xref: /netbsd-src/sys/kern/kern_exec.c (revision ecf6466c633518f478c293c388551b29e46729cc)
1 /*	$NetBSD: kern_exec.c,v 1.488 2020/01/12 22:03:22 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008, 2019 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
34  * Copyright (C) 1992 Wolfgang Solfrank.
35  * Copyright (C) 1992 TooLs GmbH.
36  * All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by TooLs GmbH.
49  * 4. The name of TooLs GmbH may not be used to endorse or promote products
50  *    derived from this software without specific prior written permission.
51  *
52  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
53  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
54  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
55  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
56  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
57  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
58  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
59  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
60  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
61  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
62  */
63 
64 #include <sys/cdefs.h>
65 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.488 2020/01/12 22:03:22 ad Exp $");
66 
67 #include "opt_exec.h"
68 #include "opt_execfmt.h"
69 #include "opt_ktrace.h"
70 #include "opt_modular.h"
71 #include "opt_syscall_debug.h"
72 #include "veriexec.h"
73 #include "opt_pax.h"
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/filedesc.h>
78 #include <sys/kernel.h>
79 #include <sys/proc.h>
80 #include <sys/ptrace.h>
81 #include <sys/mount.h>
82 #include <sys/kmem.h>
83 #include <sys/namei.h>
84 #include <sys/vnode.h>
85 #include <sys/file.h>
86 #include <sys/filedesc.h>
87 #include <sys/acct.h>
88 #include <sys/atomic.h>
89 #include <sys/exec.h>
90 #include <sys/ktrace.h>
91 #include <sys/uidinfo.h>
92 #include <sys/wait.h>
93 #include <sys/mman.h>
94 #include <sys/ras.h>
95 #include <sys/signalvar.h>
96 #include <sys/stat.h>
97 #include <sys/syscall.h>
98 #include <sys/kauth.h>
99 #include <sys/lwpctl.h>
100 #include <sys/pax.h>
101 #include <sys/cpu.h>
102 #include <sys/module.h>
103 #include <sys/syscallvar.h>
104 #include <sys/syscallargs.h>
105 #if NVERIEXEC > 0
106 #include <sys/verified_exec.h>
107 #endif /* NVERIEXEC > 0 */
108 #include <sys/sdt.h>
109 #include <sys/spawn.h>
110 #include <sys/prot.h>
111 #include <sys/cprng.h>
112 
113 #include <uvm/uvm_extern.h>
114 
115 #include <machine/reg.h>
116 
117 #include <compat/common/compat_util.h>
118 
119 #ifndef MD_TOPDOWN_INIT
120 #ifdef __USE_TOPDOWN_VM
121 #define	MD_TOPDOWN_INIT(epp)	(epp)->ep_flags |= EXEC_TOPDOWN_VM
122 #else
123 #define	MD_TOPDOWN_INIT(epp)
124 #endif
125 #endif
126 
127 struct execve_data;
128 
129 extern int user_va0_disable;
130 
131 static size_t calcargs(struct execve_data * restrict, const size_t);
132 static size_t calcstack(struct execve_data * restrict, const size_t);
133 static int copyoutargs(struct execve_data * restrict, struct lwp *,
134     char * const);
135 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
136 static int copyinargs(struct execve_data * restrict, char * const *,
137     char * const *, execve_fetch_element_t, char **);
138 static int copyinargstrs(struct execve_data * restrict, char * const *,
139     execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
140 static int exec_sigcode_map(struct proc *, const struct emul *);
141 
142 #if defined(DEBUG) && !defined(DEBUG_EXEC)
143 #define DEBUG_EXEC
144 #endif
145 #ifdef DEBUG_EXEC
146 #define DPRINTF(a) printf a
147 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
148     __LINE__, (s), (a), (b))
149 static void dump_vmcmds(const struct exec_package * const, size_t, int);
150 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
151 #else
152 #define DPRINTF(a)
153 #define COPYPRINTF(s, a, b)
154 #define DUMPVMCMDS(p, x, e) do {} while (0)
155 #endif /* DEBUG_EXEC */
156 
157 /*
158  * DTrace SDT provider definitions
159  */
160 SDT_PROVIDER_DECLARE(proc);
161 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
162 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
163 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
164 
165 /*
166  * Exec function switch:
167  *
168  * Note that each makecmds function is responsible for loading the
169  * exec package with the necessary functions for any exec-type-specific
170  * handling.
171  *
172  * Functions for specific exec types should be defined in their own
173  * header file.
174  */
175 static const struct execsw	**execsw = NULL;
176 static int			nexecs;
177 
178 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
179 
180 /* list of dynamically loaded execsw entries */
181 static LIST_HEAD(execlist_head, exec_entry) ex_head =
182     LIST_HEAD_INITIALIZER(ex_head);
183 struct exec_entry {
184 	LIST_ENTRY(exec_entry)	ex_list;
185 	SLIST_ENTRY(exec_entry)	ex_slist;
186 	const struct execsw	*ex_sw;
187 };
188 
189 #ifndef __HAVE_SYSCALL_INTERN
190 void	syscall(void);
191 #endif
192 
193 /* NetBSD autoloadable syscalls */
194 #ifdef MODULAR
195 #include <kern/syscalls_autoload.c>
196 #endif
197 
198 /* NetBSD emul struct */
199 struct emul emul_netbsd = {
200 	.e_name =		"netbsd",
201 #ifdef EMUL_NATIVEROOT
202 	.e_path =		EMUL_NATIVEROOT,
203 #else
204 	.e_path =		NULL,
205 #endif
206 #ifndef __HAVE_MINIMAL_EMUL
207 	.e_flags =		EMUL_HAS_SYS___syscall,
208 	.e_errno =		NULL,
209 	.e_nosys =		SYS_syscall,
210 	.e_nsysent =		SYS_NSYSENT,
211 #endif
212 #ifdef MODULAR
213 	.e_sc_autoload =	netbsd_syscalls_autoload,
214 #endif
215 	.e_sysent =		sysent,
216 	.e_nomodbits =		sysent_nomodbits,
217 #ifdef SYSCALL_DEBUG
218 	.e_syscallnames =	syscallnames,
219 #else
220 	.e_syscallnames =	NULL,
221 #endif
222 	.e_sendsig =		sendsig,
223 	.e_trapsignal =		trapsignal,
224 	.e_sigcode =		NULL,
225 	.e_esigcode =		NULL,
226 	.e_sigobject =		NULL,
227 	.e_setregs =		setregs,
228 	.e_proc_exec =		NULL,
229 	.e_proc_fork =		NULL,
230 	.e_proc_exit =		NULL,
231 	.e_lwp_fork =		NULL,
232 	.e_lwp_exit =		NULL,
233 #ifdef __HAVE_SYSCALL_INTERN
234 	.e_syscall_intern =	syscall_intern,
235 #else
236 	.e_syscall =		syscall,
237 #endif
238 	.e_sysctlovly =		NULL,
239 	.e_vm_default_addr =	uvm_default_mapaddr,
240 	.e_usertrap =		NULL,
241 	.e_ucsize =		sizeof(ucontext_t),
242 	.e_startlwp =		startlwp
243 };
244 
245 /*
246  * Exec lock. Used to control access to execsw[] structures.
247  * This must not be static so that netbsd32 can access it, too.
