xref: /netbsd-src/sys/kern/kern_exec.c (revision c9496f6b604074a9451a67df576a5b423068e71e)
1 /*	$NetBSD: kern_exec.c,v 1.455 2018/01/09 20:55:43 maya 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 1.455 2018/01/09 20:55:43 maya 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/kmem.h>
79 #include <sys/namei.h>
80 #include <sys/vnode.h>
81 #include <sys/file.h>
82 #include <sys/filedesc.h>
83 #include <sys/acct.h>
84 #include <sys/atomic.h>
85 #include <sys/exec.h>
86 #include <sys/ktrace.h>
87 #include <sys/uidinfo.h>
88 #include <sys/wait.h>
89 #include <sys/mman.h>
90 #include <sys/ras.h>
91 #include <sys/signalvar.h>
92 #include <sys/stat.h>
93 #include <sys/syscall.h>
94 #include <sys/kauth.h>
95 #include <sys/lwpctl.h>
96 #include <sys/pax.h>
97 #include <sys/cpu.h>
98 #include <sys/module.h>
99 #include <sys/syscallvar.h>
100 #include <sys/syscallargs.h>
101 #if NVERIEXEC > 0
102 #include <sys/verified_exec.h>
103 #endif /* NVERIEXEC > 0 */
104 #include <sys/sdt.h>
105 #include <sys/spawn.h>
106 #include <sys/prot.h>
107 #include <sys/cprng.h>
108 
109 #include <uvm/uvm_extern.h>
110 
111 #include <machine/reg.h>
112 
113 #include <compat/common/compat_util.h>
114 
115 #ifndef MD_TOPDOWN_INIT
116 #ifdef __USE_TOPDOWN_VM
117 #define	MD_TOPDOWN_INIT(epp)	(epp)->ep_flags |= EXEC_TOPDOWN_VM
118 #else
119 #define	MD_TOPDOWN_INIT(epp)
120 #endif
121 #endif
122 
123 struct execve_data;
124 
125 extern int user_va0_disable;
126 
127 static size_t calcargs(struct execve_data * restrict, const size_t);
128 static size_t calcstack(struct execve_data * restrict, const size_t);
129 static int copyoutargs(struct execve_data * restrict, struct lwp *,
130     char * const);
131 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
132 static int copyinargs(struct execve_data * restrict, char * const *,
133     char * const *, execve_fetch_element_t, char **);
134 static int copyinargstrs(struct execve_data * restrict, char * const *,
135     execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
136 static int exec_sigcode_map(struct proc *, const struct emul *);
137 
138 #if defined(DEBUG) && !defined(DEBUG_EXEC)
139 #define DEBUG_EXEC
140 #endif
141 #ifdef DEBUG_EXEC
142 #define DPRINTF(a) printf a
143 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
144     __LINE__, (s), (a), (b))
145 static void dump_vmcmds(const struct exec_package * const, size_t, int);
146 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
147 #else
148 #define DPRINTF(a)
149 #define COPYPRINTF(s, a, b)
150 #define DUMPVMCMDS(p, x, e) do {} while (0)
151 #endif /* DEBUG_EXEC */
152 
153 /*
154  * DTrace SDT provider definitions
155  */
156 SDT_PROVIDER_DECLARE(proc);
157 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
158 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
159 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
160 
161 /*
162  * Exec function switch:
163  *
164  * Note that each makecmds function is responsible for loading the
165  * exec package with the necessary functions for any exec-type-specific
166  * handling.
167  *
168  * Functions for specific exec types should be defined in their own
169  * header file.
170  */
171 static const struct execsw	**execsw = NULL;
172 static int			nexecs;
173 
174 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
175 
176 /* list of dynamically loaded execsw entries */
177 static LIST_HEAD(execlist_head, exec_entry) ex_head =
178     LIST_HEAD_INITIALIZER(ex_head);
179 struct exec_entry {
180 	LIST_ENTRY(exec_entry)	ex_list;
181 	SLIST_ENTRY(exec_entry)	ex_slist;
182 	const struct execsw	*ex_sw;
183 };
184 
185 #ifndef __HAVE_SYSCALL_INTERN
186 void	syscall(void);
187 #endif
188 
189 /* NetBSD autoloadable syscalls */
190 #ifdef MODULAR
191 #include <kern/syscalls_autoload.c>
192 #endif
193 
194 /* NetBSD emul struct */
195 struct emul emul_netbsd = {
196 	.e_name =		"netbsd",
197 #ifdef EMUL_NATIVEROOT
198 	.e_path =		EMUL_NATIVEROOT,
199 #else
200 	.e_path =		NULL,
201 #endif
202 #ifndef __HAVE_MINIMAL_EMUL
203 	.e_flags =		EMUL_HAS_SYS___syscall,
204 	.e_errno =		NULL,
205 	.e_nosys =		SYS_syscall,
206 	.e_nsysent =		SYS_NSYSENT,
207 #endif
208 #ifdef MODULAR
209 	.e_sc_autoload =	netbsd_syscalls_autoload,
210 #endif
211 	.e_sysent =		sysent,
212 #ifdef SYSCALL_DEBUG
213 	.e_syscallnames =	syscallnames,
214 #else
215 	.e_syscallnames =	NULL,
216 #endif
217 	.e_sendsig =		sendsig,
218 	.e_trapsignal =		trapsignal,
219 	.e_tracesig =		NULL,
220 	.e_sigcode =		NULL,
221 	.e_esigcode =		NULL,
222 	.e_sigobject =		NULL,
223 	.e_setregs =		setregs,
224 	.e_proc_exec =		NULL,
225 	.e_proc_fork =		NULL,
226 	.e_proc_exit =		NULL,
227 	.e_lwp_fork =		NULL,
228 	.e_lwp_exit =		NULL,
229 #ifdef __HAVE_SYSCALL_INTERN
230 	.e_syscall_intern =	syscall_intern,
231 #else
232 	.e_syscall =		syscall,
233 #endif
234 	.e_sysctlovly =		NULL,
235 	.e_vm_default_addr =	uvm_default_mapaddr,
236 	.e_usertrap =		NULL,
237 	.e_ucsize =		sizeof(ucontext_t),
238 	.e_startlwp =		startlwp
239 };
240 
241 /*
242  * Exec lock. Used to control access to execsw[] structures.
243  * This must not be static so that netbsd32 can access it, too.
244  */
245 krwlock_t exec_lock;
246 
247 static kmutex_t sigobject_lock;
248 
249 /*
250  * Data used between a loadvm and execve part of an "exec" operation
251  */
252 struct execve_data {
253 	struct exec_package	ed_pack;
254 	struct pathbuf		*ed_pathbuf;
255 	struct vattr		ed_attr;
256 	struct ps_strings	ed_arginfo;
257 	char			*ed_argp;
258 	const char		*ed_pathstring;
259 	char			*ed_resolvedpathbuf;
260 	size_t			ed_ps_strings_sz;
261 	int			ed_szsigcode;
262 	size_t			ed_argslen;
263 	long			ed_argc;
264 	long			ed_envc;
265 };
266 
267 /*
268  * data passed from parent lwp to child during a posix_spawn()
269  */
270 struct spawn_exec_data {
271 	struct execve_data	sed_exec;
272 	struct posix_spawn_file_actions
273 				*sed_actions;
274 	struct posix_spawnattr	*sed_attrs;
275 	struct proc		*sed_parent;
276 	kcondvar_t		sed_cv_child_ready;
277 	kmutex_t		sed_mtx_child;
278 	int			sed_error;
279 	volatile uint32_t	sed_refcnt;
280 };
281 
282 static struct vm_map *exec_map;
283 static struct pool exec_pool;
284 
285 static void *
286 exec_pool_alloc(struct pool *pp, int flags)
287 {
288 
289 	return (void *)uvm_km_alloc(exec_map, NCARGS, 0,
290 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
291 }
292 
293 static void
294 exec_pool_free(struct pool *pp, void *addr)
295 {
296 
297 	uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
298 }
299 
300 static struct pool_allocator exec_palloc = {
301 	.pa_alloc = exec_pool_alloc,
302 	.pa_free = exec_pool_free,
303 	.pa_pagesz = NCARGS
304 };
305 
306 /*
307  * check exec:
308  * given an "executable" described in the exec package's namei info,
309  * see what we can do with it.
310  *
311  * ON ENTRY:
312  *	exec package with appropriate namei info
313  *	lwp pointer of exec'ing lwp
314  *	NO SELF-LOCKED VNODES
315  *
316  * ON EXIT:
317  *	error:	nothing held, etc.  exec header still allocated.
318  *	ok:	filled exec package, executable's vnode (unlocked).
319  *
320  * EXEC SWITCH ENTRY:
321  * 	Locked vnode to check, exec package, proc.
322  *
323  * EXEC SWITCH EXIT:
324  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
325  *	error:	destructive:
326  *			everything deallocated execept exec header.
327  *		non-destructive:
328  *			error code, executable's vnode (unlocked),
329  *			exec header unmodified.
330  */
331 int
332 /*ARGSUSED*/
333 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
334 {
335 	int		error, i;
336 	struct vnode	*vp;
337 	struct nameidata nd;
338 	size_t		resid;
339 
340 #if 1
341 	// grab the absolute pathbuf here before namei() trashes it.
342 	pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX);
343 #endif
344 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
345 
346 	/* first get the vnode */
347 	if ((error = namei(&nd)) != 0)
348 		return error;
349 	epp->ep_vp = vp = nd.ni_vp;
350 #if 0
351 	/*
352 	 * XXX: can't use nd.ni_pnbuf, because although pb contains an
353 	 * absolute path, nd.ni_pnbuf does not if the path contains symlinks.
354 	 */
355 	/* normally this can't fail */
356 	error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
357 	KASSERT(error == 0);
358 #endif
359 
360 #ifdef DIAGNOSTIC
361 	/* paranoia (take this out once namei stuff stabilizes) */
362 	memset(nd.ni_pnbuf, '~', PATH_MAX);
363 #endif
364 
365 	/* check access and type */
366 	if (vp->v_type != VREG) {
367 		error = EACCES;
368 		goto bad1;
369 	}
370 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
371 		goto bad1;
372 
373 	/* get attributes */
374 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
375 		goto bad1;
376 
377 	/* Check mount point */
378 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
379 		error = EACCES;
380 		goto bad1;
381 	}
382 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
383 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
384 
385 	/* try to open it */
386 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
387 		goto bad1;
388 
389 	/* unlock vp, since we need it unlocked from here on out. */
390 	VOP_UNLOCK(vp);
391 
392 #if NVERIEXEC > 0
393 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
394 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
395 	    NULL);
396 	if (error)
397 		goto bad2;
398 #endif /* NVERIEXEC > 0 */
399 
400 #ifdef PAX_SEGVGUARD
401 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
402 	if (error)
403 		goto bad2;
404 #endif /* PAX_SEGVGUARD */
405 
406 	/* now we have the file, get the exec header */
407 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
408 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
409 	if (error)
410 		goto bad2;
411 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
412 
413 	/*
414 	 * Set up default address space limits.  Can be overridden
415 	 * by individual exec packages.
