xref: /netbsd-src/sys/kern/kern_exec.c (revision d3d2abdc28a790079ac0d9b337b15bd97b65d751)
1 /*	$NetBSD: kern_exec.c,v 1.478 2019/07/05 17:14:48 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 1.478 2019/07/05 17:14:48 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/ptrace.h>
78 #include <sys/mount.h>
79 #include <sys/kmem.h>
80 #include <sys/namei.h>
81 #include <sys/vnode.h>
82 #include <sys/file.h>
83 #include <sys/filedesc.h>
84 #include <sys/acct.h>
85 #include <sys/atomic.h>
86 #include <sys/exec.h>
87 #include <sys/ktrace.h>
88 #include <sys/uidinfo.h>
89 #include <sys/wait.h>
90 #include <sys/mman.h>
91 #include <sys/ras.h>
92 #include <sys/signalvar.h>
93 #include <sys/stat.h>
94 #include <sys/syscall.h>
95 #include <sys/kauth.h>
96 #include <sys/lwpctl.h>
97 #include <sys/pax.h>
98 #include <sys/cpu.h>
99 #include <sys/module.h>
100 #include <sys/syscallvar.h>
101 #include <sys/syscallargs.h>
102 #if NVERIEXEC > 0
103 #include <sys/verified_exec.h>
104 #endif /* NVERIEXEC > 0 */
105 #include <sys/sdt.h>
106 #include <sys/spawn.h>
107 #include <sys/prot.h>
108 #include <sys/cprng.h>
109 
110 #include <uvm/uvm_extern.h>
111 
112 #include <machine/reg.h>
113 
114 #include <compat/common/compat_util.h>
115 
116 #ifndef MD_TOPDOWN_INIT
117 #ifdef __USE_TOPDOWN_VM
118 #define	MD_TOPDOWN_INIT(epp)	(epp)->ep_flags |= EXEC_TOPDOWN_VM
119 #else
120 #define	MD_TOPDOWN_INIT(epp)
121 #endif
122 #endif
123 
124 struct execve_data;
125 
126 extern int user_va0_disable;
127 
128 static size_t calcargs(struct execve_data * restrict, const size_t);
129 static size_t calcstack(struct execve_data * restrict, const size_t);
130 static int copyoutargs(struct execve_data * restrict, struct lwp *,
131     char * const);
132 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
133 static int copyinargs(struct execve_data * restrict, char * const *,
134     char * const *, execve_fetch_element_t, char **);
135 static int copyinargstrs(struct execve_data * restrict, char * const *,
136     execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
137 static int exec_sigcode_map(struct proc *, const struct emul *);
138 
139 #if defined(DEBUG) && !defined(DEBUG_EXEC)
140 #define DEBUG_EXEC
141 #endif
142 #ifdef DEBUG_EXEC
143 #define DPRINTF(a) printf a
144 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
145     __LINE__, (s), (a), (b))
146 static void dump_vmcmds(const struct exec_package * const, size_t, int);
147 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
148 #else
149 #define DPRINTF(a)
150 #define COPYPRINTF(s, a, b)
151 #define DUMPVMCMDS(p, x, e) do {} while (0)
152 #endif /* DEBUG_EXEC */
153 
154 /*
155  * DTrace SDT provider definitions
156  */
157 SDT_PROVIDER_DECLARE(proc);
158 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
159 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
160 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
161 
162 /*
163  * Exec function switch:
164  *
165  * Note that each makecmds function is responsible for loading the
166  * exec package with the necessary functions for any exec-type-specific
167  * handling.
168  *
169  * Functions for specific exec types should be defined in their own
170  * header file.
171  */
172 static const struct execsw	**execsw = NULL;
173 static int			nexecs;
174 
175 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
176 
177 /* list of dynamically loaded execsw entries */
178 static LIST_HEAD(execlist_head, exec_entry) ex_head =
179     LIST_HEAD_INITIALIZER(ex_head);
180 struct exec_entry {
181 	LIST_ENTRY(exec_entry)	ex_list;
182 	SLIST_ENTRY(exec_entry)	ex_slist;
183 	const struct execsw	*ex_sw;
184 };
185 
186 #ifndef __HAVE_SYSCALL_INTERN
187 void	syscall(void);
188 #endif
189 
190 /* NetBSD autoloadable syscalls */
191 #ifdef MODULAR
192 #include <kern/syscalls_autoload.c>
193 #endif
194 
195 /* NetBSD emul struct */
196 struct emul emul_netbsd = {
197 	.e_name =		"netbsd",
198 #ifdef EMUL_NATIVEROOT
199 	.e_path =		EMUL_NATIVEROOT,
200 #else
201 	.e_path =		NULL,
202 #endif
203 #ifndef __HAVE_MINIMAL_EMUL
204 	.e_flags =		EMUL_HAS_SYS___syscall,
205 	.e_errno =		NULL,
206 	.e_nosys =		SYS_syscall,
207 	.e_nsysent =		SYS_NSYSENT,
208 #endif
209 #ifdef MODULAR
210 	.e_sc_autoload =	netbsd_syscalls_autoload,
211 #endif
212 	.e_sysent =		sysent,
213 	.e_nomodbits =		sysent_nomodbits,
214 #ifdef SYSCALL_DEBUG
215 	.e_syscallnames =	syscallnames,
216 #else
217 	.e_syscallnames =	NULL,
218 #endif
219 	.e_sendsig =		sendsig,
220 	.e_trapsignal =		trapsignal,
221 	.e_sigcode =		NULL,
222 	.e_esigcode =		NULL,
223 	.e_sigobject =		NULL,
224 	.e_setregs =		setregs,
225 	.e_proc_exec =		NULL,
226 	.e_proc_fork =		NULL,
227 	.e_proc_exit =		NULL,
228 	.e_lwp_fork =		NULL,
229 	.e_lwp_exit =		NULL,
230 #ifdef __HAVE_SYSCALL_INTERN
231 	.e_syscall_intern =	syscall_intern,
232 #else
233 	.e_syscall =		syscall,
234 #endif
235 	.e_sysctlovly =		NULL,
236 	.e_vm_default_addr =	uvm_default_mapaddr,
237 	.e_usertrap =		NULL,
238 	.e_ucsize =		sizeof(ucontext_t),
239 	.e_startlwp =		startlwp
240 };
241 
242 /*
243  * Exec lock. Used to control access to execsw[] structures.
244  * This must not be static so that netbsd32 can access it, too.
245  */
246 krwlock_t exec_lock;
247 
248 static kmutex_t sigobject_lock;
249 
250 /*
251  * Data used between a loadvm and execve part of an "exec" operation
252  */
253 struct execve_data {
254 	struct exec_package	ed_pack;
255 	struct pathbuf		*ed_pathbuf;
256 	struct vattr		ed_attr;
257 	struct ps_strings	ed_arginfo;
258 	char			*ed_argp;
259 	const char		*ed_pathstring;
260 	char			*ed_resolvedpathbuf;
261 	size_t			ed_ps_strings_sz;
262 	int			ed_szsigcode;
263 	size_t			ed_argslen;
264 	long			ed_argc;
265 	long			ed_envc;
266 };
267 
268 /*
269  * data passed from parent lwp to child during a posix_spawn()
270  */
271 struct spawn_exec_data {
272 	struct execve_data	sed_exec;
273 	struct posix_spawn_file_actions
274 				*sed_actions;
275 	struct posix_spawnattr	*sed_attrs;
276 	struct proc		*sed_parent;
277 	kcondvar_t		sed_cv_child_ready;
278 	kmutex_t		sed_mtx_child;
279 	int			sed_error;
280 	volatile uint32_t	sed_refcnt;
281 };
282 
283 static struct vm_map *exec_map;
284 static struct pool exec_pool;
285 
286 static void *
287 exec_pool_alloc(struct pool *pp, int flags)
288 {
289 
290 	return (void *)uvm_km_alloc(exec_map, NCARGS, 0,
291 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
292 }
293 
294 static void
295 exec_pool_free(struct pool *pp, void *addr)
296 {
297 
298 	uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
299 }
300 
301 static struct pool_allocator exec_palloc = {
302 	.pa_alloc = exec_pool_alloc,
303 	.pa_free = exec_pool_free,
304 	.pa_pagesz = NCARGS
305 };
306 
307 /*
308  * check exec:
309  * given an "executable" described in the exec package's namei info,
310  * see what we can do with it.
311  *
312  * ON ENTRY:
313  *	exec package with appropriate namei info
314  *	lwp pointer of exec'ing lwp
315  *	NO SELF-LOCKED VNODES
316  *
317  * ON EXIT:
318  *	error:	nothing held, etc.  exec header still allocated.
319  *	ok:	filled exec package, executable's vnode (unlocked).
320  *
321  * EXEC SWITCH ENTRY:
322  * 	Locked vnode to check, exec package, proc.
323  *
324  * EXEC SWITCH EXIT:
325  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
326  *	error:	destructive:
327  *			everything deallocated execept exec header.
328  *		non-destructive:
329  *			error code, executable's vnode (unlocked),
330  *			exec header unmodified.
331  */
332 int
333 /*ARGSUSED*/
334 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
335 {
336 	int		error, i;
337 	struct vnode	*vp;
338 	struct nameidata nd;
339 	size_t		resid;
340 
341 #if 1
342 	// grab the absolute pathbuf here before namei() trashes it.
343 	pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX);
344 #endif
345 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
346 
347 	/* first get the vnode */
348 	if ((error = namei(&nd)) != 0)
349 		return error;
350 	epp->ep_vp = vp = nd.ni_vp;
351 #if 0
352 	/*
353 	 * XXX: can't use nd.ni_pnbuf, because although pb contains an
354 	 * absolute path, nd.ni_pnbuf does not if the path contains symlinks.
355 	 */
356 	/* normally this can't fail */
357 	error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
358 	KASSERT(error == 0);
359 #endif
360 
361 #ifdef DIAGNOSTIC
362 	/* paranoia (take this out once namei stuff stabilizes) */
363 	memset(nd.ni_pnbuf, '~', PATH_MAX);
364 #endif
365 
366 	/* check access and type */
367 	if (vp->v_type != VREG) {
368 		error = EACCES;
369 		goto bad1;
370 	}
371 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
372 		goto bad1;
373 
374 	/* get attributes */
375 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
376 		goto bad1;
377 
378 	/* Check mount point */
379 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
380 		error = EACCES;
381 		goto bad1;
382 	}
383 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
384 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
385 
386 	/* try to open it */
387 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
388 		goto bad1;
389 
390 	/* unlock vp, since we need it unlocked from here on out. */
391 	VOP_UNLOCK(vp);
392 
393 #if NVERIEXEC > 0
394 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
395 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
396 	    NULL);
397 	if (error)
398 		goto bad2;
399 #endif /* NVERIEXEC > 0 */
400 
401 #ifdef PAX_SEGVGUARD
402 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
403 	if (error)
404 		goto bad2;
405 #endif /* PAX_SEGVGUARD */
406 
407 	/* now we have the file, get the exec header */
408 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
409 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
410 	if (error)
411 		goto bad2;
412 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
413 
414 	/*
415 	 * Set up default address space limits.  Can be overridden
416 	 * by individual exec packages.