248  */
249 krwlock_t exec_lock;
250 
251 static kmutex_t sigobject_lock;
252 
253 /*
254  * Data used between a loadvm and execve part of an "exec" operation
255  */
256 struct execve_data {
257 	struct exec_package	ed_pack;
258 	struct pathbuf		*ed_pathbuf;
259 	struct vattr		ed_attr;
260 	struct ps_strings	ed_arginfo;
261 	char			*ed_argp;
262 	const char		*ed_pathstring;
263 	char			*ed_resolvedname;
264 	size_t			ed_ps_strings_sz;
265 	int			ed_szsigcode;
266 	size_t			ed_argslen;
267 	long			ed_argc;
268 	long			ed_envc;
269 };
270 
271 /*
272  * data passed from parent lwp to child during a posix_spawn()
273  */
274 struct spawn_exec_data {
275 	struct execve_data	sed_exec;
276 	struct posix_spawn_file_actions
277 				*sed_actions;
278 	struct posix_spawnattr	*sed_attrs;
279 	struct proc		*sed_parent;
280 	kcondvar_t		sed_cv_child_ready;
281 	kmutex_t		sed_mtx_child;
282 	int			sed_error;
283 	volatile uint32_t	sed_refcnt;
284 };
285 
286 static struct vm_map *exec_map;
287 static struct pool exec_pool;
288 
289 static void *
290 exec_pool_alloc(struct pool *pp, int flags)
291 {
292 
293 	return (void *)uvm_km_alloc(exec_map, NCARGS, 0,
294 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
295 }
296 
297 static void
298 exec_pool_free(struct pool *pp, void *addr)
299 {
300 
301 	uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
302 }
303 
304 static struct pool_allocator exec_palloc = {
305 	.pa_alloc = exec_pool_alloc,
306 	.pa_free = exec_pool_free,
307 	.pa_pagesz = NCARGS
308 };
309 
310 static void
311 exec_path_free(struct execve_data *data)
312 {
313 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
314 	pathbuf_destroy(data->ed_pathbuf);
315 	if (data->ed_resolvedname)
316 		PNBUF_PUT(data->ed_resolvedname);
317 }
318 
319 static void
320 exec_resolvename(struct lwp *l, struct exec_package *epp, struct vnode *vp,
321     char **rpath)
322 {
323 	int error;
324 	char *p;
325 
326 	KASSERT(rpath != NULL);
327 
328 	*rpath = PNBUF_GET();
329 	error = vnode_to_path(*rpath, MAXPATHLEN, vp, l, l->l_proc);
330 	if (error) {
331 		PNBUF_PUT(*rpath);
332 		*rpath = NULL;
333 		return;
334 	}
335 	epp->ep_resolvedname = *rpath;
336 	if ((p = strrchr(*rpath, '/')) != NULL)
337 		epp->ep_kname = p + 1;
338 }
339 
340 
341 /*
342  * check exec:
343  * given an "executable" described in the exec package's namei info,
344  * see what we can do with it.
345  *
346  * ON ENTRY:
347  *	exec package with appropriate namei info
348  *	lwp pointer of exec'ing lwp
349  *	NO SELF-LOCKED VNODES
350  *
351  * ON EXIT:
352  *	error:	nothing held, etc.  exec header still allocated.
353  *	ok:	filled exec package, executable's vnode (unlocked).
354  *
355  * EXEC SWITCH ENTRY:
356  * 	Locked vnode to check, exec package, proc.
357  *
358  * EXEC SWITCH EXIT:
359  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
360  *	error:	destructive:
361  *			everything deallocated execept exec header.
362  *		non-destructive:
363  *			error code, executable's vnode (unlocked),
364  *			exec header unmodified.
365  */
366 int
367 /*ARGSUSED*/
368 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb,
369     char **rpath)
370 {
371 	int		error, i;
372 	struct vnode	*vp;
373 	size_t		resid;
374 
375 	if (epp->ep_resolvedname) {
376 		struct nameidata nd;
377 
378 		// grab the absolute pathbuf here before namei() trashes it.
379 		pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX);
380 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
381 
382 		/* first get the vnode */
383 		if ((error = namei(&nd)) != 0)
384 			return error;
385 
386 		epp->ep_vp = vp = nd.ni_vp;
387 #ifdef DIAGNOSTIC
388 		/* paranoia (take this out once namei stuff stabilizes) */
389 		memset(nd.ni_pnbuf, '~', PATH_MAX);
390 #endif
391 	} else {
392 		struct file *fp;
393 
394 		if ((error = fd_getvnode(epp->ep_xfd, &fp)) != 0)
395 			return error;
396 		epp->ep_vp = vp = fp->f_vnode;
397 		vref(vp);
398 		fd_putfile(epp->ep_xfd);
399 		exec_resolvename(l, epp, vp, rpath);
400 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
401 	}
402 
403 	/* check access and type */
404 	if (vp->v_type != VREG) {
405 		error = EACCES;
406 		goto bad1;
407 	}
408 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
409 		goto bad1;
410 
411 	/* get attributes */
412 	/* XXX VOP_GETATTR is the only thing that needs LK_EXCLUSIVE here */
413 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
414 		goto bad1;
415 
416 	/* Check mount point */
417 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
418 		error = EACCES;
419 		goto bad1;
420 	}
421 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
422 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
423 
424 	/* try to open it */
425 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
426 		goto bad1;
427 
428 	/* now we have the file, get the exec header */
429 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
430 			UIO_SYSSPACE, IO_NODELOCKED, l->l_cred, &resid, NULL);
431 	if (error)
432 		goto bad1;
433 
434 	/* unlock vp, since we need it unlocked from here on out. */
435 	VOP_UNLOCK(vp);
436 
437 #if NVERIEXEC > 0
438 	error = veriexec_verify(l, vp,
439 	    epp->ep_resolvedname ? epp->ep_resolvedname : epp->ep_kname,
440 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
441 	    NULL);
442 	if (error)
443 		goto bad2;
444 #endif /* NVERIEXEC > 0 */
445 
446 #ifdef PAX_SEGVGUARD
447 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
448 	if (error)
449 		goto bad2;
450 #endif /* PAX_SEGVGUARD */
451 
452 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
453 
454 	/*
455 	 * Set up default address space limits.  Can be overridden
456 	 * by individual exec packages.
457 	 */
458 	epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
459 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
460 
461 	/*
462 	 * set up the vmcmds for creation of the process
463 	 * address space
464 	 */
465 	error = ENOEXEC;
466 	for (i = 0; i < nexecs; i++) {
467 		int newerror;
468 
469 		epp->ep_esch = execsw[i];
470 		newerror = (*execsw[i]->es_makecmds)(l, epp);
471 
472 		if (!newerror) {
473 			/* Seems ok: check that entry point is not too high */
474 			if (epp->ep_entry >= epp->ep_vm_maxaddr) {
475 #ifdef DIAGNOSTIC
476 				printf("%s: rejecting %p due to "
477 				    "too high entry address (>= %p)\n",
478 					 __func__, (void *)epp->ep_entry,
479 					 (void *)epp->ep_vm_maxaddr);
480 #endif
481 				error = ENOEXEC;
482 				break;
483 			}
484 			/* Seems ok: check that entry point is not too low */
485 			if (epp->ep_entry < epp->ep_vm_minaddr) {
486 #ifdef DIAGNOSTIC
487 				printf("%s: rejecting %p due to "
488 				    "too low entry address (< %p)\n",
489 				     __func__, (void *)epp->ep_entry,
490 				     (void *)epp->ep_vm_minaddr);
491 #endif
492 				error = ENOEXEC;
493 				break;
494 			}
495 
496 			/* check limits */
497 			if ((epp->ep_tsize > MAXTSIZ) ||
498 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
499 						    [RLIMIT_DATA].rlim_cur)) {
500 #ifdef DIAGNOSTIC
501 				printf("%s: rejecting due to "
502 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
503 				    __func__,
504 				    (unsigned long long)epp->ep_tsize,
505 				    (unsigned long long)MAXTSIZ,
506 				    (unsigned long long)epp->ep_dsize,
507 				    (unsigned long long)
508 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
509 #endif
510 				error = ENOMEM;
511 				break;
512 			}
513 			return 0;
514 		}
515 
516 		/*
517 		 * Reset all the fields that may have been modified by the
518 		 * loader.
519 		 */
520 		KASSERT(epp->ep_emul_arg == NULL);
521 		if (epp->ep_emul_root != NULL) {
522 			vrele(epp->ep_emul_root);
523 			epp->ep_emul_root = NULL;
524 		}
525 		if (epp->ep_interp != NULL) {
526 			vrele(epp->ep_interp);
527 			epp->ep_interp = NULL;
528 		}
529 		epp->ep_pax_flags = 0;
530 
531 		/* make sure the first "interesting" error code is saved. */
532 		if (error == ENOEXEC)
533 			error = newerror;
534 
535 		if (epp->ep_flags & EXEC_DESTR)
536 			/* Error from "#!" code, tidied up by recursive call */
537 			return error;
538 	}
539 
540 	/* not found, error */
541 
542 	/*
543 	 * free any vmspace-creation commands,
544 	 * and release their references
545 	 */
546 	kill_vmcmds(&epp->ep_vmcmds);
547 
548 #if NVERIEXEC > 0 || defined(PAX_SEGVGUARD)
549 bad2:
550 #endif
551 	/*
552 	 * close and release the vnode, restore the old one, free the
553 	 * pathname buf, and punt.
