
1194 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1195 	 * exited and exec()/exit() are the only places it will be cleared.
1196 	 */
1197 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1198 #if 0
1199 		lwp_t *lp;
1200 
1201 		mutex_enter(proc_lock);
1202 		lp = p->p_vforklwp;
1203 		p->p_vforklwp = NULL;
1204 
1205 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1206 		p->p_lflag &= ~PL_PPWAIT;
1207 
1208 		lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
1209 		cv_broadcast(&lp->l_waitcv);
1210 		mutex_exit(proc_lock);
1211 #else
1212 		mutex_enter(proc_lock);
1213 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1214 		p->p_lflag &= ~PL_PPWAIT;
1215 		cv_broadcast(&p->p_pptr->p_waitcv);
1216 		mutex_exit(proc_lock);
1217 #endif
1218 	}
1219 
1220 	/*
1221 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
1222 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
1223 	 * out additional references on the process for the moment.
1224 	 */
1225 	if ((p->p_slflag & PSL_TRACED) == 0 &&
1226 
1227 	    (((data->ed_attr.va_mode & S_ISUID) != 0 &&
1228 	      kauth_cred_geteuid(l->l_cred) != data->ed_attr.va_uid) ||
1229 
1230 	     ((data->ed_attr.va_mode & S_ISGID) != 0 &&
1231 	      kauth_cred_getegid(l->l_cred) != data->ed_attr.va_gid))) {
1232 		/*
1233 		 * Mark the process as SUGID before we do
1234 		 * anything that might block.
1235 		 */
1236 		proc_crmod_enter();
1237 		proc_crmod_leave(NULL, NULL, true);
1238 
1239 		/* Make sure file descriptors 0..2 are in use. */
1240 		if ((error = fd_checkstd()) != 0) {
1241 			DPRINTF(("%s: fdcheckstd failed %d\n",
1242 			    __func__, error));
1243 			goto exec_abort;
1244 		}
1245 
1246 		/*
1247 		 * Copy the credential so other references don't see our
1248 		 * changes.
1249 		 */
1250 		l->l_cred = kauth_cred_copy(l->l_cred);
1251 #ifdef KTRACE
1252 		/*
1253 		 * If the persistent trace flag isn't set, turn off.
1254 		 */
1255 		if (p->p_tracep) {
1256 			mutex_enter(&ktrace_lock);
1257 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1258 				ktrderef(p);
1259 			mutex_exit(&ktrace_lock);
1260 		}
1261 #endif
1262 		if (data->ed_attr.va_mode & S_ISUID)
1263 			kauth_cred_seteuid(l->l_cred, data->ed_attr.va_uid);
1264 		if (data->ed_attr.va_mode & S_ISGID)
1265 			kauth_cred_setegid(l->l_cred, data->ed_attr.va_gid);
1266 	} else {
1267 		if (kauth_cred_geteuid(l->l_cred) ==
1268 		    kauth_cred_getuid(l->l_cred) &&
1269 		    kauth_cred_getegid(l->l_cred) ==
1270 		    kauth_cred_getgid(l->l_cred))
1271 			p->p_flag &= ~PK_SUGID;
1272 	}
1273 
1274 	/*
1275 	 * Copy the credential so other references don't see our changes.
1276 	 * Test to see if this is necessary first, since in the common case
1277 	 * we won't need a private reference.
1278 	 */
1279 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1280 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1281 		l->l_cred = kauth_cred_copy(l->l_cred);
1282 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1283 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1284 	}
1285 
1286 	/* Update the master credentials. */
1287 	if (l->l_cred != p->p_cred) {
1288 		kauth_cred_t ocred;
1289 
1290 		kauth_cred_hold(l->l_cred);
1291 		mutex_enter(p->p_lock);
1292 		ocred = p->p_cred;
1293 		p->p_cred = l->l_cred;
1294 		mutex_exit(p->p_lock);
1295 		kauth_cred_free(ocred);
1296 	}
1297 
1298 #if defined(__HAVE_RAS)
1299 	/*
1300 	 * Remove all RASs from the address space.
1301 	 */
1302 	ras_purgeall();
1303 #endif
1304 
1305 	doexechooks(p);
1306 
1307 	/* setup new registers and do misc. setup. */
1308 	(*data->ed_pack.ep_esch->es_emul->e_setregs)(l, &data->ed_pack,
1309 	     (vaddr_t)stack);
1310 	if (data->ed_pack.ep_esch->es_setregs)
1311 		(*data->ed_pack.ep_esch->es_setregs)(l, &data->ed_pack,
1312 		    (vaddr_t)stack);
1313 
1314 	/* Provide a consistent LWP private setting */
1315 	(void)lwp_setprivate(l, NULL);
1316 
1317 	/* Discard all PCU state; need to start fresh */
1318 	pcu_discard_all(l);
1319 
1320 	/* map the process's signal trampoline code */
1321 	if ((error = exec_sigcode_map(p, data->ed_pack.ep_esch->es_emul)) != 0) {
1322 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1323 		goto exec_abort;
1324 	}
1325 
1326 	pool_put(&exec_pool, data->ed_argp);
1327 
1328 	/* notify others that we exec'd */
1329 	KNOTE(&p->p_klist, NOTE_EXEC);
1330 
1331 	kmem_free(data->ed_pack.ep_hdr, data->ed_pack.ep_hdrlen);
1332 
1333 	SDT_PROBE(proc,,,exec_success, data->ed_pack.ep_name, 0, 0, 0, 0);
1334 
1335 	/* The emulation root will usually have been found when we looked
1336 	 * for the elf interpreter (or similar), if not look now. */
1337 	if (data->ed_pack.ep_esch->es_emul->e_path != NULL &&
1338 	    data->ed_pack.ep_emul_root == NULL)
1339 		emul_find_root(l, &data->ed_pack);
1340 
1341 	/* Any old emulation root got removed by fdcloseexec */
1342 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1343 	p->p_cwdi->cwdi_edir = data->ed_pack.ep_emul_root;
1344 	rw_exit(&p->p_cwdi->cwdi_lock);
1345 	data->ed_pack.ep_emul_root = NULL;
1346 	if (data->ed_pack.ep_interp != NULL)
1347 		vrele(data->ed_pack.ep_interp);
1348 
1349 	/*
1350 	 * Call emulation specific exec hook. This can setup per-process
1351 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1352 	 *
1353 	 * If we are executing process of different emulation than the
1354 	 * original forked process, call e_proc_exit() of the old emulation
1355 	 * first, then e_proc_exec() of new emulation. If the emulation is
1356 	 * same, the exec hook code should deallocate any old emulation
1357 	 * resources held previously by this process.