417 	 */
418 	epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
419 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
420 
421 	/*
422 	 * set up the vmcmds for creation of the process
423 	 * address space
424 	 */
425 	error = ENOEXEC;
426 	for (i = 0; i < nexecs; i++) {
427 		int newerror;
428 
429 		epp->ep_esch = execsw[i];
430 		newerror = (*execsw[i]->es_makecmds)(l, epp);
431 
432 		if (!newerror) {
433 			/* Seems ok: check that entry point is not too high */
434 			if (epp->ep_entry >= epp->ep_vm_maxaddr) {
435 #ifdef DIAGNOSTIC
436 				printf("%s: rejecting %p due to "
437 				    "too high entry address (>= %p)\n",
438 					 __func__, (void *)epp->ep_entry,
439 					 (void *)epp->ep_vm_maxaddr);
440 #endif
441 				error = ENOEXEC;
442 				break;
443 			}
444 			/* Seems ok: check that entry point is not too low */
445 			if (epp->ep_entry < epp->ep_vm_minaddr) {
446 #ifdef DIAGNOSTIC
447 				printf("%s: rejecting %p due to "
448 				    "too low entry address (< %p)\n",
449 				     __func__, (void *)epp->ep_entry,
450 				     (void *)epp->ep_vm_minaddr);
451 #endif
452 				error = ENOEXEC;
453 				break;
454 			}
455 
456 			/* check limits */
457 			if ((epp->ep_tsize > MAXTSIZ) ||
458 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
459 						    [RLIMIT_DATA].rlim_cur)) {
460 #ifdef DIAGNOSTIC
461 				printf("%s: rejecting due to "
462 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
463 				    __func__,
464 				    (unsigned long long)epp->ep_tsize,
465 				    (unsigned long long)MAXTSIZ,
466 				    (unsigned long long)epp->ep_dsize,
467 				    (unsigned long long)
468 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
469 #endif
470 				error = ENOMEM;
471 				break;
472 			}
473 			return 0;
474 		}
475 
476 		/*
477 		 * Reset all the fields that may have been modified by the
478 		 * loader.
479 		 */
480 		KASSERT(epp->ep_emul_arg == NULL);
481 		if (epp->ep_emul_root != NULL) {
482 			vrele(epp->ep_emul_root);
483 			epp->ep_emul_root = NULL;
484 		}
485 		if (epp->ep_interp != NULL) {
486 			vrele(epp->ep_interp);
487 			epp->ep_interp = NULL;
488 		}
489 		epp->ep_pax_flags = 0;
490 
491 		/* make sure the first "interesting" error code is saved. */
492 		if (error == ENOEXEC)
493 			error = newerror;
494 
495 		if (epp->ep_flags & EXEC_DESTR)
496 			/* Error from "#!" code, tidied up by recursive call */
497 			return error;
498 	}
499 
500 	/* not found, error */
501 
502 	/*
503 	 * free any vmspace-creation commands,
504 	 * and release their references
505 	 */
506 	kill_vmcmds(&epp->ep_vmcmds);
507 
508 bad2:
509 	/*
510 	 * close and release the vnode, restore the old one, free the
511 	 * pathname buf, and punt.
512 	 */
513 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
514 	VOP_CLOSE(vp, FREAD, l->l_cred);
515 	vput(vp);
516 	return error;
517 
518 bad1:
519 	/*
520 	 * free the namei pathname buffer, and put the vnode
521 	 * (which we don't yet have open).
522 	 */
523 	vput(vp);				/* was still locked */
524 	return error;
525 }
526 
527 #ifdef __MACHINE_STACK_GROWS_UP
528 #define STACK_PTHREADSPACE NBPG
529 #else
530 #define STACK_PTHREADSPACE 0
531 #endif
532 
533 static int
534 execve_fetch_element(char * const *array, size_t index, char **value)
535 {
536 	return copyin(array + index, value, sizeof(*value));
537 }
538 
539 /*
540  * exec system call
541  */
542 int
543 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
544 {
545 	/* {
546 		syscallarg(const char *)	path;
547 		syscallarg(char * const *)	argp;
548 		syscallarg(char * const *)	envp;
549 	} */
550 
551 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
552 	    SCARG(uap, envp), execve_fetch_element);
553 }
554 
555 int
556 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
557     register_t *retval)
558 {
559 	/* {
560 		syscallarg(int)			fd;
561 		syscallarg(char * const *)	argp;
562 		syscallarg(char * const *)	envp;
563 	} */
564 
565 	return ENOSYS;
566 }
567 
568 /*
569  * Load modules to try and execute an image that we do not understand.
570  * If no execsw entries are present, we load those likely to be needed
571  * in order to run native images only.  Otherwise, we autoload all
572  * possible modules that could let us run the binary.  XXX lame
573  */
574 static void
575 exec_autoload(void)
576 {
577 #ifdef MODULAR
578 	static const char * const native[] = {
579 		"exec_elf32",
580 		"exec_elf64",
581 		"exec_script",
582 		NULL
583 	};
584 	static const char * const compat[] = {
585 		"exec_elf32",
586 		"exec_elf64",
587 		"exec_script",
588 		"exec_aout",
589 		"exec_coff",
590 		"exec_ecoff",
591 		"compat_aoutm68k",
592 		"compat_netbsd32",
593 		"compat_sunos",
594 		"compat_sunos32",
595 		"compat_ultrix",
596 		NULL
597 	};
598 	char const * const *list;
599 	int i;
600 
601 	list = (nexecs == 0 ? native : compat);
602 	for (i = 0; list[i] != NULL; i++) {
603 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
604 			continue;
605 		}
606 		yield();
607 	}
608 #endif
609 }
610 
611 /*
612  * Copy the user or kernel supplied upath to the allocated pathbuffer pbp
613  * making it absolute in the process, by prepending the current working
614  * directory if it is not. If offs is supplied it will contain the offset
615  * where the original supplied copy of upath starts.
616  */
617 int
618 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg,
619     struct pathbuf **pbp, size_t *offs)
620 {
621 	char *path, *bp;
622 	size_t len, tlen;
623 	int error;
624 	struct cwdinfo *cwdi;
625 
626 	path = PNBUF_GET();
627 	if (seg == UIO_SYSSPACE) {
628 		error = copystr(upath, path, MAXPATHLEN, &len);
629 	} else {
630 		error = copyinstr(upath, path, MAXPATHLEN, &len);
631 	}
632 	if (error)
633 		goto err;
634 
635 	if (path[0] == '/') {
636 		if (offs)
637 			*offs = 0;
638 		goto out;
639 	}
640 
641 	len++;
642 	if (len + 1 >= MAXPATHLEN) {
643 		error = ENAMETOOLONG;
644 		goto err;
645 	}
646 	bp = path + MAXPATHLEN - len;
647 	memmove(bp, path, len);
648 	*(--bp) = '/';
649 
650 	cwdi = l->l_proc->p_cwdi;
651 	rw_enter(&cwdi->cwdi_lock, RW_READER);
652 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
653 	    GETCWD_CHECK_ACCESS, l);
654 	rw_exit(&cwdi->cwdi_lock);
655 
656 	if (error)
657 		goto err;
658 	tlen = path + MAXPATHLEN - bp;
659 
660 	memmove(path, bp, tlen);
661 	path[tlen - 1] = '\0';
662 	if (offs)
663 		*offs = tlen - len;
664 out:
665 	*pbp = pathbuf_assimilate(path);
666 	return 0;
667 err:
668 	PNBUF_PUT(path);
669 	return error;
670 }
671 
672 vaddr_t
673 exec_vm_minaddr(vaddr_t va_min)
674 {
675 	/*
676 	 * Increase va_min if we don't want NULL to be mappable by the
677 	 * process.
678 	 */
679 #define VM_MIN_GUARD	PAGE_SIZE
680 	if (user_va0_disable && (va_min < VM_MIN_GUARD))
681 		return VM_MIN_GUARD;
682 	return va_min;
683 }
684 
685 static int
686 execve_loadvm(struct lwp *l, const char *path, char * const *args,
687 	char * const *envs, execve_fetch_element_t fetch_element,
688 	struct execve_data * restrict data)
689 {
690 	struct exec_package	* const epp = &data->ed_pack;
691 	int			error;
692 	struct proc		*p;
693 	char			*dp;
694 	u_int			modgen;
695 	size_t			offs;
696 
697 	KASSERT(data != NULL);
698 
699 	p = l->l_proc;
700 	modgen = 0;
701 
702 	SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
703 
704 	/*
705 	 * Check if we have exceeded our number of processes limit.
706 	 * This is so that we handle the case where a root daemon
707 	 * forked, ran setuid to become the desired user and is trying
708 	 * to exec. The obvious place to do the reference counting check
709 	 * is setuid(), but we don't do the reference counting check there
710 	 * like other OS's do because then all the programs that use setuid()
711 	 * must be modified to check the return code of setuid() and exit().
712 	 * It is dangerous to make setuid() fail, because it fails open and
713 	 * the program will continue to run as root. If we make it succeed
714 	 * and return an error code, again we are not enforcing the limit.
715 	 * The best place to enforce the limit is here, when the process tries
716 	 * to execute a new image, because eventually the process will need
717 	 * to call exec in order to do something useful.
718 	 */
719  retry:
720 	if (p->p_flag & PK_SUGID) {
721 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
722 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
723 		     &p->p_rlimit[RLIMIT_NPROC],
724 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
725 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
726 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
727 		return EAGAIN;
728 	}
729 
730 	/*
731 	 * Drain existing references and forbid new ones.  The process
732 	 * should be left alone until we're done here.  This is necessary
733 	 * to avoid race conditions - e.g. in ptrace() - that might allow
734 	 * a local user to illicitly obtain elevated privileges.
735 	 */
736 	rw_enter(&p->p_reflock, RW_WRITER);
737 
738 	/*
739 	 * Init the namei data to point the file user's program name.
740 	 * This is done here rather than in check_exec(), so that it's
741 	 * possible to override this settings if any of makecmd/probe
742 	 * functions call check_exec() recursively - for example,
743 	 * see exec_script_makecmds().
744 	 */
745 	if ((error = exec_makepathbuf(l, path, UIO_USERSPACE,
746 	    &data->ed_pathbuf, &offs)) != 0)
747 		goto clrflg;
748 	data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
749 	data->ed_resolvedpathbuf = PNBUF_GET();
750 
751 	/*
752 	 * initialize the fields of the exec package.