416 	 */
417 	epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
418 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
419 
420 	/*
421 	 * set up the vmcmds for creation of the process
422 	 * address space
423 	 */
424 	error = ENOEXEC;
425 	for (i = 0; i < nexecs; i++) {
426 		int newerror;
427 
428 		epp->ep_esch = execsw[i];
429 		newerror = (*execsw[i]->es_makecmds)(l, epp);
430 
431 		if (!newerror) {
432 			/* Seems ok: check that entry point is not too high */
433 			if (epp->ep_entry > epp->ep_vm_maxaddr) {
434 #ifdef DIAGNOSTIC
435 				printf("%s: rejecting %p due to "
436 				    "too high entry address (> %p)\n",
437 					 __func__, (void *)epp->ep_entry,
438 					 (void *)epp->ep_vm_maxaddr);
439 #endif
440 				error = ENOEXEC;
441 				break;
442 			}
443 			/* Seems ok: check that entry point is not too low */
444 			if (epp->ep_entry < epp->ep_vm_minaddr) {
445 #ifdef DIAGNOSTIC
446 				printf("%s: rejecting %p due to "
447 				    "too low entry address (< %p)\n",
448 				     __func__, (void *)epp->ep_entry,
449 				     (void *)epp->ep_vm_minaddr);
450 #endif
451 				error = ENOEXEC;
452 				break;
453 			}
454 
455 			/* check limits */
456 			if ((epp->ep_tsize > MAXTSIZ) ||
457 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
458 						    [RLIMIT_DATA].rlim_cur)) {
459 #ifdef DIAGNOSTIC
460 				printf("%s: rejecting due to "
461 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
462 				    __func__,
463 				    (unsigned long long)epp->ep_tsize,
464 				    (unsigned long long)MAXTSIZ,
465 				    (unsigned long long)epp->ep_dsize,
466 				    (unsigned long long)
467 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
468 #endif
469 				error = ENOMEM;
470 				break;
471 			}
472 			return 0;
473 		}
474 
475 		/*
476 		 * Reset all the fields that may have been modified by the
477 		 * loader.
478 		 */
479 		KASSERT(epp->ep_emul_arg == NULL);
480 		if (epp->ep_emul_root != NULL) {
481 			vrele(epp->ep_emul_root);
482 			epp->ep_emul_root = NULL;
483 		}
484 		if (epp->ep_interp != NULL) {
485 			vrele(epp->ep_interp);
486 			epp->ep_interp = NULL;
487 		}
488 		epp->ep_pax_flags = 0;
489 
490 		/* make sure the first "interesting" error code is saved. */
491 		if (error == ENOEXEC)
492 			error = newerror;
493 
494 		if (epp->ep_flags & EXEC_DESTR)
495 			/* Error from "#!" code, tidied up by recursive call */
496 			return error;
497 	}
498 
499 	/* not found, error */
500 
501 	/*
502 	 * free any vmspace-creation commands,
503 	 * and release their references
504 	 */
505 	kill_vmcmds(&epp->ep_vmcmds);
506 
507 bad2:
508 	/*
509 	 * close and release the vnode, restore the old one, free the
510 	 * pathname buf, and punt.
511 	 */
512 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
513 	VOP_CLOSE(vp, FREAD, l->l_cred);
514 	vput(vp);
515 	return error;
516 
517 bad1:
518 	/*
519 	 * free the namei pathname buffer, and put the vnode
520 	 * (which we don't yet have open).
521 	 */
522 	vput(vp);				/* was still locked */
523 	return error;
524 }
525 
526 #ifdef __MACHINE_STACK_GROWS_UP
527 #define STACK_PTHREADSPACE NBPG
528 #else
529 #define STACK_PTHREADSPACE 0
530 #endif
531 
532 static int
533 execve_fetch_element(char * const *array, size_t index, char **value)
534 {
535 	return copyin(array + index, value, sizeof(*value));
536 }
537 
538 /*
539  * exec system call
540  */
541 int
542 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
543 {
544 	/* {
545 		syscallarg(const char *)	path;
546 		syscallarg(char * const *)	argp;
547 		syscallarg(char * const *)	envp;
548 	} */
549 
550 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
551 	    SCARG(uap, envp), execve_fetch_element);
552 }
553 
554 int
555 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
556     register_t *retval)
557 {
558 	/* {
559 		syscallarg(int)			fd;
560 		syscallarg(char * const *)	argp;
561 		syscallarg(char * const *)	envp;
562 	} */
563 
564 	return ENOSYS;
565 }
566 
567 /*
568  * Load modules to try and execute an image that we do not understand.
569  * If no execsw entries are present, we load those likely to be needed
570  * in order to run native images only.  Otherwise, we autoload all
571  * possible modules that could let us run the binary.  XXX lame
572  */
573 static void
574 exec_autoload(void)
575 {
576 #ifdef MODULAR
577 	static const char * const native[] = {
578 		"exec_elf32",
579 		"exec_elf64",
580 		"exec_script",
581 		NULL
582 	};
583 	static const char * const compat[] = {
584 		"exec_elf32",
585 		"exec_elf64",
586 		"exec_script",
587 		"exec_aout",
588 		"exec_coff",
589 		"exec_ecoff",
590 		"compat_aoutm68k",
591 		"compat_netbsd32",
592 		"compat_sunos",
593 		"compat_sunos32",
594 		"compat_ultrix",
595 		NULL
596 	};
597 	char const * const *list;
598 	int i;
599 
600 	list = (nexecs == 0 ? native : compat);
601 	for (i = 0; list[i] != NULL; i++) {
602 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
603 			continue;
604 		}
605 		yield();
606 	}
607 #endif
608 }
609 
610 static int
611 makepathbuf(struct lwp *l, const char *upath, struct pathbuf **pbp,
612     size_t *offs)
613 {
614 	char *path, *bp;
615 	size_t len, tlen;
616 	int error;
617 	struct cwdinfo *cwdi;
618 
619 	path = PNBUF_GET();
620 	error = copyinstr(upath, path, MAXPATHLEN, &len);
621 	if (error) {
622 		PNBUF_PUT(path);
623 		DPRINTF(("%s: copyin path @%p %d\n", __func__, upath, error));
624 		return error;
625 	}
626 
627 	if (path[0] == '/') {
628 		*offs = 0;
629 		goto out;
630 	}
631 
632 	len++;
633 	if (len + 1 >= MAXPATHLEN)
634 		goto out;
635 	bp = path + MAXPATHLEN - len;
636 	memmove(bp, path, len);
637 	*(--bp) = '/';
638 
639 	cwdi = l->l_proc->p_cwdi;
640 	rw_enter(&cwdi->cwdi_lock, RW_READER);
641 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
642 	    GETCWD_CHECK_ACCESS, l);
643 	rw_exit(&cwdi->cwdi_lock);
644 
645 	if (error) {
646 		DPRINTF(("%s: getcwd_common path %s %d\n", __func__, path,
647 		    error));
648 		goto out;
649 	}
650 	tlen = path + MAXPATHLEN - bp;
651 
652 	memmove(path, bp, tlen);
653 	path[tlen] = '\0';
654 	*offs = tlen - len;
655 out:
656 	*pbp = pathbuf_assimilate(path);
657 	return 0;
658 }
659 
660 vaddr_t
661 exec_vm_minaddr(vaddr_t va_min)
662 {
663 	/*
664 	 * Increase va_min if we don't want NULL to be mappable by the
665 	 * process.
666 	 */
667 #define VM_MIN_GUARD	PAGE_SIZE
668 	if (user_va0_disable && (va_min < VM_MIN_GUARD))
669 		return VM_MIN_GUARD;
670 	return va_min;
671 }
672 
673 static int
674 execve_loadvm(struct lwp *l, const char *path, char * const *args,
675 	char * const *envs, execve_fetch_element_t fetch_element,
676 	struct execve_data * restrict data)
677 {
678 	struct exec_package	* const epp = &data->ed_pack;
679 	int			error;
680 	struct proc		*p;
681 	char			*dp;
682 	u_int			modgen;
683 	size_t			offs = 0;	// XXX: GCC
684 
685 	KASSERT(data != NULL);
686 
687 	p = l->l_proc;
688 	modgen = 0;
689 
690 	SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
691 
692 	/*
693 	 * Check if we have exceeded our number of processes limit.
694 	 * This is so that we handle the case where a root daemon
695 	 * forked, ran setuid to become the desired user and is trying
696 	 * to exec. The obvious place to do the reference counting check
697 	 * is setuid(), but we don't do the reference counting check there
698 	 * like other OS's do because then all the programs that use setuid()
699 	 * must be modified to check the return code of setuid() and exit().
700 	 * It is dangerous to make setuid() fail, because it fails open and
701 	 * the program will continue to run as root. If we make it succeed
702 	 * and return an error code, again we are not enforcing the limit.
703 	 * The best place to enforce the limit is here, when the process tries
704 	 * to execute a new image, because eventually the process will need
705 	 * to call exec in order to do something useful.
706 	 */
707  retry:
708 	if (p->p_flag & PK_SUGID) {
709 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
710 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
711 		     &p->p_rlimit[RLIMIT_NPROC],
712 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
713 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
714 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
715 		return EAGAIN;
716 	}
717 
718 	/*
719 	 * Drain existing references and forbid new ones.  The process
720 	 * should be left alone until we're done here.  This is necessary
721 	 * to avoid race conditions - e.g. in ptrace() - that might allow
722 	 * a local user to illicitly obtain elevated privileges.
723 	 */
724 	rw_enter(&p->p_reflock, RW_WRITER);
725 
726 	/*
727 	 * Init the namei data to point the file user's program name.
728 	 * This is done here rather than in check_exec(), so that it's
729 	 * possible to override this settings if any of makecmd/probe
730 	 * functions call check_exec() recursively - for example,
731 	 * see exec_script_makecmds().
732 	 */
733 	if ((error = makepathbuf(l, path, &data->ed_pathbuf, &offs)) != 0)
734 		goto clrflg;
735 	data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
736 	data->ed_resolvedpathbuf = PNBUF_GET();
737 
738 	/*
739 	 * initialize the fields of the exec package.