1358 	 */
1359 	if (p->p_emul && p->p_emul->e_proc_exit
1360 	    && p->p_emul != data->ed_pack.ep_esch->es_emul)
1361 		(*p->p_emul->e_proc_exit)(p);
1362 
1363 	/*
1364 	 * This is now LWP 1.
1365 	 */
1366 	mutex_enter(p->p_lock);
1367 	p->p_nlwpid = 1;
1368 	l->l_lid = 1;
1369 	mutex_exit(p->p_lock);
1370 
1371 	/*
1372 	 * Call exec hook. Emulation code may NOT store reference to anything
1373 	 * from &pack.
1374 	 */
1375 	if (data->ed_pack.ep_esch->es_emul->e_proc_exec)
1376 		(*data->ed_pack.ep_esch->es_emul->e_proc_exec)(p, &data->ed_pack);
1377 
1378 	/* update p_emul, the old value is no longer needed */
1379 	p->p_emul = data->ed_pack.ep_esch->es_emul;
1380 
1381 	/* ...and the same for p_execsw */
1382 	p->p_execsw = data->ed_pack.ep_esch;
1383 
1384 #ifdef __HAVE_SYSCALL_INTERN
1385 	(*p->p_emul->e_syscall_intern)(p);
1386 #endif
1387 	ktremul();
1388 
1389 	/* Allow new references from the debugger/procfs. */
1390 	rw_exit(&p->p_reflock);
1391 	if (!no_local_exec_lock)
1392 		rw_exit(&exec_lock);
1393 
1394 	mutex_enter(proc_lock);
1395 
1396 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1397 		KSI_INIT_EMPTY(&ksi);
1398 		ksi.ksi_signo = SIGTRAP;
1399 		ksi.ksi_lid = l->l_lid;
1400 		kpsignal(p, &ksi, NULL);
1401 	}
1402 
1403 	if (p->p_sflag & PS_STOPEXEC) {
1404 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1405 		p->p_pptr->p_nstopchild++;
1406 		p->p_pptr->p_waited = 0;
1407 		mutex_enter(p->p_lock);
1408 		ksiginfo_queue_init(&kq);
1409 		sigclearall(p, &contsigmask, &kq);
1410 		lwp_lock(l);
1411 		l->l_stat = LSSTOP;
1412 		p->p_stat = SSTOP;
1413 		p->p_nrlwps--;
1414 		lwp_unlock(l);
1415 		mutex_exit(p->p_lock);
1416 		mutex_exit(proc_lock);
1417 		lwp_lock(l);
1418 		mi_switch(l);
1419 		ksiginfo_queue_drain(&kq);
1420 		KERNEL_LOCK(l->l_biglocks, l);
1421 	} else {
1422 		mutex_exit(proc_lock);
1423 	}
1424 
1425 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1426 	pathbuf_destroy(data->ed_pathbuf);
1427 	PNBUF_PUT(data->ed_resolvedpathbuf);
1428 	DPRINTF(("%s finished\n", __func__));
1429 	return EJUSTRETURN;
1430 
1431  exec_abort:
1432 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
1433 	rw_exit(&p->p_reflock);
1434 	if (!no_local_exec_lock)
1435 		rw_exit(&exec_lock);
1436 
1437 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1438 	pathbuf_destroy(data->ed_pathbuf);
1439 	PNBUF_PUT(data->ed_resolvedpathbuf);
1440 
1441 	/*
1442 	 * the old process doesn't exist anymore.  exit gracefully.