753 	 */
754 	epp->ep_kname = data->ed_pathstring + offs;
755 	epp->ep_resolvedname = data->ed_resolvedpathbuf;
756 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
757 	epp->ep_hdrlen = exec_maxhdrsz;
758 	epp->ep_hdrvalid = 0;
759 	epp->ep_emul_arg = NULL;
760 	epp->ep_emul_arg_free = NULL;
761 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
762 	epp->ep_vap = &data->ed_attr;
763 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
764 	MD_TOPDOWN_INIT(epp);
765 	epp->ep_emul_root = NULL;
766 	epp->ep_interp = NULL;
767 	epp->ep_esch = NULL;
768 	epp->ep_pax_flags = 0;
769 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
770 
771 	rw_enter(&exec_lock, RW_READER);
772 
773 	/* see if we can run it. */
774 	if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
775 		if (error != ENOENT && error != EACCES && error != ENOEXEC) {
776 			DPRINTF(("%s: check exec failed for %s, error %d\n",
777 			    __func__, epp->ep_kname, error));
778 		}
779 		goto freehdr;
780 	}
781 
782 	/* allocate an argument buffer */
783 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
784 	KASSERT(data->ed_argp != NULL);
785 	dp = data->ed_argp;
786 
787 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
788 		goto bad;
789 	}
790 
791 	/*
792 	 * Calculate the new stack size.
793 	 */
794 
795 #ifdef __MACHINE_STACK_GROWS_UP
796 /*
797  * copyargs() fills argc/argv/envp from the lower address even on
798  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
799  * so that _rtld() use it.
800  */
801 #define	RTLD_GAP	32
802 #else
803 #define	RTLD_GAP	0
804 #endif
805 
806 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
807 
808 	data->ed_argslen = calcargs(data, argenvstrlen);
809 
810 	const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
811 
812 	if (len > epp->ep_ssize) {
813 		/* in effect, compare to initial limit */
814 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
815 		error = ENOMEM;
816 		goto bad;
817 	}
818 	/* adjust "active stack depth" for process VSZ */
819 	epp->ep_ssize = len;
820 
821 	return 0;
822 
823  bad:
824 	/* free the vmspace-creation commands, and release their references */
825 	kill_vmcmds(&epp->ep_vmcmds);
826 	/* kill any opened file descriptor, if necessary */
827 	if (epp->ep_flags & EXEC_HASFD) {
828 		epp->ep_flags &= ~EXEC_HASFD;
829 		fd_close(epp->ep_fd);
830 	}
831 	/* close and put the exec'd file */
832 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
833 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
834 	vput(epp->ep_vp);
835 	pool_put(&exec_pool, data->ed_argp);
836 
837  freehdr:
838 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
839 	if (epp->ep_emul_root != NULL)
840 		vrele(epp->ep_emul_root);
841 	if (epp->ep_interp != NULL)
842 		vrele(epp->ep_interp);
843 
844 	rw_exit(&exec_lock);
845 
846 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
847 	pathbuf_destroy(data->ed_pathbuf);
848 	PNBUF_PUT(data->ed_resolvedpathbuf);
849 
850  clrflg:
851 	rw_exit(&p->p_reflock);
852 
853 	if (modgen != module_gen && error == ENOEXEC) {
854 		modgen = module_gen;
855 		exec_autoload();
856 		goto retry;
857 	}
858 
859 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
860 	return error;
861 }
862 
863 static int
864 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
865 {
866 	struct exec_package	* const epp = &data->ed_pack;
867 	struct proc		*p = l->l_proc;
868 	struct exec_vmcmd	*base_vcp;
869 	int			error = 0;
870 	size_t			i;
871 
872 	/* record proc's vnode, for use by procfs and others */
873 	if (p->p_textvp)
874 		vrele(p->p_textvp);
875 	vref(epp->ep_vp);
876 	p->p_textvp = epp->ep_vp;
877 
878 	/* create the new process's VM space by running the vmcmds */
879 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
880 
881 #ifdef TRACE_EXEC
882 	DUMPVMCMDS(epp, 0, 0);
883 #endif
884 
885 	base_vcp = NULL;
886 
887 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
888 		struct exec_vmcmd *vcp;
889 
890 		vcp = &epp->ep_vmcmds.evs_cmds[i];
891 		if (vcp->ev_flags & VMCMD_RELATIVE) {
892 			KASSERTMSG(base_vcp != NULL,
893 			    "%s: relative vmcmd with no base", __func__);
894 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
895 			    "%s: illegal base & relative vmcmd", __func__);
896 			vcp->ev_addr += base_vcp->ev_addr;
897 		}
898 		error = (*vcp->ev_proc)(l, vcp);
899 		if (error)
900 			DUMPVMCMDS(epp, i, error);
901 		if (vcp->ev_flags & VMCMD_BASE)
902 			base_vcp = vcp;
903 	}
904 
905 	/* free the vmspace-creation commands, and release their references */
906 	kill_vmcmds(&epp->ep_vmcmds);
907 
908 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
909 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
910 	vput(epp->ep_vp);
911 
912 	/* if an error happened, deallocate and punt */
913 	if (error != 0) {
914 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
915 	}
916 	return error;
917 }
918 
919 static void
920 execve_free_data(struct execve_data *data)
921 {
922 	struct exec_package	* const epp = &data->ed_pack;
923 
924 	/* free the vmspace-creation commands, and release their references */
925 	kill_vmcmds(&epp->ep_vmcmds);
926 	/* kill any opened file descriptor, if necessary */
927 	if (epp->ep_flags & EXEC_HASFD) {
928 		epp->ep_flags &= ~EXEC_HASFD;
929 		fd_close(epp->ep_fd);
930 	}
931 
932 	/* close and put the exec'd file */
933 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
934 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
935 	vput(epp->ep_vp);
936 	pool_put(&exec_pool, data->ed_argp);
937 
938 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
939 	if (epp->ep_emul_root != NULL)
940 		vrele(epp->ep_emul_root);
941 	if (epp->ep_interp != NULL)
942 		vrele(epp->ep_interp);
943 
944 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
945 	pathbuf_destroy(data->ed_pathbuf);
946 	PNBUF_PUT(data->ed_resolvedpathbuf);
947 }
948 
949 static void
950 pathexec(struct proc *p, const char *resolvedname)
951 {
952 	KASSERT(resolvedname[0] == '/');
953 
954 	/* set command name & other accounting info */
955 	strlcpy(p->p_comm, strrchr(resolvedname, '/') + 1, sizeof(p->p_comm));
956 
957 	kmem_strfree(p->p_path);
958 	p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
959 }
960 
961 /* XXX elsewhere */
962 static int
963 credexec(struct lwp *l, struct vattr *attr)
964 {
965 	struct proc *p = l->l_proc;
966 	int error;
967 
968 	/*
969 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
970 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
971 	 * out additional references on the process for the moment.
972 	 */
973 	if ((p->p_slflag & PSL_TRACED) == 0 &&
974 
975 	    (((attr->va_mode & S_ISUID) != 0 &&
976 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
977 
978 	     ((attr->va_mode & S_ISGID) != 0 &&
979 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
980 		/*
981 		 * Mark the process as SUGID before we do
982 		 * anything that might block.
983 		 */
984 		proc_crmod_enter();
985 		proc_crmod_leave(NULL, NULL, true);
986 
987 		/* Make sure file descriptors 0..2 are in use. */
988 		if ((error = fd_checkstd()) != 0) {
989 			DPRINTF(("%s: fdcheckstd failed %d\n",
990 			    __func__, error));
991 			return error;
992 		}
993 
994 		/*
995 		 * Copy the credential so other references don't see our
996 		 * changes.
997 		 */
998 		l->l_cred = kauth_cred_copy(l->l_cred);
999 #ifdef KTRACE
1000 		/*
1001 		 * If the persistent trace flag isn't set, turn off.
1002 		 */
1003 		if (p->p_tracep) {
1004 			mutex_enter(&ktrace_lock);
1005 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1006 				ktrderef(p);
1007 			mutex_exit(&ktrace_lock);
1008 		}
1009 #endif
1010 		if (attr->va_mode & S_ISUID)
1011 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
1012 		if (attr->va_mode & S_ISGID)
1013 			kauth_cred_setegid(l->l_cred, attr->va_gid);
1014 	} else {
1015 		if (kauth_cred_geteuid(l->l_cred) ==
1016 		    kauth_cred_getuid(l->l_cred) &&
1017 		    kauth_cred_getegid(l->l_cred) ==
1018 		    kauth_cred_getgid(l->l_cred))
1019 			p->p_flag &= ~PK_SUGID;
1020 	}
1021 
1022 	/*
1023 	 * Copy the credential so other references don't see our changes.
1024 	 * Test to see if this is necessary first, since in the common case
1025 	 * we won't need a private reference.
1026 	 */
1027 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1028 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1029 		l->l_cred = kauth_cred_copy(l->l_cred);
1030 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1031 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1032 	}
1033 
1034 	/* Update the master credentials. */
1035 	if (l->l_cred != p->p_cred) {
1036 		kauth_cred_t ocred;
1037 
1038 		kauth_cred_hold(l->l_cred);
1039 		mutex_enter(p->p_lock);
1040 		ocred = p->p_cred;
1041 		p->p_cred = l->l_cred;
1042 		mutex_exit(p->p_lock);
1043 		kauth_cred_free(ocred);
1044 	}
1045 
1046 	return 0;
1047 }
1048 
1049 static void
1050 emulexec(struct lwp *l, struct exec_package *epp)
1051 {
1052 	struct proc		*p = l->l_proc;
1053 
1054 	/* The emulation root will usually have been found when we looked
1055 	 * for the elf interpreter (or similar), if not look now. */
1056 	if (epp->ep_esch->es_emul->e_path != NULL &&
1057 	    epp->ep_emul_root == NULL)
1058 		emul_find_root(l, epp);
1059 
1060 	/* Any old emulation root got removed by fdcloseexec */
1061 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1062 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1063 	rw_exit(&p->p_cwdi->cwdi_lock);
1064 	epp->ep_emul_root = NULL;
1065 	if (epp->ep_interp != NULL)
1066 		vrele(epp->ep_interp);
1067 
1068 	/*
1069 	 * Call emulation specific exec hook. This can setup per-process
1070 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1071 	 *
1072 	 * If we are executing process of different emulation than the
1073 	 * original forked process, call e_proc_exit() of the old emulation
1074 	 * first, then e_proc_exec() of new emulation. If the emulation is
1075 	 * same, the exec hook code should deallocate any old emulation
1076 	 * resources held previously by this process.