554 	 */
555 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
556 	VOP_CLOSE(vp, FREAD, l->l_cred);
557 	vput(vp);
558 	return error;
559 
560 bad1:
561 	/*
562 	 * free the namei pathname buffer, and put the vnode
563 	 * (which we don't yet have open).
564 	 */
565 	vput(vp);				/* was still locked */
566 	return error;
567 }
568 
569 #ifdef __MACHINE_STACK_GROWS_UP
570 #define STACK_PTHREADSPACE NBPG
571 #else
572 #define STACK_PTHREADSPACE 0
573 #endif
574 
575 static int
576 execve_fetch_element(char * const *array, size_t index, char **value)
577 {
578 	return copyin(array + index, value, sizeof(*value));
579 }
580 
581 /*
582  * exec system call
583  */
584 int
585 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
586 {
587 	/* {
588 		syscallarg(const char *)	path;
589 		syscallarg(char * const *)	argp;
590 		syscallarg(char * const *)	envp;
591 	} */
592 
593 	return execve1(l, true, SCARG(uap, path), -1, SCARG(uap, argp),
594 	    SCARG(uap, envp), execve_fetch_element);
595 }
596 
597 int
598 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
599     register_t *retval)
600 {
601 	/* {
602 		syscallarg(int)			fd;
603 		syscallarg(char * const *)	argp;
604 		syscallarg(char * const *)	envp;
605 	} */
606 
607 	return execve1(l, false, NULL, SCARG(uap, fd), SCARG(uap, argp),
608 	    SCARG(uap, envp), execve_fetch_element);
609 }
610 
611 /*
612  * Load modules to try and execute an image that we do not understand.
613  * If no execsw entries are present, we load those likely to be needed
614  * in order to run native images only.  Otherwise, we autoload all
615  * possible modules that could let us run the binary.  XXX lame
616  */
617 static void
618 exec_autoload(void)
619 {
620 #ifdef MODULAR
621 	static const char * const native[] = {
622 		"exec_elf32",
623 		"exec_elf64",
624 		"exec_script",
625 		NULL
626 	};
627 	static const char * const compat[] = {
628 		"exec_elf32",
629 		"exec_elf64",
630 		"exec_script",
631 		"exec_aout",
632 		"exec_coff",
633 		"exec_ecoff",
634 		"compat_aoutm68k",
635 		"compat_netbsd32",
636 		"compat_sunos",
637 		"compat_sunos32",
638 		"compat_ultrix",
639 		NULL
640 	};
641 	char const * const *list;
642 	int i;
643 
644 	list = (nexecs == 0 ? native : compat);
645 	for (i = 0; list[i] != NULL; i++) {
646 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
647 			continue;
648 		}
649 		yield();
650 	}
651 #endif
652 }
653 
654 /*
655  * Copy the user or kernel supplied upath to the allocated pathbuffer pbp
656  * making it absolute in the process, by prepending the current working
657  * directory if it is not. If offs is supplied it will contain the offset
658  * where the original supplied copy of upath starts.
659  */
660 int
661 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg,
662     struct pathbuf **pbp, size_t *offs)
663 {
664 	char *path, *bp;
665 	size_t len, tlen;
666 	int error;
667 	struct cwdinfo *cwdi;
668 
669 	path = PNBUF_GET();
670 	if (seg == UIO_SYSSPACE) {
671 		error = copystr(upath, path, MAXPATHLEN, &len);
672 	} else {
673 		error = copyinstr(upath, path, MAXPATHLEN, &len);
674 	}
675 	if (error)
676 		goto err;
677 
678 	if (path[0] == '/') {
679 		if (offs)
680 			*offs = 0;
681 		goto out;
682 	}
683 
684 	len++;
685 	if (len + 1 >= MAXPATHLEN) {
686 		error = ENAMETOOLONG;
687 		goto err;
688 	}
689 	bp = path + MAXPATHLEN - len;
690 	memmove(bp, path, len);
691 	*(--bp) = '/';
692 
693 	cwdi = l->l_proc->p_cwdi;
694 	rw_enter(&cwdi->cwdi_lock, RW_READER);
695 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
696 	    GETCWD_CHECK_ACCESS, l);
697 	rw_exit(&cwdi->cwdi_lock);
698 
699 	if (error)
700 		goto err;
701 	tlen = path + MAXPATHLEN - bp;
702 
703 	memmove(path, bp, tlen);
704 	path[tlen - 1] = '\0';
705 	if (offs)
706 		*offs = tlen - len;
707 out:
708 	*pbp = pathbuf_assimilate(path);
709 	return 0;
710 err:
711 	PNBUF_PUT(path);
712 	return error;
713 }
714 
715 vaddr_t
716 exec_vm_minaddr(vaddr_t va_min)
717 {
718 	/*
719 	 * Increase va_min if we don't want NULL to be mappable by the
720 	 * process.
721 	 */
722 #define VM_MIN_GUARD	PAGE_SIZE
723 	if (user_va0_disable && (va_min < VM_MIN_GUARD))
724 		return VM_MIN_GUARD;
725 	return va_min;
726 }
727 
728 static int
729 execve_loadvm(struct lwp *l, bool has_path, const char *path, int fd,
730 	char * const *args, char * const *envs,
731 	execve_fetch_element_t fetch_element,
732 	struct execve_data * restrict data)
733 {
734 	struct exec_package	* const epp = &data->ed_pack;
735 	int			error;
736 	struct proc		*p;
737 	char			*dp;
738 	u_int			modgen;
739 
740 	KASSERT(data != NULL);
741 
742 	p = l->l_proc;
743 	modgen = 0;
744 
745 	SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
746 
747 	/*
748 	 * Check if we have exceeded our number of processes limit.
749 	 * This is so that we handle the case where a root daemon
750 	 * forked, ran setuid to become the desired user and is trying
751 	 * to exec. The obvious place to do the reference counting check
752 	 * is setuid(), but we don't do the reference counting check there
753 	 * like other OS's do because then all the programs that use setuid()
754 	 * must be modified to check the return code of setuid() and exit().
755 	 * It is dangerous to make setuid() fail, because it fails open and
756 	 * the program will continue to run as root. If we make it succeed
757 	 * and return an error code, again we are not enforcing the limit.
758 	 * The best place to enforce the limit is here, when the process tries
759 	 * to execute a new image, because eventually the process will need
760 	 * to call exec in order to do something useful.
761 	 */
762  retry:
763 	if (p->p_flag & PK_SUGID) {
764 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
765 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
766 		     &p->p_rlimit[RLIMIT_NPROC],
767 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
768 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
769 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
770 		return EAGAIN;
771 	}
772 
773 	/*
774 	 * Drain existing references and forbid new ones.  The process
775 	 * should be left alone until we're done here.  This is necessary
776 	 * to avoid race conditions - e.g. in ptrace() - that might allow
777 	 * a local user to illicitly obtain elevated privileges.
778 	 */
779 	rw_enter(&p->p_reflock, RW_WRITER);
780 
781 	if (has_path) {
782 		size_t	offs;
783 		/*
784 		 * Init the namei data to point the file user's program name.
785 		 * This is done here rather than in check_exec(), so that it's
786 		 * possible to override this settings if any of makecmd/probe
787 		 * functions call check_exec() recursively - for example,
788 		 * see exec_script_makecmds().
789 		 */
790 		if ((error = exec_makepathbuf(l, path, UIO_USERSPACE,
791 		    &data->ed_pathbuf, &offs)) != 0)
792 			goto clrflg;
793 		data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
794 		epp->ep_kname = data->ed_pathstring + offs;
795 		data->ed_resolvedname = PNBUF_GET();
796 		epp->ep_resolvedname = data->ed_resolvedname;
797 		epp->ep_xfd = -1;
798 	} else {
799 		data->ed_pathbuf = pathbuf_assimilate(strcpy(PNBUF_GET(), "/"));
800 		data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
801 		epp->ep_kname = "*fexecve*";
802 		data->ed_resolvedname = NULL;
803 		epp->ep_resolvedname = NULL;
804 		epp->ep_xfd = fd;
805 	}
806 
807 
808 	/*
809 	 * initialize the fields of the exec package.