1443 	 * get rid of the (new) address space we have created, if any, get rid
1444 	 * of our namei data and vnode, and exit noting failure
1445 	 */
1446 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1447 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1448 
1449 	exec_free_emul_arg(&data->ed_pack);
1450 	pool_put(&exec_pool, data->ed_argp);
1451 	kmem_free(data->ed_pack.ep_hdr, data->ed_pack.ep_hdrlen);
1452 	if (data->ed_pack.ep_emul_root != NULL)
1453 		vrele(data->ed_pack.ep_emul_root);
1454 	if (data->ed_pack.ep_interp != NULL)
1455 		vrele(data->ed_pack.ep_interp);
1456 
1457 	/* Acquire the sched-state mutex (exit1() will release it). */
1458 	if (!is_spawn) {
1459 		mutex_enter(p->p_lock);
1460 		exit1(l, W_EXITCODE(error, SIGABRT));
1461 	}
1462 
1463 	return error;
1464 }
1465 
1466 int
1467 execve1(struct lwp *l, const char *path, char * const *args,
1468     char * const *envs, execve_fetch_element_t fetch_element)
1469 {
1470 	struct execve_data data;
1471 	int error;
1472 
1473 	error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1474 	if (error)
1475 		return error;
1476 	error = execve_runproc(l, &data, false, false);
1477 	return error;
1478 }
1479 
1480 int
1481 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1482     char **stackp, void *argp)
1483 {
1484 	char	**cpp, *dp, *sp;
1485 	size_t	len;
1486 	void	*nullp;
1487 	long	argc, envc;
1488 	int	error;
1489 
1490 	cpp = (char **)*stackp;
1491 	nullp = NULL;
1492 	argc = arginfo->ps_nargvstr;
1493 	envc = arginfo->ps_nenvstr;
1494 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1495 		COPYPRINTF("", cpp - 1, sizeof(argc));
1496 		return error;
1497 	}
1498 
1499 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen);
1500 	sp = argp;
1501 
1502 	/* XXX don't copy them out, remap them! */
1503 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1504 
1505 	for (; --argc >= 0; sp += len, dp += len) {
1506 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1507 			COPYPRINTF("", cpp - 1, sizeof(dp));
1508 			return error;
1509 		}
1510 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1511 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1512 			return error;
1513 		}
1514 	}
1515 
1516 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1517 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1518 		return error;
1519 	}
1520 
1521 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1522 
1523 	for (; --envc >= 0; sp += len, dp += len) {
1524 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1525 			COPYPRINTF("", cpp - 1, sizeof(dp));
1526 			return error;
1527 		}
1528 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1529 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1530 			return error;
1531 		}
1532 
1533 	}
1534 
1535 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1536 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1537 		return error;
1538 	}
1539 
1540 	*stackp = (char *)cpp;
1541 	return 0;
1542 }
1543 
1544 
1545 /*
1546  * Add execsw[] entries.
1547  */
1548 int
1549 exec_add(struct execsw *esp, int count)
1550 {
1551 	struct exec_entry	*it;
1552 	int			i;
1553 
1554 	if (count == 0) {
1555 		return 0;
1556 	}
1557 
1558 	/* Check for duplicates. */
1559 	rw_enter(&exec_lock, RW_WRITER);
1560 	for (i = 0; i < count; i++) {
1561 		LIST_FOREACH(it, &ex_head, ex_list) {
1562 			/* assume unique (makecmds, probe_func, emulation) */
1563 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1564 			    it->ex_sw->u.elf_probe_func ==
1565 			    esp[i].u.elf_probe_func &&
1566 			    it->ex_sw->es_emul == esp[i].es_emul) {
1567 				rw_exit(&exec_lock);
1568 				return EEXIST;
1569 			}
1570 		}
1571 	}
1572 
1573 	/* Allocate new entries. */
1574 	for (i = 0; i < count; i++) {
1575 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1576 		it->ex_sw = &esp[i];
1577 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1578 	}
1579 
1580 	/* update execsw[] */
1581 	exec_init(0);
1582 	rw_exit(&exec_lock);
1583 	return 0;
1584 }
1585 
1586 /*
1587  * Remove execsw[] entry.
1588  */
1589 int
1590 exec_remove(struct execsw *esp, int count)
1591 {
1592 	struct exec_entry	*it, *next;
1593 	int			i;
1594 	const struct proclist_desc *pd;
1595 	proc_t			*p;
1596 
1597 	if (count == 0) {
1598 		return 0;
1599 	}
1600 
1601 	/* Abort if any are busy. */
1602 	rw_enter(&exec_lock, RW_WRITER);
1603 	for (i = 0; i < count; i++) {
1604 		mutex_enter(proc_lock);
1605 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1606 			PROCLIST_FOREACH(p, pd->pd_list) {
1607 				if (p->p_execsw == &esp[i]) {
1608 					mutex_exit(proc_lock);
1609 					rw_exit(&exec_lock);
1610 					return EBUSY;
1611 				}
1612 			}
1613 		}
1614 		mutex_exit(proc_lock);
1615 	}
1616 
1617 	/* None are busy, so remove them all. */
1618 	for (i = 0; i < count; i++) {
1619 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1620 			next = LIST_NEXT(it, ex_list);
1621 			if (it->ex_sw == &esp[i]) {
1622 				LIST_REMOVE(it, ex_list);
1623 				kmem_free(it, sizeof(*it));
1624 				break;
1625 			}
1626 		}
1627 	}
1628 
1629 	/* update execsw[] */
1630 	exec_init(0);
1631 	rw_exit(&exec_lock);
1632 	return 0;
1633 }
1634 
1635 /*
1636  * Initialize exec structures. If init_boot is true, also does necessary
1637  * one-time initialization (it's called from main() that way).
1638  * Once system is multiuser, this should be called with exec_lock held,
1639  * i.e. via exec_{add|remove}().
1640  */
1641 int
1642 exec_init(int init_boot)
1643 {
1644 	const struct execsw 	**sw;
1645 	struct exec_entry	*ex;
1646 	SLIST_HEAD(,exec_entry)	first;
1647 	SLIST_HEAD(,exec_entry)	any;
1648 	SLIST_HEAD(,exec_entry)	last;
1649 	int			i, sz;
1650 
1651 	if (init_boot) {
1652 		/* do one-time initializations */
1653 		rw_init(&exec_lock);
1654 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1655 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1656 		    "execargs", &exec_palloc, IPL_NONE);
1657 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1658 	} else {
1659 		KASSERT(rw_write_held(&exec_lock));
1660 	}
1661 
1662 	/* Sort each entry onto the appropriate queue. */
1663 	SLIST_INIT(&first);
1664 	SLIST_INIT(&any);
1665 	SLIST_INIT(&last);
1666 	sz = 0;
1667 	LIST_FOREACH(ex, &ex_head, ex_list) {
1668 		switch(ex->ex_sw->es_prio) {
1669 		case EXECSW_PRIO_FIRST:
1670 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1671 			break;
1672 		case EXECSW_PRIO_ANY:
1673 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1674 			break;
1675 		case EXECSW_PRIO_LAST:
1676 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1677 			break;
1678 		default:
1679 			panic("%s", __func__);
1680 			break;
1681 		}
1682 		sz++;
1683 	}
1684 
1685 	/*
1686 	 * Create new execsw[].  Ensure we do not try a zero-sized
1687 	 * allocation.