1077 	 */
1078 	if (p->p_emul && p->p_emul->e_proc_exit
1079 	    && p->p_emul != epp->ep_esch->es_emul)
1080 		(*p->p_emul->e_proc_exit)(p);
1081 
1082 	/*
1083 	 * This is now LWP 1.
1084 	 */
1085 	/* XXX elsewhere */
1086 	mutex_enter(p->p_lock);
1087 	p->p_nlwpid = 1;
1088 	l->l_lid = 1;
1089 	mutex_exit(p->p_lock);
1090 
1091 	/*
1092 	 * Call exec hook. Emulation code may NOT store reference to anything
1093 	 * from &pack.
1094 	 */
1095 	if (epp->ep_esch->es_emul->e_proc_exec)
1096 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1097 
1098 	/* update p_emul, the old value is no longer needed */
1099 	p->p_emul = epp->ep_esch->es_emul;
1100 
1101 	/* ...and the same for p_execsw */
1102 	p->p_execsw = epp->ep_esch;
1103 
1104 #ifdef __HAVE_SYSCALL_INTERN
1105 	(*p->p_emul->e_syscall_intern)(p);
1106 #endif
1107 	ktremul();
1108 }
1109 
1110 static int
1111 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1112 	bool no_local_exec_lock, bool is_spawn)
1113 {
1114 	struct exec_package	* const epp = &data->ed_pack;
1115 	int error = 0;
1116 	struct proc		*p;
1117 
1118 	/*
1119 	 * In case of a posix_spawn operation, the child doing the exec
1120 	 * might not hold the reader lock on exec_lock, but the parent
1121 	 * will do this instead.
1122 	 */
1123 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1124 	KASSERT(!no_local_exec_lock || is_spawn);
1125 	KASSERT(data != NULL);
1126 
1127 	p = l->l_proc;
1128 
1129 	/* Get rid of other LWPs. */
1130 	if (p->p_nlwps > 1) {
1131 		mutex_enter(p->p_lock);
1132 		exit_lwps(l);
1133 		mutex_exit(p->p_lock);
1134 	}
1135 	KDASSERT(p->p_nlwps == 1);
1136 
1137 	/* Destroy any lwpctl info. */
1138 	if (p->p_lwpctl != NULL)
1139 		lwp_ctl_exit();
1140 
1141 	/* Remove POSIX timers */
1142 	timers_free(p, TIMERS_POSIX);
1143 
1144 	/* Set the PaX flags. */
1145 	pax_set_flags(epp, p);
1146 
1147 	/*
1148 	 * Do whatever is necessary to prepare the address space
1149 	 * for remapping.  Note that this might replace the current
1150 	 * vmspace with another!
1151 	 */
1152 	if (is_spawn)
1153 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1154 		    epp->ep_vm_maxaddr,
1155 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1156 	else
1157 		uvmspace_exec(l, epp->ep_vm_minaddr,
1158 		    epp->ep_vm_maxaddr,
1159 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1160 
1161 	struct vmspace		*vm;
1162 	vm = p->p_vmspace;
1163 	vm->vm_taddr = (void *)epp->ep_taddr;
1164 	vm->vm_tsize = btoc(epp->ep_tsize);
1165 	vm->vm_daddr = (void*)epp->ep_daddr;
1166 	vm->vm_dsize = btoc(epp->ep_dsize);
1167 	vm->vm_ssize = btoc(epp->ep_ssize);
1168 	vm->vm_issize = 0;
1169 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1170 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1171 
1172 	pax_aslr_init_vm(l, vm, epp);
1173 
1174 	/* Now map address space. */
1175 	error = execve_dovmcmds(l, data);
1176 	if (error != 0)
1177 		goto exec_abort;
1178 
1179 	pathexec(p, epp->ep_resolvedname);
1180 
1181 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1182 
1183 	error = copyoutargs(data, l, newstack);
1184 	if (error != 0)
1185 		goto exec_abort;
1186 
1187 	cwdexec(p);
1188 	fd_closeexec();		/* handle close on exec */
1189 
1190 	if (__predict_false(ktrace_on))
1191 		fd_ktrexecfd();
1192 
1193 	execsigs(p);		/* reset caught signals */
1194 
1195 	mutex_enter(p->p_lock);
1196 	l->l_ctxlink = NULL;	/* reset ucontext link */
1197 	p->p_acflag &= ~AFORK;
1198 	p->p_flag |= PK_EXEC;
1199 	mutex_exit(p->p_lock);
1200 
1201 	/*
1202 	 * Stop profiling.
1203 	 */
1204 	if ((p->p_stflag & PST_PROFIL) != 0) {
1205 		mutex_spin_enter(&p->p_stmutex);
1206 		stopprofclock(p);
1207 		mutex_spin_exit(&p->p_stmutex);
1208 	}
1209 
1210 	/*
1211 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1212 	 * exited and exec()/exit() are the only places it will be cleared.
1213 	 */
1214 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1215 		lwp_t *lp;
1216 
1217 		mutex_enter(proc_lock);
1218 		lp = p->p_vforklwp;
1219 		p->p_vforklwp = NULL;
1220 
1221 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1222 		p->p_lflag &= ~PL_PPWAIT;
1223 		lp->l_vforkwaiting = false;
1224 
1225 		cv_broadcast(&lp->l_waitcv);
1226 		mutex_exit(proc_lock);
1227 	}
1228 
1229 	error = credexec(l, &data->ed_attr);
1230 	if (error)
1231 		goto exec_abort;
1232 
1233 #if defined(__HAVE_RAS)
1234 	/*
1235 	 * Remove all RASs from the address space.
1236 	 */
1237 	ras_purgeall();
1238 #endif
1239 
1240 	doexechooks(p);
1241 
1242 	/*
1243 	 * Set initial SP at the top of the stack.
1244 	 *
1245 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1246 	 * the end of arg/env strings.  Userland guesses the address of argc
1247 	 * via ps_strings::ps_argvstr.
1248 	 */
1249 
1250 	/* Setup new registers and do misc. setup. */
1251 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1252 	if (epp->ep_esch->es_setregs)
1253 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1254 
1255 	/* Provide a consistent LWP private setting */
1256 	(void)lwp_setprivate(l, NULL);
1257 
1258 	/* Discard all PCU state; need to start fresh */
1259 	pcu_discard_all(l);
1260 
1261 	/* map the process's signal trampoline code */
1262 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1263 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1264 		goto exec_abort;
1265 	}
1266 
1267 	pool_put(&exec_pool, data->ed_argp);
1268 
1269 	/* notify others that we exec'd */
1270 	KNOTE(&p->p_klist, NOTE_EXEC);
1271 
1272 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1273 
1274 	SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1275 
1276 	emulexec(l, epp);
1277 
1278 	/* Allow new references from the debugger/procfs. */
1279 	rw_exit(&p->p_reflock);
1280 	if (!no_local_exec_lock)
1281 		rw_exit(&exec_lock);
1282 
1283 	mutex_enter(proc_lock);
1284 
1285 	/* posix_spawn(3) reports a single event with implied exec(3) */
1286 	if ((p->p_slflag & PSL_TRACED) && !is_spawn) {
1287 		mutex_enter(p->p_lock);
1288 		eventswitch(TRAP_EXEC);
1289 		mutex_enter(proc_lock);
1290 	}
1291 
1292 	if (p->p_sflag & PS_STOPEXEC) {
1293 		ksiginfoq_t kq;
1294 
1295 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1296 		p->p_pptr->p_nstopchild++;
1297 		p->p_waited = 0;
1298 		mutex_enter(p->p_lock);
1299 		ksiginfo_queue_init(&kq);
1300 		sigclearall(p, &contsigmask, &kq);
1301 		lwp_lock(l);
1302 		l->l_stat = LSSTOP;
1303 		p->p_stat = SSTOP;
1304 		p->p_nrlwps--;
1305 		lwp_unlock(l);
1306 		mutex_exit(p->p_lock);
1307 		mutex_exit(proc_lock);
1308 		lwp_lock(l);
1309 		mi_switch(l);
1310 		ksiginfo_queue_drain(&kq);
1311 		KERNEL_LOCK(l->l_biglocks, l);
1312 	} else {
1313 		mutex_exit(proc_lock);
1314 	}
1315 
1316 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1317 	pathbuf_destroy(data->ed_pathbuf);
1318 	PNBUF_PUT(data->ed_resolvedpathbuf);
1319 #ifdef TRACE_EXEC
1320 	DPRINTF(("%s finished\n", __func__));
1321 #endif
1322 	return EJUSTRETURN;
1323 
1324  exec_abort:
1325 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1326 	rw_exit(&p->p_reflock);
1327 	if (!no_local_exec_lock)
1328 		rw_exit(&exec_lock);
1329 
1330 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1331 	pathbuf_destroy(data->ed_pathbuf);
1332 	PNBUF_PUT(data->ed_resolvedpathbuf);
1333 
1334 	/*
1335 	 * the old process doesn't exist anymore.  exit gracefully.
1336 	 * get rid of the (new) address space we have created, if any, get rid
1337 	 * of our namei data and vnode, and exit noting failure
1338 	 */
1339 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1340 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1341 
1342 	exec_free_emul_arg(epp);
1343 	pool_put(&exec_pool, data->ed_argp);
1344 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1345 	if (epp->ep_emul_root != NULL)
1346 		vrele(epp->ep_emul_root);
1347 	if (epp->ep_interp != NULL)
1348 		vrele(epp->ep_interp);
1349 
1350 	/* Acquire the sched-state mutex (exit1() will release it). */
1351 	if (!is_spawn) {
1352 		mutex_enter(p->p_lock);
1353 		exit1(l, error, SIGABRT);
1354 	}
1355 
1356 	return error;
1357 }
1358 
1359 int
1360 execve1(struct lwp *l, const char *path, char * const *args,
1361     char * const *envs, execve_fetch_element_t fetch_element)
1362 {
1363 	struct execve_data data;
1364 	int error;
1365 
1366 	error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1367 	if (error)
1368 		return error;
1369 	error = execve_runproc(l, &data, false, false);
1370 	return error;
1371 }
1372 
1373 static size_t
1374 fromptrsz(const struct exec_package *epp)
1375 {
1376 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1377 }
1378 
1379 static size_t
1380 ptrsz(const struct exec_package *epp)
1381 {
1382 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1383 }
1384 
1385 static size_t
1386 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1387 {
1388 	struct exec_package	* const epp = &data->ed_pack;
1389 
1390 	const size_t nargenvptrs =
1391 	    1 +				/* long argc */
1392 	    data->ed_argc +		/* char *argv[] */
1393 	    1 +				/* \0 */
1394 	    data->ed_envc +		/* char *env[] */
1395 	    1;				/* \0 */
1396 
1397 	return (nargenvptrs * ptrsz(epp))	/* pointers */
1398 	    + argenvstrlen			/* strings */
1399 	    + epp->ep_esch->es_arglen;		/* auxinfo */
1400 }
1401 
1402 static size_t
1403 calcstack(struct execve_data * restrict data, const size_t gaplen)
1404 {
1405 	struct exec_package	* const epp = &data->ed_pack;
1406 
1407 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1408 	    epp->ep_esch->es_emul->e_sigcode;
1409 
1410 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1411 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1412 
1413 	const size_t sigcode_psstr_sz =
1414 	    data->ed_szsigcode +	/* sigcode */
1415 	    data->ed_ps_strings_sz +	/* ps_strings */
1416 	    STACK_PTHREADSPACE;		/* pthread space */
1417 
1418 	const size_t stacklen =
1419 	    data->ed_argslen +
1420 	    gaplen +
1421 	    sigcode_psstr_sz;
1422 
1423 	/* make the stack "safely" aligned */
1424 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1425 }
1426 
1427 static int
1428 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1429     char * const newstack)
1430 {
1431 	struct exec_package	* const epp = &data->ed_pack;
1432 	struct proc		*p = l->l_proc;
1433 	int			error;
1434 
1435 	memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo));
1436 
1437 	/* remember information about the process */
1438 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1439 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1440 
1441 	/*
1442 	 * Allocate the stack address passed to the newly execve()'ed process.
