xref: /netbsd-src/sys/kern/kern_exec.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: kern_exec.c,v 1.459 2018/05/28 11:32:20 kamil 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 1.459 2018/05/28 11:32:20 kamil 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_sigcode =		NULL,
220 	.e_esigcode =		NULL,
221 	.e_sigobject =		NULL,
222 	.e_setregs =		setregs,
223 	.e_proc_exec =		NULL,
224 	.e_proc_fork =		NULL,
225 	.e_proc_exit =		NULL,
226 	.e_lwp_fork =		NULL,
227 	.e_lwp_exit =		NULL,
228 #ifdef __HAVE_SYSCALL_INTERN
229 	.e_syscall_intern =	syscall_intern,
230 #else
231 	.e_syscall =		syscall,
232 #endif
233 	.e_sysctlovly =		NULL,
234 	.e_vm_default_addr =	uvm_default_mapaddr,
235 	.e_usertrap =		NULL,
236 	.e_ucsize =		sizeof(ucontext_t),
237 	.e_startlwp =		startlwp
238 };
239 
240 /*
241  * Exec lock. Used to control access to execsw[] structures.
242  * This must not be static so that netbsd32 can access it, too.
243  */
244 krwlock_t exec_lock;
245 
246 static kmutex_t sigobject_lock;
247 
248 /*
249  * Data used between a loadvm and execve part of an "exec" operation
250  */
251 struct execve_data {
252 	struct exec_package	ed_pack;
253 	struct pathbuf		*ed_pathbuf;
254 	struct vattr		ed_attr;
255 	struct ps_strings	ed_arginfo;
256 	char			*ed_argp;
257 	const char		*ed_pathstring;
258 	char			*ed_resolvedpathbuf;
259 	size_t			ed_ps_strings_sz;
260 	int			ed_szsigcode;
261 	size_t			ed_argslen;
262 	long			ed_argc;
263 	long			ed_envc;
264 };
265 
266 /*
267  * data passed from parent lwp to child during a posix_spawn()
268  */
269 struct spawn_exec_data {
270 	struct execve_data	sed_exec;
271 	struct posix_spawn_file_actions
272 				*sed_actions;
273 	struct posix_spawnattr	*sed_attrs;
274 	struct proc		*sed_parent;
275 	kcondvar_t		sed_cv_child_ready;
276 	kmutex_t		sed_mtx_child;
277 	int			sed_error;
278 	volatile uint32_t	sed_refcnt;
279 };
280 
281 static struct vm_map *exec_map;
282 static struct pool exec_pool;
283 
284 static void *
285 exec_pool_alloc(struct pool *pp, int flags)
286 {
287 
288 	return (void *)uvm_km_alloc(exec_map, NCARGS, 0,
289 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
290 }
291 
292 static void
293 exec_pool_free(struct pool *pp, void *addr)
294 {
295 
296 	uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
297 }
298 
299 static struct pool_allocator exec_palloc = {
300 	.pa_alloc = exec_pool_alloc,
301 	.pa_free = exec_pool_free,
302 	.pa_pagesz = NCARGS
303 };
304 
305 /*
306  * check exec:
307  * given an "executable" described in the exec package's namei info,
308  * see what we can do with it.
309  *
310  * ON ENTRY:
311  *	exec package with appropriate namei info
312  *	lwp pointer of exec'ing lwp
313  *	NO SELF-LOCKED VNODES
314  *
315  * ON EXIT:
316  *	error:	nothing held, etc.  exec header still allocated.
317  *	ok:	filled exec package, executable's vnode (unlocked).
318  *
319  * EXEC SWITCH ENTRY:
320  * 	Locked vnode to check, exec package, proc.
321  *
322  * EXEC SWITCH EXIT:
323  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
324  *	error:	destructive:
325  *			everything deallocated execept exec header.
326  *		non-destructive:
327  *			error code, executable's vnode (unlocked),
328  *			exec header unmodified.
329  */
330 int
331 /*ARGSUSED*/
332 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
333 {
334 	int		error, i;
335 	struct vnode	*vp;
336 	struct nameidata nd;
337 	size_t		resid;
338 
339 #if 1
340 	// grab the absolute pathbuf here before namei() trashes it.
341 	pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX);
342 #endif
343 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
344 
345 	/* first get the vnode */
346 	if ((error = namei(&nd)) != 0)
347 		return error;
348 	epp->ep_vp = vp = nd.ni_vp;
349 #if 0
350 	/*
351 	 * XXX: can't use nd.ni_pnbuf, because although pb contains an
352 	 * absolute path, nd.ni_pnbuf does not if the path contains symlinks.
353 	 */
354 	/* normally this can't fail */
355 	error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
356 	KASSERT(error == 0);
357 #endif
358 
359 #ifdef DIAGNOSTIC
360 	/* paranoia (take this out once namei stuff stabilizes) */
361 	memset(nd.ni_pnbuf, '~', PATH_MAX);
362 #endif
363 
364 	/* check access and type */
365 	if (vp->v_type != VREG) {
366 		error = EACCES;
367 		goto bad1;
368 	}
369 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
370 		goto bad1;
371 
372 	/* get attributes */
373 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
374 		goto bad1;
375 
376 	/* Check mount point */
377 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
378 		error = EACCES;
379 		goto bad1;
380 	}
381 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
382 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
383 
384 	/* try to open it */
385 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
386 		goto bad1;
387 
388 	/* unlock vp, since we need it unlocked from here on out. */
389 	VOP_UNLOCK(vp);
390 
391 #if NVERIEXEC > 0
392 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
393 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
394 	    NULL);
395 	if (error)
396 		goto bad2;
397 #endif /* NVERIEXEC > 0 */
398 
399 #ifdef PAX_SEGVGUARD
400 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
401 	if (error)
402 		goto bad2;
403 #endif /* PAX_SEGVGUARD */
404 
405 	/* now we have the file, get the exec header */
406 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
407 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
408 	if (error)
409 		goto bad2;
410 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
411 
412 	/*
413 	 * Set up default address space limits.  Can be overridden
414 	 * by individual exec packages.
415 	 */
416 	epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
417 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
418 
419 	/*
420 	 * set up the vmcmds for creation of the process
421 	 * address space
422 	 */
423 	error = ENOEXEC;
424 	for (i = 0; i < nexecs; i++) {
425 		int newerror;
426 
427 		epp->ep_esch = execsw[i];
428 		newerror = (*execsw[i]->es_makecmds)(l, epp);
429 
430 		if (!newerror) {
431 			/* Seems ok: check that entry point is not too high */
432 			if (epp->ep_entry >= epp->ep_vm_maxaddr) {
433 #ifdef DIAGNOSTIC
434 				printf("%s: rejecting %p due to "
435 				    "too high entry address (>= %p)\n",
436 					 __func__, (void *)epp->ep_entry,
437 					 (void *)epp->ep_vm_maxaddr);
438 #endif
439 				error = ENOEXEC;
440 				break;
441 			}
442 			/* Seems ok: check that entry point is not too low */
443 			if (epp->ep_entry < epp->ep_vm_minaddr) {
444 #ifdef DIAGNOSTIC
445 				printf("%s: rejecting %p due to "
446 				    "too low entry address (< %p)\n",
447 				     __func__, (void *)epp->ep_entry,
448 				     (void *)epp->ep_vm_minaddr);
449 #endif
450 				error = ENOEXEC;
451 				break;
452 			}
453 
454 			/* check limits */
455 			if ((epp->ep_tsize > MAXTSIZ) ||
456 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
457 						    [RLIMIT_DATA].rlim_cur)) {
458 #ifdef DIAGNOSTIC
459 				printf("%s: rejecting due to "
460 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
461 				    __func__,
462 				    (unsigned long long)epp->ep_tsize,
463 				    (unsigned long long)MAXTSIZ,
464 				    (unsigned long long)epp->ep_dsize,
465 				    (unsigned long long)
466 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
467 #endif
468 				error = ENOMEM;
469 				break;
470 			}
471 			return 0;
472 		}
473 
474 		/*
475 		 * Reset all the fields that may have been modified by the
476 		 * loader.
477 		 */
478 		KASSERT(epp->ep_emul_arg == NULL);
479 		if (epp->ep_emul_root != NULL) {
480 			vrele(epp->ep_emul_root);
481 			epp->ep_emul_root = NULL;
482 		}
483 		if (epp->ep_interp != NULL) {
484 			vrele(epp->ep_interp);
485 			epp->ep_interp = NULL;
486 		}
487 		epp->ep_pax_flags = 0;
488 
489 		/* make sure the first "interesting" error code is saved. */
490 		if (error == ENOEXEC)
491 			error = newerror;
492 
493 		if (epp->ep_flags & EXEC_DESTR)
494 			/* Error from "#!" code, tidied up by recursive call */
495 			return error;
496 	}
497 
498 	/* not found, error */
499 
500 	/*
501 	 * free any vmspace-creation commands,
502 	 * and release their references
503 	 */
504 	kill_vmcmds(&epp->ep_vmcmds);
505 
506 bad2:
507 	/*
508 	 * close and release the vnode, restore the old one, free the
509 	 * pathname buf, and punt.
510 	 */
511 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
512 	VOP_CLOSE(vp, FREAD, l->l_cred);
513 	vput(vp);
514 	return error;
515 
516 bad1:
517 	/*
518 	 * free the namei pathname buffer, and put the vnode
519 	 * (which we don't yet have open).
520 	 */
521 	vput(vp);				/* was still locked */
522 	return error;
523 }
524 
525 #ifdef __MACHINE_STACK_GROWS_UP
526 #define STACK_PTHREADSPACE NBPG
527 #else
528 #define STACK_PTHREADSPACE 0
529 #endif
530 
531 static int
532 execve_fetch_element(char * const *array, size_t index, char **value)
533 {
534 	return copyin(array + index, value, sizeof(*value));
535 }
536 
537 /*
538  * exec system call
539  */
540 int
541 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
542 {
543 	/* {
544 		syscallarg(const char *)	path;
545 		syscallarg(char * const *)	argp;
546 		syscallarg(char * const *)	envp;
547 	} */
548 
549 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
550 	    SCARG(uap, envp), execve_fetch_element);
551 }
552 
553 int
554 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
555     register_t *retval)
556 {
557 	/* {
558 		syscallarg(int)			fd;
559 		syscallarg(char * const *)	argp;
560 		syscallarg(char * const *)	envp;
561 	} */
562 
563 	return ENOSYS;
564 }
565 
566 /*
567  * Load modules to try and execute an image that we do not understand.