810 	 */
811 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
812 	epp->ep_hdrlen = exec_maxhdrsz;
813 	epp->ep_hdrvalid = 0;
814 	epp->ep_emul_arg = NULL;
815 	epp->ep_emul_arg_free = NULL;
816 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
817 	epp->ep_vap = &data->ed_attr;
818 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
819 	MD_TOPDOWN_INIT(epp);
820 	epp->ep_emul_root = NULL;
821 	epp->ep_interp = NULL;
822 	epp->ep_esch = NULL;
823 	epp->ep_pax_flags = 0;
824 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
825 
826 	rw_enter(&exec_lock, RW_READER);
827 
828 	/* see if we can run it. */
829 	if ((error = check_exec(l, epp, data->ed_pathbuf,
830 	    &data->ed_resolvedname)) != 0) {
831 		if (error != ENOENT && error != EACCES && error != ENOEXEC) {
832 			DPRINTF(("%s: check exec failed for %s, error %d\n",
833 			    __func__, epp->ep_kname, error));
834 		}
835 		goto freehdr;
836 	}
837 
838 	/* allocate an argument buffer */
839 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
840 	KASSERT(data->ed_argp != NULL);
841 	dp = data->ed_argp;
842 
843 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
844 		goto bad;
845 	}
846 
847 	/*
848 	 * Calculate the new stack size.
849 	 */
850 
851 #ifdef __MACHINE_STACK_GROWS_UP
852 /*
853  * copyargs() fills argc/argv/envp from the lower address even on
854  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
855  * so that _rtld() use it.
856  */
857 #define	RTLD_GAP	32
858 #else
859 #define	RTLD_GAP	0
860 #endif
861 
862 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
863 
864 	data->ed_argslen = calcargs(data, argenvstrlen);
865 
866 	const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
867 
868 	if (len > epp->ep_ssize) {
869 		/* in effect, compare to initial limit */
870 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
871 		error = ENOMEM;
872 		goto bad;
873 	}
874 	/* adjust "active stack depth" for process VSZ */
875 	epp->ep_ssize = len;
876 
877 	return 0;
878 
879  bad:
880 	/* free the vmspace-creation commands, and release their references */
881 	kill_vmcmds(&epp->ep_vmcmds);
882 	/* kill any opened file descriptor, if necessary */
883 	if (epp->ep_flags & EXEC_HASFD) {
884 		epp->ep_flags &= ~EXEC_HASFD;
885 		fd_close(epp->ep_fd);
886 	}
887 	/* close and put the exec'd file */
888 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
889 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
890 	vput(epp->ep_vp);
891 	pool_put(&exec_pool, data->ed_argp);
892 
893  freehdr:
894 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
895 	if (epp->ep_emul_root != NULL)
896 		vrele(epp->ep_emul_root);
897 	if (epp->ep_interp != NULL)
898 		vrele(epp->ep_interp);
899 
900 	rw_exit(&exec_lock);
901 
902 	exec_path_free(data);
903 
904  clrflg:
905 	rw_exit(&p->p_reflock);
906 
907 	if (modgen != module_gen && error == ENOEXEC) {
908 		modgen = module_gen;
909 		exec_autoload();
910 		goto retry;
911 	}
912 
913 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
914 	return error;
915 }
916 
917 static int
918 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
919 {
920 	struct exec_package	* const epp = &data->ed_pack;
921 	struct proc		*p = l->l_proc;
922 	struct exec_vmcmd	*base_vcp;
923 	int			error = 0;
924 	size_t			i;
925 
926 	/* record proc's vnode, for use by procfs and others */
927 	if (p->p_textvp)
928 		vrele(p->p_textvp);
929 	vref(epp->ep_vp);
930 	p->p_textvp = epp->ep_vp;
931 
932 	/* create the new process's VM space by running the vmcmds */
933 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
934 
935 #ifdef TRACE_EXEC
936 	DUMPVMCMDS(epp, 0, 0);
937 #endif
938 
939 	base_vcp = NULL;
940 
941 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
942 		struct exec_vmcmd *vcp;
943 
944 		vcp = &epp->ep_vmcmds.evs_cmds[i];
945 		if (vcp->ev_flags & VMCMD_RELATIVE) {
946 			KASSERTMSG(base_vcp != NULL,
947 			    "%s: relative vmcmd with no base", __func__);
948 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
949 			    "%s: illegal base & relative vmcmd", __func__);
950 			vcp->ev_addr += base_vcp->ev_addr;
951 		}
952 		error = (*vcp->ev_proc)(l, vcp);
953 		if (error)
954 			DUMPVMCMDS(epp, i, error);
955 		if (vcp->ev_flags & VMCMD_BASE)
956 			base_vcp = vcp;
957 	}
958 
959 	/* free the vmspace-creation commands, and release their references */
960 	kill_vmcmds(&epp->ep_vmcmds);
961 
962 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
963 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
964 	vput(epp->ep_vp);
965 
966 	/* if an error happened, deallocate and punt */
967 	if (error != 0) {
968 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
969 	}
970 	return error;
971 }
972 
973 static void
974 execve_free_data(struct execve_data *data)
975 {
976 	struct exec_package	* const epp = &data->ed_pack;
977 
978 	/* free the vmspace-creation commands, and release their references */
979 	kill_vmcmds(&epp->ep_vmcmds);
980 	/* kill any opened file descriptor, if necessary */
981 	if (epp->ep_flags & EXEC_HASFD) {
982 		epp->ep_flags &= ~EXEC_HASFD;
983 		fd_close(epp->ep_fd);
984 	}
985 
986 	/* close and put the exec'd file */
987 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
988 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
989 	vput(epp->ep_vp);
990 	pool_put(&exec_pool, data->ed_argp);
991 
992 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
993 	if (epp->ep_emul_root != NULL)
994 		vrele(epp->ep_emul_root);
995 	if (epp->ep_interp != NULL)
996 		vrele(epp->ep_interp);
997 
998 	exec_path_free(data);
999 }
1000 
1001 static void
1002 pathexec(struct proc *p, const char *resolvedname)
1003 {
1004 	/* set command name & other accounting info */
1005 	const char *cmdname;
1006 
1007 	if (resolvedname == NULL) {
1008 		cmdname = "*fexecve*";
1009 		resolvedname = "/";
1010 	} else {
1011 		cmdname = strrchr(resolvedname, '/') + 1;
1012 	}
1013 	KASSERTMSG(resolvedname[0] == '/', "bad resolvedname `%s'",
1014 	    resolvedname);
1015 
1016 	strlcpy(p->p_comm, cmdname, sizeof(p->p_comm));
1017 
1018 	kmem_strfree(p->p_path);
1019 	p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
1020 }
1021 
1022 /* XXX elsewhere */
1023 static int
1024 credexec(struct lwp *l, struct vattr *attr)
1025 {
1026 	struct proc *p = l->l_proc;
1027 	int error;
1028 
1029 	/*
1030 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
1031 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
1032 	 * out additional references on the process for the moment.
1033 	 */
1034 	if ((p->p_slflag & PSL_TRACED) == 0 &&
1035 
1036 	    (((attr->va_mode & S_ISUID) != 0 &&
1037 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
1038 
1039 	     ((attr->va_mode & S_ISGID) != 0 &&
1040 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
1041 		/*
1042 		 * Mark the process as SUGID before we do
1043 		 * anything that might block.
1044 		 */
1045 		proc_crmod_enter();
1046 		proc_crmod_leave(NULL, NULL, true);
1047 
1048 		/* Make sure file descriptors 0..2 are in use. */
1049 		if ((error = fd_checkstd()) != 0) {
1050 			DPRINTF(("%s: fdcheckstd failed %d\n",
1051 			    __func__, error));
1052 			return error;
1053 		}
1054 
1055 		/*
1056 		 * Copy the credential so other references don't see our
1057 		 * changes.
1058 		 */
1059 		l->l_cred = kauth_cred_copy(l->l_cred);
1060 #ifdef KTRACE
1061 		/*
1062 		 * If the persistent trace flag isn't set, turn off.