1443 	 *
1444 	 * The new stack address will be set to the SP (stack pointer) register
1445 	 * in setregs().
1446 	 */
1447 
1448 	char *newargs = STACK_ALLOC(
1449 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1450 
1451 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1452 	    &data->ed_arginfo, &newargs, data->ed_argp);
1453 
1454 	if (error) {
1455 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1456 		return error;
1457 	}
1458 
1459 	error = copyoutpsstrs(data, p);
1460 	if (error != 0)
1461 		return error;
1462 
1463 	return 0;
1464 }
1465 
1466 static int
1467 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1468 {
1469 	struct exec_package	* const epp = &data->ed_pack;
1470 	struct ps_strings32	arginfo32;
1471 	void			*aip;
1472 	int			error;
1473 
1474 	/* fill process ps_strings info */
1475 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1476 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1477 
1478 	if (epp->ep_flags & EXEC_32) {
1479 		aip = &arginfo32;
1480 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1481 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1482 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1483 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1484 	} else
1485 		aip = &data->ed_arginfo;
1486 
1487 	/* copy out the process's ps_strings structure */
1488 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1489 	    != 0) {
1490 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1491 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1492 		return error;
1493 	}
1494 
1495 	return 0;
1496 }
1497 
1498 static int
1499 copyinargs(struct execve_data * restrict data, char * const *args,
1500     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1501 {
1502 	struct exec_package	* const epp = &data->ed_pack;
1503 	char			*dp;
1504 	size_t			i;
1505 	int			error;
1506 
1507 	dp = *dpp;
1508 
1509 	data->ed_argc = 0;
1510 
1511 	/* copy the fake args list, if there's one, freeing it as we go */
1512 	if (epp->ep_flags & EXEC_HASARGL) {
1513 		struct exec_fakearg	*fa = epp->ep_fa;
1514 
1515 		while (fa->fa_arg != NULL) {
1516 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1517 			size_t len;
1518 
1519 			len = strlcpy(dp, fa->fa_arg, maxlen);
1520 			/* Count NUL into len. */
1521 			if (len < maxlen)
1522 				len++;
1523 			else {
1524 				while (fa->fa_arg != NULL) {
1525 					kmem_free(fa->fa_arg, fa->fa_len);
1526 					fa++;
1527 				}
1528 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1529 				epp->ep_flags &= ~EXEC_HASARGL;
1530 				return E2BIG;
1531 			}
1532 			ktrexecarg(fa->fa_arg, len - 1);
1533 			dp += len;
1534 
1535 			kmem_free(fa->fa_arg, fa->fa_len);
1536 			fa++;
1537 			data->ed_argc++;
1538 		}
1539 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1540 		epp->ep_flags &= ~EXEC_HASARGL;
1541 	}
1542 
1543 	/*
1544 	 * Read and count argument strings from user.
1545 	 */
1546 
1547 	if (args == NULL) {
1548 		DPRINTF(("%s: null args\n", __func__));
1549 		return EINVAL;
1550 	}
1551 	if (epp->ep_flags & EXEC_SKIPARG)
1552 		args = (const void *)((const char *)args + fromptrsz(epp));
1553 	i = 0;
1554 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1555 	if (error != 0) {
1556 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1557 		return error;
1558 	}
1559 	data->ed_argc += i;
1560 
1561 	/*
1562 	 * Read and count environment strings from user.
1563 	 */
1564 
1565 	data->ed_envc = 0;
1566 	/* environment need not be there */
1567 	if (envs == NULL)
1568 		goto done;
1569 	i = 0;
1570 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1571 	if (error != 0) {
1572 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1573 		return error;
1574 	}
1575 	data->ed_envc += i;
1576 
1577 done:
1578 	*dpp = dp;
1579 
1580 	return 0;
1581 }
1582 
1583 static int
1584 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1585     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1586     void (*ktr)(const void *, size_t))
1587 {
1588 	char			*dp, *sp;
1589 	size_t			i;
1590 	int			error;
1591 
1592 	dp = *dpp;
1593 
1594 	i = 0;
1595 	while (1) {
1596 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1597 		size_t len;
1598 
1599 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1600 			return error;
1601 		}
1602 		if (!sp)
1603 			break;
1604 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1605 			if (error == ENAMETOOLONG)
1606 				error = E2BIG;
1607 			return error;
1608 		}
1609 		if (__predict_false(ktrace_on))
1610 			(*ktr)(dp, len - 1);
1611 		dp += len;
1612 		i++;
1613 	}
1614 
1615 	*dpp = dp;
1616 	*ip = i;
1617 
1618 	return 0;
1619 }
1620 
1621 /*
1622  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1623  * Those strings are located just after auxinfo.
1624  */
1625 int
1626 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1627     char **stackp, void *argp)
1628 {
1629 	char	**cpp, *dp, *sp;
1630 	size_t	len;
1631 	void	*nullp;
1632 	long	argc, envc;
1633 	int	error;
1634 
1635 	cpp = (char **)*stackp;
1636 	nullp = NULL;
1637 	argc = arginfo->ps_nargvstr;
1638 	envc = arginfo->ps_nenvstr;
1639 
1640 	/* argc on stack is long */
1641 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1642 
1643 	dp = (char *)(cpp +
1644 	    1 +				/* long argc */
1645 	    argc +			/* char *argv[] */
1646 	    1 +				/* \0 */
1647 	    envc +			/* char *env[] */
1648 	    1) +			/* \0 */
1649 	    pack->ep_esch->es_arglen;	/* auxinfo */
1650 	sp = argp;
1651 
1652 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1653 		COPYPRINTF("", cpp - 1, sizeof(argc));
1654 		return error;
1655 	}
1656 
1657 	/* XXX don't copy them out, remap them! */
1658 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1659 
1660 	for (; --argc >= 0; sp += len, dp += len) {
1661 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1662 			COPYPRINTF("", cpp - 1, sizeof(dp));
1663 			return error;
1664 		}
1665 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1666 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1667 			return error;
1668 		}
1669 	}
1670 
1671 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1672 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1673 		return error;
1674 	}
1675 
1676 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1677 
1678 	for (; --envc >= 0; sp += len, dp += len) {
1679 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1680 			COPYPRINTF("", cpp - 1, sizeof(dp));
1681 			return error;
1682 		}
1683 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1684 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1685 			return error;
1686 		}
1687 
1688 	}
1689 
1690 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1691 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1692 		return error;
1693 	}
1694 
1695 	*stackp = (char *)cpp;
1696 	return 0;
1697 }
1698 
1699 
1700 /*
1701  * Add execsw[] entries.
1702  */
1703 int
1704 exec_add(struct execsw *esp, int count)
1705 {
1706 	struct exec_entry	*it;
1707 	int			i;
1708 
1709 	if (count == 0) {
1710 		return 0;
1711 	}
1712 
1713 	/* Check for duplicates. */
1714 	rw_enter(&exec_lock, RW_WRITER);
1715 	for (i = 0; i < count; i++) {
1716 		LIST_FOREACH(it, &ex_head, ex_list) {
1717 			/* assume unique (makecmds, probe_func, emulation) */
1718 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1719 			    it->ex_sw->u.elf_probe_func ==
1720 			    esp[i].u.elf_probe_func &&
1721 			    it->ex_sw->es_emul == esp[i].es_emul) {
1722 				rw_exit(&exec_lock);
1723 				return EEXIST;
1724 			}
1725 		}
1726 	}
1727 
1728 	/* Allocate new entries. */
1729 	for (i = 0; i < count; i++) {
1730 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1731 		it->ex_sw = &esp[i];
1732 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1733 	}
1734 
1735 	/* update execsw[] */
1736 	exec_init(0);
1737 	rw_exit(&exec_lock);
1738 	return 0;
1739 }
1740 
1741 /*
1742  * Remove execsw[] entry.
1743  */
1744 int
1745 exec_remove(struct execsw *esp, int count)
1746 {
1747 	struct exec_entry	*it, *next;
1748 	int			i;
1749 	const struct proclist_desc *pd;
1750 	proc_t			*p;
1751 
1752 	if (count == 0) {
1753 		return 0;
1754 	}
1755 
1756 	/* Abort if any are busy. */
1757 	rw_enter(&exec_lock, RW_WRITER);
1758 	for (i = 0; i < count; i++) {
1759 		mutex_enter(proc_lock);
1760 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1761 			PROCLIST_FOREACH(p, pd->pd_list) {
1762 				if (p->p_execsw == &esp[i]) {
1763 					mutex_exit(proc_lock);
1764 					rw_exit(&exec_lock);
1765 					return EBUSY;
1766 				}
1767 			}
1768 		}
1769 		mutex_exit(proc_lock);
1770 	}
1771 
1772 	/* None are busy, so remove them all. */
1773 	for (i = 0; i < count; i++) {
1774 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1775 			next = LIST_NEXT(it, ex_list);
1776 			if (it->ex_sw == &esp[i]) {
1777 				LIST_REMOVE(it, ex_list);
1778 				kmem_free(it, sizeof(*it));
1779 				break;
1780 			}
1781 		}
1782 	}
1783 
1784 	/* update execsw[] */
1785 	exec_init(0);
1786 	rw_exit(&exec_lock);
1787 	return 0;
1788 }
1789 
1790 /*
1791  * Initialize exec structures. If init_boot is true, also does necessary
1792  * one-time initialization (it's called from main() that way).
1793  * Once system is multiuser, this should be called with exec_lock held,
1794  * i.e. via exec_{add|remove}().