568  * If no execsw entries are present, we load those likely to be needed
569  * in order to run native images only.  Otherwise, we autoload all
570  * possible modules that could let us run the binary.  XXX lame
571  */
572 static void
573 exec_autoload(void)
574 {
575 #ifdef MODULAR
576 	static const char * const native[] = {
577 		"exec_elf32",
578 		"exec_elf64",
579 		"exec_script",
580 		NULL
581 	};
582 	static const char * const compat[] = {
583 		"exec_elf32",
584 		"exec_elf64",
585 		"exec_script",
586 		"exec_aout",
587 		"exec_coff",
588 		"exec_ecoff",
589 		"compat_aoutm68k",
590 		"compat_netbsd32",
591 		"compat_sunos",
592 		"compat_sunos32",
593 		"compat_ultrix",
594 		NULL
595 	};
596 	char const * const *list;
597 	int i;
598 
599 	list = (nexecs == 0 ? native : compat);
600 	for (i = 0; list[i] != NULL; i++) {
601 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
602 			continue;
603 		}
604 		yield();
605 	}
606 #endif
607 }
608 
609 int
610 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg,
611     struct pathbuf **pbp, size_t *offs)
612 {
613 	char *path, *bp;
614 	size_t len, tlen;
615 	int error;
616 	struct cwdinfo *cwdi;
617 
618 	path = PNBUF_GET();
619 	if (seg == UIO_SYSSPACE) {
620 		error = copystr(upath, path, MAXPATHLEN, &len);
621 	} else {
622 		error = copyinstr(upath, path, MAXPATHLEN, &len);
623 	}
624 	if (error) {
625 		PNBUF_PUT(path);
626 		DPRINTF(("%s: copyin path @%p %d\n", __func__, upath, error));
627 		return error;
628 	}
629 
630 	if (path[0] == '/') {
631 		if (offs)
632 			*offs = 0;
633 		goto out;
634 	}
635 
636 	len++;
637 	if (len + 1 >= MAXPATHLEN)
638 		goto out;
639 	bp = path + MAXPATHLEN - len;
640 	memmove(bp, path, len);
641 	*(--bp) = '/';
642 
643 	cwdi = l->l_proc->p_cwdi;
644 	rw_enter(&cwdi->cwdi_lock, RW_READER);
645 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
646 	    GETCWD_CHECK_ACCESS, l);
647 	rw_exit(&cwdi->cwdi_lock);
648 
649 	if (error) {
650 		DPRINTF(("%s: getcwd_common path %s %d\n", __func__, path,
651 		    error));
652 		goto out;
653 	}
654 	tlen = path + MAXPATHLEN - bp;
655 
656 	memmove(path, bp, tlen);
657 	path[tlen] = '\0';
658 	if (offs)
659 		*offs = tlen - len;
660 out:
661 	*pbp = pathbuf_assimilate(path);
662 	return 0;
663 }
664 
665 vaddr_t
666 exec_vm_minaddr(vaddr_t va_min)
667 {
668 	/*
669 	 * Increase va_min if we don't want NULL to be mappable by the
670 	 * process.
671 	 */
672 #define VM_MIN_GUARD	PAGE_SIZE
673 	if (user_va0_disable && (va_min < VM_MIN_GUARD))
674 		return VM_MIN_GUARD;
675 	return va_min;
676 }
677 
678 static int
679 execve_loadvm(struct lwp *l, const char *path, char * const *args,
680 	char * const *envs, execve_fetch_element_t fetch_element,
681 	struct execve_data * restrict data)
682 {
683 	struct exec_package	* const epp = &data->ed_pack;
684 	int			error;
685 	struct proc		*p;
686 	char			*dp;
687 	u_int			modgen;
688 	size_t			offs = 0;	// XXX: GCC
689 
690 	KASSERT(data != NULL);
691 
692 	p = l->l_proc;
693 	modgen = 0;
694 
695 	SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
696 
697 	/*
698 	 * Check if we have exceeded our number of processes limit.
699 	 * This is so that we handle the case where a root daemon
700 	 * forked, ran setuid to become the desired user and is trying
701 	 * to exec. The obvious place to do the reference counting check
702 	 * is setuid(), but we don't do the reference counting check there
703 	 * like other OS's do because then all the programs that use setuid()
704 	 * must be modified to check the return code of setuid() and exit().
705 	 * It is dangerous to make setuid() fail, because it fails open and
706 	 * the program will continue to run as root. If we make it succeed
707 	 * and return an error code, again we are not enforcing the limit.
708 	 * The best place to enforce the limit is here, when the process tries
709 	 * to execute a new image, because eventually the process will need
710 	 * to call exec in order to do something useful.
711 	 */
712  retry:
713 	if (p->p_flag & PK_SUGID) {
714 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
715 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
716 		     &p->p_rlimit[RLIMIT_NPROC],
717 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
718 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
719 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
720 		return EAGAIN;
721 	}
722 
723 	/*
724 	 * Drain existing references and forbid new ones.  The process
725 	 * should be left alone until we're done here.  This is necessary
726 	 * to avoid race conditions - e.g. in ptrace() - that might allow
727 	 * a local user to illicitly obtain elevated privileges.
728 	 */
729 	rw_enter(&p->p_reflock, RW_WRITER);
730 
731 	/*
732 	 * Init the namei data to point the file user's program name.
733 	 * This is done here rather than in check_exec(), so that it's
734 	 * possible to override this settings if any of makecmd/probe
735 	 * functions call check_exec() recursively - for example,
736 	 * see exec_script_makecmds().
737 	 */
738 	if ((error = exec_makepathbuf(l, path, UIO_USERSPACE,
739 	    &data->ed_pathbuf, &offs)) != 0)
740 		goto clrflg;
741 	data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
742 	data->ed_resolvedpathbuf = PNBUF_GET();
743 
744 	/*
745 	 * initialize the fields of the exec package.
746 	 */
747 	epp->ep_kname = data->ed_pathstring + offs;
748 	epp->ep_resolvedname = data->ed_resolvedpathbuf;
749 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
750 	epp->ep_hdrlen = exec_maxhdrsz;
751 	epp->ep_hdrvalid = 0;
752 	epp->ep_emul_arg = NULL;
753 	epp->ep_emul_arg_free = NULL;
754 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
755 	epp->ep_vap = &data->ed_attr;
756 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
757 	MD_TOPDOWN_INIT(epp);
758 	epp->ep_emul_root = NULL;
759 	epp->ep_interp = NULL;
760 	epp->ep_esch = NULL;
761 	epp->ep_pax_flags = 0;
762 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
763 
764 	rw_enter(&exec_lock, RW_READER);
765 
766 	/* see if we can run it. */
767 	if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
768 		if (error != ENOENT && error != EACCES && error != ENOEXEC) {
769 			DPRINTF(("%s: check exec failed for %s, error %d\n",
770 			    __func__, epp->ep_kname, error));
771 		}
772 		goto freehdr;
773 	}
774 
775 	/* allocate an argument buffer */
776 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
777 	KASSERT(data->ed_argp != NULL);
778 	dp = data->ed_argp;
779 
780 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
781 		goto bad;
782 	}
783 
784 	/*
785 	 * Calculate the new stack size.
786 	 */
787 
788 #ifdef __MACHINE_STACK_GROWS_UP
789 /*
790  * copyargs() fills argc/argv/envp from the lower address even on
791  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
792  * so that _rtld() use it.
