xref: /netbsd-src/sys/kern/kern_exec.c (revision 6a493d6bc668897c91594964a732d38505b70cbb)
1 /*	$NetBSD: kern_exec.c,v 1.368 2013/12/24 14:47:04 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*-
30  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
31  * Copyright (C) 1992 Wolfgang Solfrank.
32  * Copyright (C) 1992 TooLs GmbH.
33  * All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. All advertising materials mentioning features or use of this software
44  *    must display the following acknowledgement:
45  *	This product includes software developed by TooLs GmbH.
46  * 4. The name of TooLs GmbH may not be used to endorse or promote products
47  *    derived from this software without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
50  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
51  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
52  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
53  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
54  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
55  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
56  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
57  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
58  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
59  */
60 
61 #include <sys/cdefs.h>
62 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.368 2013/12/24 14:47:04 christos 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 __USING_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 	.e_path =		NULL,
182 #ifndef __HAVE_MINIMAL_EMUL
183 	.e_flags =		EMUL_HAS_SYS___syscall,
184 	.e_errno =		NULL,
185 	.e_nosys =		SYS_syscall,
186 	.e_nsysent =		SYS_NSYSENT,
187 #endif
188 	.e_sysent =		sysent,
189 #ifdef SYSCALL_DEBUG
190 	.e_syscallnames =	syscallnames,
191 #else
192 	.e_syscallnames =	NULL,
193 #endif
194 	.e_sendsig =		sendsig,
195 	.e_trapsignal =		trapsignal,
196 	.e_tracesig =		NULL,
197 	.e_sigcode =		NULL,
198 	.e_esigcode =		NULL,
199 	.e_sigobject =		NULL,
200 	.e_setregs =		setregs,
201 	.e_proc_exec =		NULL,
202 	.e_proc_fork =		NULL,
203 	.e_proc_exit =		NULL,
204 	.e_lwp_fork =		NULL,
205 	.e_lwp_exit =		NULL,
206 #ifdef __HAVE_SYSCALL_INTERN
207 	.e_syscall_intern =	syscall_intern,
208 #else
209 	.e_syscall =		syscall,
210 #endif
211 	.e_sysctlovly =		NULL,
212 	.e_fault =		NULL,
213 	.e_vm_default_addr =	uvm_default_mapaddr,
214 	.e_usertrap =		NULL,
215 	.e_ucsize =		sizeof(ucontext_t),
216 	.e_startlwp =		startlwp
217 };
218 
219 /*
220  * Exec lock. Used to control access to execsw[] structures.
221  * This must not be static so that netbsd32 can access it, too.
222  */
223 krwlock_t exec_lock;
224 
225 static kmutex_t sigobject_lock;
226 
227 /*
228  * Data used between a loadvm and execve part of an "exec" operation
229  */
230 struct execve_data {
231 	struct exec_package	ed_pack;
232 	struct pathbuf		*ed_pathbuf;
233 	struct vattr		ed_attr;
234 	struct ps_strings	ed_arginfo;
235 	char			*ed_argp;
236 	const char		*ed_pathstring;
237 	char			*ed_resolvedpathbuf;
238 	size_t			ed_ps_strings_sz;
239 	int			ed_szsigcode;
240 	long			ed_argc;
241 	long			ed_envc;
242 };
243 
244 /*
245  * data passed from parent lwp to child during a posix_spawn()
246  */
247 struct spawn_exec_data {
248 	struct execve_data	sed_exec;
249 	struct posix_spawn_file_actions
250 				*sed_actions;
251 	struct posix_spawnattr	*sed_attrs;
252 	struct proc		*sed_parent;
253 	kcondvar_t		sed_cv_child_ready;
254 	kmutex_t		sed_mtx_child;
255 	int			sed_error;
256 	volatile uint32_t	sed_refcnt;
257 };
258 
259 static void *
260 exec_pool_alloc(struct pool *pp, int flags)
261 {
262 
263 	return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
264 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
265 }
266 
267 static void
268 exec_pool_free(struct pool *pp, void *addr)
269 {
270 
271 	uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
272 }
273 
274 static struct pool exec_pool;
275 
276 static struct pool_allocator exec_palloc = {
277 	.pa_alloc = exec_pool_alloc,
278 	.pa_free = exec_pool_free,
279 	.pa_pagesz = NCARGS
280 };
281 
282 /*
283  * check exec:
284  * given an "executable" described in the exec package's namei info,
285  * see what we can do with it.
286  *
287  * ON ENTRY:
288  *	exec package with appropriate namei info
289  *	lwp pointer of exec'ing lwp
290  *	NO SELF-LOCKED VNODES
291  *
292  * ON EXIT:
293  *	error:	nothing held, etc.  exec header still allocated.
294  *	ok:	filled exec package, executable's vnode (unlocked).
295  *
296  * EXEC SWITCH ENTRY:
297  * 	Locked vnode to check, exec package, proc.
298  *
299  * EXEC SWITCH EXIT:
300  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
301  *	error:	destructive:
302  *			everything deallocated execept exec header.
303  *		non-destructive:
304  *			error code, executable's vnode (unlocked),
305  *			exec header unmodified.
306  */
307 int
308 /*ARGSUSED*/
309 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
310 {
311 	int		error, i;
312 	struct vnode	*vp;
313 	struct nameidata nd;
314 	size_t		resid;
315 
316 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
317 
318 	/* first get the vnode */
319 	if ((error = namei(&nd)) != 0)
320 		return error;
321 	epp->ep_vp = vp = nd.ni_vp;
322 	/* normally this can't fail */
323 	if ((error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL)))
324 		goto bad1;
325 
326 #ifdef DIAGNOSTIC
327 	/* paranoia (take this out once namei stuff stabilizes) */
328 	memset(nd.ni_pnbuf, '~', PATH_MAX);
329 #endif
330 
331 	/* check access and type */
332 	if (vp->v_type != VREG) {
333 		error = EACCES;
334 		goto bad1;
335 	}
336 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
337 		goto bad1;
338 
339 	/* get attributes */
340 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
341 		goto bad1;
342 
343 	/* Check mount point */
344 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
345 		error = EACCES;
346 		goto bad1;
347 	}
348 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
349 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
350 
351 	/* try to open it */
352 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
353 		goto bad1;
354 
355 	/* unlock vp, since we need it unlocked from here on out. */
356 	VOP_UNLOCK(vp);
357 
358 #if NVERIEXEC > 0
359 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
360 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
361 	    NULL);
362 	if (error)
363 		goto bad2;
364 #endif /* NVERIEXEC > 0 */
365 
366 #ifdef PAX_SEGVGUARD
367 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
368 	if (error)
369 		goto bad2;
370 #endif /* PAX_SEGVGUARD */
371 
372 	/* now we have the file, get the exec header */
373 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
374 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
375 	if (error)
376 		goto bad2;
377 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
378 
379 	/*
380 	 * Set up default address space limits.  Can be overridden
381 	 * by individual exec packages.
382 	 *
383 	 * XXX probably should be all done in the exec packages.
384 	 */
385 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
386 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
387 	/*
388 	 * set up the vmcmds for creation of the process
389 	 * address space
390 	 */
391 	error = ENOEXEC;
392 	for (i = 0; i < nexecs; i++) {
393 		int newerror;
394 
395 		epp->ep_esch = execsw[i];
396 		newerror = (*execsw[i]->es_makecmds)(l, epp);
397 
398 		if (!newerror) {
399 			/* Seems ok: check that entry point is not too high */
400 			if (epp->ep_entry > epp->ep_vm_maxaddr) {
401 #ifdef DIAGNOSTIC
402 				printf("%s: rejecting %p due to "
403 				    "too high entry address (> %p)\n",
404 					 __func__, (void *)epp->ep_entry,
405 					 (void *)epp->ep_vm_maxaddr);
406 #endif
407 				error = ENOEXEC;
408 				break;
409 			}
410 			/* Seems ok: check that entry point is not too low */
411 			if (epp->ep_entry < epp->ep_vm_minaddr) {
412 #ifdef DIAGNOSTIC
413 				printf("%s: rejecting %p due to "
414 				    "too low entry address (< %p)\n",
415 				     __func__, (void *)epp->ep_entry,
416 				     (void *)epp->ep_vm_minaddr);
417 #endif
418 				error = ENOEXEC;
419 				break;
420 			}
421 
422 			/* check limits */
423 			if ((epp->ep_tsize > MAXTSIZ) ||
424 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
425 						    [RLIMIT_DATA].rlim_cur)) {
426 #ifdef DIAGNOSTIC
427 				printf("%s: rejecting due to "
428 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
429 				    __func__,
430 				    (unsigned long long)epp->ep_tsize,
431 				    (unsigned long long)MAXTSIZ,
432 				    (unsigned long long)epp->ep_dsize,
433 				    (unsigned long long)
434 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
435 #endif
436 				error = ENOMEM;
437 				break;
438 			}
439 			return 0;
440 		}
441 
442 		if (epp->ep_emul_root != NULL) {
443 			vrele(epp->ep_emul_root);
444 			epp->ep_emul_root = NULL;
445 		}
446 		if (epp->ep_interp != NULL) {
447 			vrele(epp->ep_interp);
448 			epp->ep_interp = NULL;
449 		}
450 
451 		/* make sure the first "interesting" error code is saved. */
452 		if (error == ENOEXEC)
453 			error = newerror;
454 
455 		if (epp->ep_flags & EXEC_DESTR)
456 			/* Error from "#!" code, tidied up by recursive call */
457 			return error;
458 	}
459 
460 	/* not found, error */
461 
462 	/*
463 	 * free any vmspace-creation commands,
464 	 * and release their references
465 	 */
466 	kill_vmcmds(&epp->ep_vmcmds);
467 
468 bad2:
469 	/*
470 	 * close and release the vnode, restore the old one, free the
471 	 * pathname buf, and punt.
472 	 */
473 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
474 	VOP_CLOSE(vp, FREAD, l->l_cred);
475 	vput(vp);
476 	return error;
477 
478 bad1:
479 	/*
480 	 * free the namei pathname buffer, and put the vnode
481 	 * (which we don't yet have open).