740 	 */
741 	epp->ep_kname = data->ed_pathstring + offs;
742 	epp->ep_resolvedname = data->ed_resolvedpathbuf;
743 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
744 	epp->ep_hdrlen = exec_maxhdrsz;
745 	epp->ep_hdrvalid = 0;
746 	epp->ep_emul_arg = NULL;
747 	epp->ep_emul_arg_free = NULL;
748 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
749 	epp->ep_vap = &data->ed_attr;
750 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
751 	MD_TOPDOWN_INIT(epp);
752 	epp->ep_emul_root = NULL;
753 	epp->ep_interp = NULL;
754 	epp->ep_esch = NULL;
755 	epp->ep_pax_flags = 0;
756 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
757 
758 	rw_enter(&exec_lock, RW_READER);
759 
760 	/* see if we can run it. */
761 	if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
762 		if (error != ENOENT && error != EACCES && error != ENOEXEC) {
763 			DPRINTF(("%s: check exec failed for %s, error %d\n",
764 			    __func__, epp->ep_kname, error));
765 		}
766 		goto freehdr;
767 	}
768 
769 	/* allocate an argument buffer */
770 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
771 	KASSERT(data->ed_argp != NULL);
772 	dp = data->ed_argp;
773 
774 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
775 		goto bad;
776 	}
777 
778 	/*
779 	 * Calculate the new stack size.
780 	 */
781 
782 #ifdef __MACHINE_STACK_GROWS_UP
783 /*
784  * copyargs() fills argc/argv/envp from the lower address even on
785  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
786  * so that _rtld() use it.
787  */
788 #define	RTLD_GAP	32
789 #else
790 #define	RTLD_GAP	0
791 #endif
792 
793 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
794 
795 	data->ed_argslen = calcargs(data, argenvstrlen);
796 
797 	const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
798 
799 	if (len > epp->ep_ssize) {
800 		/* in effect, compare to initial limit */
801 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
802 		error = ENOMEM;
803 		goto bad;
804 	}
805 	/* adjust "active stack depth" for process VSZ */
806 	epp->ep_ssize = len;
807 
808 	return 0;
809 
810  bad:
811 	/* free the vmspace-creation commands, and release their references */
812 	kill_vmcmds(&epp->ep_vmcmds);
813 	/* kill any opened file descriptor, if necessary */
814 	if (epp->ep_flags & EXEC_HASFD) {
815 		epp->ep_flags &= ~EXEC_HASFD;
816 		fd_close(epp->ep_fd);
817 	}
818 	/* close and put the exec'd file */
819 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
820 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
821 	vput(epp->ep_vp);
822 	pool_put(&exec_pool, data->ed_argp);
823 
824  freehdr:
825 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
826 	if (epp->ep_emul_root != NULL)
827 		vrele(epp->ep_emul_root);
828 	if (epp->ep_interp != NULL)
829 		vrele(epp->ep_interp);
830 
831 	rw_exit(&exec_lock);
832 
833 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
834 	pathbuf_destroy(data->ed_pathbuf);
835 	PNBUF_PUT(data->ed_resolvedpathbuf);
836 
837  clrflg:
838 	rw_exit(&p->p_reflock);
839 
840 	if (modgen != module_gen && error == ENOEXEC) {
841 		modgen = module_gen;
842 		exec_autoload();
843 		goto retry;
844 	}
845 
846 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
847 	return error;
848 }
849 
850 static int
851 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
852 {
853 	struct exec_package	* const epp = &data->ed_pack;
854 	struct proc		*p = l->l_proc;
855 	struct exec_vmcmd	*base_vcp;
856 	int			error = 0;
857 	size_t			i;
858 
859 	/* record proc's vnode, for use by procfs and others */
860 	if (p->p_textvp)
861 		vrele(p->p_textvp);
862 	vref(epp->ep_vp);
863 	p->p_textvp = epp->ep_vp;
864 
865 	/* create the new process's VM space by running the vmcmds */
866 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
867 
868 #ifdef TRACE_EXEC
869 	DUMPVMCMDS(epp, 0, 0);
870 #endif
871 
872 	base_vcp = NULL;
873 
874 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
875 		struct exec_vmcmd *vcp;
876 
877 		vcp = &epp->ep_vmcmds.evs_cmds[i];
878 		if (vcp->ev_flags & VMCMD_RELATIVE) {
879 			KASSERTMSG(base_vcp != NULL,
880 			    "%s: relative vmcmd with no base", __func__);
881 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
882 			    "%s: illegal base & relative vmcmd", __func__);
883 			vcp->ev_addr += base_vcp->ev_addr;
884 		}
885 		error = (*vcp->ev_proc)(l, vcp);
886 		if (error)
887 			DUMPVMCMDS(epp, i, error);
888 		if (vcp->ev_flags & VMCMD_BASE)
889 			base_vcp = vcp;
890 	}
891 
892 	/* free the vmspace-creation commands, and release their references */
893 	kill_vmcmds(&epp->ep_vmcmds);
894 
895 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
896 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
897 	vput(epp->ep_vp);
898 
899 	/* if an error happened, deallocate and punt */
900 	if (error != 0) {
901 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
902 	}
903 	return error;
904 }
905 
906 static void
907 execve_free_data(struct execve_data *data)
908 {
909 	struct exec_package	* const epp = &data->ed_pack;
910 
911 	/* free the vmspace-creation commands, and release their references */
912 	kill_vmcmds(&epp->ep_vmcmds);
913 	/* kill any opened file descriptor, if necessary */
914 	if (epp->ep_flags & EXEC_HASFD) {
915 		epp->ep_flags &= ~EXEC_HASFD;
916 		fd_close(epp->ep_fd);
917 	}
918 
919 	/* close and put the exec'd file */
920 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
921 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
922 	vput(epp->ep_vp);
923 	pool_put(&exec_pool, data->ed_argp);
924 
925 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
926 	if (epp->ep_emul_root != NULL)
927 		vrele(epp->ep_emul_root);
928 	if (epp->ep_interp != NULL)
929 		vrele(epp->ep_interp);
930 
931 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
932 	pathbuf_destroy(data->ed_pathbuf);
933 	PNBUF_PUT(data->ed_resolvedpathbuf);
934 }
935 
936 static void
937 pathexec(struct proc *p, const char *resolvedname)
938 {
939 	KASSERT(resolvedname[0] == '/');
940 
941 	/* set command name & other accounting info */
942 	strlcpy(p->p_comm, strrchr(resolvedname, '/') + 1, sizeof(p->p_comm));
943 
944 	kmem_strfree(p->p_path);
945 	p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
946 }
947 
948 /* XXX elsewhere */
949 static int
950 credexec(struct lwp *l, struct vattr *attr)
951 {
952 	struct proc *p = l->l_proc;
953 	int error;
954 
955 	/*
956 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
957 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
958 	 * out additional references on the process for the moment.
959 	 */
960 	if ((p->p_slflag & PSL_TRACED) == 0 &&
961 
962 	    (((attr->va_mode & S_ISUID) != 0 &&
963 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
964 
965 	     ((attr->va_mode & S_ISGID) != 0 &&
966 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
967 		/*
968 		 * Mark the process as SUGID before we do
969 		 * anything that might block.
970 		 */
971 		proc_crmod_enter();
972 		proc_crmod_leave(NULL, NULL, true);
973 
974 		/* Make sure file descriptors 0..2 are in use. */
975 		if ((error = fd_checkstd()) != 0) {
976 			DPRINTF(("%s: fdcheckstd failed %d\n",
977 			    __func__, error));
978 			return error;
979 		}
980 
981 		/*
982 		 * Copy the credential so other references don't see our
983 		 * changes.
984 		 */
985 		l->l_cred = kauth_cred_copy(l->l_cred);
986 #ifdef KTRACE
987 		/*
988 		 * If the persistent trace flag isn't set, turn off.
989 		 */
990 		if (p->p_tracep) {
991 			mutex_enter(&ktrace_lock);
992 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
993 				ktrderef(p);
994 			mutex_exit(&ktrace_lock);
995 		}
996 #endif
997 		if (attr->va_mode & S_ISUID)
998 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
999 		if (attr->va_mode & S_ISGID)
1000 			kauth_cred_setegid(l->l_cred, attr->va_gid);
1001 	} else {
1002 		if (kauth_cred_geteuid(l->l_cred) ==
1003 		    kauth_cred_getuid(l->l_cred) &&
1004 		    kauth_cred_getegid(l->l_cred) ==
1005 		    kauth_cred_getgid(l->l_cred))
1006 			p->p_flag &= ~PK_SUGID;
1007 	}
1008 
1009 	/*
1010 	 * Copy the credential so other references don't see our changes.
1011 	 * Test to see if this is necessary first, since in the common case
1012 	 * we won't need a private reference.
1013 	 */
1014 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1015 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1016 		l->l_cred = kauth_cred_copy(l->l_cred);
1017 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1018 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1019 	}
1020 
1021 	/* Update the master credentials. */
1022 	if (l->l_cred != p->p_cred) {
1023 		kauth_cred_t ocred;
1024 
1025 		kauth_cred_hold(l->l_cred);
1026 		mutex_enter(p->p_lock);
1027 		ocred = p->p_cred;
1028 		p->p_cred = l->l_cred;
1029 		mutex_exit(p->p_lock);
1030 		kauth_cred_free(ocred);
1031 	}
1032 
1033 	return 0;
1034 }
1035 
1036 static void
1037 emulexec(struct lwp *l, struct exec_package *epp)
1038 {
1039 	struct proc		*p = l->l_proc;
1040 
1041 	/* The emulation root will usually have been found when we looked
1042 	 * for the elf interpreter (or similar), if not look now. */
1043 	if (epp->ep_esch->es_emul->e_path != NULL &&
1044 	    epp->ep_emul_root == NULL)
1045 		emul_find_root(l, epp);
1046 
1047 	/* Any old emulation root got removed by fdcloseexec */
1048 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1049 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1050 	rw_exit(&p->p_cwdi->cwdi_lock);
1051 	epp->ep_emul_root = NULL;
1052 	if (epp->ep_interp != NULL)
1053 		vrele(epp->ep_interp);
1054 
1055 	/*
1056 	 * Call emulation specific exec hook. This can setup per-process
1057 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1058 	 *
1059 	 * If we are executing process of different emulation than the
1060 	 * original forked process, call e_proc_exit() of the old emulation
1061 	 * first, then e_proc_exec() of new emulation. If the emulation is
1062 	 * same, the exec hook code should deallocate any old emulation
1063 	 * resources held previously by this process.