793  */
794 #define	RTLD_GAP	32
795 #else
796 #define	RTLD_GAP	0
797 #endif
798 
799 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
800 
801 	data->ed_argslen = calcargs(data, argenvstrlen);
802 
803 	const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
804 
805 	if (len > epp->ep_ssize) {
806 		/* in effect, compare to initial limit */
807 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
808 		error = ENOMEM;
809 		goto bad;
810 	}
811 	/* adjust "active stack depth" for process VSZ */
812 	epp->ep_ssize = len;
813 
814 	return 0;
815 
816  bad:
817 	/* free the vmspace-creation commands, and release their references */
818 	kill_vmcmds(&epp->ep_vmcmds);
819 	/* kill any opened file descriptor, if necessary */
820 	if (epp->ep_flags & EXEC_HASFD) {
821 		epp->ep_flags &= ~EXEC_HASFD;
822 		fd_close(epp->ep_fd);
823 	}
824 	/* close and put the exec'd file */
825 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
826 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
827 	vput(epp->ep_vp);
828 	pool_put(&exec_pool, data->ed_argp);
829 
830  freehdr:
831 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
832 	if (epp->ep_emul_root != NULL)
833 		vrele(epp->ep_emul_root);
834 	if (epp->ep_interp != NULL)
835 		vrele(epp->ep_interp);
836 
837 	rw_exit(&exec_lock);
838 
839 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
840 	pathbuf_destroy(data->ed_pathbuf);
841 	PNBUF_PUT(data->ed_resolvedpathbuf);
842 
843  clrflg:
844 	rw_exit(&p->p_reflock);
845 
846 	if (modgen != module_gen && error == ENOEXEC) {
847 		modgen = module_gen;
848 		exec_autoload();
849 		goto retry;
850 	}
851 
852 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
853 	return error;
854 }
855 
856 static int
857 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
858 {
859 	struct exec_package	* const epp = &data->ed_pack;
860 	struct proc		*p = l->l_proc;
861 	struct exec_vmcmd	*base_vcp;
862 	int			error = 0;
863 	size_t			i;
864 
865 	/* record proc's vnode, for use by procfs and others */
866 	if (p->p_textvp)
867 		vrele(p->p_textvp);
868 	vref(epp->ep_vp);
869 	p->p_textvp = epp->ep_vp;
870 
871 	/* create the new process's VM space by running the vmcmds */
872 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
873 
874 #ifdef TRACE_EXEC
875 	DUMPVMCMDS(epp, 0, 0);
876 #endif
877 
878 	base_vcp = NULL;
879 
880 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
881 		struct exec_vmcmd *vcp;
882 
883 		vcp = &epp->ep_vmcmds.evs_cmds[i];
884 		if (vcp->ev_flags & VMCMD_RELATIVE) {
885 			KASSERTMSG(base_vcp != NULL,
886 			    "%s: relative vmcmd with no base", __func__);
887 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
888 			    "%s: illegal base & relative vmcmd", __func__);
889 			vcp->ev_addr += base_vcp->ev_addr;
890 		}
891 		error = (*vcp->ev_proc)(l, vcp);
892 		if (error)
893 			DUMPVMCMDS(epp, i, error);
894 		if (vcp->ev_flags & VMCMD_BASE)
895 			base_vcp = vcp;
896 	}
897 
898 	/* free the vmspace-creation commands, and release their references */
899 	kill_vmcmds(&epp->ep_vmcmds);
900 
901 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
902 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
903 	vput(epp->ep_vp);
904 
905 	/* if an error happened, deallocate and punt */
906 	if (error != 0) {
907 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
908 	}
909 	return error;
910 }
911 
912 static void
913 execve_free_data(struct execve_data *data)
914 {
915 	struct exec_package	* const epp = &data->ed_pack;
916 
917 	/* free the vmspace-creation commands, and release their references */
918 	kill_vmcmds(&epp->ep_vmcmds);
919 	/* kill any opened file descriptor, if necessary */
920 	if (epp->ep_flags & EXEC_HASFD) {
921 		epp->ep_flags &= ~EXEC_HASFD;
922 		fd_close(epp->ep_fd);
923 	}
924 
925 	/* close and put the exec'd file */
926 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
927 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
928 	vput(epp->ep_vp);
929 	pool_put(&exec_pool, data->ed_argp);
930 
931 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
932 	if (epp->ep_emul_root != NULL)
933 		vrele(epp->ep_emul_root);
934 	if (epp->ep_interp != NULL)
935 		vrele(epp->ep_interp);
936 
937 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
938 	pathbuf_destroy(data->ed_pathbuf);
939 	PNBUF_PUT(data->ed_resolvedpathbuf);
940 }
941 
942 static void
943 pathexec(struct proc *p, const char *resolvedname)
944 {
945 	KASSERT(resolvedname[0] == '/');
946 
947 	/* set command name & other accounting info */
948 	strlcpy(p->p_comm, strrchr(resolvedname, '/') + 1, sizeof(p->p_comm));
949 
950 	kmem_strfree(p->p_path);
951 	p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
952 }
953 
954 /* XXX elsewhere */
955 static int
956 credexec(struct lwp *l, struct vattr *attr)
957 {
958 	struct proc *p = l->l_proc;
959 	int error;
960 
961 	/*
962 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
963 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
964 	 * out additional references on the process for the moment.
965 	 */
966 	if ((p->p_slflag & PSL_TRACED) == 0 &&
967 
968 	    (((attr->va_mode & S_ISUID) != 0 &&
969 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
970 
971 	     ((attr->va_mode & S_ISGID) != 0 &&
972 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
973 		/*
974 		 * Mark the process as SUGID before we do
975 		 * anything that might block.
976 		 */
977 		proc_crmod_enter();
978 		proc_crmod_leave(NULL, NULL, true);
979 
980 		/* Make sure file descriptors 0..2 are in use. */
981 		if ((error = fd_checkstd()) != 0) {
982 			DPRINTF(("%s: fdcheckstd failed %d\n",
983 			    __func__, error));
984 			return error;
985 		}
986 
987 		/*
988 		 * Copy the credential so other references don't see our
989 		 * changes.
990 		 */
991 		l->l_cred = kauth_cred_copy(l->l_cred);
992 #ifdef KTRACE
993 		/*
994 		 * If the persistent trace flag isn't set, turn off.
995 		 */
996 		if (p->p_tracep) {
997 			mutex_enter(&ktrace_lock);
998 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
999 				ktrderef(p);
1000 			mutex_exit(&ktrace_lock);
1001 		}
1002 #endif
1003 		if (attr->va_mode & S_ISUID)
1004 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
1005 		if (attr->va_mode & S_ISGID)
1006 			kauth_cred_setegid(l->l_cred, attr->va_gid);
1007 	} else {
1008 		if (kauth_cred_geteuid(l->l_cred) ==
1009 		    kauth_cred_getuid(l->l_cred) &&
1010 		    kauth_cred_getegid(l->l_cred) ==
1011 		    kauth_cred_getgid(l->l_cred))
1012 			p->p_flag &= ~PK_SUGID;
1013 	}
1014 
1015 	/*
1016 	 * Copy the credential so other references don't see our changes.
1017 	 * Test to see if this is necessary first, since in the common case
1018 	 * we won't need a private reference.
1019 	 */
1020 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1021 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1022 		l->l_cred = kauth_cred_copy(l->l_cred);
1023 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1024 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1025 	}
1026 
1027 	/* Update the master credentials. */
1028 	if (l->l_cred != p->p_cred) {
1029 		kauth_cred_t ocred;
1030 
1031 		kauth_cred_hold(l->l_cred);
1032 		mutex_enter(p->p_lock);
1033 		ocred = p->p_cred;
1034 		p->p_cred = l->l_cred;
1035 		mutex_exit(p->p_lock);
1036 		kauth_cred_free(ocred);
1037 	}
1038 
1039 	return 0;
1040 }
1041 
1042 static void
1043 emulexec(struct lwp *l, struct exec_package *epp)
1044 {
1045 	struct proc		*p = l->l_proc;
1046 
1047 	/* The emulation root will usually have been found when we looked
1048 	 * for the elf interpreter (or similar), if not look now. */
1049 	if (epp->ep_esch->es_emul->e_path != NULL &&
1050 	    epp->ep_emul_root == NULL)
1051 		emul_find_root(l, epp);
1052 
1053 	/* Any old emulation root got removed by fdcloseexec */
1054 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1055 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1056 	rw_exit(&p->p_cwdi->cwdi_lock);
1057 	epp->ep_emul_root = NULL;
1058 	if (epp->ep_interp != NULL)
1059 		vrele(epp->ep_interp);
1060 
1061 	/*
1062 	 * Call emulation specific exec hook. This can setup per-process
1063 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1064 	 *
1065 	 * If we are executing process of different emulation than the
1066 	 * original forked process, call e_proc_exit() of the old emulation
1067 	 * first, then e_proc_exec() of new emulation. If the emulation is
1068 	 * same, the exec hook code should deallocate any old emulation
1069 	 * resources held previously by this process.
1070 	 */
1071 	if (p->p_emul && p->p_emul->e_proc_exit
1072 	    && p->p_emul != epp->ep_esch->es_emul)
1073 		(*p->p_emul->e_proc_exit)(p);
1074 
1075 	/*
1076 	 * This is now LWP 1.
1077 	 */
1078 	/* XXX elsewhere */
1079 	mutex_enter(p->p_lock);
1080 	p->p_nlwpid = 1;
1081 	l->l_lid = 1;
1082 	mutex_exit(p->p_lock);
1083 
1084 	/*
1085 	 * Call exec hook. Emulation code may NOT store reference to anything
1086 	 * from &pack.
1087 	 */
1088 	if (epp->ep_esch->es_emul->e_proc_exec)
1089 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1090 
1091 	/* update p_emul, the old value is no longer needed */
1092 	p->p_emul = epp->ep_esch->es_emul;
1093 
1094 	/* ...and the same for p_execsw */
1095 	p->p_execsw = epp->ep_esch;
1096 
1097 #ifdef __HAVE_SYSCALL_INTERN
1098 	(*p->p_emul->e_syscall_intern)(p);
1099 #endif
1100 	ktremul();
1101 }
1102 
1103 static int
1104 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1105 	bool no_local_exec_lock, bool is_spawn)
1106 {
1107 	struct exec_package	* const epp = &data->ed_pack;
1108 	int error = 0;
1109 	struct proc		*p;
1110 
1111 	/*
1112 	 * In case of a posix_spawn operation, the child doing the exec
1113 	 * might not hold the reader lock on exec_lock, but the parent
1114 	 * will do this instead.
1115 	 */
1116 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1117 	KASSERT(!no_local_exec_lock || is_spawn);
1118 	KASSERT(data != NULL);
1119 
1120 	p = l->l_proc;
1121 
1122 	/* Get rid of other LWPs. */
1123 	if (p->p_nlwps > 1) {
1124 		mutex_enter(p->p_lock);
1125 		exit_lwps(l);
1126 		mutex_exit(p->p_lock);
1127 	}
1128 	KDASSERT(p->p_nlwps == 1);
1129 
1130 	/* Destroy any lwpctl info. */
1131 	if (p->p_lwpctl != NULL)
1132 		lwp_ctl_exit();
1133 
1134 	/* Remove POSIX timers */
1135 	timers_free(p, TIMERS_POSIX);
1136 
1137 	/* Set the PaX flags. */
1138 	pax_set_flags(epp, p);
1139 
1140 	/*
1141 	 * Do whatever is necessary to prepare the address space
1142 	 * for remapping.  Note that this might replace the current
1143 	 * vmspace with another!