482 	 */
483 	vput(vp);				/* was still locked */
484 	return error;
485 }
486 
487 #ifdef __MACHINE_STACK_GROWS_UP
488 #define STACK_PTHREADSPACE NBPG
489 #else
490 #define STACK_PTHREADSPACE 0
491 #endif
492 
493 static int
494 execve_fetch_element(char * const *array, size_t index, char **value)
495 {
496 	return copyin(array + index, value, sizeof(*value));
497 }
498 
499 /*
500  * exec system call
501  */
502 int
503 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
504 {
505 	/* {
506 		syscallarg(const char *)	path;
507 		syscallarg(char * const *)	argp;
508 		syscallarg(char * const *)	envp;
509 	} */
510 
511 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
512 	    SCARG(uap, envp), execve_fetch_element);
513 }
514 
515 int
516 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
517     register_t *retval)
518 {
519 	/* {
520 		syscallarg(int)			fd;
521 		syscallarg(char * const *)	argp;
522 		syscallarg(char * const *)	envp;
523 	} */
524 
525 	return ENOSYS;
526 }
527 
528 /*
529  * Load modules to try and execute an image that we do not understand.
530  * If no execsw entries are present, we load those likely to be needed
531  * in order to run native images only.  Otherwise, we autoload all
532  * possible modules that could let us run the binary.  XXX lame
533  */
534 static void
535 exec_autoload(void)
536 {
537 #ifdef MODULAR
538 	static const char * const native[] = {
539 		"exec_elf32",
540 		"exec_elf64",
541 		"exec_script",
542 		NULL
543 	};
544 	static const char * const compat[] = {
545 		"exec_elf32",
546 		"exec_elf64",
547 		"exec_script",
548 		"exec_aout",
549 		"exec_coff",
550 		"exec_ecoff",
551 		"compat_aoutm68k",
552 		"compat_freebsd",
553 		"compat_ibcs2",
554 		"compat_linux",
555 		"compat_linux32",
556 		"compat_netbsd32",
557 		"compat_sunos",
558 		"compat_sunos32",
559 		"compat_svr4",
560 		"compat_svr4_32",
561 		"compat_ultrix",
562 		NULL
563 	};
564 	char const * const *list;
565 	int i;
566 
567 	list = (nexecs == 0 ? native : compat);
568 	for (i = 0; list[i] != NULL; i++) {
569 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
570 		    	continue;
571 		}
572 	   	yield();
573 	}
574 #endif
575 }
576 
577 static int
578 execve_loadvm(struct lwp *l, const char *path, char * const *args,
579 	char * const *envs, execve_fetch_element_t fetch_element,
580 	struct execve_data * restrict data)
581 {
582 	int			error;
583 	struct proc		*p;
584 	char			*dp, *sp;
585 	size_t			i, len;
586 	struct exec_fakearg	*tmpfap;
587 	u_int			modgen;
588 
589 	KASSERT(data != NULL);
590 
591 	p = l->l_proc;
592  	modgen = 0;
593 
594 	SDT_PROBE(proc,,,exec, path, 0, 0, 0, 0);
595 
596 	/*
597 	 * Check if we have exceeded our number of processes limit.
598 	 * This is so that we handle the case where a root daemon
599 	 * forked, ran setuid to become the desired user and is trying
600 	 * to exec. The obvious place to do the reference counting check
601 	 * is setuid(), but we don't do the reference counting check there
602 	 * like other OS's do because then all the programs that use setuid()
603 	 * must be modified to check the return code of setuid() and exit().
604 	 * It is dangerous to make setuid() fail, because it fails open and
605 	 * the program will continue to run as root. If we make it succeed
606 	 * and return an error code, again we are not enforcing the limit.
607 	 * The best place to enforce the limit is here, when the process tries
608 	 * to execute a new image, because eventually the process will need
609 	 * to call exec in order to do something useful.
610 	 */
611  retry:
612 	if (p->p_flag & PK_SUGID) {
613 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
614 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
615 		     &p->p_rlimit[RLIMIT_NPROC],
616 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
617 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
618 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
619 		return EAGAIN;
620 	}
621 
622 	/*
623 	 * Drain existing references and forbid new ones.  The process
624 	 * should be left alone until we're done here.  This is necessary
625 	 * to avoid race conditions - e.g. in ptrace() - that might allow
626 	 * a local user to illicitly obtain elevated privileges.
627 	 */
628 	rw_enter(&p->p_reflock, RW_WRITER);
629 
630 	/*
631 	 * Init the namei data to point the file user's program name.
632 	 * This is done here rather than in check_exec(), so that it's
633 	 * possible to override this settings if any of makecmd/probe
634 	 * functions call check_exec() recursively - for example,
635 	 * see exec_script_makecmds().
636 	 */
637 	error = pathbuf_copyin(path, &data->ed_pathbuf);
638 	if (error) {
639 		DPRINTF(("%s: pathbuf_copyin path @%p %d\n", __func__,
640 		    path, error));
641 		goto clrflg;
642 	}
643 	data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
644 	data->ed_resolvedpathbuf = PNBUF_GET();
645 
646 	/*
647 	 * initialize the fields of the exec package.
648 	 */
649 	data->ed_pack.ep_name = path;
650 	data->ed_pack.ep_kname = data->ed_pathstring;
651 	data->ed_pack.ep_resolvedname = data->ed_resolvedpathbuf;
652 	data->ed_pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
653 	data->ed_pack.ep_hdrlen = exec_maxhdrsz;
654 	data->ed_pack.ep_hdrvalid = 0;
655 	data->ed_pack.ep_emul_arg = NULL;
656 	data->ed_pack.ep_emul_arg_free = NULL;
657 	data->ed_pack.ep_vmcmds.evs_cnt = 0;
658 	data->ed_pack.ep_vmcmds.evs_used = 0;
659 	data->ed_pack.ep_vap = &data->ed_attr;
660 	data->ed_pack.ep_flags = 0;
661 	MD_TOPDOWN_INIT(&data->ed_pack);
662 	data->ed_pack.ep_emul_root = NULL;
663 	data->ed_pack.ep_interp = NULL;
664 	data->ed_pack.ep_esch = NULL;
665 	data->ed_pack.ep_pax_flags = 0;
666 	memset(data->ed_pack.ep_machine_arch, 0,
667 	    sizeof(data->ed_pack.ep_machine_arch));
668 
669 	rw_enter(&exec_lock, RW_READER);
670 
671 	/* see if we can run it. */
672 	if ((error = check_exec(l, &data->ed_pack, data->ed_pathbuf)) != 0) {
673 		if (error != ENOENT) {
674 			DPRINTF(("%s: check exec failed %d\n",
675 			    __func__, error));
676 		}
677 		goto freehdr;
678 	}
679 
680 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
681 
682 	/* allocate an argument buffer */
683 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
684 	KASSERT(data->ed_argp != NULL);
685 	dp = data->ed_argp;
686 	data->ed_argc = 0;
687 
688 	/* copy the fake args list, if there's one, freeing it as we go */
689 	if (data->ed_pack.ep_flags & EXEC_HASARGL) {
690 		tmpfap = data->ed_pack.ep_fa;
691 		while (tmpfap->fa_arg != NULL) {
692 			const char *cp;
693 
694 			cp = tmpfap->fa_arg;
695 			while (*cp)
696 				*dp++ = *cp++;
697 			*dp++ = '\0';
698 			ktrexecarg(tmpfap->fa_arg, cp - tmpfap->fa_arg);
699 
700 			kmem_free(tmpfap->fa_arg, tmpfap->fa_len);
701 			tmpfap++; data->ed_argc++;
702 		}
703 		kmem_free(data->ed_pack.ep_fa, data->ed_pack.ep_fa_len);
704 		data->ed_pack.ep_flags &= ~EXEC_HASARGL;
705 	}
706 
707 	/* Now get argv & environment */
708 	if (args == NULL) {
709 		DPRINTF(("%s: null args\n", __func__));
710 		error = EINVAL;
711 		goto bad;
712 	}
713 	/* 'i' will index the argp/envp element to be retrieved */
714 	i = 0;
715 	if (data->ed_pack.ep_flags & EXEC_SKIPARG)
716 		i++;
717 
718 	while (1) {
719 		len = data->ed_argp + ARG_MAX - dp;
720 		if ((error = (*fetch_element)(args, i, &sp)) != 0) {
721 			DPRINTF(("%s: fetch_element args %d\n",
722 			    __func__, error));
723 			goto bad;
724 		}
725 		if (!sp)
726 			break;
727 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
728 			DPRINTF(("%s: copyinstr args %d\n", __func__, error));
729 			if (error == ENAMETOOLONG)
730 				error = E2BIG;
731 			goto bad;
732 		}
733 		ktrexecarg(dp, len - 1);
734 		dp += len;
735 		i++;
736 		data->ed_argc++;
737 	}
738 
739 	data->ed_envc = 0;
740 	/* environment need not be there */
741 	if (envs != NULL) {
742 		i = 0;
743 		while (1) {
744 			len = data->ed_argp + ARG_MAX - dp;
745 			if ((error = (*fetch_element)(envs, i, &sp)) != 0) {
746 				DPRINTF(("%s: fetch_element env %d\n",
747 				    __func__, error));
748 				goto bad;
749 			}
750 			if (!sp)
751 				break;
752 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
753 				DPRINTF(("%s: copyinstr env %d\n",
754 				    __func__, error));
755 				if (error == ENAMETOOLONG)
756 					error = E2BIG;
757 				goto bad;
758 			}
759 
760 			ktrexecenv(dp, len - 1);
761 			dp += len;
762 			i++;
763 			data->ed_envc++;
764 		}