1063 		 */
1064 		if (p->p_tracep) {
1065 			mutex_enter(&ktrace_lock);
1066 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1067 				ktrderef(p);
1068 			mutex_exit(&ktrace_lock);
1069 		}
1070 #endif
1071 		if (attr->va_mode & S_ISUID)
1072 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
1073 		if (attr->va_mode & S_ISGID)
1074 			kauth_cred_setegid(l->l_cred, attr->va_gid);
1075 	} else {
1076 		if (kauth_cred_geteuid(l->l_cred) ==
1077 		    kauth_cred_getuid(l->l_cred) &&
1078 		    kauth_cred_getegid(l->l_cred) ==
1079 		    kauth_cred_getgid(l->l_cred))
1080 			p->p_flag &= ~PK_SUGID;
1081 	}
1082 
1083 	/*
1084 	 * Copy the credential so other references don't see our changes.
1085 	 * Test to see if this is necessary first, since in the common case
1086 	 * we won't need a private reference.
1087 	 */
1088 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1089 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1090 		l->l_cred = kauth_cred_copy(l->l_cred);
1091 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1092 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1093 	}
1094 
1095 	/* Update the master credentials. */
1096 	if (l->l_cred != p->p_cred) {
1097 		kauth_cred_t ocred;
1098 
1099 		kauth_cred_hold(l->l_cred);
1100 		mutex_enter(p->p_lock);
1101 		ocred = p->p_cred;
1102 		p->p_cred = l->l_cred;
1103 		mutex_exit(p->p_lock);
1104 		kauth_cred_free(ocred);
1105 	}
1106 
1107 	return 0;
1108 }
1109 
1110 static void
1111 emulexec(struct lwp *l, struct exec_package *epp)
1112 {
1113 	struct proc		*p = l->l_proc;
1114 
1115 	/* The emulation root will usually have been found when we looked
1116 	 * for the elf interpreter (or similar), if not look now. */
1117 	if (epp->ep_esch->es_emul->e_path != NULL &&
1118 	    epp->ep_emul_root == NULL)
1119 		emul_find_root(l, epp);
1120 
1121 	/* Any old emulation root got removed by fdcloseexec */
1122 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1123 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1124 	rw_exit(&p->p_cwdi->cwdi_lock);
1125 	epp->ep_emul_root = NULL;
1126 	if (epp->ep_interp != NULL)
1127 		vrele(epp->ep_interp);
1128 
1129 	/*
1130 	 * Call emulation specific exec hook. This can setup per-process
1131 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1132 	 *
1133 	 * If we are executing process of different emulation than the
1134 	 * original forked process, call e_proc_exit() of the old emulation
1135 	 * first, then e_proc_exec() of new emulation. If the emulation is
1136 	 * same, the exec hook code should deallocate any old emulation
1137 	 * resources held previously by this process.
1138 	 */
1139 	if (p->p_emul && p->p_emul->e_proc_exit
1140 	    && p->p_emul != epp->ep_esch->es_emul)
1141 		(*p->p_emul->e_proc_exit)(p);
1142 
1143 	/*
1144 	 * This is now LWP 1.
1145 	 */
1146 	/* XXX elsewhere */
1147 	mutex_enter(p->p_lock);
1148 	p->p_nlwpid = 1;
1149 	l->l_lid = 1;
1150 	mutex_exit(p->p_lock);
1151 
1152 	/*
1153 	 * Call exec hook. Emulation code may NOT store reference to anything
1154 	 * from &pack.
1155 	 */
1156 	if (epp->ep_esch->es_emul->e_proc_exec)
1157 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1158 
1159 	/* update p_emul, the old value is no longer needed */
1160 	p->p_emul = epp->ep_esch->es_emul;
1161 
1162 	/* ...and the same for p_execsw */
1163 	p->p_execsw = epp->ep_esch;
1164 
1165 #ifdef __HAVE_SYSCALL_INTERN
1166 	(*p->p_emul->e_syscall_intern)(p);
1167 #endif
1168 	ktremul();
1169 }
1170 
1171 static int
1172 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1173 	bool no_local_exec_lock, bool is_spawn)
1174 {
1175 	struct exec_package	* const epp = &data->ed_pack;
1176 	int error = 0;
1177 	struct proc		*p;
1178 	struct vmspace		*vm;
1179 
1180 	/*
1181 	 * In case of a posix_spawn operation, the child doing the exec
1182 	 * might not hold the reader lock on exec_lock, but the parent
1183 	 * will do this instead.
1184 	 */
1185 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1186 	KASSERT(!no_local_exec_lock || is_spawn);
1187 	KASSERT(data != NULL);
1188 
1189 	p = l->l_proc;
1190 
1191 	/* Get rid of other LWPs. */
1192 	if (p->p_nlwps > 1) {
1193 		mutex_enter(p->p_lock);
1194 		exit_lwps(l);
1195 		mutex_exit(p->p_lock);
1196 	}
1197 	KDASSERT(p->p_nlwps == 1);
1198 
1199 	/* Destroy any lwpctl info. */
1200 	if (p->p_lwpctl != NULL)
1201 		lwp_ctl_exit();
1202 
1203 	/* Remove POSIX timers */
1204 	timers_free(p, TIMERS_POSIX);
1205 
1206 	/* Set the PaX flags. */
1207 	pax_set_flags(epp, p);
1208 
1209 	/*
1210 	 * Do whatever is necessary to prepare the address space
1211 	 * for remapping.  Note that this might replace the current
1212 	 * vmspace with another!
1213 	 *
1214 	 * vfork(): do not touch any user space data in the new child
1215 	 * until we have awoken the parent below, or it will defeat
1216 	 * lazy pmap switching (on x86).
1217 	 */
1218 	if (is_spawn)
1219 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1220 		    epp->ep_vm_maxaddr,
1221 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1222 	else
1223 		uvmspace_exec(l, epp->ep_vm_minaddr,
1224 		    epp->ep_vm_maxaddr,
1225 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1226 	vm = p->p_vmspace;
1227 
1228 	vm->vm_taddr = (void *)epp->ep_taddr;
1229 	vm->vm_tsize = btoc(epp->ep_tsize);
1230 	vm->vm_daddr = (void*)epp->ep_daddr;
1231 	vm->vm_dsize = btoc(epp->ep_dsize);
1232 	vm->vm_ssize = btoc(epp->ep_ssize);
1233 	vm->vm_issize = 0;
1234 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1235 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1236 
1237 	pax_aslr_init_vm(l, vm, epp);
1238 
1239 	cwdexec(p);
1240 	fd_closeexec();		/* handle close on exec */
1241 
1242 	if (__predict_false(ktrace_on))
1243 		fd_ktrexecfd();
1244 
1245 	execsigs(p);		/* reset caught signals */
1246 
1247 	mutex_enter(p->p_lock);
1248 	l->l_ctxlink = NULL;	/* reset ucontext link */
1249 	p->p_acflag &= ~AFORK;
1250 	p->p_flag |= PK_EXEC;
1251 	mutex_exit(p->p_lock);
1252 
1253 	error = credexec(l, &data->ed_attr);
1254 	if (error)
1255 		goto exec_abort;
1256 
1257 #if defined(__HAVE_RAS)
1258 	/*
1259 	 * Remove all RASs from the address space.
1260 	 */
1261 	ras_purgeall();
1262 #endif
1263 
1264 	/*
1265 	 * Stop profiling.
1266 	 */
1267 	if ((p->p_stflag & PST_PROFIL) != 0) {
1268 		mutex_spin_enter(&p->p_stmutex);
1269 		stopprofclock(p);
1270 		mutex_spin_exit(&p->p_stmutex);
1271 	}
1272 
1273 	/*
1274 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1275 	 * exited and exec()/exit() are the only places it will be cleared.
1276 	 *
1277 	 * Once the parent has been awoken, curlwp may teleport to a new CPU
1278 	 * in sched_vforkexec(), and it's then OK to start messing with user
1279 	 * data.  See comment above.
1280 	 */
1281 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1282 		bool samecpu;
1283 		lwp_t *lp;
1284 
1285 		mutex_enter(proc_lock);
1286 		lp = p->p_vforklwp;
1287 		p->p_vforklwp = NULL;
1288 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1289 		cv_broadcast(&lp->l_waitcv);
1290 
1291 		/* Clear flags after cv_broadcast() (scheduler needs them). */
1292 		p->p_lflag &= ~PL_PPWAIT;
1293 		lp->l_vforkwaiting = false;
1294 
1295 		/* If parent is still on same CPU, teleport curlwp elsewhere. */
1296 		samecpu = (lp->l_cpu == curlwp->l_cpu);
1297 		mutex_exit(proc_lock);
1298 
1299 		/* Give the parent its CPU back - find a new home. */
1300 		KASSERT(!is_spawn);
1301 		sched_vforkexec(l, samecpu);
1302 	}
1303 
1304 	/* Now map address space. */
1305 	error = execve_dovmcmds(l, data);
1306 	if (error != 0)
1307 		goto exec_abort;
1308 
1309 	pathexec(p, epp->ep_resolvedname);
1310 
1311 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1312 
1313 	error = copyoutargs(data, l, newstack);
1314 	if (error != 0)
1315 		goto exec_abort;
1316 
1317 	doexechooks(p);
1318 
1319 	/*
1320 	 * Set initial SP at the top of the stack.