1144 	 */
1145 	if (is_spawn)
1146 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1147 		    epp->ep_vm_maxaddr,
1148 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1149 	else
1150 		uvmspace_exec(l, epp->ep_vm_minaddr,
1151 		    epp->ep_vm_maxaddr,
1152 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1153 
1154 	struct vmspace		*vm;
1155 	vm = p->p_vmspace;
1156 	vm->vm_taddr = (void *)epp->ep_taddr;
1157 	vm->vm_tsize = btoc(epp->ep_tsize);
1158 	vm->vm_daddr = (void*)epp->ep_daddr;
1159 	vm->vm_dsize = btoc(epp->ep_dsize);
1160 	vm->vm_ssize = btoc(epp->ep_ssize);
1161 	vm->vm_issize = 0;
1162 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1163 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1164 
1165 	pax_aslr_init_vm(l, vm, epp);
1166 
1167 	/* Now map address space. */
1168 	error = execve_dovmcmds(l, data);
1169 	if (error != 0)
1170 		goto exec_abort;
1171 
1172 	pathexec(p, epp->ep_resolvedname);
1173 
1174 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1175 
1176 	error = copyoutargs(data, l, newstack);
1177 	if (error != 0)
1178 		goto exec_abort;
1179 
1180 	cwdexec(p);
1181 	fd_closeexec();		/* handle close on exec */
1182 
1183 	if (__predict_false(ktrace_on))
1184 		fd_ktrexecfd();
1185 
1186 	execsigs(p);		/* reset caught signals */
1187 
1188 	mutex_enter(p->p_lock);
1189 	l->l_ctxlink = NULL;	/* reset ucontext link */
1190 	p->p_acflag &= ~AFORK;
1191 	p->p_flag |= PK_EXEC;
1192 	mutex_exit(p->p_lock);
1193 
1194 	/*
1195 	 * Stop profiling.
1196 	 */
1197 	if ((p->p_stflag & PST_PROFIL) != 0) {
1198 		mutex_spin_enter(&p->p_stmutex);
1199 		stopprofclock(p);
1200 		mutex_spin_exit(&p->p_stmutex);
1201 	}
1202 
1203 	/*
1204 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1205 	 * exited and exec()/exit() are the only places it will be cleared.
1206 	 */
1207 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1208 		mutex_enter(proc_lock);
1209 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1210 		p->p_lflag &= ~PL_PPWAIT;
1211 		cv_broadcast(&p->p_pptr->p_waitcv);
1212 		mutex_exit(proc_lock);
1213 	}
1214 
1215 	error = credexec(l, &data->ed_attr);
1216 	if (error)
1217 		goto exec_abort;
1218 
1219 #if defined(__HAVE_RAS)
1220 	/*
1221 	 * Remove all RASs from the address space.
1222 	 */
1223 	ras_purgeall();
1224 #endif
1225 
1226 	doexechooks(p);
1227 
1228 	/*
1229 	 * Set initial SP at the top of the stack.
1230 	 *
1231 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1232 	 * the end of arg/env strings.  Userland guesses the address of argc
1233 	 * via ps_strings::ps_argvstr.
1234 	 */
1235 
1236 	/* Setup new registers and do misc. setup. */
1237 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1238 	if (epp->ep_esch->es_setregs)
1239 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1240 
1241 	/* Provide a consistent LWP private setting */
1242 	(void)lwp_setprivate(l, NULL);
1243 
1244 	/* Discard all PCU state; need to start fresh */
1245 	pcu_discard_all(l);
1246 
1247 	/* map the process's signal trampoline code */
1248 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1249 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1250 		goto exec_abort;
1251 	}
1252 
1253 	pool_put(&exec_pool, data->ed_argp);
1254 
1255 	/* notify others that we exec'd */
1256 	KNOTE(&p->p_klist, NOTE_EXEC);
1257 
1258 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1259 
1260 	SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1261 
1262 	emulexec(l, epp);
1263 
1264 	/* Allow new references from the debugger/procfs. */
1265 	rw_exit(&p->p_reflock);
1266 	if (!no_local_exec_lock)
1267 		rw_exit(&exec_lock);
1268 
1269 	mutex_enter(proc_lock);
1270 
1271 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1272 		mutex_enter(p->p_lock);
1273 		p->p_xsig = SIGTRAP;
1274 		p->p_sigctx.ps_faked = true; // XXX
1275 		p->p_sigctx.ps_info._signo = p->p_xsig;
1276 		p->p_sigctx.ps_info._code = TRAP_EXEC;
1277 		sigswitch(0, SIGTRAP, false);
1278 		// XXX ktrpoint(KTR_PSIG)
1279 		mutex_exit(p->p_lock);
1280 		mutex_enter(proc_lock);
1281 	}
1282 
1283 	if (p->p_sflag & PS_STOPEXEC) {
1284 		ksiginfoq_t kq;
1285 
1286 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1287 		p->p_pptr->p_nstopchild++;
1288 		p->p_waited = 0;
1289 		mutex_enter(p->p_lock);
1290 		ksiginfo_queue_init(&kq);
1291 		sigclearall(p, &contsigmask, &kq);
1292 		lwp_lock(l);
1293 		l->l_stat = LSSTOP;
1294 		p->p_stat = SSTOP;
1295 		p->p_nrlwps--;
1296 		lwp_unlock(l);
1297 		mutex_exit(p->p_lock);
1298 		mutex_exit(proc_lock);
1299 		lwp_lock(l);
1300 		mi_switch(l);
1301 		ksiginfo_queue_drain(&kq);
1302 		KERNEL_LOCK(l->l_biglocks, l);
1303 	} else {
1304 		mutex_exit(proc_lock);
1305 	}
1306 
1307 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1308 	pathbuf_destroy(data->ed_pathbuf);
1309 	PNBUF_PUT(data->ed_resolvedpathbuf);
1310 #ifdef TRACE_EXEC
1311 	DPRINTF(("%s finished\n", __func__));
1312 #endif
1313 	return EJUSTRETURN;
1314 
1315  exec_abort:
1316 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1317 	rw_exit(&p->p_reflock);
1318 	if (!no_local_exec_lock)
1319 		rw_exit(&exec_lock);
1320 
1321 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1322 	pathbuf_destroy(data->ed_pathbuf);
1323 	PNBUF_PUT(data->ed_resolvedpathbuf);
1324 
1325 	/*
1326 	 * the old process doesn't exist anymore.  exit gracefully.
1327 	 * get rid of the (new) address space we have created, if any, get rid
1328 	 * of our namei data and vnode, and exit noting failure
1329 	 */
1330 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1331 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1332 
1333 	exec_free_emul_arg(epp);
1334 	pool_put(&exec_pool, data->ed_argp);
1335 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1336 	if (epp->ep_emul_root != NULL)
1337 		vrele(epp->ep_emul_root);
1338 	if (epp->ep_interp != NULL)
1339 		vrele(epp->ep_interp);
1340 
1341 	/* Acquire the sched-state mutex (exit1() will release it). */
1342 	if (!is_spawn) {
1343 		mutex_enter(p->p_lock);
1344 		exit1(l, error, SIGABRT);
1345 	}
1346 
1347 	return error;
1348 }
1349 
1350 int
1351 execve1(struct lwp *l, const char *path, char * const *args,
1352     char * const *envs, execve_fetch_element_t fetch_element)
1353 {
1354 	struct execve_data data;
1355 	int error;
1356 
1357 	error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1358 	if (error)
1359 		return error;
1360 	error = execve_runproc(l, &data, false, false);
1361 	return error;
1362 }
1363 
1364 static size_t
1365 fromptrsz(const struct exec_package *epp)
1366 {
1367 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1368 }
1369 
1370 static size_t
1371 ptrsz(const struct exec_package *epp)
1372 {
1373 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1374 }
1375 
1376 static size_t
1377 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1378 {
1379 	struct exec_package	* const epp = &data->ed_pack;
1380 
1381 	const size_t nargenvptrs =
1382 	    1 +				/* long argc */
1383 	    data->ed_argc +		/* char *argv[] */
1384 	    1 +				/* \0 */
1385 	    data->ed_envc +		/* char *env[] */
1386 	    1;				/* \0 */
1387 
1388 	return (nargenvptrs * ptrsz(epp))	/* pointers */
1389 	    + argenvstrlen			/* strings */
1390 	    + epp->ep_esch->es_arglen;		/* auxinfo */
1391 }
1392 
1393 static size_t
1394 calcstack(struct execve_data * restrict data, const size_t gaplen)
1395 {
1396 	struct exec_package	* const epp = &data->ed_pack;
1397 
1398 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1399 	    epp->ep_esch->es_emul->e_sigcode;
1400 
1401 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1402 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1403 
1404 	const size_t sigcode_psstr_sz =
1405 	    data->ed_szsigcode +	/* sigcode */
1406 	    data->ed_ps_strings_sz +	/* ps_strings */
1407 	    STACK_PTHREADSPACE;		/* pthread space */
1408 
1409 	const size_t stacklen =
1410 	    data->ed_argslen +
1411 	    gaplen +
1412 	    sigcode_psstr_sz;
1413 
1414 	/* make the stack "safely" aligned */
1415 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1416 }
1417 
1418 static int
1419 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1420     char * const newstack)
1421 {
1422 	struct exec_package	* const epp = &data->ed_pack;
1423 	struct proc		*p = l->l_proc;
1424 	int			error;
1425 
1426 	/* remember information about the process */
1427 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1428 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1429 
1430 	/*
1431 	 * Allocate the stack address passed to the newly execve()'ed process.
1432 	 *
1433 	 * The new stack address will be set to the SP (stack pointer) register
1434 	 * in setregs().