765 	}
766 
767 	dp = (char *) ALIGN(dp);
768 
769 	data->ed_szsigcode = data->ed_pack.ep_esch->es_emul->e_esigcode -
770 	    data->ed_pack.ep_esch->es_emul->e_sigcode;
771 
772 #ifdef __MACHINE_STACK_GROWS_UP
773 /* See big comment lower down */
774 #define	RTLD_GAP	32
775 #else
776 #define	RTLD_GAP	0
777 #endif
778 
779 	/* Now check if args & environ fit into new stack */
780 	if (data->ed_pack.ep_flags & EXEC_32) {
781 		data->ed_ps_strings_sz = sizeof(struct ps_strings32);
782 		len = ((data->ed_argc + data->ed_envc + 2 +
783 		    data->ed_pack.ep_esch->es_arglen) *
784 		    sizeof(int) + sizeof(int) + dp + RTLD_GAP +
785 		    data->ed_szsigcode + data->ed_ps_strings_sz + STACK_PTHREADSPACE)
786 		    - data->ed_argp;
787 	} else {
788 		data->ed_ps_strings_sz = sizeof(struct ps_strings);
789 		len = ((data->ed_argc + data->ed_envc + 2 +
790 		    data->ed_pack.ep_esch->es_arglen) *
791 		    sizeof(char *) + sizeof(int) + dp + RTLD_GAP +
792 		    data->ed_szsigcode + data->ed_ps_strings_sz + STACK_PTHREADSPACE)
793 		    - data->ed_argp;
794 	}
795 
796 #ifdef PAX_ASLR
797 	if (pax_aslr_active(l))
798 		len += (cprng_fast32() % PAGE_SIZE);
799 #endif /* PAX_ASLR */
800 
801 	/* make the stack "safely" aligned */
802 	len = STACK_LEN_ALIGN(len, STACK_ALIGNBYTES);
803 
804 	if (len > data->ed_pack.ep_ssize) {
805 		/* in effect, compare to initial limit */
806 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
807 		goto bad;
808 	}
809 	/* adjust "active stack depth" for process VSZ */
810 	data->ed_pack.ep_ssize = len;
811 
812 	return 0;
813 
814  bad:
815 	/* free the vmspace-creation commands, and release their references */
816 	kill_vmcmds(&data->ed_pack.ep_vmcmds);
817 	/* kill any opened file descriptor, if necessary */
818 	if (data->ed_pack.ep_flags & EXEC_HASFD) {
819 		data->ed_pack.ep_flags &= ~EXEC_HASFD;
820 		fd_close(data->ed_pack.ep_fd);
821 	}
822 	/* close and put the exec'd file */
823 	vn_lock(data->ed_pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
824 	VOP_CLOSE(data->ed_pack.ep_vp, FREAD, l->l_cred);
825 	vput(data->ed_pack.ep_vp);
826 	pool_put(&exec_pool, data->ed_argp);
827 
828  freehdr:
829 	kmem_free(data->ed_pack.ep_hdr, data->ed_pack.ep_hdrlen);
830 	if (data->ed_pack.ep_emul_root != NULL)
831 		vrele(data->ed_pack.ep_emul_root);
832 	if (data->ed_pack.ep_interp != NULL)
833 		vrele(data->ed_pack.ep_interp);
834 
835 	rw_exit(&exec_lock);
836 
837 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
838 	pathbuf_destroy(data->ed_pathbuf);
839 	PNBUF_PUT(data->ed_resolvedpathbuf);
840 
841  clrflg:
842 	rw_exit(&p->p_reflock);
843 
844 	if (modgen != module_gen && error == ENOEXEC) {
845 		modgen = module_gen;
846 		exec_autoload();
847 		goto retry;
848 	}
849 
850 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
851 	return error;
852 }
853 
854 static void
855 execve_free_data(struct execve_data *data)
856 {
857 
858 	/* free the vmspace-creation commands, and release their references */
859 	kill_vmcmds(&data->ed_pack.ep_vmcmds);
860 	/* kill any opened file descriptor, if necessary */
861 	if (data->ed_pack.ep_flags & EXEC_HASFD) {
862 		data->ed_pack.ep_flags &= ~EXEC_HASFD;
863 		fd_close(data->ed_pack.ep_fd);
864 	}
865 
866 	/* close and put the exec'd file */
867 	vn_lock(data->ed_pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
868 	VOP_CLOSE(data->ed_pack.ep_vp, FREAD, curlwp->l_cred);
869 	vput(data->ed_pack.ep_vp);
870 	pool_put(&exec_pool, data->ed_argp);
871 
872 	kmem_free(data->ed_pack.ep_hdr, data->ed_pack.ep_hdrlen);
873 	if (data->ed_pack.ep_emul_root != NULL)
874 		vrele(data->ed_pack.ep_emul_root);
875 	if (data->ed_pack.ep_interp != NULL)
876 		vrele(data->ed_pack.ep_interp);
877 
878 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
879 	pathbuf_destroy(data->ed_pathbuf);
880 	PNBUF_PUT(data->ed_resolvedpathbuf);
881 }
882 
883 static int
884 execve_runproc(struct lwp *l, struct execve_data * restrict data,
885 	bool no_local_exec_lock, bool is_spawn)
886 {
887 	int error = 0;
888 	struct proc		*p;
889 	size_t			i;
890 	char			*stack, *dp;
891 	const char		*commandname;
892 	struct ps_strings32	arginfo32;
893 	struct exec_vmcmd	*base_vcp;
894 	void			*aip;
895 	struct vmspace		*vm;
896 	ksiginfo_t		ksi;
897 	ksiginfoq_t		kq;
898 
899 	/*
900 	 * In case of a posix_spawn operation, the child doing the exec
901 	 * might not hold the reader lock on exec_lock, but the parent
902 	 * will do this instead.
903 	 */
904 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
905 	KASSERT(data != NULL);
906 	if (data == NULL)
907 		return (EINVAL);
908 
909 	p = l->l_proc;
910 	if (no_local_exec_lock)
911 		KASSERT(is_spawn);
912 
913 	base_vcp = NULL;
914 
915 	if (data->ed_pack.ep_flags & EXEC_32)
916 		aip = &arginfo32;
917 	else
918 		aip = &data->ed_arginfo;
919 
920 	/* Get rid of other LWPs. */
921 	if (p->p_nlwps > 1) {
922 		mutex_enter(p->p_lock);
923 		exit_lwps(l);
924 		mutex_exit(p->p_lock);
925 	}
926 	KDASSERT(p->p_nlwps == 1);
927 
928 	/* Destroy any lwpctl info. */
929 	if (p->p_lwpctl != NULL)
930 		lwp_ctl_exit();
931 
932 	/* Remove POSIX timers */
933 	timers_free(p, TIMERS_POSIX);
934 
935 	/*
936 	 * Do whatever is necessary to prepare the address space
937 	 * for remapping.  Note that this might replace the current
938 	 * vmspace with another!
939 	 */
940 	if (is_spawn)
941 		uvmspace_spawn(l, data->ed_pack.ep_vm_minaddr,
942 		    data->ed_pack.ep_vm_maxaddr,
943 		    data->ed_pack.ep_flags & EXEC_TOPDOWN_VM);
944 	else
945 		uvmspace_exec(l, data->ed_pack.ep_vm_minaddr,
946 		    data->ed_pack.ep_vm_maxaddr,
947 		    data->ed_pack.ep_flags & EXEC_TOPDOWN_VM);
948 
949 	/* record proc's vnode, for use by procfs and others */
950         if (p->p_textvp)
951                 vrele(p->p_textvp);
952 	vref(data->ed_pack.ep_vp);
953 	p->p_textvp = data->ed_pack.ep_vp;
954 
955 	/* Now map address space */
956 	vm = p->p_vmspace;
957 	vm->vm_taddr = (void *)data->ed_pack.ep_taddr;
958 	vm->vm_tsize = btoc(data->ed_pack.ep_tsize);
959 	vm->vm_daddr = (void*)data->ed_pack.ep_daddr;
960 	vm->vm_dsize = btoc(data->ed_pack.ep_dsize);
961 	vm->vm_ssize = btoc(data->ed_pack.ep_ssize);
962 	vm->vm_issize = 0;
963 	vm->vm_maxsaddr = (void *)data->ed_pack.ep_maxsaddr;
964 	vm->vm_minsaddr = (void *)data->ed_pack.ep_minsaddr;
965 
966 #ifdef PAX_ASLR
967 	pax_aslr_init(l, vm);
968 #endif /* PAX_ASLR */
969 
970 	/* create the new process's VM space by running the vmcmds */
971 #ifdef DIAGNOSTIC
972 	if (data->ed_pack.ep_vmcmds.evs_used == 0)
973 		panic("%s: no vmcmds", __func__);
974 #endif
975 
976 #ifdef DEBUG_EXEC
977 	{
978 		size_t j;
979 		struct exec_vmcmd *vp = &data->ed_pack.ep_vmcmds.evs_cmds[0];
980 		DPRINTF(("vmcmds %u\n", data->ed_pack.ep_vmcmds.evs_used));
981 		for (j = 0; j < data->ed_pack.ep_vmcmds.evs_used; j++) {
982 			DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
983 			    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
984 			    PRIxVSIZE" prot=0%o flags=%d\n", j,
985 			    vp[j].ev_proc == vmcmd_map_pagedvn ?
986 			    "pagedvn" :
987 			    vp[j].ev_proc == vmcmd_map_readvn ?
988 			    "readvn" :
989 			    vp[j].ev_proc == vmcmd_map_zero ?
990 			    "zero" : "*unknown*",
991 			    vp[j].ev_addr, vp[j].ev_len,
992 			    vp[j].ev_offset, vp[j].ev_prot,
993 			    vp[j].ev_flags));
994 		}
995 	}
996 #endif	/* DEBUG_EXEC */
997 
998 	for (i = 0; i < data->ed_pack.ep_vmcmds.evs_used && !error; i++) {
999 		struct exec_vmcmd *vcp;
1000 
1001 		vcp = &data->ed_pack.ep_vmcmds.evs_cmds[i];
1002 		if (vcp->ev_flags & VMCMD_RELATIVE) {
1003 #ifdef DIAGNOSTIC
1004 			if (base_vcp == NULL)
1005 				panic("%s: relative vmcmd with no base",
1006 				    __func__);
1007 			if (vcp->ev_flags & VMCMD_BASE)
1008 				panic("%s: illegal base & relative vmcmd",
1009 				    __func__);
1010 #endif
1011 			vcp->ev_addr += base_vcp->ev_addr;
1012 		}
1013 		error = (*vcp->ev_proc)(l, vcp);
1014 #ifdef DEBUG_EXEC
1015 		if (error) {
1016 			size_t j;
1017 			struct exec_vmcmd *vp =
1018 			    &data->ed_pack.ep_vmcmds.evs_cmds[0];
1019 			DPRINTF(("vmcmds %zu/%u, error %d\n", i,
1020 			    data->ed_pack.ep_vmcmds.evs_used, error));
1021 			for (j = 0; j < data->ed_pack.ep_vmcmds.evs_used; j++) {
1022 				DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
1023 				    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
1024 				    PRIxVSIZE" prot=0%o flags=%d\n", j,
1025 				    vp[j].ev_proc == vmcmd_map_pagedvn ?