1321 	 *
1322 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1323 	 * the end of arg/env strings.  Userland guesses the address of argc
1324 	 * via ps_strings::ps_argvstr.
1325 	 */
1326 
1327 	/* Setup new registers and do misc. setup. */
1328 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1329 	if (epp->ep_esch->es_setregs)
1330 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1331 
1332 	/* Provide a consistent LWP private setting */
1333 	(void)lwp_setprivate(l, NULL);
1334 
1335 	/* Discard all PCU state; need to start fresh */
1336 	pcu_discard_all(l);
1337 
1338 	/* map the process's signal trampoline code */
1339 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1340 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1341 		goto exec_abort;
1342 	}
1343 
1344 	pool_put(&exec_pool, data->ed_argp);
1345 
1346 	/* notify others that we exec'd */
1347 	KNOTE(&p->p_klist, NOTE_EXEC);
1348 
1349 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1350 
1351 	SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1352 
1353 	emulexec(l, epp);
1354 
1355 	/* Allow new references from the debugger/procfs. */
1356 	rw_exit(&p->p_reflock);
1357 	if (!no_local_exec_lock)
1358 		rw_exit(&exec_lock);
1359 
1360 	mutex_enter(proc_lock);
1361 
1362 	/* posix_spawn(3) reports a single event with implied exec(3) */
1363 	if ((p->p_slflag & PSL_TRACED) && !is_spawn) {
1364 		mutex_enter(p->p_lock);
1365 		eventswitch(TRAP_EXEC, 0, 0);
1366 		mutex_enter(proc_lock);
1367 	}
1368 
1369 	if (p->p_sflag & PS_STOPEXEC) {
1370 		ksiginfoq_t kq;
1371 
1372 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1373 		p->p_pptr->p_nstopchild++;
1374 		p->p_waited = 0;
1375 		mutex_enter(p->p_lock);
1376 		ksiginfo_queue_init(&kq);
1377 		sigclearall(p, &contsigmask, &kq);
1378 		lwp_lock(l);
1379 		l->l_stat = LSSTOP;
1380 		p->p_stat = SSTOP;
1381 		p->p_nrlwps--;
1382 		lwp_unlock(l);
1383 		mutex_exit(p->p_lock);
1384 		mutex_exit(proc_lock);
1385 		lwp_lock(l);
1386 		spc_lock(l->l_cpu);
1387 		mi_switch(l);
1388 		ksiginfo_queue_drain(&kq);
1389 	} else {
1390 		mutex_exit(proc_lock);
1391 	}
1392 
1393 	exec_path_free(data);
1394 #ifdef TRACE_EXEC
1395 	DPRINTF(("%s finished\n", __func__));
1396 #endif
1397 	return EJUSTRETURN;
1398 
1399  exec_abort:
1400 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1401 	rw_exit(&p->p_reflock);
1402 	if (!no_local_exec_lock)
1403 		rw_exit(&exec_lock);
1404 
1405 	exec_path_free(data);
1406 
1407 	/*
1408 	 * the old process doesn't exist anymore.  exit gracefully.
1409 	 * get rid of the (new) address space we have created, if any, get rid
1410 	 * of our namei data and vnode, and exit noting failure
1411 	 */
1412 	if (vm != NULL) {
1413 		uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1414 			VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1415 	}
1416 
1417 	exec_free_emul_arg(epp);
1418 	pool_put(&exec_pool, data->ed_argp);
1419 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1420 	if (epp->ep_emul_root != NULL)
1421 		vrele(epp->ep_emul_root);
1422 	if (epp->ep_interp != NULL)
1423 		vrele(epp->ep_interp);
1424 
1425 	/* Acquire the sched-state mutex (exit1() will release it). */
1426 	if (!is_spawn) {
1427 		mutex_enter(p->p_lock);
1428 		exit1(l, error, SIGABRT);
1429 	}
1430 
1431 	return error;
1432 }
1433 
1434 int
1435 execve1(struct lwp *l, bool has_path, const char *path, int fd,
1436     char * const *args, char * const *envs,
1437     execve_fetch_element_t fetch_element)
1438 {
1439 	struct execve_data data;
1440 	int error;
1441 
1442 	error = execve_loadvm(l, has_path, path, fd, args, envs, fetch_element,
1443 	    &data);
1444 	if (error)
1445 		return error;
1446 	error = execve_runproc(l, &data, false, false);
1447 	return error;
1448 }
1449 
1450 static size_t
1451 fromptrsz(const struct exec_package *epp)
1452 {
1453 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1454 }
1455 
1456 static size_t
1457 ptrsz(const struct exec_package *epp)
1458 {
1459 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1460 }
1461 
1462 static size_t
1463 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1464 {
1465 	struct exec_package	* const epp = &data->ed_pack;
1466 
1467 	const size_t nargenvptrs =
1468 	    1 +				/* long argc */
1469 	    data->ed_argc +		/* char *argv[] */
1470 	    1 +				/* \0 */
1471 	    data->ed_envc +		/* char *env[] */
1472 	    1;				/* \0 */
1473 
1474 	return (nargenvptrs * ptrsz(epp))	/* pointers */
1475 	    + argenvstrlen			/* strings */
1476 	    + epp->ep_esch->es_arglen;		/* auxinfo */
1477 }
1478 
1479 static size_t
1480 calcstack(struct execve_data * restrict data, const size_t gaplen)
1481 {
1482 	struct exec_package	* const epp = &data->ed_pack;
1483 
1484 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1485 	    epp->ep_esch->es_emul->e_sigcode;
1486 
1487 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1488 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1489 
1490 	const size_t sigcode_psstr_sz =
1491 	    data->ed_szsigcode +	/* sigcode */
1492 	    data->ed_ps_strings_sz +	/* ps_strings */
1493 	    STACK_PTHREADSPACE;		/* pthread space */
1494 
1495 	const size_t stacklen =
1496 	    data->ed_argslen +
1497 	    gaplen +
1498 	    sigcode_psstr_sz;
1499 
1500 	/* make the stack "safely" aligned */
1501 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1502 }
1503 
1504 static int
1505 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1506     char * const newstack)
1507 {
1508 	struct exec_package	* const epp = &data->ed_pack;
1509 	struct proc		*p = l->l_proc;
1510 	int			error;
1511 
1512 	memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo));
1513 
1514 	/* remember information about the process */
1515 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1516 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1517 
1518 	/*
1519 	 * Allocate the stack address passed to the newly execve()'ed process.
1520 	 *
1521 	 * The new stack address will be set to the SP (stack pointer) register
1522 	 * in setregs().