1435 	 */
1436 
1437 	char *newargs = STACK_ALLOC(
1438 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1439 
1440 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1441 	    &data->ed_arginfo, &newargs, data->ed_argp);
1442 
1443 	if (error) {
1444 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1445 		return error;
1446 	}
1447 
1448 	error = copyoutpsstrs(data, p);
1449 	if (error != 0)
1450 		return error;
1451 
1452 	return 0;
1453 }
1454 
1455 static int
1456 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1457 {
1458 	struct exec_package	* const epp = &data->ed_pack;
1459 	struct ps_strings32	arginfo32;
1460 	void			*aip;
1461 	int			error;
1462 
1463 	/* fill process ps_strings info */
1464 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1465 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1466 
1467 	if (epp->ep_flags & EXEC_32) {
1468 		aip = &arginfo32;
1469 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1470 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1471 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1472 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1473 	} else
1474 		aip = &data->ed_arginfo;
1475 
1476 	/* copy out the process's ps_strings structure */
1477 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1478 	    != 0) {
1479 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1480 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1481 		return error;
1482 	}
1483 
1484 	return 0;
1485 }
1486 
1487 static int
1488 copyinargs(struct execve_data * restrict data, char * const *args,
1489     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1490 {
1491 	struct exec_package	* const epp = &data->ed_pack;
1492 	char			*dp;
1493 	size_t			i;
1494 	int			error;
1495 
1496 	dp = *dpp;
1497 
1498 	data->ed_argc = 0;
1499 
1500 	/* copy the fake args list, if there's one, freeing it as we go */
1501 	if (epp->ep_flags & EXEC_HASARGL) {
1502 		struct exec_fakearg	*fa = epp->ep_fa;
1503 
1504 		while (fa->fa_arg != NULL) {
1505 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1506 			size_t len;
1507 
1508 			len = strlcpy(dp, fa->fa_arg, maxlen);
1509 			/* Count NUL into len. */
1510 			if (len < maxlen)
1511 				len++;
1512 			else {
1513 				while (fa->fa_arg != NULL) {
1514 					kmem_free(fa->fa_arg, fa->fa_len);
1515 					fa++;
1516 				}
1517 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1518 				epp->ep_flags &= ~EXEC_HASARGL;
1519 				return E2BIG;
1520 			}
1521 			ktrexecarg(fa->fa_arg, len - 1);
1522 			dp += len;
1523 
1524 			kmem_free(fa->fa_arg, fa->fa_len);
1525 			fa++;
1526 			data->ed_argc++;
1527 		}
1528 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1529 		epp->ep_flags &= ~EXEC_HASARGL;
1530 	}
1531 
1532 	/*
1533 	 * Read and count argument strings from user.
1534 	 */
1535 
1536 	if (args == NULL) {
1537 		DPRINTF(("%s: null args\n", __func__));
1538 		return EINVAL;
1539 	}
1540 	if (epp->ep_flags & EXEC_SKIPARG)
1541 		args = (const void *)((const char *)args + fromptrsz(epp));
1542 	i = 0;
1543 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1544 	if (error != 0) {
1545 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1546 		return error;
1547 	}
1548 	data->ed_argc += i;
1549 
1550 	/*
1551 	 * Read and count environment strings from user.
1552 	 */
1553 
1554 	data->ed_envc = 0;
1555 	/* environment need not be there */
1556 	if (envs == NULL)
1557 		goto done;
1558 	i = 0;
1559 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1560 	if (error != 0) {
1561 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1562 		return error;
1563 	}
1564 	data->ed_envc += i;
1565 
1566 done:
1567 	*dpp = dp;
1568 
1569 	return 0;
1570 }
1571 
1572 static int
1573 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1574     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1575     void (*ktr)(const void *, size_t))
1576 {
1577 	char			*dp, *sp;
1578 	size_t			i;
1579 	int			error;
1580 
1581 	dp = *dpp;
1582 
1583 	i = 0;
1584 	while (1) {
1585 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1586 		size_t len;
1587 
1588 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1589 			return error;
1590 		}
1591 		if (!sp)
1592 			break;
1593 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1594 			if (error == ENAMETOOLONG)
1595 				error = E2BIG;
1596 			return error;
1597 		}
1598 		if (__predict_false(ktrace_on))
1599 			(*ktr)(dp, len - 1);
1600 		dp += len;
1601 		i++;
1602 	}
1603 
1604 	*dpp = dp;
1605 	*ip = i;
1606 
1607 	return 0;
1608 }
1609 
1610 /*
1611  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1612  * Those strings are located just after auxinfo.
1613  */
1614 int
1615 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1616     char **stackp, void *argp)
1617 {
1618 	char	**cpp, *dp, *sp;
1619 	size_t	len;
1620 	void	*nullp;
1621 	long	argc, envc;
1622 	int	error;
1623 
1624 	cpp = (char **)*stackp;
1625 	nullp = NULL;
1626 	argc = arginfo->ps_nargvstr;
1627 	envc = arginfo->ps_nenvstr;
1628 
1629 	/* argc on stack is long */
1630 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1631 
1632 	dp = (char *)(cpp +
1633 	    1 +				/* long argc */
1634 	    argc +			/* char *argv[] */
1635 	    1 +				/* \0 */
1636 	    envc +			/* char *env[] */
1637 	    1) +			/* \0 */
1638 	    pack->ep_esch->es_arglen;	/* auxinfo */
1639 	sp = argp;
1640 
1641 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1642 		COPYPRINTF("", cpp - 1, sizeof(argc));
1643 		return error;
1644 	}
1645 
1646 	/* XXX don't copy them out, remap them! */
1647 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1648 
1649 	for (; --argc >= 0; sp += len, dp += len) {
1650 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1651 			COPYPRINTF("", cpp - 1, sizeof(dp));
1652 			return error;
1653 		}
1654 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1655 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1656 			return error;
1657 		}
1658 	}
1659 
1660 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1661 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1662 		return error;
1663 	}
1664 
1665 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1666 
1667 	for (; --envc >= 0; sp += len, dp += len) {
1668 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1669 			COPYPRINTF("", cpp - 1, sizeof(dp));
1670 			return error;
1671 		}
1672 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1673 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1674 			return error;
1675 		}
1676 
1677 	}
1678 
1679 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1680 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1681 		return error;
1682 	}
1683 
1684 	*stackp = (char *)cpp;
1685 	return 0;
1686 }
1687 
1688 
1689 /*
1690  * Add execsw[] entries.
1691  */
1692 int
1693 exec_add(struct execsw *esp, int count)
1694 {
1695 	struct exec_entry	*it;
1696 	int			i;
1697 
1698 	if (count == 0) {
1699 		return 0;
1700 	}
1701 
1702 	/* Check for duplicates. */
1703 	rw_enter(&exec_lock, RW_WRITER);
1704 	for (i = 0; i < count; i++) {
1705 		LIST_FOREACH(it, &ex_head, ex_list) {
1706 			/* assume unique (makecmds, probe_func, emulation) */
1707 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1708 			    it->ex_sw->u.elf_probe_func ==
1709 			    esp[i].u.elf_probe_func &&
1710 			    it->ex_sw->es_emul == esp[i].es_emul) {
1711 				rw_exit(&exec_lock);
1712 				return EEXIST;
1713 			}
1714 		}
1715 	}
1716 
1717 	/* Allocate new entries. */
1718 	for (i = 0; i < count; i++) {
1719 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1720 		it->ex_sw = &esp[i];
1721 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1722 	}
1723 
1724 	/* update execsw[] */
1725 	exec_init(0);
1726 	rw_exit(&exec_lock);
1727 	return 0;
1728 }
1729 
1730 /*
1731  * Remove execsw[] entry.
1732  */
1733 int
1734 exec_remove(struct execsw *esp, int count)
1735 {
1736 	struct exec_entry	*it, *next;
1737 	int			i;
1738 	const struct proclist_desc *pd;
1739 	proc_t			*p;
1740 
1741 	if (count == 0) {
1742 		return 0;
1743 	}
1744 
1745 	/* Abort if any are busy. */
1746 	rw_enter(&exec_lock, RW_WRITER);
1747 	for (i = 0; i < count; i++) {
1748 		mutex_enter(proc_lock);
1749 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1750 			PROCLIST_FOREACH(p, pd->pd_list) {
1751 				if (p->p_execsw == &esp[i]) {
1752 					mutex_exit(proc_lock);
1753 					rw_exit(&exec_lock);
1754 					return EBUSY;
1755 				}
1756 			}
1757 		}
1758 		mutex_exit(proc_lock);
1759 	}
1760 
1761 	/* None are busy, so remove them all. */
1762 	for (i = 0; i < count; i++) {
1763 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1764 			next = LIST_NEXT(it, ex_list);
1765 			if (it->ex_sw == &esp[i]) {
1766 				LIST_REMOVE(it, ex_list);
1767 				kmem_free(it, sizeof(*it));
1768 				break;
1769 			}
1770 		}
1771 	}
1772 
1773 	/* update execsw[] */
1774 	exec_init(0);
1775 	rw_exit(&exec_lock);
1776 	return 0;
1777 }
1778 
1779 /*
1780  * Initialize exec structures. If init_boot is true, also does necessary
1781  * one-time initialization (it's called from main() that way).
1782  * Once system is multiuser, this should be called with exec_lock held,
1783  * i.e. via exec_{add|remove}().
1784  */
1785 int
1786 exec_init(int init_boot)
1787 {
1788 	const struct execsw 	**sw;
1789 	struct exec_entry	*ex;
1790 	SLIST_HEAD(,exec_entry)	first;
1791 	SLIST_HEAD(,exec_entry)	any;
1792 	SLIST_HEAD(,exec_entry)	last;
1793 	int			i, sz;
1794 
1795 	if (init_boot) {
1796 		/* do one-time initializations */
1797 		vaddr_t vmin = 0, vmax;
1798 
1799 		rw_init(&exec_lock);
1800 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1801 		exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1802 		    maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1803 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1804 		    "execargs", &exec_palloc, IPL_NONE);
1805 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1806 	} else {
1807 		KASSERT(rw_write_held(&exec_lock));
1808 	}
1809 
1810 	/* Sort each entry onto the appropriate queue. */
1811 	SLIST_INIT(&first);
1812 	SLIST_INIT(&any);
1813 	SLIST_INIT(&last);
1814 	sz = 0;
1815 	LIST_FOREACH(ex, &ex_head, ex_list) {
1816 		switch(ex->ex_sw->es_prio) {
1817 		case EXECSW_PRIO_FIRST:
1818 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1819 			break;
1820 		case EXECSW_PRIO_ANY:
1821 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1822 			break;
1823 		case EXECSW_PRIO_LAST:
1824 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1825 			break;
1826 		default:
1827 			panic("%s", __func__);
1828 			break;
1829 		}
1830 		sz++;
1831 	}
1832 
1833 	/*
1834 	 * Create new execsw[].  Ensure we do not try a zero-sized
1835 	 * allocation.