1026 				    "pagedvn" :
1027 				    vp[j].ev_proc == vmcmd_map_readvn ?
1028 				    "readvn" :
1029 				    vp[j].ev_proc == vmcmd_map_zero ?
1030 				    "zero" : "*unknown*",
1031 				    vp[j].ev_addr, vp[j].ev_len,
1032 				    vp[j].ev_offset, vp[j].ev_prot,
1033 				    vp[j].ev_flags));
1034 				if (j == i)
1035 					DPRINTF(("     ^--- failed\n"));
1036 			}
1037 		}
1038 #endif /* DEBUG_EXEC */
1039 		if (vcp->ev_flags & VMCMD_BASE)
1040 			base_vcp = vcp;
1041 	}
1042 
1043 	/* free the vmspace-creation commands, and release their references */
1044 	kill_vmcmds(&data->ed_pack.ep_vmcmds);
1045 
1046 	vn_lock(data->ed_pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
1047 	VOP_CLOSE(data->ed_pack.ep_vp, FREAD, l->l_cred);
1048 	vput(data->ed_pack.ep_vp);
1049 
1050 	/* if an error happened, deallocate and punt */
1051 	if (error) {
1052 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
1053 		goto exec_abort;
1054 	}
1055 
1056 	/* remember information about the process */
1057 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1058 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1059 
1060 	/* set command name & other accounting info */
1061 	commandname = strrchr(data->ed_pack.ep_resolvedname, '/');
1062 	if (commandname != NULL) {
1063 		commandname++;
1064 	} else {
1065 		commandname = data->ed_pack.ep_resolvedname;
1066 	}
1067 	i = min(strlen(commandname), MAXCOMLEN);
1068 	(void)memcpy(p->p_comm, commandname, i);
1069 	p->p_comm[i] = '\0';
1070 
1071 	dp = PNBUF_GET();
1072 	/*
1073 	 * If the path starts with /, we don't need to do any work.
1074 	 * This handles the majority of the cases.
1075 	 * In the future perhaps we could canonicalize it?
1076 	 */
1077 	if (data->ed_pathstring[0] == '/')
1078 		(void)strlcpy(data->ed_pack.ep_path = dp, data->ed_pathstring,
1079 		    MAXPATHLEN);
1080 #ifdef notyet
1081 	/*
1082 	 * Although this works most of the time [since the entry was just
1083 	 * entered in the cache] we don't use it because it will fail for
1084 	 * entries that are not placed in the cache because their name is
1085 	 * longer than NCHNAMLEN and it is not the cleanest interface,
1086 	 * because there could be races. When the namei cache is re-written,
1087 	 * this can be changed to use the appropriate function.
1088 	 */
1089 	else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p)))
1090 		data->ed_pack.ep_path = dp;
1091 #endif
1092 	else {
1093 #ifdef notyet
1094 		printf("Cannot get path for pid %d [%s] (error %d)\n",
1095 		    (int)p->p_pid, p->p_comm, error);
1096 #endif
1097 		data->ed_pack.ep_path = NULL;
1098 		PNBUF_PUT(dp);
1099 	}
1100 
1101 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
1102 		STACK_PTHREADSPACE + data->ed_ps_strings_sz + data->ed_szsigcode),
1103 		data->ed_pack.ep_ssize - (data->ed_ps_strings_sz + data->ed_szsigcode));
1104 
1105 #ifdef __MACHINE_STACK_GROWS_UP
1106 	/*
1107 	 * The copyargs call always copies into lower addresses
1108 	 * first, moving towards higher addresses, starting with
1109 	 * the stack pointer that we give.  When the stack grows
1110 	 * down, this puts argc/argv/envp very shallow on the
1111 	 * stack, right at the first user stack pointer.
1112 	 * When the stack grows up, the situation is reversed.
1113 	 *
1114 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
1115 	 * function expects to be called with a single pointer to
1116 	 * a region that has a few words it can stash values into,
1117 	 * followed by argc/argv/envp.  When the stack grows down,
1118 	 * it's easy to decrement the stack pointer a little bit to
1119 	 * allocate the space for these few words and pass the new
1120 	 * stack pointer to _rtld.  When the stack grows up, however,
1121 	 * a few words before argc is part of the signal trampoline, XXX
1122 	 * so we have a problem.
1123 	 *
1124 	 * Instead of changing how _rtld works, we take the easy way
1125 	 * out and steal 32 bytes before we call copyargs.
1126 	 * This extra space was allowed for when 'pack.ep_ssize' was calculated.
1127 	 */
1128 	stack += RTLD_GAP;
1129 #endif /* __MACHINE_STACK_GROWS_UP */
1130 
1131 	/* Now copy argc, args & environ to new stack */
1132 	error = (*data->ed_pack.ep_esch->es_copyargs)(l, &data->ed_pack,
1133 	    &data->ed_arginfo, &stack, data->ed_argp);
1134 
1135 	if (data->ed_pack.ep_path) {
1136 		PNBUF_PUT(data->ed_pack.ep_path);
1137 		data->ed_pack.ep_path = NULL;
1138 	}
1139 	if (error) {
1140 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1141 		goto exec_abort;
1142 	}
1143 	/* Move the stack back to original point */
1144 	stack = (char *)STACK_GROW(vm->vm_minsaddr, data->ed_pack.ep_ssize);
1145 
1146 	/* fill process ps_strings info */
1147 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
1148 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1149 
1150 	if (data->ed_pack.ep_flags & EXEC_32) {
1151 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1152 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1153 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1154 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1155 	}
1156 
1157 	/* copy out the process's ps_strings structure */
1158 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1159 	    != 0) {
1160 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1161 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1162 		goto exec_abort;
1163 	}
1164 
1165 	cwdexec(p);
1166 	fd_closeexec();		/* handle close on exec */
1167 
1168 	if (__predict_false(ktrace_on))
1169 		fd_ktrexecfd();
1170 
1171 	execsigs(p);		/* reset catched signals */
1172 
1173 	l->l_ctxlink = NULL;	/* reset ucontext link */
1174 
1175 
1176 	p->p_acflag &= ~AFORK;
1177 	mutex_enter(p->p_lock);
1178 	p->p_flag |= PK_EXEC;
1179 	mutex_exit(p->p_lock);
1180 
1181 	/*
1182 	 * Stop profiling.
1183 	 */
1184 	if ((p->p_stflag & PST_PROFIL) != 0) {
1185 		mutex_spin_enter(&p->p_stmutex);
1186 		stopprofclock(p);
1187 		mutex_spin_exit(&p->p_stmutex);
1188 	}
1189 
1190 	/*
1191 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1192 	 * exited and exec()/exit() are the only places it will be cleared.
1193 	 */
1194 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1195 #if 0
1196 		lwp_t *lp;
1197 
1198 		mutex_enter(proc_lock);
1199 		lp = p->p_vforklwp;
1200 		p->p_vforklwp = NULL;
1201 
1202 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1203 		p->p_lflag &= ~PL_PPWAIT;
1204 
1205 		lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
1206 		cv_broadcast(&lp->l_waitcv);
1207 		mutex_exit(proc_lock);
1208 #else
1209 		mutex_enter(proc_lock);
1210 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1211 		p->p_lflag &= ~PL_PPWAIT;
1212 		cv_broadcast(&p->p_pptr->p_waitcv);
1213 		mutex_exit(proc_lock);
1214 #endif
1215 	}
1216 
1217 	/*
1218 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
1219 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
1220 	 * out additional references on the process for the moment.
1221 	 */
1222 	if ((p->p_slflag & PSL_TRACED) == 0 &&
1223 
1224 	    (((data->ed_attr.va_mode & S_ISUID) != 0 &&
1225 	      kauth_cred_geteuid(l->l_cred) != data->ed_attr.va_uid) ||
1226 
1227 	     ((data->ed_attr.va_mode & S_ISGID) != 0 &&
1228 	      kauth_cred_getegid(l->l_cred) != data->ed_attr.va_gid))) {
1229 		/*
1230 		 * Mark the process as SUGID before we do
1231 		 * anything that might block.
1232 		 */
1233 		proc_crmod_enter();
1234 		proc_crmod_leave(NULL, NULL, true);
1235 
1236 		/* Make sure file descriptors 0..2 are in use. */
1237 		if ((error = fd_checkstd()) != 0) {
1238 			DPRINTF(("%s: fdcheckstd failed %d\n",
1239 			    __func__, error));
1240 			goto exec_abort;
1241 		}
1242 
1243 		/*
1244 		 * Copy the credential so other references don't see our
1245 		 * changes.
1246 		 */
1247 		l->l_cred = kauth_cred_copy(l->l_cred);
1248 #ifdef KTRACE
1249 		/*
1250 		 * If the persistent trace flag isn't set, turn off.
1251 		 */
1252 		if (p->p_tracep) {
1253 			mutex_enter(&ktrace_lock);
1254 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1255 				ktrderef(p);
1256 			mutex_exit(&ktrace_lock);
1257 		}
1258 #endif
1259 		if (data->ed_attr.va_mode & S_ISUID)
1260 			kauth_cred_seteuid(l->l_cred, data->ed_attr.va_uid);
1261 		if (data->ed_attr.va_mode & S_ISGID)
1262 			kauth_cred_setegid(l->l_cred, data->ed_attr.va_gid);
1263 	} else {
1264 		if (kauth_cred_geteuid(l->l_cred) ==
1265 		    kauth_cred_getuid(l->l_cred) &&
1266 		    kauth_cred_getegid(l->l_cred) ==
1267 		    kauth_cred_getgid(l->l_cred))
1268 			p->p_flag &= ~PK_SUGID;
1269 	}
1270 
1271 	/*
1272 	 * Copy the credential so other references don't see our changes.
1273 	 * Test to see if this is necessary first, since in the common case
1274 	 * we won't need a private reference.