1688 	 */
1689 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1690 	i = 0;
1691 	SLIST_FOREACH(ex, &first, ex_slist) {
1692 		sw[i++] = ex->ex_sw;
1693 	}
1694 	SLIST_FOREACH(ex, &any, ex_slist) {
1695 		sw[i++] = ex->ex_sw;
1696 	}
1697 	SLIST_FOREACH(ex, &last, ex_slist) {
1698 		sw[i++] = ex->ex_sw;
1699 	}
1700 
1701 	/* Replace old execsw[] and free used memory. */
1702 	if (execsw != NULL) {
1703 		kmem_free(__UNCONST(execsw),
1704 		    nexecs * sizeof(struct execsw *) + 1);
1705 	}
1706 	execsw = sw;
1707 	nexecs = sz;
1708 
1709 	/* Figure out the maximum size of an exec header. */
1710 	exec_maxhdrsz = sizeof(int);
1711 	for (i = 0; i < nexecs; i++) {
1712 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1713 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1714 	}
1715 
1716 	return 0;
1717 }
1718 
1719 static int
1720 exec_sigcode_map(struct proc *p, const struct emul *e)
1721 {
1722 	vaddr_t va;
1723 	vsize_t sz;
1724 	int error;
1725 	struct uvm_object *uobj;
1726 
1727 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1728 
1729 	if (e->e_sigobject == NULL || sz == 0) {
1730 		return 0;
1731 	}
1732 
1733 	/*
1734 	 * If we don't have a sigobject for this emulation, create one.
1735 	 *
1736 	 * sigobject is an anonymous memory object (just like SYSV shared
1737 	 * memory) that we keep a permanent reference to and that we map
1738 	 * in all processes that need this sigcode. The creation is simple,
1739 	 * we create an object, add a permanent reference to it, map it in
1740 	 * kernel space, copy out the sigcode to it and unmap it.
1741 	 * We map it with PROT_READ|PROT_EXEC into the process just
1742 	 * the way sys_mmap() would map it.
1743 	 */
1744 
1745 	uobj = *e->e_sigobject;
1746 	if (uobj == NULL) {
1747 		mutex_enter(&sigobject_lock);
1748 		if ((uobj = *e->e_sigobject) == NULL) {
1749 			uobj = uao_create(sz, 0);
1750 			(*uobj->pgops->pgo_reference)(uobj);
1751 			va = vm_map_min(kernel_map);
1752 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1753 			    uobj, 0, 0,
1754 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1755 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1756 				printf("kernel mapping failed %d\n", error);
1757 				(*uobj->pgops->pgo_detach)(uobj);
1758 				mutex_exit(&sigobject_lock);
1759 				return error;
1760 			}
1761 			memcpy((void *)va, e->e_sigcode, sz);
1762 #ifdef PMAP_NEED_PROCWR
1763 			pmap_procwr(&proc0, va, sz);
1764 #endif
1765 			uvm_unmap(kernel_map, va, va + round_page(sz));
1766 			*e->e_sigobject = uobj;
1767 		}
1768 		mutex_exit(&sigobject_lock);
1769 	}
1770 
1771 	/* Just a hint to uvm_map where to put it. */
1772 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1773 	    round_page(sz));
1774 
1775 #ifdef __alpha__
1776 	/*
1777 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1778 	 * which causes the above calculation to put the sigcode at
1779 	 * an invalid address.  Put it just below the text instead.
1780 	 */
1781 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1782 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1783 	}
1784 #endif
1785 
1786 	(*uobj->pgops->pgo_reference)(uobj);
1787 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1788 			uobj, 0, 0,
1789 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1790 				    UVM_ADV_RANDOM, 0));
1791 	if (error) {
1792 		DPRINTF(("%s, %d: map %p "
1793 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1794 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1795 		    va, error));
1796 		(*uobj->pgops->pgo_detach)(uobj);
1797 		return error;
1798 	}
1799 	p->p_sigctx.ps_sigcode = (void *)va;
1800 	return 0;
1801 }
1802 
1803 /*
1804  * Release a refcount on spawn_exec_data and destroy memory, if this
1805  * was the last one.
1806  */
1807 static void
1808 spawn_exec_data_release(struct spawn_exec_data *data)
1809 {
1810 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1811 		return;
1812 
1813 	cv_destroy(&data->sed_cv_child_ready);
1814 	mutex_destroy(&data->sed_mtx_child);
1815 
1816 	if (data->sed_actions)
1817 		posix_spawn_fa_free(data->sed_actions,
1818 		    data->sed_actions->len);
1819 	if (data->sed_attrs)
1820 		kmem_free(data->sed_attrs,
1821 		    sizeof(*data->sed_attrs));
1822 	kmem_free(data, sizeof(*data));
1823 }
1824 
1825 /*
1826  * A child lwp of a posix_spawn operation starts here and ends up in
1827  * cpu_spawn_return, dealing with all filedescriptor and scheduler
1828  * manipulations in between.
1829  * The parent waits for the child, as it is not clear whether the child
1830  * will be able to acquire its own exec_lock. If it can, the parent can
1831  * be released early and continue running in parallel. If not (or if the
1832  * magic debug flag is passed in the scheduler attribute struct), the
1833  * child rides on the parent's exec lock until it is ready to return to
1834  * to userland - and only then releases the parent. This method loses
1835  * concurrency, but improves error reporting.