1836 	 */
1837 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1838 	i = 0;
1839 	SLIST_FOREACH(ex, &first, ex_slist) {
1840 		sw[i++] = ex->ex_sw;
1841 	}
1842 	SLIST_FOREACH(ex, &any, ex_slist) {
1843 		sw[i++] = ex->ex_sw;
1844 	}
1845 	SLIST_FOREACH(ex, &last, ex_slist) {
1846 		sw[i++] = ex->ex_sw;
1847 	}
1848 
1849 	/* Replace old execsw[] and free used memory. */
1850 	if (execsw != NULL) {
1851 		kmem_free(__UNCONST(execsw),
1852 		    nexecs * sizeof(struct execsw *) + 1);
1853 	}
1854 	execsw = sw;
1855 	nexecs = sz;
1856 
1857 	/* Figure out the maximum size of an exec header. */
1858 	exec_maxhdrsz = sizeof(int);
1859 	for (i = 0; i < nexecs; i++) {
1860 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1861 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1862 	}
1863 
1864 	return 0;
1865 }
1866 
1867 static int
1868 exec_sigcode_map(struct proc *p, const struct emul *e)
1869 {
1870 	vaddr_t va;
1871 	vsize_t sz;
1872 	int error;
1873 	struct uvm_object *uobj;
1874 
1875 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1876 
1877 	if (e->e_sigobject == NULL || sz == 0) {
1878 		return 0;
1879 	}
1880 
1881 	/*
1882 	 * If we don't have a sigobject for this emulation, create one.
1883 	 *
1884 	 * sigobject is an anonymous memory object (just like SYSV shared
1885 	 * memory) that we keep a permanent reference to and that we map
1886 	 * in all processes that need this sigcode. The creation is simple,
1887 	 * we create an object, add a permanent reference to it, map it in
1888 	 * kernel space, copy out the sigcode to it and unmap it.
1889 	 * We map it with PROT_READ|PROT_EXEC into the process just
1890 	 * the way sys_mmap() would map it.
1891 	 */
1892 
1893 	uobj = *e->e_sigobject;
1894 	if (uobj == NULL) {
1895 		mutex_enter(&sigobject_lock);
1896 		if ((uobj = *e->e_sigobject) == NULL) {
1897 			uobj = uao_create(sz, 0);
1898 			(*uobj->pgops->pgo_reference)(uobj);
1899 			va = vm_map_min(kernel_map);
1900 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1901 			    uobj, 0, 0,
1902 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1903 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1904 				printf("kernel mapping failed %d\n", error);
1905 				(*uobj->pgops->pgo_detach)(uobj);
1906 				mutex_exit(&sigobject_lock);
1907 				return error;
1908 			}
1909 			memcpy((void *)va, e->e_sigcode, sz);
1910 #ifdef PMAP_NEED_PROCWR
1911 			pmap_procwr(&proc0, va, sz);
1912 #endif
1913 			uvm_unmap(kernel_map, va, va + round_page(sz));
1914 			*e->e_sigobject = uobj;
1915 		}
1916 		mutex_exit(&sigobject_lock);
1917 	}
1918 
1919 	/* Just a hint to uvm_map where to put it. */
1920 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1921 	    round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
1922 
1923 #ifdef __alpha__
1924 	/*
1925 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1926 	 * which causes the above calculation to put the sigcode at
1927 	 * an invalid address.  Put it just below the text instead.
1928 	 */
1929 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1930 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1931 	}
1932 #endif
1933 
1934 	(*uobj->pgops->pgo_reference)(uobj);
1935 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1936 			uobj, 0, 0,
1937 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1938 				    UVM_ADV_RANDOM, 0));
1939 	if (error) {
1940 		DPRINTF(("%s, %d: map %p "
1941 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1942 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1943 		    va, error));
1944 		(*uobj->pgops->pgo_detach)(uobj);
1945 		return error;
1946 	}
1947 	p->p_sigctx.ps_sigcode = (void *)va;
1948 	return 0;
1949 }
1950 
1951 /*
1952  * Release a refcount on spawn_exec_data and destroy memory, if this
1953  * was the last one.
1954  */
1955 static void
1956 spawn_exec_data_release(struct spawn_exec_data *data)
1957 {
1958 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1959 		return;
1960 
1961 	cv_destroy(&data->sed_cv_child_ready);
1962 	mutex_destroy(&data->sed_mtx_child);
1963 
1964 	if (data->sed_actions)
1965 		posix_spawn_fa_free(data->sed_actions,
1966 		    data->sed_actions->len);
1967 	if (data->sed_attrs)
1968 		kmem_free(data->sed_attrs,
1969 		    sizeof(*data->sed_attrs));
1970 	kmem_free(data, sizeof(*data));
1971 }
1972 
1973 /*
1974  * A child lwp of a posix_spawn operation starts here and ends up in
1975  * cpu_spawn_return, dealing with all filedescriptor and scheduler
1976  * manipulations in between.
1977  * The parent waits for the child, as it is not clear whether the child
1978  * will be able to acquire its own exec_lock. If it can, the parent can
1979  * be released early and continue running in parallel. If not (or if the
1980  * magic debug flag is passed in the scheduler attribute struct), the
1981  * child rides on the parent's exec lock until it is ready to return to
1982  * to userland - and only then releases the parent. This method loses
1983  * concurrency, but improves error reporting.
1984  */
1985 static void
1986 spawn_return(void *arg)
1987 {
1988 	struct spawn_exec_data *spawn_data = arg;
1989 	struct lwp *l = curlwp;
1990 	int error, newfd;
1991 	int ostat;
1992 	size_t i;
1993 	const struct posix_spawn_file_actions_entry *fae;
1994 	pid_t ppid;
1995 	register_t retval;
1996 	bool have_reflock;
1997 	bool parent_is_waiting = true;
1998 
1999 	/*
2000 	 * Check if we can release parent early.
2001 	 * We either need to have no sed_attrs, or sed_attrs does not
2002 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2003 	 * safe access to the parent proc (passed in sed_parent).
2004 	 * We then try to get the exec_lock, and only if that works, we can
2005 	 * release the parent here already.
2006 	 */
2007 	ppid = spawn_data->sed_parent->p_pid;
2008 	if ((!spawn_data->sed_attrs
2009 	    || (spawn_data->sed_attrs->sa_flags
2010 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2011 	    && rw_tryenter(&exec_lock, RW_READER)) {
2012 		parent_is_waiting = false;
2013 		mutex_enter(&spawn_data->sed_mtx_child);
2014 		cv_signal(&spawn_data->sed_cv_child_ready);
2015 		mutex_exit(&spawn_data->sed_mtx_child);
2016 	}
2017 
2018 	/* don't allow debugger access yet */
2019 	rw_enter(&l->l_proc->p_reflock, RW_WRITER);
2020 	have_reflock = true;
2021 
2022 	error = 0;
2023 	/* handle posix_spawn_file_actions */
2024 	if (spawn_data->sed_actions != NULL) {
2025 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
2026 			fae = &spawn_data->sed_actions->fae[i];
2027 			switch (fae->fae_action) {
2028 			case FAE_OPEN:
2029 				if (fd_getfile(fae->fae_fildes) != NULL) {
2030 					error = fd_close(fae->fae_fildes);
2031 					if (error)
2032 						break;
2033 				}
2034 				error = fd_open(fae->fae_path, fae->fae_oflag,
2035 				    fae->fae_mode, &newfd);
2036 				if (error)
2037 					break;
2038 				if (newfd != fae->fae_fildes) {
2039 					error = dodup(l, newfd,
2040 					    fae->fae_fildes, 0, &retval);
2041 					if (fd_getfile(newfd) != NULL)
2042 						fd_close(newfd);
2043 				}
2044 				break;
2045 			case FAE_DUP2:
2046 				error = dodup(l, fae->fae_fildes,
2047 				    fae->fae_newfildes, 0, &retval);
2048 				break;
2049 			case FAE_CLOSE:
2050 				if (fd_getfile(fae->fae_fildes) == NULL) {
2051 					error = EBADF;
2052 					break;
2053 				}
2054 				error = fd_close(fae->fae_fildes);
2055 				break;
2056 			}
2057 			if (error)
2058 				goto report_error;
2059 		}
2060 	}
2061 
2062 	/* handle posix_spawnattr */
2063 	if (spawn_data->sed_attrs != NULL) {
2064 		struct sigaction sigact;
2065 		sigact._sa_u._sa_handler = SIG_DFL;
2066 		sigact.sa_flags = 0;
2067 
2068 		/*
2069 		 * set state to SSTOP so that this proc can be found by pid.
2070 		 * see proc_enterprp, do_sched_setparam below
2071 		 */
2072 		mutex_enter(proc_lock);
2073 		/*
2074 		 * p_stat should be SACTIVE, so we need to adjust the
2075 		 * parent's p_nstopchild here.  For safety, just make
2076 		 * we're on the good side of SDEAD before we adjust.