1064 	 */
1065 	if (p->p_emul && p->p_emul->e_proc_exit
1066 	    && p->p_emul != epp->ep_esch->es_emul)
1067 		(*p->p_emul->e_proc_exit)(p);
1068 
1069 	/*
1070 	 * This is now LWP 1.
1071 	 */
1072 	/* XXX elsewhere */
1073 	mutex_enter(p->p_lock);
1074 	p->p_nlwpid = 1;
1075 	l->l_lid = 1;
1076 	mutex_exit(p->p_lock);
1077 
1078 	/*
1079 	 * Call exec hook. Emulation code may NOT store reference to anything
1080 	 * from &pack.
1081 	 */
1082 	if (epp->ep_esch->es_emul->e_proc_exec)
1083 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1084 
1085 	/* update p_emul, the old value is no longer needed */
1086 	p->p_emul = epp->ep_esch->es_emul;
1087 
1088 	/* ...and the same for p_execsw */
1089 	p->p_execsw = epp->ep_esch;
1090 
1091 #ifdef __HAVE_SYSCALL_INTERN
1092 	(*p->p_emul->e_syscall_intern)(p);
1093 #endif
1094 	ktremul();
1095 }
1096 
1097 static int
1098 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1099 	bool no_local_exec_lock, bool is_spawn)
1100 {
1101 	struct exec_package	* const epp = &data->ed_pack;
1102 	int error = 0;
1103 	struct proc		*p;
1104 
1105 	/*
1106 	 * In case of a posix_spawn operation, the child doing the exec
1107 	 * might not hold the reader lock on exec_lock, but the parent
1108 	 * will do this instead.
1109 	 */
1110 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1111 	KASSERT(!no_local_exec_lock || is_spawn);
1112 	KASSERT(data != NULL);
1113 
1114 	p = l->l_proc;
1115 
1116 	/* Get rid of other LWPs. */
1117 	if (p->p_nlwps > 1) {
1118 		mutex_enter(p->p_lock);
1119 		exit_lwps(l);
1120 		mutex_exit(p->p_lock);
1121 	}
1122 	KDASSERT(p->p_nlwps == 1);
1123 
1124 	/* Destroy any lwpctl info. */
1125 	if (p->p_lwpctl != NULL)
1126 		lwp_ctl_exit();
1127 
1128 	/* Remove POSIX timers */
1129 	timers_free(p, TIMERS_POSIX);
1130 
1131 	/* Set the PaX flags. */
1132 	pax_set_flags(epp, p);
1133 
1134 	/*
1135 	 * Do whatever is necessary to prepare the address space
1136 	 * for remapping.  Note that this might replace the current
1137 	 * vmspace with another!
1138 	 */
1139 	if (is_spawn)
1140 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1141 		    epp->ep_vm_maxaddr,
1142 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1143 	else
1144 		uvmspace_exec(l, epp->ep_vm_minaddr,
1145 		    epp->ep_vm_maxaddr,
1146 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1147 
1148 	struct vmspace		*vm;
1149 	vm = p->p_vmspace;
1150 	vm->vm_taddr = (void *)epp->ep_taddr;
1151 	vm->vm_tsize = btoc(epp->ep_tsize);
1152 	vm->vm_daddr = (void*)epp->ep_daddr;
1153 	vm->vm_dsize = btoc(epp->ep_dsize);
1154 	vm->vm_ssize = btoc(epp->ep_ssize);
1155 	vm->vm_issize = 0;
1156 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1157 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1158 
1159 	pax_aslr_init_vm(l, vm, epp);
1160 
1161 	/* Now map address space. */
1162 	error = execve_dovmcmds(l, data);
1163 	if (error != 0)
1164 		goto exec_abort;
1165 
1166 	pathexec(p, epp->ep_resolvedname);
1167 
1168 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1169 
1170 	error = copyoutargs(data, l, newstack);
1171 	if (error != 0)
1172 		goto exec_abort;
1173 
1174 	cwdexec(p);
1175 	fd_closeexec();		/* handle close on exec */
1176 
1177 	if (__predict_false(ktrace_on))
1178 		fd_ktrexecfd();
1179 
1180 	execsigs(p);		/* reset caught signals */
1181 
1182 	mutex_enter(p->p_lock);
1183 	l->l_ctxlink = NULL;	/* reset ucontext link */
1184 	p->p_acflag &= ~AFORK;
1185 	p->p_flag |= PK_EXEC;
1186 	mutex_exit(p->p_lock);
1187 
1188 	/*
1189 	 * Stop profiling.
1190 	 */
1191 	if ((p->p_stflag & PST_PROFIL) != 0) {
1192 		mutex_spin_enter(&p->p_stmutex);
1193 		stopprofclock(p);
1194 		mutex_spin_exit(&p->p_stmutex);
1195 	}
1196 
1197 	/*
1198 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1199 	 * exited and exec()/exit() are the only places it will be cleared.
1200 	 */
1201 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1202 		mutex_enter(proc_lock);
1203 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1204 		p->p_lflag &= ~PL_PPWAIT;
1205 		cv_broadcast(&p->p_pptr->p_waitcv);
1206 		mutex_exit(proc_lock);
1207 	}
1208 
1209 	error = credexec(l, &data->ed_attr);
1210 	if (error)
1211 		goto exec_abort;
1212 
1213 #if defined(__HAVE_RAS)
1214 	/*
1215 	 * Remove all RASs from the address space.
1216 	 */
1217 	ras_purgeall();
1218 #endif
1219 
1220 	doexechooks(p);
1221 
1222 	/*
1223 	 * Set initial SP at the top of the stack.
1224 	 *
1225 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1226 	 * the end of arg/env strings.  Userland guesses the address of argc
1227 	 * via ps_strings::ps_argvstr.
1228 	 */
1229 
1230 	/* Setup new registers and do misc. setup. */
1231 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1232 	if (epp->ep_esch->es_setregs)
1233 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1234 
1235 	/* Provide a consistent LWP private setting */
1236 	(void)lwp_setprivate(l, NULL);
1237 
1238 	/* Discard all PCU state; need to start fresh */
1239 	pcu_discard_all(l);
1240 
1241 	/* map the process's signal trampoline code */
1242 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1243 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1244 		goto exec_abort;
1245 	}
1246 
1247 	pool_put(&exec_pool, data->ed_argp);
1248 
1249 	/* notify others that we exec'd */
1250 	KNOTE(&p->p_klist, NOTE_EXEC);
1251 
1252 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1253 
1254 	SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1255 
1256 	emulexec(l, epp);
1257 
1258 	/* Allow new references from the debugger/procfs. */
1259 	rw_exit(&p->p_reflock);
1260 	if (!no_local_exec_lock)
1261 		rw_exit(&exec_lock);
1262 
1263 	mutex_enter(proc_lock);
1264 
1265 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1266 		ksiginfo_t ksi;
1267 
1268 		KSI_INIT_EMPTY(&ksi);
1269 		ksi.ksi_signo = SIGTRAP;
1270 		ksi.ksi_code = TRAP_EXEC;
1271 		ksi.ksi_lid = l->l_lid;
1272 		kpsignal(p, &ksi, NULL);
1273 	}
1274 
1275 	if (p->p_sflag & PS_STOPEXEC) {
1276 		ksiginfoq_t kq;
1277 
1278 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1279 		p->p_pptr->p_nstopchild++;
1280 		p->p_waited = 0;
1281 		mutex_enter(p->p_lock);
1282 		ksiginfo_queue_init(&kq);
1283 		sigclearall(p, &contsigmask, &kq);
1284 		lwp_lock(l);
1285 		l->l_stat = LSSTOP;
1286 		p->p_stat = SSTOP;
1287 		p->p_nrlwps--;
1288 		lwp_unlock(l);
1289 		mutex_exit(p->p_lock);
1290 		mutex_exit(proc_lock);
1291 		lwp_lock(l);
1292 		mi_switch(l);
1293 		ksiginfo_queue_drain(&kq);
1294 		KERNEL_LOCK(l->l_biglocks, l);
1295 	} else {
1296 		mutex_exit(proc_lock);
1297 	}
1298 
1299 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1300 	pathbuf_destroy(data->ed_pathbuf);
1301 	PNBUF_PUT(data->ed_resolvedpathbuf);
1302 #ifdef TRACE_EXEC
1303 	DPRINTF(("%s finished\n", __func__));
1304 #endif
1305 	return EJUSTRETURN;
1306 
1307  exec_abort:
1308 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1309 	rw_exit(&p->p_reflock);
1310 	if (!no_local_exec_lock)
1311 		rw_exit(&exec_lock);
1312 
1313 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1314 	pathbuf_destroy(data->ed_pathbuf);
1315 	PNBUF_PUT(data->ed_resolvedpathbuf);
1316 
1317 	/*
1318 	 * the old process doesn't exist anymore.  exit gracefully.
1319 	 * get rid of the (new) address space we have created, if any, get rid
1320 	 * of our namei data and vnode, and exit noting failure
1321 	 */
1322 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1323 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1324 
1325 	exec_free_emul_arg(epp);
1326 	pool_put(&exec_pool, data->ed_argp);
1327 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1328 	if (epp->ep_emul_root != NULL)
1329 		vrele(epp->ep_emul_root);
1330 	if (epp->ep_interp != NULL)
1331 		vrele(epp->ep_interp);
1332 
1333 	/* Acquire the sched-state mutex (exit1() will release it). */
1334 	if (!is_spawn) {
1335 		mutex_enter(p->p_lock);
1336 		exit1(l, error, SIGABRT);
1337 	}
1338 
1339 	return error;
1340 }
1341 
1342 int
1343 execve1(struct lwp *l, const char *path, char * const *args,
1344     char * const *envs, execve_fetch_element_t fetch_element)
1345 {
1346 	struct execve_data data;
1347 	int error;
1348 
1349 	error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1350 	if (error)
1351 		return error;
1352 	error = execve_runproc(l, &data, false, false);
1353 	return error;
1354 }
1355 
1356 static size_t
1357 fromptrsz(const struct exec_package *epp)
1358 {
1359 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1360 }
1361 
1362 static size_t
1363 ptrsz(const struct exec_package *epp)
1364 {
1365 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1366 }
1367 
1368 static size_t
1369 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1370 {
1371 	struct exec_package	* const epp = &data->ed_pack;
1372 
1373 	const size_t nargenvptrs =
1374 	    1 +				/* long argc */
1375 	    data->ed_argc +		/* char *argv[] */
1376 	    1 +				/* \0 */
1377 	    data->ed_envc +		/* char *env[] */
1378 	    1;				/* \0 */
1379 
1380 	return (nargenvptrs * ptrsz(epp))	/* pointers */
1381 	    + argenvstrlen			/* strings */
1382 	    + epp->ep_esch->es_arglen;		/* auxinfo */
1383 }
1384 
1385 static size_t
1386 calcstack(struct execve_data * restrict data, const size_t gaplen)
1387 {
1388 	struct exec_package	* const epp = &data->ed_pack;
1389 
1390 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1391 	    epp->ep_esch->es_emul->e_sigcode;
1392 
1393 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1394 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1395 
1396 	const size_t sigcode_psstr_sz =
1397 	    data->ed_szsigcode +	/* sigcode */
1398 	    data->ed_ps_strings_sz +	/* ps_strings */
1399 	    STACK_PTHREADSPACE;		/* pthread space */
1400 
1401 	const size_t stacklen =
1402 	    data->ed_argslen +
1403 	    gaplen +
1404 	    sigcode_psstr_sz;
1405 
1406 	/* make the stack "safely" aligned */
1407 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1408 }
1409 
1410 static int
1411 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1412     char * const newstack)
1413 {
1414 	struct exec_package	* const epp = &data->ed_pack;
1415 	struct proc		*p = l->l_proc;
1416 	int			error;
1417 
1418 	/* remember information about the process */
1419 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1420 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1421 
1422 	/*
1423 	 * Allocate the stack address passed to the newly execve()'ed process.