1523 	 */
1524 
1525 	char *newargs = STACK_ALLOC(
1526 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1527 
1528 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1529 	    &data->ed_arginfo, &newargs, data->ed_argp);
1530 
1531 	if (error) {
1532 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1533 		return error;
1534 	}
1535 
1536 	error = copyoutpsstrs(data, p);
1537 	if (error != 0)
1538 		return error;
1539 
1540 	return 0;
1541 }
1542 
1543 static int
1544 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1545 {
1546 	struct exec_package	* const epp = &data->ed_pack;
1547 	struct ps_strings32	arginfo32;
1548 	void			*aip;
1549 	int			error;
1550 
1551 	/* fill process ps_strings info */
1552 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1553 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1554 
1555 	if (epp->ep_flags & EXEC_32) {
1556 		aip = &arginfo32;
1557 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1558 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1559 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1560 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1561 	} else
1562 		aip = &data->ed_arginfo;
1563 
1564 	/* copy out the process's ps_strings structure */
1565 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1566 	    != 0) {
1567 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1568 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1569 		return error;
1570 	}
1571 
1572 	return 0;
1573 }
1574 
1575 static int
1576 copyinargs(struct execve_data * restrict data, char * const *args,
1577     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1578 {
1579 	struct exec_package	* const epp = &data->ed_pack;
1580 	char			*dp;
1581 	size_t			i;
1582 	int			error;
1583 
1584 	dp = *dpp;
1585 
1586 	data->ed_argc = 0;
1587 
1588 	/* copy the fake args list, if there's one, freeing it as we go */
1589 	if (epp->ep_flags & EXEC_HASARGL) {
1590 		struct exec_fakearg	*fa = epp->ep_fa;
1591 
1592 		while (fa->fa_arg != NULL) {
1593 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1594 			size_t len;
1595 
1596 			len = strlcpy(dp, fa->fa_arg, maxlen);
1597 			/* Count NUL into len. */
1598 			if (len < maxlen)
1599 				len++;
1600 			else {
1601 				while (fa->fa_arg != NULL) {
1602 					kmem_free(fa->fa_arg, fa->fa_len);
1603 					fa++;
1604 				}
1605 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1606 				epp->ep_flags &= ~EXEC_HASARGL;
1607 				return E2BIG;
1608 			}
1609 			ktrexecarg(fa->fa_arg, len - 1);
1610 			dp += len;
1611 
1612 			kmem_free(fa->fa_arg, fa->fa_len);
1613 			fa++;
1614 			data->ed_argc++;
1615 		}
1616 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1617 		epp->ep_flags &= ~EXEC_HASARGL;
1618 	}
1619 
1620 	/*
1621 	 * Read and count argument strings from user.
1622 	 */
1623 
1624 	if (args == NULL) {
1625 		DPRINTF(("%s: null args\n", __func__));
1626 		return EINVAL;
1627 	}
1628 	if (epp->ep_flags & EXEC_SKIPARG)
1629 		args = (const void *)((const char *)args + fromptrsz(epp));
1630 	i = 0;
1631 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1632 	if (error != 0) {
1633 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1634 		return error;
1635 	}
1636 	data->ed_argc += i;
1637 
1638 	/*
1639 	 * Read and count environment strings from user.
1640 	 */
1641 
1642 	data->ed_envc = 0;
1643 	/* environment need not be there */
1644 	if (envs == NULL)
1645 		goto done;
1646 	i = 0;
1647 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1648 	if (error != 0) {
1649 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1650 		return error;
1651 	}
1652 	data->ed_envc += i;
1653 
1654 done:
1655 	*dpp = dp;
1656 
1657 	return 0;
1658 }
1659 
1660 static int
1661 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1662     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1663     void (*ktr)(const void *, size_t))
1664 {
1665 	char			*dp, *sp;
1666 	size_t			i;
1667 	int			error;
1668 
1669 	dp = *dpp;
1670 
1671 	i = 0;
1672 	while (1) {
1673 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1674 		size_t len;
1675 
1676 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1677 			return error;
1678 		}
1679 		if (!sp)
1680 			break;
1681 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1682 			if (error == ENAMETOOLONG)
1683 				error = E2BIG;
1684 			return error;
1685 		}
1686 		if (__predict_false(ktrace_on))
1687 			(*ktr)(dp, len - 1);
1688 		dp += len;
1689 		i++;
1690 	}
1691 
1692 	*dpp = dp;
1693 	*ip = i;
1694 
1695 	return 0;
1696 }
1697 
1698 /*
1699  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1700  * Those strings are located just after auxinfo.
1701  */
1702 int
1703 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1704     char **stackp, void *argp)
1705 {
1706 	char	**cpp, *dp, *sp;
1707 	size_t	len;
1708 	void	*nullp;
1709 	long	argc, envc;
1710 	int	error;
1711 
1712 	cpp = (char **)*stackp;
1713 	nullp = NULL;
1714 	argc = arginfo->ps_nargvstr;
1715 	envc = arginfo->ps_nenvstr;
1716 
1717 	/* argc on stack is long */
1718 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1719 
1720 	dp = (char *)(cpp +
1721 	    1 +				/* long argc */
1722 	    argc +			/* char *argv[] */
1723 	    1 +				/* \0 */
1724 	    envc +			/* char *env[] */
1725 	    1) +			/* \0 */
1726 	    pack->ep_esch->es_arglen;	/* auxinfo */
1727 	sp = argp;
1728 
1729 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1730 		COPYPRINTF("", cpp - 1, sizeof(argc));
1731 		return error;
1732 	}
1733 
1734 	/* XXX don't copy them out, remap them! */
1735 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1736 
1737 	for (; --argc >= 0; sp += len, dp += len) {
1738 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1739 			COPYPRINTF("", cpp - 1, sizeof(dp));
1740 			return error;
1741 		}
1742 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1743 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1744 			return error;
1745 		}
1746 	}
1747 
1748 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1749 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1750 		return error;
1751 	}
1752 
1753 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1754 
1755 	for (; --envc >= 0; sp += len, dp += len) {
1756 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1757 			COPYPRINTF("", cpp - 1, sizeof(dp));
1758 			return error;
1759 		}
1760 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1761 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1762 			return error;
1763 		}
1764 
1765 	}
1766 
1767 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1768 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1769 		return error;
1770 	}
1771 
1772 	*stackp = (char *)cpp;
1773 	return 0;
1774 }
1775 
1776 
1777 /*
1778  * Add execsw[] entries.
1779  */
1780 int
1781 exec_add(struct execsw *esp, int count)
1782 {
1783 	struct exec_entry	*it;
1784 	int			i;
1785 
1786 	if (count == 0) {
1787 		return 0;
1788 	}
1789 
1790 	/* Check for duplicates. */
1791 	rw_enter(&exec_lock, RW_WRITER);
1792 	for (i = 0; i < count; i++) {
1793 		LIST_FOREACH(it, &ex_head, ex_list) {
1794 			/* assume unique (makecmds, probe_func, emulation) */
1795 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1796 			    it->ex_sw->u.elf_probe_func ==
1797 			    esp[i].u.elf_probe_func &&
1798 			    it->ex_sw->es_emul == esp[i].es_emul) {
1799 				rw_exit(&exec_lock);
1800 				return EEXIST;
1801 			}
1802 		}
1803 	}
1804 
1805 	/* Allocate new entries. */
1806 	for (i = 0; i < count; i++) {
1807 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1808 		it->ex_sw = &esp[i];
1809 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1810 	}
1811 
1812 	/* update execsw[] */
1813 	exec_init(0);
1814 	rw_exit(&exec_lock);
1815 	return 0;
1816 }
1817 
1818 /*
1819  * Remove execsw[] entry.
1820  */
1821 int
1822 exec_remove(struct execsw *esp, int count)
1823 {
1824 	struct exec_entry	*it, *next;
1825 	int			i;
1826 	const struct proclist_desc *pd;
1827 	proc_t			*p;
1828 
1829 	if (count == 0) {
1830 		return 0;
1831 	}
1832 
1833 	/* Abort if any are busy. */
1834 	rw_enter(&exec_lock, RW_WRITER);
1835 	for (i = 0; i < count; i++) {
1836 		mutex_enter(proc_lock);
1837 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1838 			PROCLIST_FOREACH(p, pd->pd_list) {
1839 				if (p->p_execsw == &esp[i]) {
1840 					mutex_exit(proc_lock);
1841 					rw_exit(&exec_lock);
1842 					return EBUSY;
1843 				}
1844 			}
1845 		}
1846 		mutex_exit(proc_lock);
1847 	}
1848 
1849 	/* None are busy, so remove them all. */
1850 	for (i = 0; i < count; i++) {
1851 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1852 			next = LIST_NEXT(it, ex_list);
1853 			if (it->ex_sw == &esp[i]) {
1854 				LIST_REMOVE(it, ex_list);
1855 				kmem_free(it, sizeof(*it));
1856 				break;
1857 			}
1858 		}
1859 	}
1860 
1861 	/* update execsw[] */
1862 	exec_init(0);
1863 	rw_exit(&exec_lock);
1864 	return 0;
1865 }
1866 
1867 /*
1868  * Initialize exec structures. If init_boot is true, also does necessary
1869  * one-time initialization (it's called from main() that way).
1870  * Once system is multiuser, this should be called with exec_lock held,
1871  * i.e. via exec_{add|remove}().