1275 	 */
1276 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1277 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1278 		l->l_cred = kauth_cred_copy(l->l_cred);
1279 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1280 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1281 	}
1282 
1283 	/* Update the master credentials. */
1284 	if (l->l_cred != p->p_cred) {
1285 		kauth_cred_t ocred;
1286 
1287 		kauth_cred_hold(l->l_cred);
1288 		mutex_enter(p->p_lock);
1289 		ocred = p->p_cred;
1290 		p->p_cred = l->l_cred;
1291 		mutex_exit(p->p_lock);
1292 		kauth_cred_free(ocred);
1293 	}
1294 
1295 #if defined(__HAVE_RAS)
1296 	/*
1297 	 * Remove all RASs from the address space.
1298 	 */
1299 	ras_purgeall();
1300 #endif
1301 
1302 	doexechooks(p);
1303 
1304 	/* setup new registers and do misc. setup. */
1305 	(*data->ed_pack.ep_esch->es_emul->e_setregs)(l, &data->ed_pack,
1306 	     (vaddr_t)stack);
1307 	if (data->ed_pack.ep_esch->es_setregs)
1308 		(*data->ed_pack.ep_esch->es_setregs)(l, &data->ed_pack,
1309 		    (vaddr_t)stack);
1310 
1311 	/* Provide a consistent LWP private setting */
1312 	(void)lwp_setprivate(l, NULL);
1313 
1314 	/* Discard all PCU state; need to start fresh */
1315 	pcu_discard_all(l);
1316 
1317 	/* map the process's signal trampoline code */
1318 	if ((error = exec_sigcode_map(p, data->ed_pack.ep_esch->es_emul)) != 0) {
1319 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1320 		goto exec_abort;
1321 	}
1322 
1323 	pool_put(&exec_pool, data->ed_argp);
1324 
1325 	/* notify others that we exec'd */
1326 	KNOTE(&p->p_klist, NOTE_EXEC);
1327 
1328 	kmem_free(data->ed_pack.ep_hdr, data->ed_pack.ep_hdrlen);
1329 
1330 	SDT_PROBE(proc,,,exec_success, data->ed_pack.ep_name, 0, 0, 0, 0);
1331 
1332 	/* The emulation root will usually have been found when we looked
1333 	 * for the elf interpreter (or similar), if not look now. */
1334 	if (data->ed_pack.ep_esch->es_emul->e_path != NULL &&
1335 	    data->ed_pack.ep_emul_root == NULL)
1336 		emul_find_root(l, &data->ed_pack);
1337 
1338 	/* Any old emulation root got removed by fdcloseexec */
1339 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1340 	p->p_cwdi->cwdi_edir = data->ed_pack.ep_emul_root;
1341 	rw_exit(&p->p_cwdi->cwdi_lock);
1342 	data->ed_pack.ep_emul_root = NULL;
1343 	if (data->ed_pack.ep_interp != NULL)
1344 		vrele(data->ed_pack.ep_interp);
1345 
1346 	/*
1347 	 * Call emulation specific exec hook. This can setup per-process
1348 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1349 	 *
1350 	 * If we are executing process of different emulation than the
1351 	 * original forked process, call e_proc_exit() of the old emulation
1352 	 * first, then e_proc_exec() of new emulation. If the emulation is
1353 	 * same, the exec hook code should deallocate any old emulation
1354 	 * resources held previously by this process.
1355 	 */
1356 	if (p->p_emul && p->p_emul->e_proc_exit
1357 	    && p->p_emul != data->ed_pack.ep_esch->es_emul)
1358 		(*p->p_emul->e_proc_exit)(p);
1359 
1360 	/*
1361 	 * This is now LWP 1.
1362 	 */
1363 	mutex_enter(p->p_lock);
1364 	p->p_nlwpid = 1;
1365 	l->l_lid = 1;
1366 	mutex_exit(p->p_lock);
1367 
1368 	/*
1369 	 * Call exec hook. Emulation code may NOT store reference to anything
1370 	 * from &pack.
1371 	 */
1372 	if (data->ed_pack.ep_esch->es_emul->e_proc_exec)
1373 		(*data->ed_pack.ep_esch->es_emul->e_proc_exec)(p, &data->ed_pack);
1374 
1375 	/* update p_emul, the old value is no longer needed */
1376 	p->p_emul = data->ed_pack.ep_esch->es_emul;
1377 
1378 	/* ...and the same for p_execsw */
1379 	p->p_execsw = data->ed_pack.ep_esch;
1380 
1381 #ifdef __HAVE_SYSCALL_INTERN
1382 	(*p->p_emul->e_syscall_intern)(p);
1383 #endif
1384 	ktremul();
1385 
1386 	/* Allow new references from the debugger/procfs. */
1387 	rw_exit(&p->p_reflock);
1388 	if (!no_local_exec_lock)
1389 		rw_exit(&exec_lock);
1390 
1391 	mutex_enter(proc_lock);
1392 
1393 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1394 		KSI_INIT_EMPTY(&ksi);
1395 		ksi.ksi_signo = SIGTRAP;
1396 		ksi.ksi_lid = l->l_lid;
1397 		kpsignal(p, &ksi, NULL);
1398 	}
1399 
1400 	if (p->p_sflag & PS_STOPEXEC) {
1401 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1402 		p->p_pptr->p_nstopchild++;
1403 		p->p_pptr->p_waited = 0;
1404 		mutex_enter(p->p_lock);
1405 		ksiginfo_queue_init(&kq);
1406 		sigclearall(p, &contsigmask, &kq);
1407 		lwp_lock(l);
1408 		l->l_stat = LSSTOP;
1409 		p->p_stat = SSTOP;
1410 		p->p_nrlwps--;
1411 		lwp_unlock(l);
1412 		mutex_exit(p->p_lock);
1413 		mutex_exit(proc_lock);
1414 		lwp_lock(l);
1415 		mi_switch(l);
1416 		ksiginfo_queue_drain(&kq);
1417 		KERNEL_LOCK(l->l_biglocks, l);
1418 	} else {
1419 		mutex_exit(proc_lock);
1420 	}
1421 
1422 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1423 	pathbuf_destroy(data->ed_pathbuf);
1424 	PNBUF_PUT(data->ed_resolvedpathbuf);
1425 	DPRINTF(("%s finished\n", __func__));
1426 	return (EJUSTRETURN);
1427 
1428  exec_abort:
1429 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
1430 	rw_exit(&p->p_reflock);
1431 	if (!no_local_exec_lock)
1432 		rw_exit(&exec_lock);
1433 
1434 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1435 	pathbuf_destroy(data->ed_pathbuf);
1436 	PNBUF_PUT(data->ed_resolvedpathbuf);
1437 
1438 	/*
1439 	 * the old process doesn't exist anymore.  exit gracefully.
1440 	 * get rid of the (new) address space we have created, if any, get rid
1441 	 * of our namei data and vnode, and exit noting failure
1442 	 */
1443 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1444 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1445 
1446 	exec_free_emul_arg(&data->ed_pack);
1447 	pool_put(&exec_pool, data->ed_argp);
1448 	kmem_free(data->ed_pack.ep_hdr, data->ed_pack.ep_hdrlen);
1449 	if (data->ed_pack.ep_emul_root != NULL)
1450 		vrele(data->ed_pack.ep_emul_root);
1451 	if (data->ed_pack.ep_interp != NULL)
1452 		vrele(data->ed_pack.ep_interp);
1453 
1454 	/* Acquire the sched-state mutex (exit1() will release it). */
1455 	if (!is_spawn) {
1456 		mutex_enter(p->p_lock);
1457 		exit1(l, W_EXITCODE(error, SIGABRT));
1458 	}
1459 
1460 	return error;
1461 }
1462 
1463 int
1464 execve1(struct lwp *l, const char *path, char * const *args,
1465     char * const *envs, execve_fetch_element_t fetch_element)
1466 {
1467 	struct execve_data data;
1468 	int error;
1469 
1470 	error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1471 	if (error)
1472 		return error;
1473 	error = execve_runproc(l, &data, false, false);
1474 	return error;
1475 }
1476 
1477 int
1478 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1479     char **stackp, void *argp)
1480 {
1481 	char	**cpp, *dp, *sp;
1482 	size_t	len;
1483 	void	*nullp;
1484 	long	argc, envc;
1485 	int	error;
1486 
1487 	cpp = (char **)*stackp;
1488 	nullp = NULL;
1489 	argc = arginfo->ps_nargvstr;
1490 	envc = arginfo->ps_nenvstr;
1491 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1492 		COPYPRINTF("", cpp - 1, sizeof(argc));
1493 		return error;
1494 	}
1495 
1496 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen);
1497 	sp = argp;
1498 
1499 	/* XXX don't copy them out, remap them! */
1500 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1501 
1502 	for (; --argc >= 0; sp += len, dp += len) {
1503 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1504 			COPYPRINTF("", cpp - 1, sizeof(dp));
1505 			return error;
1506 		}
1507 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1508 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1509 			return error;
1510 		}
1511 	}
1512 
1513 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1514 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1515 		return error;
1516 	}
1517 
1518 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1519 
1520 	for (; --envc >= 0; sp += len, dp += len) {
1521 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1522 			COPYPRINTF("", cpp - 1, sizeof(dp));
1523 			return error;
1524 		}
1525 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1526 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1527 			return error;
1528 		}
1529 
1530 	}
1531 
1532 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1533 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1534 		return error;
1535 	}
1536 
1537 	*stackp = (char *)cpp;
1538 	return 0;
1539 }
1540 
1541 
1542 /*
1543  * Add execsw[] entries.
1544  */
1545 int
1546 exec_add(struct execsw *esp, int count)
1547 {
1548 	struct exec_entry	*it;
1549 	int			i;
1550 
1551 	if (count == 0) {
1552 		return 0;
1553 	}
1554 
1555 	/* Check for duplicates. */
1556 	rw_enter(&exec_lock, RW_WRITER);
1557 	for (i = 0; i < count; i++) {
1558 		LIST_FOREACH(it, &ex_head, ex_list) {
1559 			/* assume unique (makecmds, probe_func, emulation) */
1560 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1561 			    it->ex_sw->u.elf_probe_func ==
1562 			    esp[i].u.elf_probe_func &&
1563 			    it->ex_sw->es_emul == esp[i].es_emul) {
1564 				rw_exit(&exec_lock);
1565 				return EEXIST;
1566 			}
1567 		}
1568 	}
1569 
1570 	/* Allocate new entries. */
1571 	for (i = 0; i < count; i++) {
1572 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1573 		it->ex_sw = &esp[i];
1574 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1575 	}
1576 
1577 	/* update execsw[] */
1578 	exec_init(0);
1579 	rw_exit(&exec_lock);
1580 	return 0;
1581 }
1582 
1583 /*
1584  * Remove execsw[] entry.