1424 	 *
1425 	 * The new stack address will be set to the SP (stack pointer) register
1426 	 * in setregs().
1427 	 */
1428 
1429 	char *newargs = STACK_ALLOC(
1430 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1431 
1432 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1433 	    &data->ed_arginfo, &newargs, data->ed_argp);
1434 
1435 	if (error) {
1436 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1437 		return error;
1438 	}
1439 
1440 	error = copyoutpsstrs(data, p);
1441 	if (error != 0)
1442 		return error;
1443 
1444 	return 0;
1445 }
1446 
1447 static int
1448 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1449 {
1450 	struct exec_package	* const epp = &data->ed_pack;
1451 	struct ps_strings32	arginfo32;
1452 	void			*aip;
1453 	int			error;
1454 
1455 	/* fill process ps_strings info */
1456 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1457 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1458 
1459 	if (epp->ep_flags & EXEC_32) {
1460 		aip = &arginfo32;
1461 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1462 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1463 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1464 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1465 	} else
1466 		aip = &data->ed_arginfo;
1467 
1468 	/* copy out the process's ps_strings structure */
1469 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1470 	    != 0) {
1471 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1472 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1473 		return error;
1474 	}
1475 
1476 	return 0;
1477 }
1478 
1479 static int
1480 copyinargs(struct execve_data * restrict data, char * const *args,
1481     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1482 {
1483 	struct exec_package	* const epp = &data->ed_pack;
1484 	char			*dp;
1485 	size_t			i;
1486 	int			error;
1487 
1488 	dp = *dpp;
1489 
1490 	data->ed_argc = 0;
1491 
1492 	/* copy the fake args list, if there's one, freeing it as we go */
1493 	if (epp->ep_flags & EXEC_HASARGL) {
1494 		struct exec_fakearg	*fa = epp->ep_fa;
1495 
1496 		while (fa->fa_arg != NULL) {
1497 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1498 			size_t len;
1499 
1500 			len = strlcpy(dp, fa->fa_arg, maxlen);
1501 			/* Count NUL into len. */
1502 			if (len < maxlen)
1503 				len++;
1504 			else {
1505 				while (fa->fa_arg != NULL) {
1506 					kmem_free(fa->fa_arg, fa->fa_len);
1507 					fa++;
1508 				}
1509 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1510 				epp->ep_flags &= ~EXEC_HASARGL;
1511 				return E2BIG;
1512 			}
1513 			ktrexecarg(fa->fa_arg, len - 1);
1514 			dp += len;
1515 
1516 			kmem_free(fa->fa_arg, fa->fa_len);
1517 			fa++;
1518 			data->ed_argc++;
1519 		}
1520 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1521 		epp->ep_flags &= ~EXEC_HASARGL;
1522 	}
1523 
1524 	/*
1525 	 * Read and count argument strings from user.
1526 	 */
1527 
1528 	if (args == NULL) {
1529 		DPRINTF(("%s: null args\n", __func__));
1530 		return EINVAL;
1531 	}
1532 	if (epp->ep_flags & EXEC_SKIPARG)
1533 		args = (const void *)((const char *)args + fromptrsz(epp));
1534 	i = 0;
1535 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1536 	if (error != 0) {
1537 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1538 		return error;
1539 	}
1540 	data->ed_argc += i;
1541 
1542 	/*
1543 	 * Read and count environment strings from user.
1544 	 */
1545 
1546 	data->ed_envc = 0;
1547 	/* environment need not be there */
1548 	if (envs == NULL)
1549 		goto done;
1550 	i = 0;
1551 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1552 	if (error != 0) {
1553 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1554 		return error;
1555 	}
1556 	data->ed_envc += i;
1557 
1558 done:
1559 	*dpp = dp;
1560 
1561 	return 0;
1562 }
1563 
1564 static int
1565 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1566     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1567     void (*ktr)(const void *, size_t))
1568 {
1569 	char			*dp, *sp;
1570 	size_t			i;
1571 	int			error;
1572 
1573 	dp = *dpp;
1574 
1575 	i = 0;
1576 	while (1) {
1577 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1578 		size_t len;
1579 
1580 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1581 			return error;
1582 		}
1583 		if (!sp)
1584 			break;
1585 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1586 			if (error == ENAMETOOLONG)
1587 				error = E2BIG;
1588 			return error;
1589 		}
1590 		if (__predict_false(ktrace_on))
1591 			(*ktr)(dp, len - 1);
1592 		dp += len;
1593 		i++;
1594 	}
1595 
1596 	*dpp = dp;
1597 	*ip = i;
1598 
1599 	return 0;
1600 }
1601 
1602 /*
1603  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1604  * Those strings are located just after auxinfo.
1605  */
1606 int
1607 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1608     char **stackp, void *argp)
1609 {
1610 	char	**cpp, *dp, *sp;
1611 	size_t	len;
1612 	void	*nullp;
1613 	long	argc, envc;
1614 	int	error;
1615 
1616 	cpp = (char **)*stackp;
1617 	nullp = NULL;
1618 	argc = arginfo->ps_nargvstr;
1619 	envc = arginfo->ps_nenvstr;
1620 
1621 	/* argc on stack is long */
1622 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1623 
1624 	dp = (char *)(cpp +
1625 	    1 +				/* long argc */
1626 	    argc +			/* char *argv[] */
1627 	    1 +				/* \0 */
1628 	    envc +			/* char *env[] */
1629 	    1) +			/* \0 */
1630 	    pack->ep_esch->es_arglen;	/* auxinfo */
1631 	sp = argp;
1632 
1633 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1634 		COPYPRINTF("", cpp - 1, sizeof(argc));
1635 		return error;
1636 	}
1637 
1638 	/* XXX don't copy them out, remap them! */
1639 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1640 
1641 	for (; --argc >= 0; sp += len, dp += len) {
1642 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1643 			COPYPRINTF("", cpp - 1, sizeof(dp));
1644 			return error;
1645 		}
1646 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1647 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1648 			return error;
1649 		}
1650 	}
1651 
1652 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1653 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1654 		return error;
1655 	}
1656 
1657 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1658 
1659 	for (; --envc >= 0; sp += len, dp += len) {
1660 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1661 			COPYPRINTF("", cpp - 1, sizeof(dp));
1662 			return error;
1663 		}
1664 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1665 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1666 			return error;
1667 		}
1668 
1669 	}
1670 
1671 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1672 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1673 		return error;
1674 	}
1675 
1676 	*stackp = (char *)cpp;
1677 	return 0;
1678 }
1679 
1680 
1681 /*
1682  * Add execsw[] entries.
1683  */
1684 int
1685 exec_add(struct execsw *esp, int count)
1686 {
1687 	struct exec_entry	*it;
1688 	int			i;
1689 
1690 	if (count == 0) {
1691 		return 0;
1692 	}
1693 
1694 	/* Check for duplicates. */
1695 	rw_enter(&exec_lock, RW_WRITER);
1696 	for (i = 0; i < count; i++) {
1697 		LIST_FOREACH(it, &ex_head, ex_list) {
1698 			/* assume unique (makecmds, probe_func, emulation) */
1699 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1700 			    it->ex_sw->u.elf_probe_func ==
1701 			    esp[i].u.elf_probe_func &&
1702 			    it->ex_sw->es_emul == esp[i].es_emul) {
1703 				rw_exit(&exec_lock);
1704 				return EEXIST;
1705 			}
1706 		}
1707 	}
1708 
1709 	/* Allocate new entries. */
1710 	for (i = 0; i < count; i++) {
1711 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1712 		it->ex_sw = &esp[i];
1713 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1714 	}
1715 
1716 	/* update execsw[] */
1717 	exec_init(0);
1718 	rw_exit(&exec_lock);
1719 	return 0;
1720 }
1721 
1722 /*
1723  * Remove execsw[] entry.
1724  */
1725 int
1726 exec_remove(struct execsw *esp, int count)
1727 {
1728 	struct exec_entry	*it, *next;
1729 	int			i;
1730 	const struct proclist_desc *pd;
1731 	proc_t			*p;
1732 
1733 	if (count == 0) {
1734 		return 0;
1735 	}
1736 
1737 	/* Abort if any are busy. */
1738 	rw_enter(&exec_lock, RW_WRITER);
1739 	for (i = 0; i < count; i++) {
1740 		mutex_enter(proc_lock);
1741 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1742 			PROCLIST_FOREACH(p, pd->pd_list) {
1743 				if (p->p_execsw == &esp[i]) {
1744 					mutex_exit(proc_lock);
1745 					rw_exit(&exec_lock);
1746 					return EBUSY;
1747 				}
1748 			}
1749 		}
1750 		mutex_exit(proc_lock);
1751 	}
1752 
1753 	/* None are busy, so remove them all. */
1754 	for (i = 0; i < count; i++) {
1755 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1756 			next = LIST_NEXT(it, ex_list);
1757 			if (it->ex_sw == &esp[i]) {
1758 				LIST_REMOVE(it, ex_list);
1759 				kmem_free(it, sizeof(*it));
1760 				break;
1761 			}
1762 		}
1763 	}
1764 
1765 	/* update execsw[] */
1766 	exec_init(0);
1767 	rw_exit(&exec_lock);
1768 	return 0;
1769 }
1770 
1771 /*
1772  * Initialize exec structures. If init_boot is true, also does necessary
1773  * one-time initialization (it's called from main() that way).