1836  */
1837 static void
1838 spawn_return(void *arg)
1839 {
1840 	struct spawn_exec_data *spawn_data = arg;
1841 	struct lwp *l = curlwp;
1842 	int error, newfd;
1843 	size_t i;
1844 	const struct posix_spawn_file_actions_entry *fae;
1845 	pid_t ppid;
1846 	register_t retval;
1847 	bool have_reflock;
1848 	bool parent_is_waiting = true;
1849 
1850 	/*
1851 	 * Check if we can release parent early.
1852 	 * We either need to have no sed_attrs, or sed_attrs does not
1853 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
1854 	 * safe access to the parent proc (passed in sed_parent).
1855 	 * We then try to get the exec_lock, and only if that works, we can
1856 	 * release the parent here already.
1857 	 */
1858 	ppid = spawn_data->sed_parent->p_pid;
1859 	if ((!spawn_data->sed_attrs
1860 	    || (spawn_data->sed_attrs->sa_flags
1861 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
1862 	    && rw_tryenter(&exec_lock, RW_READER)) {
1863 		parent_is_waiting = false;
1864 		mutex_enter(&spawn_data->sed_mtx_child);
1865 		cv_signal(&spawn_data->sed_cv_child_ready);
1866 		mutex_exit(&spawn_data->sed_mtx_child);
1867 	}
1868 
1869 	/* don't allow debugger access yet */
1870 	rw_enter(&l->l_proc->p_reflock, RW_WRITER);
1871 	have_reflock = true;
1872 
1873 	error = 0;
1874 	/* handle posix_spawn_file_actions */
1875 	if (spawn_data->sed_actions != NULL) {
1876 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
1877 			fae = &spawn_data->sed_actions->fae[i];
1878 			switch (fae->fae_action) {
1879 			case FAE_OPEN:
1880 				if (fd_getfile(fae->fae_fildes) != NULL) {
1881 					error = fd_close(fae->fae_fildes);
1882 					if (error)
1883 						break;
1884 				}
1885 				error = fd_open(fae->fae_path, fae->fae_oflag,
1886 				    fae->fae_mode, &newfd);
1887 				if (error)
1888 					break;
1889 				if (newfd != fae->fae_fildes) {
1890 					error = dodup(l, newfd,
1891 					    fae->fae_fildes, 0, &retval);
1892 					if (fd_getfile(newfd) != NULL)
1893 						fd_close(newfd);
1894 				}
1895 				break;
1896 			case FAE_DUP2:
1897 				error = dodup(l, fae->fae_fildes,
1898 				    fae->fae_newfildes, 0, &retval);
1899 				break;
1900 			case FAE_CLOSE:
1901 				if (fd_getfile(fae->fae_fildes) == NULL) {
1902 					error = EBADF;
1903 					break;
1904 				}
1905 				error = fd_close(fae->fae_fildes);
1906 				break;
1907 			}
1908 			if (error)
1909 				goto report_error;
1910 		}
1911 	}
1912 
1913 	/* handle posix_spawnattr */
1914 	if (spawn_data->sed_attrs != NULL) {
1915 		int ostat;
1916 		struct sigaction sigact;
1917 		sigact._sa_u._sa_handler = SIG_DFL;
1918 		sigact.sa_flags = 0;
1919 
1920 		/*
1921 		 * set state to SSTOP so that this proc can be found by pid.
1922 		 * see proc_enterprp, do_sched_setparam below
1923 		 */
1924 		ostat = l->l_proc->p_stat;
1925 		l->l_proc->p_stat = SSTOP;
1926 
1927 		/* Set process group */
1928 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
1929 			pid_t mypid = l->l_proc->p_pid,
1930 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
1931 
1932 			if (pgrp == 0)
1933 				pgrp = mypid;
1934 
1935 			error = proc_enterpgrp(spawn_data->sed_parent,
1936 			    mypid, pgrp, false);
1937 			if (error)
1938 				goto report_error;
1939 		}
1940 
1941 		/* Set scheduler policy */
1942 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
1943 			error = do_sched_setparam(l->l_proc->p_pid, 0,
1944 			    spawn_data->sed_attrs->sa_schedpolicy,
1945 			    &spawn_data->sed_attrs->sa_schedparam);
1946 		else if (spawn_data->sed_attrs->sa_flags
1947 		    & POSIX_SPAWN_SETSCHEDPARAM) {
1948 			error = do_sched_setparam(ppid, 0,
1949 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
1950 		}
1951 		if (error)
1952 			goto report_error;
1953 
1954 		/* Reset user ID's */
1955 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
1956 			error = do_setresuid(l, -1,
1957 			     kauth_cred_getgid(l->l_cred), -1,
1958 			     ID_E_EQ_R | ID_E_EQ_S);
1959 			if (error)
1960 				goto report_error;
1961 			error = do_setresuid(l, -1,
1962 			    kauth_cred_getuid(l->l_cred), -1,
1963 			    ID_E_EQ_R | ID_E_EQ_S);
1964 			if (error)
1965 				goto report_error;
1966 		}
1967 
1968 		/* Set signal masks/defaults */
1969 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
1970 			mutex_enter(l->l_proc->p_lock);
1971 			error = sigprocmask1(l, SIG_SETMASK,
1972 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
1973 			mutex_exit(l->l_proc->p_lock);
1974 			if (error)
1975 				goto report_error;
1976 		}
1977 
1978 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
1979 			/*
1980 			 * The following sigaction call is using a sigaction
1981 			 * version 0 trampoline which is in the compatibility
1982 			 * code only. This is not a problem because for SIG_DFL
1983 			 * and SIG_IGN, the trampolines are now ignored. If they
1984 			 * were not, this would be a problem because we are
1985 			 * holding the exec_lock, and the compat code needs
1986 			 * to do the same in order to replace the trampoline
1987 			 * code of the process.