1585  */
1586 int
1587 exec_remove(struct execsw *esp, int count)
1588 {
1589 	struct exec_entry	*it, *next;
1590 	int			i;
1591 	const struct proclist_desc *pd;
1592 	proc_t			*p;
1593 
1594 	if (count == 0) {
1595 		return 0;
1596 	}
1597 
1598 	/* Abort if any are busy. */
1599 	rw_enter(&exec_lock, RW_WRITER);
1600 	for (i = 0; i < count; i++) {
1601 		mutex_enter(proc_lock);
1602 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1603 			PROCLIST_FOREACH(p, pd->pd_list) {
1604 				if (p->p_execsw == &esp[i]) {
1605 					mutex_exit(proc_lock);
1606 					rw_exit(&exec_lock);
1607 					return EBUSY;
1608 				}
1609 			}
1610 		}
1611 		mutex_exit(proc_lock);
1612 	}
1613 
1614 	/* None are busy, so remove them all. */
1615 	for (i = 0; i < count; i++) {
1616 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1617 			next = LIST_NEXT(it, ex_list);
1618 			if (it->ex_sw == &esp[i]) {
1619 				LIST_REMOVE(it, ex_list);
1620 				kmem_free(it, sizeof(*it));
1621 				break;
1622 			}
1623 		}
1624 	}
1625 
1626 	/* update execsw[] */
1627 	exec_init(0);
1628 	rw_exit(&exec_lock);
1629 	return 0;
1630 }
1631 
1632 /*
1633  * Initialize exec structures. If init_boot is true, also does necessary
1634  * one-time initialization (it's called from main() that way).
1635  * Once system is multiuser, this should be called with exec_lock held,
1636  * i.e. via exec_{add|remove}().
1637  */
1638 int
1639 exec_init(int init_boot)
1640 {
1641 	const struct execsw 	**sw;
1642 	struct exec_entry	*ex;
1643 	SLIST_HEAD(,exec_entry)	first;
1644 	SLIST_HEAD(,exec_entry)	any;
1645 	SLIST_HEAD(,exec_entry)	last;
1646 	int			i, sz;
1647 
1648 	if (init_boot) {
1649 		/* do one-time initializations */
1650 		rw_init(&exec_lock);
1651 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1652 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1653 		    "execargs", &exec_palloc, IPL_NONE);
1654 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1655 	} else {
1656 		KASSERT(rw_write_held(&exec_lock));
1657 	}
1658 
1659 	/* Sort each entry onto the appropriate queue. */
1660 	SLIST_INIT(&first);
1661 	SLIST_INIT(&any);
1662 	SLIST_INIT(&last);
1663 	sz = 0;
1664 	LIST_FOREACH(ex, &ex_head, ex_list) {
1665 		switch(ex->ex_sw->es_prio) {
1666 		case EXECSW_PRIO_FIRST:
1667 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1668 			break;
1669 		case EXECSW_PRIO_ANY:
1670 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1671 			break;
1672 		case EXECSW_PRIO_LAST:
1673 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1674 			break;
1675 		default:
1676 			panic("%s", __func__);
1677 			break;
1678 		}
1679 		sz++;
1680 	}
1681 
1682 	/*
1683 	 * Create new execsw[].  Ensure we do not try a zero-sized
1684 	 * allocation.
1685 	 */
1686 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1687 	i = 0;
1688 	SLIST_FOREACH(ex, &first, ex_slist) {
1689 		sw[i++] = ex->ex_sw;
1690 	}
1691 	SLIST_FOREACH(ex, &any, ex_slist) {
1692 		sw[i++] = ex->ex_sw;
1693 	}
1694 	SLIST_FOREACH(ex, &last, ex_slist) {
1695 		sw[i++] = ex->ex_sw;
1696 	}
1697 
1698 	/* Replace old execsw[] and free used memory. */
1699 	if (execsw != NULL) {
1700 		kmem_free(__UNCONST(execsw),
1701 		    nexecs * sizeof(struct execsw *) + 1);
1702 	}
1703 	execsw = sw;
1704 	nexecs = sz;
1705 
1706 	/* Figure out the maximum size of an exec header. */
1707 	exec_maxhdrsz = sizeof(int);
1708 	for (i = 0; i < nexecs; i++) {
1709 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1710 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1711 	}
1712 
1713 	return 0;
1714 }
1715 
1716 static int
1717 exec_sigcode_map(struct proc *p, const struct emul *e)
1718 {
1719 	vaddr_t va;
1720 	vsize_t sz;
1721 	int error;
1722 	struct uvm_object *uobj;
1723 
1724 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1725 
1726 	if (e->e_sigobject == NULL || sz == 0) {
1727 		return 0;
1728 	}
1729 
1730 	/*
1731 	 * If we don't have a sigobject for this emulation, create one.
1732 	 *
1733 	 * sigobject is an anonymous memory object (just like SYSV shared
1734 	 * memory) that we keep a permanent reference to and that we map
1735 	 * in all processes that need this sigcode. The creation is simple,
1736 	 * we create an object, add a permanent reference to it, map it in
1737 	 * kernel space, copy out the sigcode to it and unmap it.
1738 	 * We map it with PROT_READ|PROT_EXEC into the process just
1739 	 * the way sys_mmap() would map it.
1740 	 */
1741 
1742 	uobj = *e->e_sigobject;
1743 	if (uobj == NULL) {
1744 		mutex_enter(&sigobject_lock);
1745 		if ((uobj = *e->e_sigobject) == NULL) {
1746 			uobj = uao_create(sz, 0);
1747 			(*uobj->pgops->pgo_reference)(uobj);
1748 			va = vm_map_min(kernel_map);
1749 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1750 			    uobj, 0, 0,
1751 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1752 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1753 				printf("kernel mapping failed %d\n", error);
1754 				(*uobj->pgops->pgo_detach)(uobj);
1755 				mutex_exit(&sigobject_lock);
1756 				return (error);
1757 			}
1758 			memcpy((void *)va, e->e_sigcode, sz);
1759 #ifdef PMAP_NEED_PROCWR
1760 			pmap_procwr(&proc0, va, sz);
1761 #endif
1762 			uvm_unmap(kernel_map, va, va + round_page(sz));
1763 			*e->e_sigobject = uobj;
1764 		}
1765 		mutex_exit(&sigobject_lock);
1766 	}
1767 
1768 	/* Just a hint to uvm_map where to put it. */
1769 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1770 	    round_page(sz));
1771 
1772 #ifdef __alpha__
1773 	/*
1774 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1775 	 * which causes the above calculation to put the sigcode at
1776 	 * an invalid address.  Put it just below the text instead.
1777 	 */
1778 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1779 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1780 	}
1781 #endif
1782 
1783 	(*uobj->pgops->pgo_reference)(uobj);
1784 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1785 			uobj, 0, 0,
1786 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1787 				    UVM_ADV_RANDOM, 0));
1788 	if (error) {
1789 		DPRINTF(("%s, %d: map %p "
1790 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1791 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1792 		    va, error));
1793 		(*uobj->pgops->pgo_detach)(uobj);
1794 		return (error);
1795 	}
1796 	p->p_sigctx.ps_sigcode = (void *)va;
1797 	return (0);
1798 }
1799 
1800 /*
1801  * Release a refcount on spawn_exec_data and destroy memory, if this
1802  * was the last one.
1803  */
1804 static void
1805 spawn_exec_data_release(struct spawn_exec_data *data)
1806 {
1807 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1808 		return;
1809 
1810 	cv_destroy(&data->sed_cv_child_ready);
1811 	mutex_destroy(&data->sed_mtx_child);
1812 
1813 	if (data->sed_actions)
1814 		posix_spawn_fa_free(data->sed_actions,
1815 		    data->sed_actions->len);
1816 	if (data->sed_attrs)
1817 		kmem_free(data->sed_attrs,
1818 		    sizeof(*data->sed_attrs));
1819 	kmem_free(data, sizeof(*data));
1820 }
1821 
1822 /*
1823  * A child lwp of a posix_spawn operation starts here and ends up in
1824  * cpu_spawn_return, dealing with all filedescriptor and scheduler
1825  * manipulations in between.
1826  * The parent waits for the child, as it is not clear wether the child
1827  * will be able to aquire its own exec_lock. If it can, the parent can
1828  * be released early and continue running in parallel. If not (or if the
1829  * magic debug flag is passed in the scheduler attribute struct), the
1830  * child rides on the parent's exec lock untill it is ready to return to
1831  * to userland - and only then releases the parent. This method loses
1832  * concurrency, but improves error reporting.
1833  */
1834 static void
1835 spawn_return(void *arg)
1836 {
1837 	struct spawn_exec_data *spawn_data = arg;
1838 	struct lwp *l = curlwp;
1839 	int error, newfd;
1840 	size_t i;
1841 	const struct posix_spawn_file_actions_entry *fae;
1842 	pid_t ppid;
1843 	register_t retval;
1844 	bool have_reflock;
1845 	bool parent_is_waiting = true;
1846 
1847 	/*
1848 	 * Check if we can release parent early.
1849 	 * We either need to have no sed_attrs, or sed_attrs does not
1850 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
1851 	 * safe access to the parent proc (passed in sed_parent).
1852 	 * We then try to get the exec_lock, and only if that works, we can
1853 	 * release the parent here already.