1774  * Once system is multiuser, this should be called with exec_lock held,
1775  * i.e. via exec_{add|remove}().
1776  */
1777 int
1778 exec_init(int init_boot)
1779 {
1780 	const struct execsw 	**sw;
1781 	struct exec_entry	*ex;
1782 	SLIST_HEAD(,exec_entry)	first;
1783 	SLIST_HEAD(,exec_entry)	any;
1784 	SLIST_HEAD(,exec_entry)	last;
1785 	int			i, sz;
1786 
1787 	if (init_boot) {
1788 		/* do one-time initializations */
1789 		vaddr_t vmin = 0, vmax;
1790 
1791 		rw_init(&exec_lock);
1792 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1793 		exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1794 		    maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1795 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1796 		    "execargs", &exec_palloc, IPL_NONE);
1797 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1798 	} else {
1799 		KASSERT(rw_write_held(&exec_lock));
1800 	}
1801 
1802 	/* Sort each entry onto the appropriate queue. */
1803 	SLIST_INIT(&first);
1804 	SLIST_INIT(&any);
1805 	SLIST_INIT(&last);
1806 	sz = 0;
1807 	LIST_FOREACH(ex, &ex_head, ex_list) {
1808 		switch(ex->ex_sw->es_prio) {
1809 		case EXECSW_PRIO_FIRST:
1810 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1811 			break;
1812 		case EXECSW_PRIO_ANY:
1813 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1814 			break;
1815 		case EXECSW_PRIO_LAST:
1816 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1817 			break;
1818 		default:
1819 			panic("%s", __func__);
1820 			break;
1821 		}
1822 		sz++;
1823 	}
1824 
1825 	/*
1826 	 * Create new execsw[].  Ensure we do not try a zero-sized
1827 	 * allocation.
1828 	 */
1829 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1830 	i = 0;
1831 	SLIST_FOREACH(ex, &first, ex_slist) {
1832 		sw[i++] = ex->ex_sw;
1833 	}
1834 	SLIST_FOREACH(ex, &any, ex_slist) {
1835 		sw[i++] = ex->ex_sw;
1836 	}
1837 	SLIST_FOREACH(ex, &last, ex_slist) {
1838 		sw[i++] = ex->ex_sw;
1839 	}
1840 
1841 	/* Replace old execsw[] and free used memory. */
1842 	if (execsw != NULL) {
1843 		kmem_free(__UNCONST(execsw),
1844 		    nexecs * sizeof(struct execsw *) + 1);
1845 	}
1846 	execsw = sw;
1847 	nexecs = sz;
1848 
1849 	/* Figure out the maximum size of an exec header. */
1850 	exec_maxhdrsz = sizeof(int);
1851 	for (i = 0; i < nexecs; i++) {
1852 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1853 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1854 	}
1855 
1856 	return 0;
1857 }
1858 
1859 static int
1860 exec_sigcode_map(struct proc *p, const struct emul *e)
1861 {
1862 	vaddr_t va;
1863 	vsize_t sz;
1864 	int error;
1865 	struct uvm_object *uobj;
1866 
1867 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1868 
1869 	if (e->e_sigobject == NULL || sz == 0) {
1870 		return 0;
1871 	}
1872 
1873 	/*
1874 	 * If we don't have a sigobject for this emulation, create one.
1875 	 *
1876 	 * sigobject is an anonymous memory object (just like SYSV shared
1877 	 * memory) that we keep a permanent reference to and that we map
1878 	 * in all processes that need this sigcode. The creation is simple,
1879 	 * we create an object, add a permanent reference to it, map it in
1880 	 * kernel space, copy out the sigcode to it and unmap it.
1881 	 * We map it with PROT_READ|PROT_EXEC into the process just
1882 	 * the way sys_mmap() would map it.
1883 	 */
1884 
1885 	uobj = *e->e_sigobject;
1886 	if (uobj == NULL) {
1887 		mutex_enter(&sigobject_lock);
1888 		if ((uobj = *e->e_sigobject) == NULL) {
1889 			uobj = uao_create(sz, 0);
1890 			(*uobj->pgops->pgo_reference)(uobj);
1891 			va = vm_map_min(kernel_map);
1892 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1893 			    uobj, 0, 0,
1894 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1895 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1896 				printf("kernel mapping failed %d\n", error);
1897 				(*uobj->pgops->pgo_detach)(uobj);
1898 				mutex_exit(&sigobject_lock);
1899 				return error;
1900 			}
1901 			memcpy((void *)va, e->e_sigcode, sz);
1902 #ifdef PMAP_NEED_PROCWR
1903 			pmap_procwr(&proc0, va, sz);
1904 #endif
1905 			uvm_unmap(kernel_map, va, va + round_page(sz));
1906 			*e->e_sigobject = uobj;
1907 		}
1908 		mutex_exit(&sigobject_lock);
1909 	}
1910 
1911 	/* Just a hint to uvm_map where to put it. */
1912 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1913 	    round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
1914 
1915 #ifdef __alpha__
1916 	/*
1917 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1918 	 * which causes the above calculation to put the sigcode at
1919 	 * an invalid address.  Put it just below the text instead.
1920 	 */
1921 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1922 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1923 	}
1924 #endif
1925 
1926 	(*uobj->pgops->pgo_reference)(uobj);
1927 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1928 			uobj, 0, 0,
1929 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1930 				    UVM_ADV_RANDOM, 0));
1931 	if (error) {
1932 		DPRINTF(("%s, %d: map %p "
1933 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1934 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1935 		    va, error));
1936 		(*uobj->pgops->pgo_detach)(uobj);
1937 		return error;
1938 	}
1939 	p->p_sigctx.ps_sigcode = (void *)va;
1940 	return 0;
1941 }
1942 
1943 /*
1944  * Release a refcount on spawn_exec_data and destroy memory, if this
1945  * was the last one.
1946  */
1947 static void
1948 spawn_exec_data_release(struct spawn_exec_data *data)
1949 {
1950 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1951 		return;
1952 
1953 	cv_destroy(&data->sed_cv_child_ready);
1954 	mutex_destroy(&data->sed_mtx_child);
1955 
1956 	if (data->sed_actions)
1957 		posix_spawn_fa_free(data->sed_actions,
1958 		    data->sed_actions->len);
1959 	if (data->sed_attrs)
1960 		kmem_free(data->sed_attrs,
1961 		    sizeof(*data->sed_attrs));
1962 	kmem_free(data, sizeof(*data));
1963 }
1964 
1965 /*
1966  * A child lwp of a posix_spawn operation starts here and ends up in
1967  * cpu_spawn_return, dealing with all filedescriptor and scheduler
1968  * manipulations in between.
1969  * The parent waits for the child, as it is not clear whether the child
1970  * will be able to acquire its own exec_lock. If it can, the parent can
1971  * be released early and continue running in parallel. If not (or if the
1972  * magic debug flag is passed in the scheduler attribute struct), the
1973  * child rides on the parent's exec lock until it is ready to return to
1974  * to userland - and only then releases the parent. This method loses
1975  * concurrency, but improves error reporting.
1976  */
1977 static void
1978 spawn_return(void *arg)
1979 {
1980 	struct spawn_exec_data *spawn_data = arg;
1981 	struct lwp *l = curlwp;
1982 	int error, newfd;
1983 	int ostat;
1984 	size_t i;
1985 	const struct posix_spawn_file_actions_entry *fae;
1986 	pid_t ppid;
1987 	register_t retval;
1988 	bool have_reflock;
1989 	bool parent_is_waiting = true;
1990 
1991 	/*
1992 	 * Check if we can release parent early.
1993 	 * We either need to have no sed_attrs, or sed_attrs does not
1994 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
1995 	 * safe access to the parent proc (passed in sed_parent).
1996 	 * We then try to get the exec_lock, and only if that works, we can
1997 	 * release the parent here already.
1998 	 */
1999 	ppid = spawn_data->sed_parent->p_pid;
2000 	if ((!spawn_data->sed_attrs
2001 	    || (spawn_data->sed_attrs->sa_flags
2002 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2003 	    && rw_tryenter(&exec_lock, RW_READER)) {
2004 		parent_is_waiting = false;
2005 		mutex_enter(&spawn_data->sed_mtx_child);
2006 		cv_signal(&spawn_data->sed_cv_child_ready);
2007 		mutex_exit(&spawn_data->sed_mtx_child);
2008 	}
2009 
2010 	/* don't allow debugger access yet */
2011 	rw_enter(&l->l_proc->p_reflock, RW_WRITER);
2012 	have_reflock = true;
2013 
2014 	error = 0;
2015 	/* handle posix_spawn_file_actions */
2016 	if (spawn_data->sed_actions != NULL) {
2017 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
2018 			fae = &spawn_data->sed_actions->fae[i];
2019 			switch (fae->fae_action) {
2020 			case FAE_OPEN:
2021 				if (fd_getfile(fae->fae_fildes) != NULL) {
2022 					error = fd_close(fae->fae_fildes);
2023 					if (error)
2024 						break;
2025 				}
2026 				error = fd_open(fae->fae_path, fae->fae_oflag,
2027 				    fae->fae_mode, &newfd);
2028 				if (error)
2029 					break;
2030 				if (newfd != fae->fae_fildes) {
2031 					error = dodup(l, newfd,
2032 					    fae->fae_fildes, 0, &retval);
2033 					if (fd_getfile(newfd) != NULL)
2034 						fd_close(newfd);
2035 				}
2036 				break;
2037 			case FAE_DUP2:
2038 				error = dodup(l, fae->fae_fildes,
2039 				    fae->fae_newfildes, 0, &retval);
2040 				break;
2041 			case FAE_CLOSE:
2042 				if (fd_getfile(fae->fae_fildes) == NULL) {
2043 					error = EBADF;
2044 					break;
2045 				}
2046 				error = fd_close(fae->fae_fildes);
2047 				break;
2048 			}
2049 			if (error)
2050 				goto report_error;
2051 		}
2052 	}
2053 
2054 	/* handle posix_spawnattr */
2055 	if (spawn_data->sed_attrs != NULL) {
2056 		struct sigaction sigact;
2057 		sigact._sa_u._sa_handler = SIG_DFL;
2058 		sigact.sa_flags = 0;
2059 
2060 		/*
2061 		 * set state to SSTOP so that this proc can be found by pid.