1795  */
1796 int
1797 exec_init(int init_boot)
1798 {
1799 	const struct execsw 	**sw;
1800 	struct exec_entry	*ex;
1801 	SLIST_HEAD(,exec_entry)	first;
1802 	SLIST_HEAD(,exec_entry)	any;
1803 	SLIST_HEAD(,exec_entry)	last;
1804 	int			i, sz;
1805 
1806 	if (init_boot) {
1807 		/* do one-time initializations */
1808 		vaddr_t vmin = 0, vmax;
1809 
1810 		rw_init(&exec_lock);
1811 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1812 		exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1813 		    maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1814 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1815 		    "execargs", &exec_palloc, IPL_NONE);
1816 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1817 	} else {
1818 		KASSERT(rw_write_held(&exec_lock));
1819 	}
1820 
1821 	/* Sort each entry onto the appropriate queue. */
1822 	SLIST_INIT(&first);
1823 	SLIST_INIT(&any);
1824 	SLIST_INIT(&last);
1825 	sz = 0;
1826 	LIST_FOREACH(ex, &ex_head, ex_list) {
1827 		switch(ex->ex_sw->es_prio) {
1828 		case EXECSW_PRIO_FIRST:
1829 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1830 			break;
1831 		case EXECSW_PRIO_ANY:
1832 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1833 			break;
1834 		case EXECSW_PRIO_LAST:
1835 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1836 			break;
1837 		default:
1838 			panic("%s", __func__);
1839 			break;
1840 		}
1841 		sz++;
1842 	}
1843 
1844 	/*
1845 	 * Create new execsw[].  Ensure we do not try a zero-sized
1846 	 * allocation.
1847 	 */
1848 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1849 	i = 0;
1850 	SLIST_FOREACH(ex, &first, ex_slist) {
1851 		sw[i++] = ex->ex_sw;
1852 	}
1853 	SLIST_FOREACH(ex, &any, ex_slist) {
1854 		sw[i++] = ex->ex_sw;
1855 	}
1856 	SLIST_FOREACH(ex, &last, ex_slist) {
1857 		sw[i++] = ex->ex_sw;
1858 	}
1859 
1860 	/* Replace old execsw[] and free used memory. */
1861 	if (execsw != NULL) {
1862 		kmem_free(__UNCONST(execsw),
1863 		    nexecs * sizeof(struct execsw *) + 1);
1864 	}
1865 	execsw = sw;
1866 	nexecs = sz;
1867 
1868 	/* Figure out the maximum size of an exec header. */
1869 	exec_maxhdrsz = sizeof(int);
1870 	for (i = 0; i < nexecs; i++) {
1871 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1872 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1873 	}
1874 
1875 	return 0;
1876 }
1877 
1878 static int
1879 exec_sigcode_map(struct proc *p, const struct emul *e)
1880 {
1881 	vaddr_t va;
1882 	vsize_t sz;
1883 	int error;
1884 	struct uvm_object *uobj;
1885 
1886 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1887 
1888 	if (e->e_sigobject == NULL || sz == 0) {
1889 		return 0;
1890 	}
1891 
1892 	/*
1893 	 * If we don't have a sigobject for this emulation, create one.
1894 	 *
1895 	 * sigobject is an anonymous memory object (just like SYSV shared
1896 	 * memory) that we keep a permanent reference to and that we map
1897 	 * in all processes that need this sigcode. The creation is simple,
1898 	 * we create an object, add a permanent reference to it, map it in
1899 	 * kernel space, copy out the sigcode to it and unmap it.
1900 	 * We map it with PROT_READ|PROT_EXEC into the process just
1901 	 * the way sys_mmap() would map it.
1902 	 */
1903 
1904 	uobj = *e->e_sigobject;
1905 	if (uobj == NULL) {
1906 		mutex_enter(&sigobject_lock);
1907 		if ((uobj = *e->e_sigobject) == NULL) {
1908 			uobj = uao_create(sz, 0);
1909 			(*uobj->pgops->pgo_reference)(uobj);
1910 			va = vm_map_min(kernel_map);
1911 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1912 			    uobj, 0, 0,
1913 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1914 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1915 				printf("kernel mapping failed %d\n", error);
1916 				(*uobj->pgops->pgo_detach)(uobj);
1917 				mutex_exit(&sigobject_lock);
1918 				return error;
1919 			}
1920 			memcpy((void *)va, e->e_sigcode, sz);
1921 #ifdef PMAP_NEED_PROCWR
1922 			pmap_procwr(&proc0, va, sz);
1923 #endif
1924 			uvm_unmap(kernel_map, va, va + round_page(sz));
1925 			*e->e_sigobject = uobj;
1926 		}
1927 		mutex_exit(&sigobject_lock);
1928 	}
1929 
1930 	/* Just a hint to uvm_map where to put it. */
1931 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1932 	    round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
1933 
1934 #ifdef __alpha__
1935 	/*
1936 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1937 	 * which causes the above calculation to put the sigcode at
1938 	 * an invalid address.  Put it just below the text instead.
1939 	 */
1940 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1941 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1942 	}
1943 #endif
1944 
1945 	(*uobj->pgops->pgo_reference)(uobj);
1946 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1947 			uobj, 0, 0,
1948 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1949 				    UVM_ADV_RANDOM, 0));
1950 	if (error) {
1951 		DPRINTF(("%s, %d: map %p "
1952 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1953 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1954 		    va, error));
1955 		(*uobj->pgops->pgo_detach)(uobj);
1956 		return error;
1957 	}
1958 	p->p_sigctx.ps_sigcode = (void *)va;
1959 	return 0;
1960 }
1961 
1962 /*
1963  * Release a refcount on spawn_exec_data and destroy memory, if this
1964  * was the last one.
1965  */
1966 static void
1967 spawn_exec_data_release(struct spawn_exec_data *data)
1968 {
1969 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1970 		return;
1971 
1972 	cv_destroy(&data->sed_cv_child_ready);
1973 	mutex_destroy(&data->sed_mtx_child);
1974 
1975 	if (data->sed_actions)
1976 		posix_spawn_fa_free(data->sed_actions,
1977 		    data->sed_actions->len);
1978 	if (data->sed_attrs)
1979 		kmem_free(data->sed_attrs,
1980 		    sizeof(*data->sed_attrs));
1981 	kmem_free(data, sizeof(*data));
1982 }
1983 
1984 /*
1985  * A child lwp of a posix_spawn operation starts here and ends up in
1986  * cpu_spawn_return, dealing with all filedescriptor and scheduler
1987  * manipulations in between.
1988  * The parent waits for the child, as it is not clear whether the child
1989  * will be able to acquire its own exec_lock. If it can, the parent can
1990  * be released early and continue running in parallel. If not (or if the
1991  * magic debug flag is passed in the scheduler attribute struct), the
1992  * child rides on the parent's exec lock until it is ready to return to
1993  * to userland - and only then releases the parent. This method loses
1994  * concurrency, but improves error reporting.
1995  */
1996 static void
1997 spawn_return(void *arg)
1998 {
1999 	struct spawn_exec_data *spawn_data = arg;
2000 	struct lwp *l = curlwp;
2001 	struct proc *p = l->l_proc;
2002 	int error, newfd;
2003 	int ostat;
2004 	size_t i;
2005 	const struct posix_spawn_file_actions_entry *fae;
2006 	pid_t ppid;
2007 	register_t retval;
2008 	bool have_reflock;
2009 	bool parent_is_waiting = true;
2010 
2011 	/*
2012 	 * Check if we can release parent early.
2013 	 * We either need to have no sed_attrs, or sed_attrs does not
2014 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2015 	 * safe access to the parent proc (passed in sed_parent).
2016 	 * We then try to get the exec_lock, and only if that works, we can
2017 	 * release the parent here already.
2018 	 */
2019 	ppid = spawn_data->sed_parent->p_pid;
2020 	if ((!spawn_data->sed_attrs
2021 	    || (spawn_data->sed_attrs->sa_flags
2022 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2023 	    && rw_tryenter(&exec_lock, RW_READER)) {
2024 		parent_is_waiting = false;
2025 		mutex_enter(&spawn_data->sed_mtx_child);
2026 		cv_signal(&spawn_data->sed_cv_child_ready);
2027 		mutex_exit(&spawn_data->sed_mtx_child);
2028 	}
2029 
2030 	/* don't allow debugger access yet */
2031 	rw_enter(&p->p_reflock, RW_WRITER);
2032 	have_reflock = true;
2033 
2034 	error = 0;
2035 	/* handle posix_spawn_file_actions */
2036 	if (spawn_data->sed_actions != NULL) {
2037 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
2038 			fae = &spawn_data->sed_actions->fae[i];
2039 			switch (fae->fae_action) {
2040 			case FAE_OPEN:
2041 				if (fd_getfile(fae->fae_fildes) != NULL) {
2042 					error = fd_close(fae->fae_fildes);
2043 					if (error)
2044 						break;
2045 				}
2046 				error = fd_open(fae->fae_path, fae->fae_oflag,
2047 				    fae->fae_mode, &newfd);
2048 				if (error)
2049 					break;
2050 				if (newfd != fae->fae_fildes) {
2051 					error = dodup(l, newfd,
2052 					    fae->fae_fildes, 0, &retval);
2053 					if (fd_getfile(newfd) != NULL)
2054 						fd_close(newfd);
2055 				}
2056 				break;
2057 			case FAE_DUP2:
2058 				error = dodup(l, fae->fae_fildes,
2059 				    fae->fae_newfildes, 0, &retval);
2060 				break;
2061 			case FAE_CLOSE:
2062 				if (fd_getfile(fae->fae_fildes) == NULL) {
2063 					error = EBADF;
2064 					break;
2065 				}
2066 				error = fd_close(fae->fae_fildes);
2067 				break;
2068 			}
2069 			if (error)
2070 				goto report_error;
2071 		}
2072 	}
2073 
2074 	/* handle posix_spawnattr */
2075 	if (spawn_data->sed_attrs != NULL) {
2076 		struct sigaction sigact;
2077 		memset(&sigact, 0, sizeof(sigact));
2078 		sigact._sa_u._sa_handler = SIG_DFL;
2079 		sigact.sa_flags = 0;
2080 
2081 		/*
2082 		 * set state to SSTOP so that this proc can be found by pid.
2083 		 * see proc_enterprp, do_sched_setparam below
2084 		 */
2085 		mutex_enter(proc_lock);
2086 		/*
2087 		 * p_stat should be SACTIVE, so we need to adjust the
2088 		 * parent's p_nstopchild here.  For safety, just make
2089 		 * we're on the good side of SDEAD before we adjust.