1854 	 */
1855 	ppid = spawn_data->sed_parent->p_pid;
1856 	if ((!spawn_data->sed_attrs
1857 	    || (spawn_data->sed_attrs->sa_flags
1858 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
1859 	    && rw_tryenter(&exec_lock, RW_READER)) {
1860 		parent_is_waiting = false;
1861 		mutex_enter(&spawn_data->sed_mtx_child);
1862 		cv_signal(&spawn_data->sed_cv_child_ready);
1863 		mutex_exit(&spawn_data->sed_mtx_child);
1864 	}
1865 
1866 	/* don't allow debugger access yet */
1867 	rw_enter(&l->l_proc->p_reflock, RW_WRITER);
1868 	have_reflock = true;
1869 
1870 	error = 0;
1871 	/* handle posix_spawn_file_actions */
1872 	if (spawn_data->sed_actions != NULL) {
1873 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
1874 			fae = &spawn_data->sed_actions->fae[i];
1875 			switch (fae->fae_action) {
1876 			case FAE_OPEN:
1877 				if (fd_getfile(fae->fae_fildes) != NULL) {
1878 					error = fd_close(fae->fae_fildes);
1879 					if (error)
1880 						break;
1881 				}
1882 				error = fd_open(fae->fae_path, fae->fae_oflag,
1883 				    fae->fae_mode, &newfd);
1884  				if (error)
1885  					break;
1886 				if (newfd != fae->fae_fildes) {
1887 					error = dodup(l, newfd,
1888 					    fae->fae_fildes, 0, &retval);
1889 					if (fd_getfile(newfd) != NULL)
1890 						fd_close(newfd);
1891 				}
1892 				break;
1893 			case FAE_DUP2:
1894 				error = dodup(l, fae->fae_fildes,
1895 				    fae->fae_newfildes, 0, &retval);
1896 				break;
1897 			case FAE_CLOSE:
1898 				if (fd_getfile(fae->fae_fildes) == NULL) {
1899 					error = EBADF;
1900 					break;
1901 				}
1902 				error = fd_close(fae->fae_fildes);
1903 				break;
1904 			}
1905 			if (error)
1906 				goto report_error;
1907 		}
1908 	}
1909 
1910 	/* handle posix_spawnattr */
1911 	if (spawn_data->sed_attrs != NULL) {
1912 		int ostat;
1913 		struct sigaction sigact;
1914 		sigact._sa_u._sa_handler = SIG_DFL;
1915 		sigact.sa_flags = 0;
1916 
1917 		/*
1918 		 * set state to SSTOP so that this proc can be found by pid.
1919 		 * see proc_enterprp, do_sched_setparam below
1920 		 */
1921 		ostat = l->l_proc->p_stat;
1922 		l->l_proc->p_stat = SSTOP;
1923 
1924 		/* Set process group */
1925 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
1926 			pid_t mypid = l->l_proc->p_pid,
1927 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
1928 
1929 			if (pgrp == 0)
1930 				pgrp = mypid;
1931 
1932 			error = proc_enterpgrp(spawn_data->sed_parent,
1933 			    mypid, pgrp, false);
1934 			if (error)
1935 				goto report_error;
1936 		}
1937 
1938 		/* Set scheduler policy */
1939 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
1940 			error = do_sched_setparam(l->l_proc->p_pid, 0,
1941 			    spawn_data->sed_attrs->sa_schedpolicy,
1942 			    &spawn_data->sed_attrs->sa_schedparam);
1943 		else if (spawn_data->sed_attrs->sa_flags
1944 		    & POSIX_SPAWN_SETSCHEDPARAM) {
1945 			error = do_sched_setparam(ppid, 0,
1946 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
1947 		}
1948 		if (error)
1949 			goto report_error;
1950 
1951 		/* Reset user ID's */
1952 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
1953 			error = do_setresuid(l, -1,
1954 			     kauth_cred_getgid(l->l_cred), -1,
1955 			     ID_E_EQ_R | ID_E_EQ_S);
1956 			if (error)
1957 				goto report_error;
1958 			error = do_setresuid(l, -1,
1959 			    kauth_cred_getuid(l->l_cred), -1,
1960 			    ID_E_EQ_R | ID_E_EQ_S);
1961 			if (error)
1962 				goto report_error;
1963 		}
1964 
1965 		/* Set signal masks/defaults */
1966 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
1967 			mutex_enter(l->l_proc->p_lock);
1968 			error = sigprocmask1(l, SIG_SETMASK,
1969 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
1970 			mutex_exit(l->l_proc->p_lock);
1971 			if (error)
1972 				goto report_error;
1973 		}
1974 
1975 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
1976 			for (i = 1; i <= NSIG; i++) {
1977 				if (sigismember(
1978 				    &spawn_data->sed_attrs->sa_sigdefault, i))
1979 					sigaction1(l, i, &sigact, NULL, NULL,
1980 					    0);
1981 			}
1982 		}
1983 		l->l_proc->p_stat = ostat;
1984 	}
1985 
1986 	/* now do the real exec */
1987 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
1988 	    true);
1989 	have_reflock = false;
1990 	if (error == EJUSTRETURN)
1991 		error = 0;
1992 	else if (error)
1993 		goto report_error;
1994 
1995 	if (parent_is_waiting) {
1996 		mutex_enter(&spawn_data->sed_mtx_child);
1997 		cv_signal(&spawn_data->sed_cv_child_ready);
1998 		mutex_exit(&spawn_data->sed_mtx_child);
1999 	}
2000 
2001 	/* release our refcount on the data */
2002 	spawn_exec_data_release(spawn_data);
2003 
2004 	/* and finaly: leave to userland for the first time */
2005 	cpu_spawn_return(l);
2006 
2007 	/* NOTREACHED */
2008 	return;
2009 
2010  report_error:
2011  	if (have_reflock) {
2012  		/*
2013 		 * We have not passed through execve_runproc(),
2014 		 * which would have released the p_reflock and also
2015 		 * taken ownership of the sed_exec part of spawn_data,
2016 		 * so release/free both here.
2017 		 */
2018 		rw_exit(&l->l_proc->p_reflock);
2019 		execve_free_data(&spawn_data->sed_exec);
2020 	}
2021 
2022 	if (parent_is_waiting) {
2023 		/* pass error to parent */
2024 		mutex_enter(&spawn_data->sed_mtx_child);
2025 		spawn_data->sed_error = error;
2026 		cv_signal(&spawn_data->sed_cv_child_ready);
2027 		mutex_exit(&spawn_data->sed_mtx_child);
2028 	} else {
2029 		rw_exit(&exec_lock);
2030 	}
2031 
2032 	/* release our refcount on the data */
2033 	spawn_exec_data_release(spawn_data);
2034 
2035 	/* done, exit */
2036 	mutex_enter(l->l_proc->p_lock);
2037 	/*
2038 	 * Posix explicitly asks for an exit code of 127 if we report
2039 	 * errors from the child process - so, unfortunately, there
2040 	 * is no way to report a more exact error code.
2041 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2042 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2043 	 */
2044 	exit1(l, W_EXITCODE(127, 0));
2045 }
2046 
2047 void
2048 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2049 {
2050 
2051 	for (size_t i = 0; i < len; i++) {
2052 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2053 		if (fae->fae_action != FAE_OPEN)
2054 			continue;
2055 		kmem_free(fae->fae_path, strlen(fae->fae_path) + 1);
2056 	}
2057 	if (fa->len > 0)
2058 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2059 	kmem_free(fa, sizeof(*fa));
2060 }
2061 
2062 static int
2063 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2064     const struct posix_spawn_file_actions *ufa)
2065 {
2066 	struct posix_spawn_file_actions *fa;
2067 	struct posix_spawn_file_actions_entry *fae;
2068 	char *pbuf = NULL;
2069 	int error;
2070 	size_t i = 0;
2071 
2072 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2073 	error = copyin(ufa, fa, sizeof(*fa));
2074 	if (error) {
2075 		fa->fae = NULL;
2076 		fa->len = 0;
2077 		goto out;
2078 	}
2079 
2080 	if (fa->len == 0) {
2081 		kmem_free(fa, sizeof(*fa));
2082 		return 0;
2083 	}
2084 
2085 	fa->size = fa->len;
2086 	size_t fal = fa->len * sizeof(*fae);
2087 	fae = fa->fae;
2088 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2089 	error = copyin(fae, fa->fae, fal);
2090 	if (error)
2091 		goto out;
2092 
2093 	pbuf = PNBUF_GET();
2094 	for (; i < fa->len; i++) {
2095 		fae = &fa->fae[i];
2096 		if (fae->fae_action != FAE_OPEN)
2097 			continue;
2098 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2099 		if (error)
2100 			goto out;
2101 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2102 		memcpy(fae->fae_path, pbuf, fal);
2103 	}
2104 	PNBUF_PUT(pbuf);
2105 
2106 	*fap = fa;
2107 	return 0;
2108 out:
2109 	if (pbuf)
2110 		PNBUF_PUT(pbuf);
2111 	posix_spawn_fa_free(fa, i);
2112 	return error;
2113 }
2114 
2115 int
2116 check_posix_spawn(struct lwp *l1)
2117 {
2118 	int error, tnprocs, count;
2119 	uid_t uid;
2120 	struct proc *p1;
2121 
2122 	p1 = l1->l_proc;
2123 	uid = kauth_cred_getuid(l1->l_cred);
2124 	tnprocs = atomic_inc_uint_nv(&nprocs);
2125 
2126 	/*
2127 	 * Although process entries are dynamically created, we still keep
2128 	 * a global limit on the maximum number we will create.
2129 	 */
2130 	if (__predict_false(tnprocs >= maxproc))
2131 		error = -1;
2132 	else
2133 		error = kauth_authorize_process(l1->l_cred,
2134 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2135 
2136 	if (error) {
2137 		atomic_dec_uint(&nprocs);
2138 		return EAGAIN;
2139 	}
2140 
2141 	/*
2142 	 * Enforce limits.
2143 	 */
2144 	count = chgproccnt(uid, 1);
2145 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2146 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2147 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2148 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2149 		(void)chgproccnt(uid, -1);
2150 		atomic_dec_uint(&nprocs);
2151 		return EAGAIN;
2152 	}
2153 
2154 	return 0;
2155 }
2156 
2157 int
2158 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2159 	struct posix_spawn_file_actions *fa,
2160 	struct posix_spawnattr *sa,
2161 	char *const *argv, char *const *envp,
2162 	execve_fetch_element_t fetch)
2163 {
2164 
2165 	struct proc *p1, *p2;
2166 	struct lwp *l2;
2167 	int error;
2168 	struct spawn_exec_data *spawn_data;
2169 	vaddr_t uaddr;
2170 	pid_t pid;
2171 	bool have_exec_lock = false;
2172 
2173 	p1 = l1->l_proc;
2174 
2175 	/* Allocate and init spawn_data */
2176 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2177 	spawn_data->sed_refcnt = 1; /* only parent so far */
2178 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2179 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2180 	mutex_enter(&spawn_data->sed_mtx_child);
2181 
2182 	/*
2183 	 * Do the first part of the exec now, collect state
2184 	 * in spawn_data.