1988 			 */
1989 			for (i = 1; i <= NSIG; i++) {
1990 				if (sigismember(
1991 				    &spawn_data->sed_attrs->sa_sigdefault, i))
1992 					sigaction1(l, i, &sigact, NULL, NULL,
1993 					    0);
1994 			}
1995 		}
1996 		l->l_proc->p_stat = ostat;
1997 	}
1998 
1999 	/* now do the real exec */
2000 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2001 	    true);
2002 	have_reflock = false;
2003 	if (error == EJUSTRETURN)
2004 		error = 0;
2005 	else if (error)
2006 		goto report_error;
2007 
2008 	if (parent_is_waiting) {
2009 		mutex_enter(&spawn_data->sed_mtx_child);
2010 		cv_signal(&spawn_data->sed_cv_child_ready);
2011 		mutex_exit(&spawn_data->sed_mtx_child);
2012 	}
2013 
2014 	/* release our refcount on the data */
2015 	spawn_exec_data_release(spawn_data);
2016 
2017 	/* and finally: leave to userland for the first time */
2018 	cpu_spawn_return(l);
2019 
2020 	/* NOTREACHED */
2021 	return;
2022 
2023  report_error:
2024 	if (have_reflock) {
2025 		/*
2026 		 * We have not passed through execve_runproc(),
2027 		 * which would have released the p_reflock and also
2028 		 * taken ownership of the sed_exec part of spawn_data,
2029 		 * so release/free both here.
2030 		 */
2031 		rw_exit(&l->l_proc->p_reflock);
2032 		execve_free_data(&spawn_data->sed_exec);
2033 	}
2034 
2035 	if (parent_is_waiting) {
2036 		/* pass error to parent */
2037 		mutex_enter(&spawn_data->sed_mtx_child);
2038 		spawn_data->sed_error = error;
2039 		cv_signal(&spawn_data->sed_cv_child_ready);
2040 		mutex_exit(&spawn_data->sed_mtx_child);
2041 	} else {
2042 		rw_exit(&exec_lock);
2043 	}
2044 
2045 	/* release our refcount on the data */
2046 	spawn_exec_data_release(spawn_data);
2047 
2048 	/* done, exit */
2049 	mutex_enter(l->l_proc->p_lock);
2050 	/*
2051 	 * Posix explicitly asks for an exit code of 127 if we report
2052 	 * errors from the child process - so, unfortunately, there
2053 	 * is no way to report a more exact error code.
2054 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2055 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2056 	 */
2057 	exit1(l, W_EXITCODE(127, 0));
2058 }
2059 
2060 void
2061 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2062 {
2063 
2064 	for (size_t i = 0; i < len; i++) {
2065 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2066 		if (fae->fae_action != FAE_OPEN)
2067 			continue;
2068 		kmem_free(fae->fae_path, strlen(fae->fae_path) + 1);
2069 	}
2070 	if (fa->len > 0)
2071 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2072 	kmem_free(fa, sizeof(*fa));
2073 }
2074 
2075 static int
2076 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2077     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2078 {
2079 	struct posix_spawn_file_actions *fa;
2080 	struct posix_spawn_file_actions_entry *fae;
2081 	char *pbuf = NULL;
2082 	int error;
2083 	size_t i = 0;
2084 
2085 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2086 	error = copyin(ufa, fa, sizeof(*fa));
2087 	if (error || fa->len == 0) {
2088 		kmem_free(fa, sizeof(*fa));
2089 		return error;	/* 0 if not an error, and len == 0 */
2090 	}
2091 
2092 	if (fa->len > lim) {
2093 		kmem_free(fa, sizeof(*fa));
2094 		return EINVAL;
2095 	}
2096 
2097 	fa->size = fa->len;
2098 	size_t fal = fa->len * sizeof(*fae);
2099 	fae = fa->fae;
2100 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2101 	error = copyin(fae, fa->fae, fal);
2102 	if (error)
2103 		goto out;
2104 
2105 	pbuf = PNBUF_GET();
2106 	for (; i < fa->len; i++) {
2107 		fae = &fa->fae[i];
2108 		if (fae->fae_action != FAE_OPEN)
2109 			continue;
2110 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2111 		if (error)
2112 			goto out;
2113 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2114 		memcpy(fae->fae_path, pbuf, fal);
2115 	}
2116 	PNBUF_PUT(pbuf);
2117 
2118 	*fap = fa;
2119 	return 0;
2120 out:
2121 	if (pbuf)
2122 		PNBUF_PUT(pbuf);
2123 	posix_spawn_fa_free(fa, i);
2124 	return error;
2125 }
2126 
2127 int
2128 check_posix_spawn(struct lwp *l1)
2129 {
2130 	int error, tnprocs, count;
2131 	uid_t uid;
2132 	struct proc *p1;
2133 
2134 	p1 = l1->l_proc;
2135 	uid = kauth_cred_getuid(l1->l_cred);
2136 	tnprocs = atomic_inc_uint_nv(&nprocs);
2137 
2138 	/*
2139 	 * Although process entries are dynamically created, we still keep
2140 	 * a global limit on the maximum number we will create.
2141 	 */
2142 	if (__predict_false(tnprocs >= maxproc))
2143 		error = -1;
2144 	else
2145 		error = kauth_authorize_process(l1->l_cred,
2146 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2147 
2148 	if (error) {
2149 		atomic_dec_uint(&nprocs);
2150 		return EAGAIN;
2151 	}
2152 
2153 	/*
2154 	 * Enforce limits.