2062 		 * see proc_enterprp, do_sched_setparam below
2063 		 */
2064 		mutex_enter(proc_lock);
2065 		/*
2066 		 * p_stat should be SACTIVE, so we need to adjust the
2067 		 * parent's p_nstopchild here.  For safety, just make
2068 		 * we're on the good side of SDEAD before we adjust.
2069 		 */
2070 		ostat = l->l_proc->p_stat;
2071 		KASSERT(ostat < SSTOP);
2072 		l->l_proc->p_stat = SSTOP;
2073 		l->l_proc->p_waited = 0;
2074 		l->l_proc->p_pptr->p_nstopchild++;
2075 		mutex_exit(proc_lock);
2076 
2077 		/* Set process group */
2078 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2079 			pid_t mypid = l->l_proc->p_pid,
2080 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
2081 
2082 			if (pgrp == 0)
2083 				pgrp = mypid;
2084 
2085 			error = proc_enterpgrp(spawn_data->sed_parent,
2086 			    mypid, pgrp, false);
2087 			if (error)
2088 				goto report_error_stopped;
2089 		}
2090 
2091 		/* Set scheduler policy */
2092 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2093 			error = do_sched_setparam(l->l_proc->p_pid, 0,
2094 			    spawn_data->sed_attrs->sa_schedpolicy,
2095 			    &spawn_data->sed_attrs->sa_schedparam);
2096 		else if (spawn_data->sed_attrs->sa_flags
2097 		    & POSIX_SPAWN_SETSCHEDPARAM) {
2098 			error = do_sched_setparam(ppid, 0,
2099 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2100 		}
2101 		if (error)
2102 			goto report_error_stopped;
2103 
2104 		/* Reset user ID's */
2105 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2106 			error = do_setresuid(l, -1,
2107 			     kauth_cred_getgid(l->l_cred), -1,
2108 			     ID_E_EQ_R | ID_E_EQ_S);
2109 			if (error)
2110 				goto report_error_stopped;
2111 			error = do_setresuid(l, -1,
2112 			    kauth_cred_getuid(l->l_cred), -1,
2113 			    ID_E_EQ_R | ID_E_EQ_S);
2114 			if (error)
2115 				goto report_error_stopped;
2116 		}
2117 
2118 		/* Set signal masks/defaults */
2119 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2120 			mutex_enter(l->l_proc->p_lock);
2121 			error = sigprocmask1(l, SIG_SETMASK,
2122 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
2123 			mutex_exit(l->l_proc->p_lock);
2124 			if (error)
2125 				goto report_error_stopped;
2126 		}
2127 
2128 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2129 			/*
2130 			 * The following sigaction call is using a sigaction
2131 			 * version 0 trampoline which is in the compatibility
2132 			 * code only. This is not a problem because for SIG_DFL
2133 			 * and SIG_IGN, the trampolines are now ignored. If they
2134 			 * were not, this would be a problem because we are
2135 			 * holding the exec_lock, and the compat code needs
2136 			 * to do the same in order to replace the trampoline
2137 			 * code of the process.
2138 			 */
2139 			for (i = 1; i <= NSIG; i++) {
2140 				if (sigismember(
2141 				    &spawn_data->sed_attrs->sa_sigdefault, i))
2142 					sigaction1(l, i, &sigact, NULL, NULL,
2143 					    0);
2144 			}
2145 		}
2146 		mutex_enter(proc_lock);
2147 		l->l_proc->p_stat = ostat;
2148 		l->l_proc->p_pptr->p_nstopchild--;
2149 		mutex_exit(proc_lock);
2150 	}
2151 
2152 	/* now do the real exec */
2153 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2154 	    true);
2155 	have_reflock = false;
2156 	if (error == EJUSTRETURN)
2157 		error = 0;
2158 	else if (error)
2159 		goto report_error;
2160 
2161 	if (parent_is_waiting) {
2162 		mutex_enter(&spawn_data->sed_mtx_child);
2163 		cv_signal(&spawn_data->sed_cv_child_ready);
2164 		mutex_exit(&spawn_data->sed_mtx_child);
2165 	}
2166 
2167 	/* release our refcount on the data */
2168 	spawn_exec_data_release(spawn_data);
2169 
2170 	/* and finally: leave to userland for the first time */
2171 	cpu_spawn_return(l);
2172 
2173 	/* NOTREACHED */
2174 	return;
2175 
2176  report_error_stopped:
2177 	mutex_enter(proc_lock);
2178 	l->l_proc->p_stat = ostat;
2179 	l->l_proc->p_pptr->p_nstopchild--;
2180 	mutex_exit(proc_lock);
2181  report_error:
2182 	if (have_reflock) {
2183 		/*
2184 		 * We have not passed through execve_runproc(),
2185 		 * which would have released the p_reflock and also
2186 		 * taken ownership of the sed_exec part of spawn_data,
2187 		 * so release/free both here.
2188 		 */
2189 		rw_exit(&l->l_proc->p_reflock);
2190 		execve_free_data(&spawn_data->sed_exec);
2191 	}
2192 
2193 	if (parent_is_waiting) {
2194 		/* pass error to parent */
2195 		mutex_enter(&spawn_data->sed_mtx_child);
2196 		spawn_data->sed_error = error;
2197 		cv_signal(&spawn_data->sed_cv_child_ready);
2198 		mutex_exit(&spawn_data->sed_mtx_child);
2199 	} else {
2200 		rw_exit(&exec_lock);
2201 	}
2202 
2203 	/* release our refcount on the data */
2204 	spawn_exec_data_release(spawn_data);
2205 
2206 	/* done, exit */
2207 	mutex_enter(l->l_proc->p_lock);
2208 	/*
2209 	 * Posix explicitly asks for an exit code of 127 if we report
2210 	 * errors from the child process - so, unfortunately, there
2211 	 * is no way to report a more exact error code.
2212 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2213 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2214 	 */
2215 	exit1(l, 127, 0);
2216 }
2217 
2218 void
2219 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2220 {
2221 
2222 	for (size_t i = 0; i < len; i++) {
2223 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2224 		if (fae->fae_action != FAE_OPEN)
2225 			continue;
2226 		kmem_strfree(fae->fae_path);
2227 	}
2228 	if (fa->len > 0)
2229 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2230 	kmem_free(fa, sizeof(*fa));
2231 }
2232 
2233 static int
2234 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2235     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2236 {
2237 	struct posix_spawn_file_actions *fa;
2238 	struct posix_spawn_file_actions_entry *fae;
2239 	char *pbuf = NULL;
2240 	int error;
2241 	size_t i = 0;
2242 
2243 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2244 	error = copyin(ufa, fa, sizeof(*fa));
2245 	if (error || fa->len == 0) {
2246 		kmem_free(fa, sizeof(*fa));
2247 		return error;	/* 0 if not an error, and len == 0 */
2248 	}
2249 
2250 	if (fa->len > lim) {
2251 		kmem_free(fa, sizeof(*fa));
2252 		return EINVAL;
2253 	}
2254 
2255 	fa->size = fa->len;
2256 	size_t fal = fa->len * sizeof(*fae);
2257 	fae = fa->fae;
2258 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2259 	error = copyin(fae, fa->fae, fal);
2260 	if (error)
2261 		goto out;
2262 
2263 	pbuf = PNBUF_GET();
2264 	for (; i < fa->len; i++) {
2265 		fae = &fa->fae[i];
2266 		if (fae->fae_action != FAE_OPEN)
2267 			continue;
2268 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2269 		if (error)
2270 			goto out;
2271 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2272 		memcpy(fae->fae_path, pbuf, fal);
2273 	}
2274 	PNBUF_PUT(pbuf);
2275 
2276 	*fap = fa;
2277 	return 0;
2278 out:
2279 	if (pbuf)
2280 		PNBUF_PUT(pbuf);
2281 	posix_spawn_fa_free(fa, i);
2282 	return error;
2283 }
2284 
2285 int
2286 check_posix_spawn(struct lwp *l1)
2287 {
2288 	int error, tnprocs, count;
2289 	uid_t uid;
2290 	struct proc *p1;
2291 
2292 	p1 = l1->l_proc;
2293 	uid = kauth_cred_getuid(l1->l_cred);
2294 	tnprocs = atomic_inc_uint_nv(&nprocs);
2295 
2296 	/*
2297 	 * Although process entries are dynamically created, we still keep
2298 	 * a global limit on the maximum number we will create.
2299 	 */
2300 	if (__predict_false(tnprocs >= maxproc))
2301 		error = -1;
2302 	else
2303 		error = kauth_authorize_process(l1->l_cred,
2304 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2305 
2306 	if (error) {
2307 		atomic_dec_uint(&nprocs);
2308 		return EAGAIN;
2309 	}
2310 
2311 	/*
2312 	 * Enforce limits.
2313 	 */
2314 	count = chgproccnt(uid, 1);
2315 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2316 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2317 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2318 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2319 		(void)chgproccnt(uid, -1);
2320 		atomic_dec_uint(&nprocs);
2321 		return EAGAIN;
2322 	}
2323 
2324 	return 0;
2325 }
2326 
2327 int
2328 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2329 	struct posix_spawn_file_actions *fa,
2330 	struct posix_spawnattr *sa,
2331 	char *const *argv, char *const *envp,
2332 	execve_fetch_element_t fetch)
2333 {
2334 
2335 	struct proc *p1, *p2;
2336 	struct lwp *l2;
2337 	int error;
2338 	struct spawn_exec_data *spawn_data;
2339 	vaddr_t uaddr;
2340 	pid_t pid;
2341 	bool have_exec_lock = false;
2342 
2343 	p1 = l1->l_proc;
2344 
2345 	/* Allocate and init spawn_data */
2346 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2347 	spawn_data->sed_refcnt = 1; /* only parent so far */
2348 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2349 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2350 	mutex_enter(&spawn_data->sed_mtx_child);
2351 
2352 	/*
2353 	 * Do the first part of the exec now, collect state
2354 	 * in spawn_data.
2355 	 */
2356 	error = execve_loadvm(l1, path, argv,
2357 	    envp, fetch, &spawn_data->sed_exec);
2358 	if (error == EJUSTRETURN)
2359 		error = 0;
2360 	else if (error)
2361 		goto error_exit;
2362 
2363 	have_exec_lock = true;
2364 
2365 	/*
2366 	 * Allocate virtual address space for the U-area now, while it
2367 	 * is still easy to abort the fork operation if we're out of
2368 	 * kernel virtual address space.