2077 		 */
2078 		ostat = l->l_proc->p_stat;
2079 		KASSERT(ostat < SSTOP);
2080 		l->l_proc->p_stat = SSTOP;
2081 		l->l_proc->p_waited = 0;
2082 		l->l_proc->p_pptr->p_nstopchild++;
2083 		mutex_exit(proc_lock);
2084 
2085 		/* Set process group */
2086 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2087 			pid_t mypid = l->l_proc->p_pid,
2088 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
2089 
2090 			if (pgrp == 0)
2091 				pgrp = mypid;
2092 
2093 			error = proc_enterpgrp(spawn_data->sed_parent,
2094 			    mypid, pgrp, false);
2095 			if (error)
2096 				goto report_error_stopped;
2097 		}
2098 
2099 		/* Set scheduler policy */
2100 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2101 			error = do_sched_setparam(l->l_proc->p_pid, 0,
2102 			    spawn_data->sed_attrs->sa_schedpolicy,
2103 			    &spawn_data->sed_attrs->sa_schedparam);
2104 		else if (spawn_data->sed_attrs->sa_flags
2105 		    & POSIX_SPAWN_SETSCHEDPARAM) {
2106 			error = do_sched_setparam(ppid, 0,
2107 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2108 		}
2109 		if (error)
2110 			goto report_error_stopped;
2111 
2112 		/* Reset user ID's */
2113 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2114 			error = do_setresuid(l, -1,
2115 			     kauth_cred_getgid(l->l_cred), -1,
2116 			     ID_E_EQ_R | ID_E_EQ_S);
2117 			if (error)
2118 				goto report_error_stopped;
2119 			error = do_setresuid(l, -1,
2120 			    kauth_cred_getuid(l->l_cred), -1,
2121 			    ID_E_EQ_R | ID_E_EQ_S);
2122 			if (error)
2123 				goto report_error_stopped;
2124 		}
2125 
2126 		/* Set signal masks/defaults */
2127 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2128 			mutex_enter(l->l_proc->p_lock);
2129 			error = sigprocmask1(l, SIG_SETMASK,
2130 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
2131 			mutex_exit(l->l_proc->p_lock);
2132 			if (error)
2133 				goto report_error_stopped;
2134 		}
2135 
2136 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2137 			/*
2138 			 * The following sigaction call is using a sigaction
2139 			 * version 0 trampoline which is in the compatibility
2140 			 * code only. This is not a problem because for SIG_DFL
2141 			 * and SIG_IGN, the trampolines are now ignored. If they
2142 			 * were not, this would be a problem because we are
2143 			 * holding the exec_lock, and the compat code needs
2144 			 * to do the same in order to replace the trampoline
2145 			 * code of the process.
2146 			 */
2147 			for (i = 1; i <= NSIG; i++) {
2148 				if (sigismember(
2149 				    &spawn_data->sed_attrs->sa_sigdefault, i))
2150 					sigaction1(l, i, &sigact, NULL, NULL,
2151 					    0);
2152 			}
2153 		}
2154 		mutex_enter(proc_lock);
2155 		l->l_proc->p_stat = ostat;
2156 		l->l_proc->p_pptr->p_nstopchild--;
2157 		mutex_exit(proc_lock);
2158 	}
2159 
2160 	/* now do the real exec */
2161 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2162 	    true);
2163 	have_reflock = false;
2164 	if (error == EJUSTRETURN)
2165 		error = 0;
2166 	else if (error)
2167 		goto report_error;
2168 
2169 	if (parent_is_waiting) {
2170 		mutex_enter(&spawn_data->sed_mtx_child);
2171 		cv_signal(&spawn_data->sed_cv_child_ready);
2172 		mutex_exit(&spawn_data->sed_mtx_child);
2173 	}
2174 
2175 	/* release our refcount on the data */
2176 	spawn_exec_data_release(spawn_data);
2177 
2178 	/* and finally: leave to userland for the first time */
2179 	cpu_spawn_return(l);
2180 
2181 	/* NOTREACHED */
2182 	return;
2183 
2184  report_error_stopped:
2185 	mutex_enter(proc_lock);
2186 	l->l_proc->p_stat = ostat;
2187 	l->l_proc->p_pptr->p_nstopchild--;
2188 	mutex_exit(proc_lock);
2189  report_error:
2190 	if (have_reflock) {
2191 		/*
2192 		 * We have not passed through execve_runproc(),
2193 		 * which would have released the p_reflock and also
2194 		 * taken ownership of the sed_exec part of spawn_data,
2195 		 * so release/free both here.
2196 		 */
2197 		rw_exit(&l->l_proc->p_reflock);
2198 		execve_free_data(&spawn_data->sed_exec);
2199 	}
2200 
2201 	if (parent_is_waiting) {
2202 		/* pass error to parent */
2203 		mutex_enter(&spawn_data->sed_mtx_child);
2204 		spawn_data->sed_error = error;
2205 		cv_signal(&spawn_data->sed_cv_child_ready);
2206 		mutex_exit(&spawn_data->sed_mtx_child);
2207 	} else {
2208 		rw_exit(&exec_lock);
2209 	}
2210 
2211 	/* release our refcount on the data */
2212 	spawn_exec_data_release(spawn_data);
2213 
2214 	/* done, exit */
2215 	mutex_enter(l->l_proc->p_lock);
2216 	/*
2217 	 * Posix explicitly asks for an exit code of 127 if we report
2218 	 * errors from the child process - so, unfortunately, there
2219 	 * is no way to report a more exact error code.
2220 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2221 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2222 	 */
2223 	exit1(l, 127, 0);
2224 }
2225 
2226 void
2227 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2228 {
2229 
2230 	for (size_t i = 0; i < len; i++) {
2231 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2232 		if (fae->fae_action != FAE_OPEN)
2233 			continue;
2234 		kmem_strfree(fae->fae_path);
2235 	}
2236 	if (fa->len > 0)
2237 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2238 	kmem_free(fa, sizeof(*fa));
2239 }
2240 
2241 static int
2242 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2243     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2244 {
2245 	struct posix_spawn_file_actions *fa;
2246 	struct posix_spawn_file_actions_entry *fae;
2247 	char *pbuf = NULL;
2248 	int error;
2249 	size_t i = 0;
2250 
2251 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2252 	error = copyin(ufa, fa, sizeof(*fa));
2253 	if (error || fa->len == 0) {
2254 		kmem_free(fa, sizeof(*fa));
2255 		return error;	/* 0 if not an error, and len == 0 */
2256 	}
2257 
2258 	if (fa->len > lim) {
2259 		kmem_free(fa, sizeof(*fa));
2260 		return EINVAL;
2261 	}
2262 
2263 	fa->size = fa->len;
2264 	size_t fal = fa->len * sizeof(*fae);
2265 	fae = fa->fae;
2266 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2267 	error = copyin(fae, fa->fae, fal);
2268 	if (error)
2269 		goto out;
2270 
2271 	pbuf = PNBUF_GET();
2272 	for (; i < fa->len; i++) {
2273 		fae = &fa->fae[i];
2274 		if (fae->fae_action != FAE_OPEN)
2275 			continue;
2276 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2277 		if (error)
2278 			goto out;
2279 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2280 		memcpy(fae->fae_path, pbuf, fal);
2281 	}
2282 	PNBUF_PUT(pbuf);
2283 
2284 	*fap = fa;
2285 	return 0;
2286 out:
2287 	if (pbuf)
2288 		PNBUF_PUT(pbuf);
2289 	posix_spawn_fa_free(fa, i);
2290 	return error;
2291 }
2292 
2293 int
2294 check_posix_spawn(struct lwp *l1)
2295 {
2296 	int error, tnprocs, count;
2297 	uid_t uid;
2298 	struct proc *p1;
2299 
2300 	p1 = l1->l_proc;
2301 	uid = kauth_cred_getuid(l1->l_cred);
2302 	tnprocs = atomic_inc_uint_nv(&nprocs);
2303 
2304 	/*
2305 	 * Although process entries are dynamically created, we still keep
2306 	 * a global limit on the maximum number we will create.
2307 	 */
2308 	if (__predict_false(tnprocs >= maxproc))
2309 		error = -1;
2310 	else
2311 		error = kauth_authorize_process(l1->l_cred,
2312 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2313 
2314 	if (error) {
2315 		atomic_dec_uint(&nprocs);
2316 		return EAGAIN;
2317 	}
2318 
2319 	/*
2320 	 * Enforce limits.
2321 	 */
2322 	count = chgproccnt(uid, 1);
2323 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2324 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2325 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2326 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2327 		(void)chgproccnt(uid, -1);
2328 		atomic_dec_uint(&nprocs);
2329 		return EAGAIN;
2330 	}
2331 
2332 	return 0;
2333 }
2334 
2335 int
2336 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2337 	struct posix_spawn_file_actions *fa,
2338 	struct posix_spawnattr *sa,
2339 	char *const *argv, char *const *envp,
2340 	execve_fetch_element_t fetch)
2341 {
2342 
2343 	struct proc *p1, *p2;
2344 	struct lwp *l2;
2345 	int error;
2346 	struct spawn_exec_data *spawn_data;
2347 	vaddr_t uaddr;
2348 	pid_t pid;
2349 	bool have_exec_lock = false;
2350 
2351 	p1 = l1->l_proc;
2352 
2353 	/* Allocate and init spawn_data */
2354 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2355 	spawn_data->sed_refcnt = 1; /* only parent so far */
2356 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2357 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2358 	mutex_enter(&spawn_data->sed_mtx_child);
2359 
2360 	/*
2361 	 * Do the first part of the exec now, collect state
2362 	 * in spawn_data.
2363 	 */
2364 	error = execve_loadvm(l1, path, argv,
2365 	    envp, fetch, &spawn_data->sed_exec);
2366 	if (error == EJUSTRETURN)
2367 		error = 0;
2368 	else if (error)
2369 		goto error_exit;
2370 
2371 	have_exec_lock = true;
2372 
2373 	/*
2374 	 * Allocate virtual address space for the U-area now, while it
2375 	 * is still easy to abort the fork operation if we're out of
2376 	 * kernel virtual address space.
2377 	 */
2378 	uaddr = uvm_uarea_alloc();
2379 	if (__predict_false(uaddr == 0)) {
2380 		error = ENOMEM;
2381 		goto error_exit;
2382 	}
2383 
2384 	/*
2385 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2386 	 * replace it with its own before returning to userland
2387 	 * in the child.
2388 	 * This is a point of no return, we will have to go through
2389 	 * the child proc to properly clean it up past this point.
2390 	 */
2391 	p2 = proc_alloc();
2392 	pid = p2->p_pid;
2393 
2394 	/*
2395 	 * Make a proc table entry for the new process.