2155 	 */
2156 	count = chgproccnt(uid, 1);
2157 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2158 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2159 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2160 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2161 		(void)chgproccnt(uid, -1);
2162 		atomic_dec_uint(&nprocs);
2163 		return EAGAIN;
2164 	}
2165 
2166 	return 0;
2167 }
2168 
2169 int
2170 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2171 	struct posix_spawn_file_actions *fa,
2172 	struct posix_spawnattr *sa,
2173 	char *const *argv, char *const *envp,
2174 	execve_fetch_element_t fetch)
2175 {
2176 
2177 	struct proc *p1, *p2;
2178 	struct lwp *l2;
2179 	int error;
2180 	struct spawn_exec_data *spawn_data;
2181 	vaddr_t uaddr;
2182 	pid_t pid;
2183 	bool have_exec_lock = false;
2184 
2185 	p1 = l1->l_proc;
2186 
2187 	/* Allocate and init spawn_data */
2188 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2189 	spawn_data->sed_refcnt = 1; /* only parent so far */
2190 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2191 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2192 	mutex_enter(&spawn_data->sed_mtx_child);
2193 
2194 	/*
2195 	 * Do the first part of the exec now, collect state
2196 	 * in spawn_data.
2197 	 */
2198 	error = execve_loadvm(l1, path, argv,
2199 	    envp, fetch, &spawn_data->sed_exec);
2200 	if (error == EJUSTRETURN)
2201 		error = 0;
2202 	else if (error)
2203 		goto error_exit;
2204 
2205 	have_exec_lock = true;
2206 
2207 	/*
2208 	 * Allocate virtual address space for the U-area now, while it
2209 	 * is still easy to abort the fork operation if we're out of
2210 	 * kernel virtual address space.
2211 	 */
2212 	uaddr = uvm_uarea_alloc();
2213 	if (__predict_false(uaddr == 0)) {
2214 		error = ENOMEM;
2215 		goto error_exit;
2216 	}
2217 
2218 	/*
2219 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2220 	 * replace it with its own before returning to userland
2221 	 * in the child.
2222 	 * This is a point of no return, we will have to go through
2223 	 * the child proc to properly clean it up past this point.
2224 	 */
2225 	p2 = proc_alloc();
2226 	pid = p2->p_pid;
2227 
2228 	/*
2229 	 * Make a proc table entry for the new process.
2230 	 * Start by zeroing the section of proc that is zero-initialized,
2231 	 * then copy the section that is copied directly from the parent.
2232 	 */
2233 	memset(&p2->p_startzero, 0,
2234 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2235 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2236 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2237 	p2->p_vmspace = proc0.p_vmspace;
2238 
2239 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2240 
2241 	LIST_INIT(&p2->p_lwps);
2242 	LIST_INIT(&p2->p_sigwaiters);
2243 
2244 	/*
2245 	 * Duplicate sub-structures as needed.
2246 	 * Increase reference counts on shared objects.
2247 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2248 	 * handling are important in order to keep a consistent behaviour
2249 	 * for the child after the fork.  If we are a 32-bit process, the
2250 	 * child will be too.
2251 	 */
2252 	p2->p_flag =
2253 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2254 	p2->p_emul = p1->p_emul;
2255 	p2->p_execsw = p1->p_execsw;
2256 
2257 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2258 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2259 	rw_init(&p2->p_reflock);
2260 	cv_init(&p2->p_waitcv, "wait");
2261 	cv_init(&p2->p_lwpcv, "lwpwait");
2262 
2263 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2264 
2265 	kauth_proc_fork(p1, p2);
2266 
2267 	p2->p_raslist = NULL;
2268 	p2->p_fd = fd_copy();
2269 
2270 	/* XXX racy */
2271 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2272 
2273 	p2->p_cwdi = cwdinit();
2274 
2275 	/*
2276 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2277 	 * we just need increase pl_refcnt.
2278 	 */
2279 	if (!p1->p_limit->pl_writeable) {
2280 		lim_addref(p1->p_limit);
2281 		p2->p_limit = p1->p_limit;
2282 	} else {
2283 		p2->p_limit = lim_copy(p1->p_limit);
2284 	}
2285 
2286 	p2->p_lflag = 0;
2287 	p2->p_sflag = 0;
2288 	p2->p_slflag = 0;
2289 	p2->p_pptr = p1;
2290 	p2->p_ppid = p1->p_pid;
2291 	LIST_INIT(&p2->p_children);
2292 
2293 	p2->p_aio = NULL;
2294 
2295 #ifdef KTRACE
2296 	/*
2297 	 * Copy traceflag and tracefile if enabled.
2298 	 * If not inherited, these were zeroed above.
2299 	 */
2300 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2301 		mutex_enter(&ktrace_lock);
2302 		p2->p_traceflag = p1->p_traceflag;
2303 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2304 			ktradref(p2);
2305 		mutex_exit(&ktrace_lock);
2306 	}
2307 #endif
2308 
2309 	/*
2310 	 * Create signal actions for the child process.
2311 	 */
2312 	p2->p_sigacts = sigactsinit(p1, 0);
2313 	mutex_enter(p1->p_lock);
2314 	p2->p_sflag |=
2315 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2316 	sched_proc_fork(p1, p2);
2317 	mutex_exit(p1->p_lock);
2318 
2319 	p2->p_stflag = p1->p_stflag;
2320 
2321 	/*
2322 	 * p_stats.
2323 	 * Copy parts of p_stats, and zero out the rest.
2324 	 */
2325 	p2->p_stats = pstatscopy(p1->p_stats);
2326 
2327 	/* copy over machdep flags to the new proc */
2328 	cpu_proc_fork(p1, p2);
2329 
2330 	/*
2331 	 * Prepare remaining parts of spawn data
2332 	 */
2333 	spawn_data->sed_actions = fa;
2334 	spawn_data->sed_attrs = sa;
2335 
2336 	spawn_data->sed_parent = p1;
2337 
2338 	/* create LWP */
2339 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2340 	    &l2, l1->l_class);
2341 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2342 
2343 	/*
2344 	 * Copy the credential so other references don't see our changes.