2369 	 */
2370 	uaddr = uvm_uarea_alloc();
2371 	if (__predict_false(uaddr == 0)) {
2372 		error = ENOMEM;
2373 		goto error_exit;
2374 	}
2375 
2376 	/*
2377 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2378 	 * replace it with its own before returning to userland
2379 	 * in the child.
2380 	 * This is a point of no return, we will have to go through
2381 	 * the child proc to properly clean it up past this point.
2382 	 */
2383 	p2 = proc_alloc();
2384 	pid = p2->p_pid;
2385 
2386 	/*
2387 	 * Make a proc table entry for the new process.
2388 	 * Start by zeroing the section of proc that is zero-initialized,
2389 	 * then copy the section that is copied directly from the parent.
2390 	 */
2391 	memset(&p2->p_startzero, 0,
2392 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2393 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2394 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2395 	p2->p_vmspace = proc0.p_vmspace;
2396 
2397 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2398 
2399 	LIST_INIT(&p2->p_lwps);
2400 	LIST_INIT(&p2->p_sigwaiters);
2401 
2402 	/*
2403 	 * Duplicate sub-structures as needed.
2404 	 * Increase reference counts on shared objects.
2405 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2406 	 * handling are important in order to keep a consistent behaviour
2407 	 * for the child after the fork.  If we are a 32-bit process, the
2408 	 * child will be too.
2409 	 */
2410 	p2->p_flag =
2411 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2412 	p2->p_emul = p1->p_emul;
2413 	p2->p_execsw = p1->p_execsw;
2414 
2415 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2416 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2417 	rw_init(&p2->p_reflock);
2418 	cv_init(&p2->p_waitcv, "wait");
2419 	cv_init(&p2->p_lwpcv, "lwpwait");
2420 
2421 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2422 
2423 	kauth_proc_fork(p1, p2);
2424 
2425 	p2->p_raslist = NULL;
2426 	p2->p_fd = fd_copy();
2427 
2428 	/* XXX racy */
2429 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2430 
2431 	p2->p_cwdi = cwdinit();
2432 
2433 	/*
2434 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2435 	 * we just need increase pl_refcnt.
2436 	 */
2437 	if (!p1->p_limit->pl_writeable) {
2438 		lim_addref(p1->p_limit);
2439 		p2->p_limit = p1->p_limit;
2440 	} else {
2441 		p2->p_limit = lim_copy(p1->p_limit);
2442 	}
2443 
2444 	p2->p_lflag = 0;
2445 	p2->p_sflag = 0;
2446 	p2->p_slflag = 0;
2447 	p2->p_pptr = p1;
2448 	p2->p_ppid = p1->p_pid;
2449 	LIST_INIT(&p2->p_children);
2450 
2451 	p2->p_aio = NULL;
2452 
2453 #ifdef KTRACE
2454 	/*
2455 	 * Copy traceflag and tracefile if enabled.
2456 	 * If not inherited, these were zeroed above.
2457 	 */
2458 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2459 		mutex_enter(&ktrace_lock);
2460 		p2->p_traceflag = p1->p_traceflag;
2461 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2462 			ktradref(p2);
2463 		mutex_exit(&ktrace_lock);
2464 	}
2465 #endif
2466 
2467 	/*
2468 	 * Create signal actions for the child process.
2469 	 */
2470 	p2->p_sigacts = sigactsinit(p1, 0);
2471 	mutex_enter(p1->p_lock);
2472 	p2->p_sflag |=
2473 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2474 	sched_proc_fork(p1, p2);
2475 	mutex_exit(p1->p_lock);
2476 
2477 	p2->p_stflag = p1->p_stflag;
2478 
2479 	/*
2480 	 * p_stats.
2481 	 * Copy parts of p_stats, and zero out the rest.
2482 	 */
2483 	p2->p_stats = pstatscopy(p1->p_stats);
2484 
2485 	/* copy over machdep flags to the new proc */
2486 	cpu_proc_fork(p1, p2);
2487 
2488 	/*
2489 	 * Prepare remaining parts of spawn data
2490 	 */
2491 	spawn_data->sed_actions = fa;
2492 	spawn_data->sed_attrs = sa;
2493 
2494 	spawn_data->sed_parent = p1;
2495 
2496 	/* create LWP */
2497 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2498 	    &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2499 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2500 
2501 	/*
2502 	 * Copy the credential so other references don't see our changes.
2503 	 * Test to see if this is necessary first, since in the common case
2504 	 * we won't need a private reference.
2505 	 */
2506 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2507 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2508 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2509 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2510 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2511 	}
2512 
2513 	/* Update the master credentials. */
2514 	if (l2->l_cred != p2->p_cred) {
2515 		kauth_cred_t ocred;
2516 
2517 		kauth_cred_hold(l2->l_cred);
2518 		mutex_enter(p2->p_lock);
2519 		ocred = p2->p_cred;
2520 		p2->p_cred = l2->l_cred;
2521 		mutex_exit(p2->p_lock);
2522 		kauth_cred_free(ocred);
2523 	}
2524 
2525 	*child_ok = true;
2526 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2527 #if 0
2528 	l2->l_nopreempt = 1; /* start it non-preemptable */
2529 #endif
2530 
2531 	/*
2532 	 * It's now safe for the scheduler and other processes to see the
2533 	 * child process.
2534 	 */
2535 	mutex_enter(proc_lock);
2536 
2537 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2538 		p2->p_lflag |= PL_CONTROLT;
2539 
2540 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2541 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2542 
2543 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2544 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2545 
2546 	p2->p_trace_enabled = trace_is_enabled(p2);
2547 #ifdef __HAVE_SYSCALL_INTERN
2548 	(*p2->p_emul->e_syscall_intern)(p2);
2549 #endif
2550 
2551 	/*
2552 	 * Make child runnable, set start time, and add to run queue except
2553 	 * if the parent requested the child to start in SSTOP state.
2554 	 */
2555 	mutex_enter(p2->p_lock);
2556 
2557 	getmicrotime(&p2->p_stats->p_start);
2558 
2559 	lwp_lock(l2);
2560 	KASSERT(p2->p_nrlwps == 1);
2561 	p2->p_nrlwps = 1;
2562 	p2->p_stat = SACTIVE;
2563 	l2->l_stat = LSRUN;
2564 	sched_enqueue(l2, false);
2565 	lwp_unlock(l2);
2566 
2567 	mutex_exit(p2->p_lock);
2568 	mutex_exit(proc_lock);
2569 
2570 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2571 	error = spawn_data->sed_error;
2572 	mutex_exit(&spawn_data->sed_mtx_child);
2573 	spawn_exec_data_release(spawn_data);
2574 
2575 	rw_exit(&p1->p_reflock);
2576 	rw_exit(&exec_lock);
2577 	have_exec_lock = false;
2578 
2579 	*pid_res = pid;
2580 	return error;
2581 
2582  error_exit:
2583 	if (have_exec_lock) {
2584 		execve_free_data(&spawn_data->sed_exec);
2585 		rw_exit(&p1->p_reflock);
2586 		rw_exit(&exec_lock);
2587 	}
2588 	mutex_exit(&spawn_data->sed_mtx_child);
2589 	spawn_exec_data_release(spawn_data);
2590 
2591 	return error;
2592 }
2593 
2594 int
2595 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2596     register_t *retval)
2597 {
2598 	/* {
2599 		syscallarg(pid_t *) pid;
2600 		syscallarg(const char *) path;
2601 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2602 		syscallarg(const struct posix_spawnattr *) attrp;
2603 		syscallarg(char *const *) argv;
2604 		syscallarg(char *const *) envp;
2605 	} */
2606 
2607 	int error;
2608 	struct posix_spawn_file_actions *fa = NULL;
2609 	struct posix_spawnattr *sa = NULL;
2610 	pid_t pid;
2611 	bool child_ok = false;
2612 	rlim_t max_fileactions;
2613 	proc_t *p = l1->l_proc;
2614 
2615 	error = check_posix_spawn(l1);
2616 	if (error) {
2617 		*retval = error;
2618 		return 0;
2619 	}
2620 
2621 	/* copy in file_actions struct */
2622 	if (SCARG(uap, file_actions) != NULL) {
2623 		max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2624 		    maxfiles);
2625 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2626 		    max_fileactions);
2627 		if (error)
2628 			goto error_exit;
2629 	}
2630 
2631 	/* copyin posix_spawnattr struct */
2632 	if (SCARG(uap, attrp) != NULL) {
2633 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2634 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2635 		if (error)
2636 			goto error_exit;
2637 	}
2638 
2639 	/*
2640 	 * Do the spawn
2641 	 */
2642 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2643 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2644 	if (error)
2645 		goto error_exit;
2646 
2647 	if (error == 0 && SCARG(uap, pid) != NULL)
2648 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2649 
2650 	*retval = error;
2651 	return 0;
2652 
2653  error_exit:
2654 	if (!child_ok) {
2655 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2656 		atomic_dec_uint(&nprocs);
2657 
2658 		if (sa)
2659 			kmem_free(sa, sizeof(*sa));
2660 		if (fa)
2661 			posix_spawn_fa_free(fa, fa->len);
2662 	}
2663 
2664 	*retval = error;
2665 	return 0;
2666 }
2667 
2668 void
2669 exec_free_emul_arg(struct exec_package *epp)
2670 {
2671 	if (epp->ep_emul_arg_free != NULL) {
2672 		KASSERT(epp->ep_emul_arg != NULL);
2673 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2674 		epp->ep_emul_arg_free = NULL;
2675 		epp->ep_emul_arg = NULL;
2676 	} else {
2677 		KASSERT(epp->ep_emul_arg == NULL);
2678 	}
2679 }
2680 
2681 #ifdef DEBUG_EXEC
2682 static void
2683 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2684 {
2685 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2686 	size_t j;
2687 
2688 	if (error == 0)
2689 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2690 	else
2691 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2692 		    epp->ep_vmcmds.evs_used, error));
2693 
2694 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2695 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2696 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2697 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2698 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2699 		    "pagedvn" :
2700 		    vp[j].ev_proc == vmcmd_map_readvn ?
2701 		    "readvn" :
2702 		    vp[j].ev_proc == vmcmd_map_zero ?
2703 		    "zero" : "*unknown*",
2704 		    vp[j].ev_addr, vp[j].ev_len,
2705 		    vp[j].ev_offset, vp[j].ev_prot,
2706 		    vp[j].ev_flags));
2707 		if (error != 0 && j == x)
2708 			DPRINTF(("     ^--- failed\n"));
2709 	}
2710 }
2711 #endif
2712