2090 		 */
2091 		ostat = p->p_stat;
2092 		KASSERT(ostat < SSTOP);
2093 		p->p_stat = SSTOP;
2094 		p->p_waited = 0;
2095 		p->p_pptr->p_nstopchild++;
2096 		mutex_exit(proc_lock);
2097 
2098 		/* Set process group */
2099 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2100 			pid_t mypid = p->p_pid,
2101 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
2102 
2103 			if (pgrp == 0)
2104 				pgrp = mypid;
2105 
2106 			error = proc_enterpgrp(spawn_data->sed_parent,
2107 			    mypid, pgrp, false);
2108 			if (error)
2109 				goto report_error_stopped;
2110 		}
2111 
2112 		/* Set scheduler policy */
2113 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2114 			error = do_sched_setparam(p->p_pid, 0,
2115 			    spawn_data->sed_attrs->sa_schedpolicy,
2116 			    &spawn_data->sed_attrs->sa_schedparam);
2117 		else if (spawn_data->sed_attrs->sa_flags
2118 		    & POSIX_SPAWN_SETSCHEDPARAM) {
2119 			error = do_sched_setparam(ppid, 0,
2120 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2121 		}
2122 		if (error)
2123 			goto report_error_stopped;
2124 
2125 		/* Reset user ID's */
2126 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2127 			error = do_setresuid(l, -1,
2128 			     kauth_cred_getgid(l->l_cred), -1,
2129 			     ID_E_EQ_R | ID_E_EQ_S);
2130 			if (error)
2131 				goto report_error_stopped;
2132 			error = do_setresuid(l, -1,
2133 			    kauth_cred_getuid(l->l_cred), -1,
2134 			    ID_E_EQ_R | ID_E_EQ_S);
2135 			if (error)
2136 				goto report_error_stopped;
2137 		}
2138 
2139 		/* Set signal masks/defaults */
2140 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2141 			mutex_enter(p->p_lock);
2142 			error = sigprocmask1(l, SIG_SETMASK,
2143 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
2144 			mutex_exit(p->p_lock);
2145 			if (error)
2146 				goto report_error_stopped;
2147 		}
2148 
2149 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2150 			/*
2151 			 * The following sigaction call is using a sigaction
2152 			 * version 0 trampoline which is in the compatibility
2153 			 * code only. This is not a problem because for SIG_DFL
2154 			 * and SIG_IGN, the trampolines are now ignored. If they
2155 			 * were not, this would be a problem because we are
2156 			 * holding the exec_lock, and the compat code needs
2157 			 * to do the same in order to replace the trampoline
2158 			 * code of the process.
2159 			 */
2160 			for (i = 1; i <= NSIG; i++) {
2161 				if (sigismember(
2162 				    &spawn_data->sed_attrs->sa_sigdefault, i))
2163 					sigaction1(l, i, &sigact, NULL, NULL,
2164 					    0);
2165 			}
2166 		}
2167 		mutex_enter(proc_lock);
2168 		p->p_stat = ostat;
2169 		p->p_pptr->p_nstopchild--;
2170 		mutex_exit(proc_lock);
2171 	}
2172 
2173 	/* now do the real exec */
2174 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2175 	    true);
2176 	have_reflock = false;
2177 	if (error == EJUSTRETURN)
2178 		error = 0;
2179 	else if (error)
2180 		goto report_error;
2181 
2182 	if (parent_is_waiting) {
2183 		mutex_enter(&spawn_data->sed_mtx_child);
2184 		cv_signal(&spawn_data->sed_cv_child_ready);
2185 		mutex_exit(&spawn_data->sed_mtx_child);
2186 	}
2187 
2188 	/* release our refcount on the data */
2189 	spawn_exec_data_release(spawn_data);
2190 
2191 	if (p->p_slflag & PSL_TRACED) {
2192 		/* Paranoid check */
2193 		mutex_enter(proc_lock);
2194 		if (!(p->p_slflag & PSL_TRACED)) {
2195 			mutex_exit(proc_lock);
2196 			goto cpu_return;
2197 		}
2198 
2199 		mutex_enter(p->p_lock);
2200 		eventswitch(TRAP_CHLD);
2201 	}
2202 
2203  cpu_return:
2204 	/* and finally: leave to userland for the first time */
2205 	cpu_spawn_return(l);
2206 
2207 	/* NOTREACHED */
2208 	return;
2209 
2210  report_error_stopped:
2211 	mutex_enter(proc_lock);
2212 	p->p_stat = ostat;
2213 	p->p_pptr->p_nstopchild--;
2214 	mutex_exit(proc_lock);
2215  report_error:
2216 	if (have_reflock) {
2217 		/*
2218 		 * We have not passed through execve_runproc(),
2219 		 * which would have released the p_reflock and also
2220 		 * taken ownership of the sed_exec part of spawn_data,
2221 		 * so release/free both here.
2222 		 */
2223 		rw_exit(&p->p_reflock);
2224 		execve_free_data(&spawn_data->sed_exec);
2225 	}
2226 
2227 	if (parent_is_waiting) {
2228 		/* pass error to parent */
2229 		mutex_enter(&spawn_data->sed_mtx_child);
2230 		spawn_data->sed_error = error;
2231 		cv_signal(&spawn_data->sed_cv_child_ready);
2232 		mutex_exit(&spawn_data->sed_mtx_child);
2233 	} else {
2234 		rw_exit(&exec_lock);
2235 	}
2236 
2237 	/* release our refcount on the data */
2238 	spawn_exec_data_release(spawn_data);
2239 
2240 	/* done, exit */
2241 	mutex_enter(p->p_lock);
2242 	/*
2243 	 * Posix explicitly asks for an exit code of 127 if we report
2244 	 * errors from the child process - so, unfortunately, there
2245 	 * is no way to report a more exact error code.
2246 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2247 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2248 	 */
2249 	exit1(l, 127, 0);
2250 }
2251 
2252 void
2253 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2254 {
2255 
2256 	for (size_t i = 0; i < len; i++) {
2257 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2258 		if (fae->fae_action != FAE_OPEN)
2259 			continue;
2260 		kmem_strfree(fae->fae_path);
2261 	}
2262 	if (fa->len > 0)
2263 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2264 	kmem_free(fa, sizeof(*fa));
2265 }
2266 
2267 static int
2268 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2269     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2270 {
2271 	struct posix_spawn_file_actions *fa;
2272 	struct posix_spawn_file_actions_entry *fae;
2273 	char *pbuf = NULL;
2274 	int error;
2275 	size_t i = 0;
2276 
2277 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2278 	error = copyin(ufa, fa, sizeof(*fa));
2279 	if (error || fa->len == 0) {
2280 		kmem_free(fa, sizeof(*fa));
2281 		return error;	/* 0 if not an error, and len == 0 */
2282 	}
2283 
2284 	if (fa->len > lim) {
2285 		kmem_free(fa, sizeof(*fa));
2286 		return EINVAL;
2287 	}
2288 
2289 	fa->size = fa->len;
2290 	size_t fal = fa->len * sizeof(*fae);
2291 	fae = fa->fae;
2292 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2293 	error = copyin(fae, fa->fae, fal);
2294 	if (error)
2295 		goto out;
2296 
2297 	pbuf = PNBUF_GET();
2298 	for (; i < fa->len; i++) {
2299 		fae = &fa->fae[i];
2300 		if (fae->fae_action != FAE_OPEN)
2301 			continue;
2302 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2303 		if (error)
2304 			goto out;
2305 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2306 		memcpy(fae->fae_path, pbuf, fal);
2307 	}
2308 	PNBUF_PUT(pbuf);
2309 
2310 	*fap = fa;
2311 	return 0;
2312 out:
2313 	if (pbuf)
2314 		PNBUF_PUT(pbuf);
2315 	posix_spawn_fa_free(fa, i);
2316 	return error;
2317 }
2318 
2319 int
2320 check_posix_spawn(struct lwp *l1)
2321 {
2322 	int error, tnprocs, count;
2323 	uid_t uid;
2324 	struct proc *p1;
2325 
2326 	p1 = l1->l_proc;
2327 	uid = kauth_cred_getuid(l1->l_cred);
2328 	tnprocs = atomic_inc_uint_nv(&nprocs);
2329 
2330 	/*
2331 	 * Although process entries are dynamically created, we still keep
2332 	 * a global limit on the maximum number we will create.
2333 	 */
2334 	if (__predict_false(tnprocs >= maxproc))
2335 		error = -1;
2336 	else
2337 		error = kauth_authorize_process(l1->l_cred,
2338 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2339 
2340 	if (error) {
2341 		atomic_dec_uint(&nprocs);
2342 		return EAGAIN;
2343 	}
2344 
2345 	/*
2346 	 * Enforce limits.
2347 	 */
2348 	count = chgproccnt(uid, 1);
2349 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2350 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2351 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2352 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2353 		(void)chgproccnt(uid, -1);
2354 		atomic_dec_uint(&nprocs);
2355 		return EAGAIN;
2356 	}
2357 
2358 	return 0;
2359 }
2360 
2361 int
2362 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2363 	struct posix_spawn_file_actions *fa,
2364 	struct posix_spawnattr *sa,
2365 	char *const *argv, char *const *envp,
2366 	execve_fetch_element_t fetch)
2367 {
2368 
2369 	struct proc *p1, *p2;
2370 	struct lwp *l2;
2371 	int error;
2372 	struct spawn_exec_data *spawn_data;
2373 	vaddr_t uaddr;
2374 	pid_t pid;
2375 	bool have_exec_lock = false;
2376 
2377 	p1 = l1->l_proc;
2378 
2379 	/* Allocate and init spawn_data */
2380 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2381 	spawn_data->sed_refcnt = 1; /* only parent so far */
2382 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2383 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2384 	mutex_enter(&spawn_data->sed_mtx_child);
2385 
2386 	/*
2387 	 * Do the first part of the exec now, collect state
2388 	 * in spawn_data.
2389 	 */
2390 	error = execve_loadvm(l1, path, argv,
2391 	    envp, fetch, &spawn_data->sed_exec);
2392 	if (error == EJUSTRETURN)
2393 		error = 0;
2394 	else if (error)
2395 		goto error_exit;
2396 
2397 	have_exec_lock = true;
2398 
2399 	/*
2400 	 * Allocate virtual address space for the U-area now, while it
2401 	 * is still easy to abort the fork operation if we're out of
2402 	 * kernel virtual address space.
2403 	 */
2404 	uaddr = uvm_uarea_alloc();
2405 	if (__predict_false(uaddr == 0)) {
2406 		error = ENOMEM;
2407 		goto error_exit;
2408 	}
2409 
2410 	/*
2411 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2412 	 * replace it with its own before returning to userland
2413 	 * in the child.
2414 	 * This is a point of no return, we will have to go through
2415 	 * the child proc to properly clean it up past this point.
2416 	 */
2417 	p2 = proc_alloc();
2418 	pid = p2->p_pid;
2419 
2420 	/*
2421 	 * Make a proc table entry for the new process.