2185 	 */
2186 	error = execve_loadvm(l1, path, argv,
2187 	    envp, fetch, &spawn_data->sed_exec);
2188 	if (error == EJUSTRETURN)
2189 		error = 0;
2190 	else if (error)
2191 		goto error_exit;
2192 
2193 	have_exec_lock = true;
2194 
2195 	/*
2196 	 * Allocate virtual address space for the U-area now, while it
2197 	 * is still easy to abort the fork operation if we're out of
2198 	 * kernel virtual address space.
2199 	 */
2200 	uaddr = uvm_uarea_alloc();
2201 	if (__predict_false(uaddr == 0)) {
2202 		error = ENOMEM;
2203 		goto error_exit;
2204 	}
2205 
2206 	/*
2207 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2208 	 * replace it with its own before returning to userland
2209 	 * in the child.
2210 	 * This is a point of no return, we will have to go through
2211 	 * the child proc to properly clean it up past this point.
2212 	 */
2213 	p2 = proc_alloc();
2214 	pid = p2->p_pid;
2215 
2216 	/*
2217 	 * Make a proc table entry for the new process.
2218 	 * Start by zeroing the section of proc that is zero-initialized,
2219 	 * then copy the section that is copied directly from the parent.
2220 	 */
2221 	memset(&p2->p_startzero, 0,
2222 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2223 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2224 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2225 	p2->p_vmspace = proc0.p_vmspace;
2226 
2227 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2228 
2229 	LIST_INIT(&p2->p_lwps);
2230 	LIST_INIT(&p2->p_sigwaiters);
2231 
2232 	/*
2233 	 * Duplicate sub-structures as needed.
2234 	 * Increase reference counts on shared objects.
2235 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2236 	 * handling are important in order to keep a consistent behaviour
2237 	 * for the child after the fork.  If we are a 32-bit process, the
2238 	 * child will be too.
2239 	 */
2240 	p2->p_flag =
2241 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2242 	p2->p_emul = p1->p_emul;
2243 	p2->p_execsw = p1->p_execsw;
2244 
2245 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2246 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2247 	rw_init(&p2->p_reflock);
2248 	cv_init(&p2->p_waitcv, "wait");
2249 	cv_init(&p2->p_lwpcv, "lwpwait");
2250 
2251 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2252 
2253 	kauth_proc_fork(p1, p2);
2254 
2255 	p2->p_raslist = NULL;
2256 	p2->p_fd = fd_copy();
2257 
2258 	/* XXX racy */
2259 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2260 
2261 	p2->p_cwdi = cwdinit();
2262 
2263 	/*
2264 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2265 	 * we just need increase pl_refcnt.
2266 	 */
2267 	if (!p1->p_limit->pl_writeable) {
2268 		lim_addref(p1->p_limit);
2269 		p2->p_limit = p1->p_limit;
2270 	} else {
2271 		p2->p_limit = lim_copy(p1->p_limit);
2272 	}
2273 
2274 	p2->p_lflag = 0;
2275 	p2->p_sflag = 0;
2276 	p2->p_slflag = 0;
2277 	p2->p_pptr = p1;
2278 	p2->p_ppid = p1->p_pid;
2279 	LIST_INIT(&p2->p_children);
2280 
2281 	p2->p_aio = NULL;
2282 
2283 #ifdef KTRACE
2284 	/*
2285 	 * Copy traceflag and tracefile if enabled.
2286 	 * If not inherited, these were zeroed above.
2287 	 */
2288 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2289 		mutex_enter(&ktrace_lock);
2290 		p2->p_traceflag = p1->p_traceflag;
2291 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2292 			ktradref(p2);
2293 		mutex_exit(&ktrace_lock);
2294 	}
2295 #endif
2296 
2297 	/*
2298 	 * Create signal actions for the child process.
2299 	 */
2300 	p2->p_sigacts = sigactsinit(p1, 0);
2301 	mutex_enter(p1->p_lock);
2302 	p2->p_sflag |=
2303 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2304 	sched_proc_fork(p1, p2);
2305 	mutex_exit(p1->p_lock);
2306 
2307 	p2->p_stflag = p1->p_stflag;
2308 
2309 	/*
2310 	 * p_stats.
2311 	 * Copy parts of p_stats, and zero out the rest.
2312 	 */
2313 	p2->p_stats = pstatscopy(p1->p_stats);
2314 
2315 	/* copy over machdep flags to the new proc */
2316 	cpu_proc_fork(p1, p2);
2317 
2318 	/*
2319 	 * Prepare remaining parts of spawn data
2320 	 */
2321 	spawn_data->sed_actions = fa;
2322 	spawn_data->sed_attrs = sa;
2323 
2324 	spawn_data->sed_parent = p1;
2325 
2326 	/* create LWP */
2327 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2328 	    &l2, l1->l_class);
2329 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2330 
2331 	/*
2332 	 * Copy the credential so other references don't see our changes.
2333 	 * Test to see if this is necessary first, since in the common case
2334 	 * we won't need a private reference.
2335 	 */
2336 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2337 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2338 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2339 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2340 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2341 	}
2342 
2343 	/* Update the master credentials. */
2344 	if (l2->l_cred != p2->p_cred) {
2345 		kauth_cred_t ocred;
2346 
2347 		kauth_cred_hold(l2->l_cred);
2348 		mutex_enter(p2->p_lock);
2349 		ocred = p2->p_cred;
2350 		p2->p_cred = l2->l_cred;
2351 		mutex_exit(p2->p_lock);
2352 		kauth_cred_free(ocred);
2353 	}
2354 
2355 	*child_ok = true;
2356 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2357 #if 0
2358 	l2->l_nopreempt = 1; /* start it non-preemptable */
2359 #endif
2360 
2361 	/*
2362 	 * It's now safe for the scheduler and other processes to see the
2363 	 * child process.
2364 	 */
2365 	mutex_enter(proc_lock);
2366 
2367 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2368 		p2->p_lflag |= PL_CONTROLT;
2369 
2370 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2371 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2372 
2373 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2374 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2375 
2376 	p2->p_trace_enabled = trace_is_enabled(p2);
2377 #ifdef __HAVE_SYSCALL_INTERN
2378 	(*p2->p_emul->e_syscall_intern)(p2);
2379 #endif
2380 
2381 	/*
2382 	 * Make child runnable, set start time, and add to run queue except
2383 	 * if the parent requested the child to start in SSTOP state.
2384 	 */
2385 	mutex_enter(p2->p_lock);
2386 
2387 	getmicrotime(&p2->p_stats->p_start);
2388 
2389 	lwp_lock(l2);
2390 	KASSERT(p2->p_nrlwps == 1);
2391 	p2->p_nrlwps = 1;
2392 	p2->p_stat = SACTIVE;
2393 	l2->l_stat = LSRUN;
2394 	sched_enqueue(l2, false);
2395 	lwp_unlock(l2);
2396 
2397 	mutex_exit(p2->p_lock);
2398 	mutex_exit(proc_lock);
2399 
2400 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2401 	error = spawn_data->sed_error;
2402 	mutex_exit(&spawn_data->sed_mtx_child);
2403 	spawn_exec_data_release(spawn_data);
2404 
2405 	rw_exit(&p1->p_reflock);
2406 	rw_exit(&exec_lock);
2407 	have_exec_lock = false;
2408 
2409 	*pid_res = pid;
2410 	return error;
2411 
2412  error_exit:
2413  	if (have_exec_lock) {
2414 		execve_free_data(&spawn_data->sed_exec);
2415 		rw_exit(&p1->p_reflock);
2416  		rw_exit(&exec_lock);
2417 	}
2418 	mutex_exit(&spawn_data->sed_mtx_child);
2419 	spawn_exec_data_release(spawn_data);
2420 
2421 	return error;
2422 }
2423 
2424 int
2425 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2426     register_t *retval)
2427 {
2428 	/* {
2429 		syscallarg(pid_t *) pid;
2430 		syscallarg(const char *) path;
2431 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2432 		syscallarg(const struct posix_spawnattr *) attrp;
2433 		syscallarg(char *const *) argv;
2434 		syscallarg(char *const *) envp;
2435 	} */
2436 
2437 	int error;
2438 	struct posix_spawn_file_actions *fa = NULL;
2439 	struct posix_spawnattr *sa = NULL;
2440 	pid_t pid;
2441 	bool child_ok = false;
2442 
2443 	error = check_posix_spawn(l1);
2444 	if (error) {
2445 		*retval = error;
2446 		return 0;
2447 	}
2448 
2449 	/* copy in file_actions struct */
2450 	if (SCARG(uap, file_actions) != NULL) {
2451 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions));
2452 		if (error)
2453 			goto error_exit;
2454 	}
2455 
2456 	/* copyin posix_spawnattr struct */
2457 	if (SCARG(uap, attrp) != NULL) {
2458 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2459 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2460 		if (error)
2461 			goto error_exit;
2462 	}
2463 
2464 	/*
2465 	 * Do the spawn
2466 	 */
2467 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2468 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2469 	if (error)
2470 		goto error_exit;
2471 
2472 	if (error == 0 && SCARG(uap, pid) != NULL)
2473 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2474 
2475 	*retval = error;
2476 	return 0;
2477 
2478  error_exit:
2479 	if (!child_ok) {
2480 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2481 		atomic_dec_uint(&nprocs);
2482 
2483 		if (sa)
2484 			kmem_free(sa, sizeof(*sa));
2485 		if (fa)
2486 			posix_spawn_fa_free(fa, fa->len);
2487 	}
2488 
2489 	*retval = error;
2490 	return 0;
2491 }
2492 
2493 void
2494 exec_free_emul_arg(struct exec_package *epp)
2495 {
2496 	if (epp->ep_emul_arg_free != NULL) {
2497 		KASSERT(epp->ep_emul_arg != NULL);
2498 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2499 		epp->ep_emul_arg_free = NULL;
2500 		epp->ep_emul_arg = NULL;
2501 	} else {
2502 		KASSERT(epp->ep_emul_arg == NULL);
2503 	}
2504 }
2505