2396 	 * Start by zeroing the section of proc that is zero-initialized,
2397 	 * then copy the section that is copied directly from the parent.
2398 	 */
2399 	memset(&p2->p_startzero, 0,
2400 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2401 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2402 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2403 	p2->p_vmspace = proc0.p_vmspace;
2404 
2405 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2406 
2407 	LIST_INIT(&p2->p_lwps);
2408 	LIST_INIT(&p2->p_sigwaiters);
2409 
2410 	/*
2411 	 * Duplicate sub-structures as needed.
2412 	 * Increase reference counts on shared objects.
2413 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2414 	 * handling are important in order to keep a consistent behaviour
2415 	 * for the child after the fork.  If we are a 32-bit process, the
2416 	 * child will be too.
2417 	 */
2418 	p2->p_flag =
2419 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2420 	p2->p_emul = p1->p_emul;
2421 	p2->p_execsw = p1->p_execsw;
2422 
2423 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2424 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2425 	rw_init(&p2->p_reflock);
2426 	cv_init(&p2->p_waitcv, "wait");
2427 	cv_init(&p2->p_lwpcv, "lwpwait");
2428 
2429 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2430 
2431 	kauth_proc_fork(p1, p2);
2432 
2433 	p2->p_raslist = NULL;
2434 	p2->p_fd = fd_copy();
2435 
2436 	/* XXX racy */
2437 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2438 
2439 	p2->p_cwdi = cwdinit();
2440 
2441 	/*
2442 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2443 	 * we just need increase pl_refcnt.
2444 	 */
2445 	if (!p1->p_limit->pl_writeable) {
2446 		lim_addref(p1->p_limit);
2447 		p2->p_limit = p1->p_limit;
2448 	} else {
2449 		p2->p_limit = lim_copy(p1->p_limit);
2450 	}
2451 
2452 	p2->p_lflag = 0;
2453 	p2->p_sflag = 0;
2454 	p2->p_slflag = 0;
2455 	p2->p_pptr = p1;
2456 	p2->p_ppid = p1->p_pid;
2457 	LIST_INIT(&p2->p_children);
2458 
2459 	p2->p_aio = NULL;
2460 
2461 #ifdef KTRACE
2462 	/*
2463 	 * Copy traceflag and tracefile if enabled.
2464 	 * If not inherited, these were zeroed above.
2465 	 */
2466 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2467 		mutex_enter(&ktrace_lock);
2468 		p2->p_traceflag = p1->p_traceflag;
2469 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2470 			ktradref(p2);
2471 		mutex_exit(&ktrace_lock);
2472 	}
2473 #endif
2474 
2475 	/*
2476 	 * Create signal actions for the child process.
2477 	 */
2478 	p2->p_sigacts = sigactsinit(p1, 0);
2479 	mutex_enter(p1->p_lock);
2480 	p2->p_sflag |=
2481 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2482 	sched_proc_fork(p1, p2);
2483 	mutex_exit(p1->p_lock);
2484 
2485 	p2->p_stflag = p1->p_stflag;
2486 
2487 	/*
2488 	 * p_stats.
2489 	 * Copy parts of p_stats, and zero out the rest.
2490 	 */
2491 	p2->p_stats = pstatscopy(p1->p_stats);
2492 
2493 	/* copy over machdep flags to the new proc */
2494 	cpu_proc_fork(p1, p2);
2495 
2496 	/*
2497 	 * Prepare remaining parts of spawn data
2498 	 */
2499 	spawn_data->sed_actions = fa;
2500 	spawn_data->sed_attrs = sa;
2501 
2502 	spawn_data->sed_parent = p1;
2503 
2504 	/* create LWP */
2505 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2506 	    &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2507 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2508 
2509 	/*
2510 	 * Copy the credential so other references don't see our changes.
2511 	 * Test to see if this is necessary first, since in the common case
2512 	 * we won't need a private reference.
2513 	 */
2514 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2515 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2516 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2517 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2518 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2519 	}
2520 
2521 	/* Update the master credentials. */
2522 	if (l2->l_cred != p2->p_cred) {
2523 		kauth_cred_t ocred;
2524 
2525 		kauth_cred_hold(l2->l_cred);
2526 		mutex_enter(p2->p_lock);
2527 		ocred = p2->p_cred;
2528 		p2->p_cred = l2->l_cred;
2529 		mutex_exit(p2->p_lock);
2530 		kauth_cred_free(ocred);
2531 	}
2532 
2533 	*child_ok = true;
2534 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2535 #if 0
2536 	l2->l_nopreempt = 1; /* start it non-preemptable */
2537 #endif
2538 
2539 	/*
2540 	 * It's now safe for the scheduler and other processes to see the
2541 	 * child process.
2542 	 */
2543 	mutex_enter(proc_lock);
2544 
2545 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2546 		p2->p_lflag |= PL_CONTROLT;
2547 
2548 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2549 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2550 
2551 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2552 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2553 
2554 	p2->p_trace_enabled = trace_is_enabled(p2);
2555 #ifdef __HAVE_SYSCALL_INTERN
2556 	(*p2->p_emul->e_syscall_intern)(p2);
2557 #endif
2558 
2559 	/*
2560 	 * Make child runnable, set start time, and add to run queue except
2561 	 * if the parent requested the child to start in SSTOP state.
2562 	 */
2563 	mutex_enter(p2->p_lock);
2564 
2565 	getmicrotime(&p2->p_stats->p_start);
2566 
2567 	lwp_lock(l2);
2568 	KASSERT(p2->p_nrlwps == 1);
2569 	p2->p_nrlwps = 1;
2570 	p2->p_stat = SACTIVE;
2571 	l2->l_stat = LSRUN;
2572 	sched_enqueue(l2, false);
2573 	lwp_unlock(l2);
2574 
2575 	mutex_exit(p2->p_lock);
2576 	mutex_exit(proc_lock);
2577 
2578 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2579 	error = spawn_data->sed_error;
2580 	mutex_exit(&spawn_data->sed_mtx_child);
2581 	spawn_exec_data_release(spawn_data);
2582 
2583 	rw_exit(&p1->p_reflock);
2584 	rw_exit(&exec_lock);
2585 	have_exec_lock = false;
2586 
2587 	*pid_res = pid;
2588 	return error;
2589 
2590  error_exit:
2591 	if (have_exec_lock) {
2592 		execve_free_data(&spawn_data->sed_exec);
2593 		rw_exit(&p1->p_reflock);
2594 		rw_exit(&exec_lock);
2595 	}
2596 	mutex_exit(&spawn_data->sed_mtx_child);
2597 	spawn_exec_data_release(spawn_data);
2598 
2599 	return error;
2600 }
2601 
2602 int
2603 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2604     register_t *retval)
2605 {
2606 	/* {
2607 		syscallarg(pid_t *) pid;
2608 		syscallarg(const char *) path;
2609 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2610 		syscallarg(const struct posix_spawnattr *) attrp;
2611 		syscallarg(char *const *) argv;
2612 		syscallarg(char *const *) envp;
2613 	} */
2614 
2615 	int error;
2616 	struct posix_spawn_file_actions *fa = NULL;
2617 	struct posix_spawnattr *sa = NULL;
2618 	pid_t pid;
2619 	bool child_ok = false;
2620 	rlim_t max_fileactions;
2621 	proc_t *p = l1->l_proc;
2622 
2623 	error = check_posix_spawn(l1);
2624 	if (error) {
2625 		*retval = error;
2626 		return 0;
2627 	}
2628 
2629 	/* copy in file_actions struct */
2630 	if (SCARG(uap, file_actions) != NULL) {
2631 		max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2632 		    maxfiles);
2633 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2634 		    max_fileactions);
2635 		if (error)
2636 			goto error_exit;
2637 	}
2638 
2639 	/* copyin posix_spawnattr struct */
2640 	if (SCARG(uap, attrp) != NULL) {
2641 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2642 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2643 		if (error)
2644 			goto error_exit;
2645 	}
2646 
2647 	/*
2648 	 * Do the spawn
2649 	 */
2650 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2651 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2652 	if (error)
2653 		goto error_exit;
2654 
2655 	if (error == 0 && SCARG(uap, pid) != NULL)
2656 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2657 
2658 	*retval = error;
2659 	return 0;
2660 
2661  error_exit:
2662 	if (!child_ok) {
2663 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2664 		atomic_dec_uint(&nprocs);
2665 
2666 		if (sa)
2667 			kmem_free(sa, sizeof(*sa));
2668 		if (fa)
2669 			posix_spawn_fa_free(fa, fa->len);
2670 	}
2671 
2672 	*retval = error;
2673 	return 0;
2674 }
2675 
2676 void
2677 exec_free_emul_arg(struct exec_package *epp)
2678 {
2679 	if (epp->ep_emul_arg_free != NULL) {
2680 		KASSERT(epp->ep_emul_arg != NULL);
2681 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2682 		epp->ep_emul_arg_free = NULL;
2683 		epp->ep_emul_arg = NULL;
2684 	} else {
2685 		KASSERT(epp->ep_emul_arg == NULL);
2686 	}
2687 }
2688 
2689 #ifdef DEBUG_EXEC
2690 static void
2691 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2692 {
2693 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2694 	size_t j;
2695 
2696 	if (error == 0)
2697 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2698 	else
2699 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2700 		    epp->ep_vmcmds.evs_used, error));
2701 
2702 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2703 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2704 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2705 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2706 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2707 		    "pagedvn" :
2708 		    vp[j].ev_proc == vmcmd_map_readvn ?
2709 		    "readvn" :
2710 		    vp[j].ev_proc == vmcmd_map_zero ?
2711 		    "zero" : "*unknown*",
2712 		    vp[j].ev_addr, vp[j].ev_len,
2713 		    vp[j].ev_offset, vp[j].ev_prot,
2714 		    vp[j].ev_flags));
2715 		if (error != 0 && j == x)
2716 			DPRINTF(("     ^--- failed\n"));
2717 	}
2718 }
2719 #endif
2720