2422 	 * Start by zeroing the section of proc that is zero-initialized,
2423 	 * then copy the section that is copied directly from the parent.
2424 	 */
2425 	memset(&p2->p_startzero, 0,
2426 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2427 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2428 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2429 	p2->p_vmspace = proc0.p_vmspace;
2430 
2431 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2432 
2433 	LIST_INIT(&p2->p_lwps);
2434 	LIST_INIT(&p2->p_sigwaiters);
2435 
2436 	/*
2437 	 * Duplicate sub-structures as needed.
2438 	 * Increase reference counts on shared objects.
2439 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2440 	 * handling are important in order to keep a consistent behaviour
2441 	 * for the child after the fork.  If we are a 32-bit process, the
2442 	 * child will be too.
2443 	 */
2444 	p2->p_flag =
2445 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2446 	p2->p_emul = p1->p_emul;
2447 	p2->p_execsw = p1->p_execsw;
2448 
2449 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2450 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2451 	rw_init(&p2->p_reflock);
2452 	cv_init(&p2->p_waitcv, "wait");
2453 	cv_init(&p2->p_lwpcv, "lwpwait");
2454 
2455 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2456 
2457 	kauth_proc_fork(p1, p2);
2458 
2459 	p2->p_raslist = NULL;
2460 	p2->p_fd = fd_copy();
2461 
2462 	/* XXX racy */
2463 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2464 
2465 	p2->p_cwdi = cwdinit();
2466 
2467 	/*
2468 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2469 	 * we just need increase pl_refcnt.
2470 	 */
2471 	if (!p1->p_limit->pl_writeable) {
2472 		lim_addref(p1->p_limit);
2473 		p2->p_limit = p1->p_limit;
2474 	} else {
2475 		p2->p_limit = lim_copy(p1->p_limit);
2476 	}
2477 
2478 	p2->p_lflag = 0;
2479 	l1->l_vforkwaiting = false;
2480 	p2->p_sflag = 0;
2481 	p2->p_slflag = 0;
2482 	p2->p_pptr = p1;
2483 	p2->p_ppid = p1->p_pid;
2484 	LIST_INIT(&p2->p_children);
2485 
2486 	p2->p_aio = NULL;
2487 
2488 #ifdef KTRACE
2489 	/*
2490 	 * Copy traceflag and tracefile if enabled.
2491 	 * If not inherited, these were zeroed above.
2492 	 */
2493 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2494 		mutex_enter(&ktrace_lock);
2495 		p2->p_traceflag = p1->p_traceflag;
2496 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2497 			ktradref(p2);
2498 		mutex_exit(&ktrace_lock);
2499 	}
2500 #endif
2501 
2502 	/*
2503 	 * Create signal actions for the child process.
2504 	 */
2505 	p2->p_sigacts = sigactsinit(p1, 0);
2506 	mutex_enter(p1->p_lock);
2507 	p2->p_sflag |=
2508 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2509 	sched_proc_fork(p1, p2);
2510 	mutex_exit(p1->p_lock);
2511 
2512 	p2->p_stflag = p1->p_stflag;
2513 
2514 	/*
2515 	 * p_stats.
2516 	 * Copy parts of p_stats, and zero out the rest.
2517 	 */
2518 	p2->p_stats = pstatscopy(p1->p_stats);
2519 
2520 	/* copy over machdep flags to the new proc */
2521 	cpu_proc_fork(p1, p2);
2522 
2523 	/*
2524 	 * Prepare remaining parts of spawn data
2525 	 */
2526 	spawn_data->sed_actions = fa;
2527 	spawn_data->sed_attrs = sa;
2528 
2529 	spawn_data->sed_parent = p1;
2530 
2531 	/* create LWP */
2532 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2533 	    &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2534 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2535 
2536 	/*
2537 	 * Copy the credential so other references don't see our changes.
2538 	 * Test to see if this is necessary first, since in the common case
2539 	 * we won't need a private reference.
2540 	 */
2541 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2542 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2543 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2544 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2545 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2546 	}
2547 
2548 	/* Update the master credentials. */
2549 	if (l2->l_cred != p2->p_cred) {
2550 		kauth_cred_t ocred;
2551 
2552 		kauth_cred_hold(l2->l_cred);
2553 		mutex_enter(p2->p_lock);
2554 		ocred = p2->p_cred;
2555 		p2->p_cred = l2->l_cred;
2556 		mutex_exit(p2->p_lock);
2557 		kauth_cred_free(ocred);
2558 	}
2559 
2560 	*child_ok = true;
2561 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2562 #if 0
2563 	l2->l_nopreempt = 1; /* start it non-preemptable */
2564 #endif
2565 
2566 	/*
2567 	 * It's now safe for the scheduler and other processes to see the
2568 	 * child process.
2569 	 */
2570 	mutex_enter(proc_lock);
2571 
2572 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2573 		p2->p_lflag |= PL_CONTROLT;
2574 
2575 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2576 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2577 
2578 	if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) ==
2579 	    (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2580 		proc_changeparent(p2, p1->p_pptr);
2581 		p1->p_pspid = p2->p_pid;
2582 		p2->p_pspid = p1->p_pid;
2583 	}
2584 
2585 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2586 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2587 
2588 	p2->p_trace_enabled = trace_is_enabled(p2);
2589 #ifdef __HAVE_SYSCALL_INTERN
2590 	(*p2->p_emul->e_syscall_intern)(p2);
2591 #endif
2592 
2593 	/*
2594 	 * Make child runnable, set start time, and add to run queue except
2595 	 * if the parent requested the child to start in SSTOP state.
2596 	 */
2597 	mutex_enter(p2->p_lock);
2598 
2599 	getmicrotime(&p2->p_stats->p_start);
2600 
2601 	lwp_lock(l2);
2602 	KASSERT(p2->p_nrlwps == 1);
2603 	p2->p_nrlwps = 1;
2604 	p2->p_stat = SACTIVE;
2605 	l2->l_stat = LSRUN;
2606 	sched_enqueue(l2, false);
2607 	lwp_unlock(l2);
2608 
2609 	mutex_exit(p2->p_lock);
2610 	mutex_exit(proc_lock);
2611 
2612 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2613 	error = spawn_data->sed_error;
2614 	mutex_exit(&spawn_data->sed_mtx_child);
2615 	spawn_exec_data_release(spawn_data);
2616 
2617 	rw_exit(&p1->p_reflock);
2618 	rw_exit(&exec_lock);
2619 	have_exec_lock = false;
2620 
2621 	*pid_res = pid;
2622 
2623 	if (error)
2624 		return error;
2625 
2626 	if (p1->p_slflag & PSL_TRACED) {
2627 		/* Paranoid check */
2628 		mutex_enter(proc_lock);
2629 		if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) !=
2630 		    (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2631 			mutex_exit(proc_lock);
2632 			return 0;
2633 		}
2634 
2635 		mutex_enter(p1->p_lock);
2636 		eventswitch(TRAP_CHLD);
2637 	}
2638 	return 0;
2639 
2640  error_exit:
2641 	if (have_exec_lock) {
2642 		execve_free_data(&spawn_data->sed_exec);
2643 		rw_exit(&p1->p_reflock);
2644 		rw_exit(&exec_lock);
2645 	}
2646 	mutex_exit(&spawn_data->sed_mtx_child);
2647 	spawn_exec_data_release(spawn_data);
2648 
2649 	return error;
2650 }
2651 
2652 int
2653 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2654     register_t *retval)
2655 {
2656 	/* {
2657 		syscallarg(pid_t *) pid;
2658 		syscallarg(const char *) path;
2659 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2660 		syscallarg(const struct posix_spawnattr *) attrp;
2661 		syscallarg(char *const *) argv;
2662 		syscallarg(char *const *) envp;
2663 	} */
2664 
2665 	int error;
2666 	struct posix_spawn_file_actions *fa = NULL;
2667 	struct posix_spawnattr *sa = NULL;
2668 	pid_t pid;
2669 	bool child_ok = false;
2670 	rlim_t max_fileactions;
2671 	proc_t *p = l1->l_proc;
2672 
2673 	error = check_posix_spawn(l1);
2674 	if (error) {
2675 		*retval = error;
2676 		return 0;
2677 	}
2678 
2679 	/* copy in file_actions struct */
2680 	if (SCARG(uap, file_actions) != NULL) {
2681 		max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2682 		    maxfiles);
2683 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2684 		    max_fileactions);
2685 		if (error)
2686 			goto error_exit;
2687 	}
2688 
2689 	/* copyin posix_spawnattr struct */
2690 	if (SCARG(uap, attrp) != NULL) {
2691 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2692 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2693 		if (error)
2694 			goto error_exit;
2695 	}
2696 
2697 	/*
2698 	 * Do the spawn
2699 	 */
2700 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2701 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2702 	if (error)
2703 		goto error_exit;
2704 
2705 	if (error == 0 && SCARG(uap, pid) != NULL)
2706 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2707 
2708 	*retval = error;
2709 	return 0;
2710 
2711  error_exit:
2712 	if (!child_ok) {
2713 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2714 		atomic_dec_uint(&nprocs);
2715 
2716 		if (sa)
2717 			kmem_free(sa, sizeof(*sa));
2718 		if (fa)
2719 			posix_spawn_fa_free(fa, fa->len);
2720 	}
2721 
2722 	*retval = error;
2723 	return 0;
2724 }
2725 
2726 void
2727 exec_free_emul_arg(struct exec_package *epp)
2728 {
2729 	if (epp->ep_emul_arg_free != NULL) {
2730 		KASSERT(epp->ep_emul_arg != NULL);
2731 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2732 		epp->ep_emul_arg_free = NULL;
2733 		epp->ep_emul_arg = NULL;
2734 	} else {
2735 		KASSERT(epp->ep_emul_arg == NULL);
2736 	}
2737 }
2738 
2739 #ifdef DEBUG_EXEC
2740 static void
2741 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2742 {
2743 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2744 	size_t j;
2745 
2746 	if (error == 0)
2747 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2748 	else
2749 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2750 		    epp->ep_vmcmds.evs_used, error));
2751 
2752 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2753 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2754 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2755 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2756 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2757 		    "pagedvn" :
2758 		    vp[j].ev_proc == vmcmd_map_readvn ?
2759 		    "readvn" :
2760 		    vp[j].ev_proc == vmcmd_map_zero ?
2761 		    "zero" : "*unknown*",
2762 		    vp[j].ev_addr, vp[j].ev_len,
2763 		    vp[j].ev_offset, vp[j].ev_prot,
2764 		    vp[j].ev_flags));
2765 		if (error != 0 && j == x)
2766 			DPRINTF(("     ^--- failed\n"));
2767 	}
2768 }
2769 #endif
2770