2345 	 * Test to see if this is necessary first, since in the common case
2346 	 * we won't need a private reference.
2347 	 */
2348 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2349 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2350 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2351 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2352 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2353 	}
2354 
2355 	/* Update the master credentials. */
2356 	if (l2->l_cred != p2->p_cred) {
2357 		kauth_cred_t ocred;
2358 
2359 		kauth_cred_hold(l2->l_cred);
2360 		mutex_enter(p2->p_lock);
2361 		ocred = p2->p_cred;
2362 		p2->p_cred = l2->l_cred;
2363 		mutex_exit(p2->p_lock);
2364 		kauth_cred_free(ocred);
2365 	}
2366 
2367 	*child_ok = true;
2368 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2369 #if 0
2370 	l2->l_nopreempt = 1; /* start it non-preemptable */
2371 #endif
2372 
2373 	/*
2374 	 * It's now safe for the scheduler and other processes to see the
2375 	 * child process.
2376 	 */
2377 	mutex_enter(proc_lock);
2378 
2379 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2380 		p2->p_lflag |= PL_CONTROLT;
2381 
2382 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2383 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2384 
2385 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2386 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2387 
2388 	p2->p_trace_enabled = trace_is_enabled(p2);
2389 #ifdef __HAVE_SYSCALL_INTERN
2390 	(*p2->p_emul->e_syscall_intern)(p2);
2391 #endif
2392 
2393 	/*
2394 	 * Make child runnable, set start time, and add to run queue except
2395 	 * if the parent requested the child to start in SSTOP state.
2396 	 */
2397 	mutex_enter(p2->p_lock);
2398 
2399 	getmicrotime(&p2->p_stats->p_start);
2400 
2401 	lwp_lock(l2);
2402 	KASSERT(p2->p_nrlwps == 1);
2403 	p2->p_nrlwps = 1;
2404 	p2->p_stat = SACTIVE;
2405 	l2->l_stat = LSRUN;
2406 	sched_enqueue(l2, false);
2407 	lwp_unlock(l2);
2408 
2409 	mutex_exit(p2->p_lock);
2410 	mutex_exit(proc_lock);
2411 
2412 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2413 	error = spawn_data->sed_error;
2414 	mutex_exit(&spawn_data->sed_mtx_child);
2415 	spawn_exec_data_release(spawn_data);
2416 
2417 	rw_exit(&p1->p_reflock);
2418 	rw_exit(&exec_lock);
2419 	have_exec_lock = false;
2420 
2421 	*pid_res = pid;
2422 	return error;
2423 
2424  error_exit:
2425 	if (have_exec_lock) {
2426 		execve_free_data(&spawn_data->sed_exec);
2427 		rw_exit(&p1->p_reflock);
2428 		rw_exit(&exec_lock);
2429 	}
2430 	mutex_exit(&spawn_data->sed_mtx_child);
2431 	spawn_exec_data_release(spawn_data);
2432 
2433 	return error;
2434 }
2435 
2436 int
2437 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2438     register_t *retval)
2439 {
2440 	/* {
2441 		syscallarg(pid_t *) pid;
2442 		syscallarg(const char *) path;
2443 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2444 		syscallarg(const struct posix_spawnattr *) attrp;
2445 		syscallarg(char *const *) argv;
2446 		syscallarg(char *const *) envp;
2447 	} */
2448 
2449 	int error;
2450 	struct posix_spawn_file_actions *fa = NULL;
2451 	struct posix_spawnattr *sa = NULL;
2452 	pid_t pid;
2453 	bool child_ok = false;
2454 	rlim_t max_fileactions;
2455 	proc_t *p = l1->l_proc;
2456 
2457 	error = check_posix_spawn(l1);
2458 	if (error) {
2459 		*retval = error;
2460 		return 0;
2461 	}
2462 
2463 	/* copy in file_actions struct */
2464 	if (SCARG(uap, file_actions) != NULL) {
2465 		max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2466 		    maxfiles);
2467 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2468 		    max_fileactions);
2469 		if (error)
2470 			goto error_exit;
2471 	}
2472 
2473 	/* copyin posix_spawnattr struct */
2474 	if (SCARG(uap, attrp) != NULL) {
2475 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2476 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2477 		if (error)
2478 			goto error_exit;
2479 	}
2480 
2481 	/*
2482 	 * Do the spawn
2483 	 */
2484 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2485 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2486 	if (error)
2487 		goto error_exit;
2488 
2489 	if (error == 0 && SCARG(uap, pid) != NULL)
2490 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2491 
2492 	*retval = error;
2493 	return 0;
2494 
2495  error_exit:
2496 	if (!child_ok) {
2497 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2498 		atomic_dec_uint(&nprocs);
2499 
2500 		if (sa)
2501 			kmem_free(sa, sizeof(*sa));
2502 		if (fa)
2503 			posix_spawn_fa_free(fa, fa->len);
2504 	}
2505 
2506 	*retval = error;
2507 	return 0;
2508 }
2509 
2510 void
2511 exec_free_emul_arg(struct exec_package *epp)
2512 {
2513 	if (epp->ep_emul_arg_free != NULL) {
2514 		KASSERT(epp->ep_emul_arg != NULL);
2515 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2516 		epp->ep_emul_arg_free = NULL;
2517 		epp->ep_emul_arg = NULL;
2518 	} else {
2519 		KASSERT(epp->ep_emul_arg == NULL);
2520 	}
2521 }
2522