1872  */
1873 int
1874 exec_init(int init_boot)
1875 {
1876 	const struct execsw 	**sw;
1877 	struct exec_entry	*ex;
1878 	SLIST_HEAD(,exec_entry)	first;
1879 	SLIST_HEAD(,exec_entry)	any;
1880 	SLIST_HEAD(,exec_entry)	last;
1881 	int			i, sz;
1882 
1883 	if (init_boot) {
1884 		/* do one-time initializations */
1885 		vaddr_t vmin = 0, vmax;
1886 
1887 		rw_init(&exec_lock);
1888 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1889 		exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1890 		    maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1891 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1892 		    "execargs", &exec_palloc, IPL_NONE);
1893 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1894 	} else {
1895 		KASSERT(rw_write_held(&exec_lock));
1896 	}
1897 
1898 	/* Sort each entry onto the appropriate queue. */
1899 	SLIST_INIT(&first);
1900 	SLIST_INIT(&any);
1901 	SLIST_INIT(&last);
1902 	sz = 0;
1903 	LIST_FOREACH(ex, &ex_head, ex_list) {
1904 		switch(ex->ex_sw->es_prio) {
1905 		case EXECSW_PRIO_FIRST:
1906 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1907 			break;
1908 		case EXECSW_PRIO_ANY:
1909 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1910 			break;
1911 		case EXECSW_PRIO_LAST:
1912 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1913 			break;
1914 		default:
1915 			panic("%s", __func__);
1916 			break;
1917 		}
1918 		sz++;
1919 	}
1920 
1921 	/*
1922 	 * Create new execsw[].  Ensure we do not try a zero-sized
1923 	 * allocation.
1924 	 */
1925 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1926 	i = 0;
1927 	SLIST_FOREACH(ex, &first, ex_slist) {
1928 		sw[i++] = ex->ex_sw;
1929 	}
1930 	SLIST_FOREACH(ex, &any, ex_slist) {
1931 		sw[i++] = ex->ex_sw;
1932 	}
1933 	SLIST_FOREACH(ex, &last, ex_slist) {
1934 		sw[i++] = ex->ex_sw;
1935 	}
1936 
1937 	/* Replace old execsw[] and free used memory. */
1938 	if (execsw != NULL) {
1939 		kmem_free(__UNCONST(execsw),
1940 		    nexecs * sizeof(struct execsw *) + 1);
1941 	}
1942 	execsw = sw;
1943 	nexecs = sz;
1944 
1945 	/* Figure out the maximum size of an exec header. */
1946 	exec_maxhdrsz = sizeof(int);
1947 	for (i = 0; i < nexecs; i++) {
1948 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1949 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1950 	}
1951 
1952 	return 0;
1953 }
1954 
1955 static int
1956 exec_sigcode_map(struct proc *p, const struct emul *e)
1957 {
1958 	vaddr_t va;
1959 	vsize_t sz;
1960 	int error;
1961 	struct uvm_object *uobj;
1962 
1963 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1964 
1965 	if (e->e_sigobject == NULL || sz == 0) {
1966 		return 0;
1967 	}
1968 
1969 	/*
1970 	 * If we don't have a sigobject for this emulation, create one.
1971 	 *
1972 	 * sigobject is an anonymous memory object (just like SYSV shared
1973 	 * memory) that we keep a permanent reference to and that we map
1974 	 * in all processes that need this sigcode. The creation is simple,
1975 	 * we create an object, add a permanent reference to it, map it in
1976 	 * kernel space, copy out the sigcode to it and unmap it.
1977 	 * We map it with PROT_READ|PROT_EXEC into the process just
1978 	 * the way sys_mmap() would map it.
1979 	 */
1980 
1981 	uobj = *e->e_sigobject;
1982 	if (uobj == NULL) {
1983 		mutex_enter(&sigobject_lock);
1984 		if ((uobj = *e->e_sigobject) == NULL) {
1985 			uobj = uao_create(sz, 0);
1986 			(*uobj->pgops->pgo_reference)(uobj);
1987 			va = vm_map_min(kernel_map);
1988 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1989 			    uobj, 0, 0,
1990 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1991 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1992 				printf("kernel mapping failed %d\n", error);
1993 				(*uobj->pgops->pgo_detach)(uobj);
1994 				mutex_exit(&sigobject_lock);
1995 				return error;
1996 			}
1997 			memcpy((void *)va, e->e_sigcode, sz);
1998 #ifdef PMAP_NEED_PROCWR
1999 			pmap_procwr(&proc0, va, sz);
2000 #endif
2001 			uvm_unmap(kernel_map, va, va + round_page(sz));
2002 			*e->e_sigobject = uobj;
2003 		}
2004 		mutex_exit(&sigobject_lock);
2005 	}
2006 
2007 	/* Just a hint to uvm_map where to put it. */
2008 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
2009 	    round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
2010 
2011 #ifdef __alpha__
2012 	/*
2013 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
2014 	 * which causes the above calculation to put the sigcode at
2015 	 * an invalid address.  Put it just below the text instead.
2016 	 */
2017 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
2018 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
2019 	}
2020 #endif
2021 
2022 	(*uobj->pgops->pgo_reference)(uobj);
2023 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
2024 			uobj, 0, 0,
2025 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
2026 				    UVM_ADV_RANDOM, 0));
2027 	if (error) {
2028 		DPRINTF(("%s, %d: map %p "
2029 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
2030 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
2031 		    va, error));
2032 		(*uobj->pgops->pgo_detach)(uobj);
2033 		return error;
2034 	}
2035 	p->p_sigctx.ps_sigcode = (void *)va;
2036 	return 0;
2037 }
2038 
2039 /*
2040  * Release a refcount on spawn_exec_data and destroy memory, if this
2041  * was the last one.
2042  */
2043 static void
2044 spawn_exec_data_release(struct spawn_exec_data *data)
2045 {
2046 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
2047 		return;
2048 
2049 	cv_destroy(&data->sed_cv_child_ready);
2050 	mutex_destroy(&data->sed_mtx_child);
2051 
2052 	if (data->sed_actions)
2053 		posix_spawn_fa_free(data->sed_actions,
2054 		    data->sed_actions->len);
2055 	if (data->sed_attrs)
2056 		kmem_free(data->sed_attrs,
2057 		    sizeof(*data->sed_attrs));
2058 	kmem_free(data, sizeof(*data));
2059 }
2060 
2061 /*
2062  * A child lwp of a posix_spawn operation starts here and ends up in
2063  * cpu_spawn_return, dealing with all filedescriptor and scheduler
2064  * manipulations in between.
2065  * The parent waits for the child, as it is not clear whether the child
2066  * will be able to acquire its own exec_lock. If it can, the parent can
2067  * be released early and continue running in parallel. If not (or if the
2068  * magic debug flag is passed in the scheduler attribute struct), the
2069  * child rides on the parent's exec lock until it is ready to return to
2070  * to userland - and only then releases the parent. This method loses
2071  * concurrency, but improves error reporting.
2072  */
2073 static void
2074 spawn_return(void *arg)
2075 {
2076 	struct spawn_exec_data *spawn_data = arg;
2077 	struct lwp *l = curlwp;
2078 	struct proc *p = l->l_proc;
2079 	int error, newfd;
2080 	int ostat;
2081 	size_t i;
2082 	const struct posix_spawn_file_actions_entry *fae;
2083 	pid_t ppid;
2084 	register_t retval;
2085 	bool have_reflock;
2086 	bool parent_is_waiting = true;
2087 
2088 	/*
2089 	 * Check if we can release parent early.
2090 	 * We either need to have no sed_attrs, or sed_attrs does not
2091 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2092 	 * safe access to the parent proc (passed in sed_parent).
2093 	 * We then try to get the exec_lock, and only if that works, we can
2094 	 * release the parent here already.
2095 	 */
2096 	ppid = spawn_data->sed_parent->p_pid;
2097 	if ((!spawn_data->sed_attrs
2098 	    || (spawn_data->sed_attrs->sa_flags
2099 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2100 	    && rw_tryenter(&exec_lock, RW_READER)) {
2101 		parent_is_waiting = false;
2102 		mutex_enter(&spawn_data->sed_mtx_child);
2103 		cv_signal(&spawn_data->sed_cv_child_ready);
2104 		mutex_exit(&spawn_data->sed_mtx_child);
2105 	}
2106 
2107 	/* don't allow debugger access yet */
2108 	rw_enter(&p->p_reflock, RW_WRITER);
2109 	have_reflock = true;
2110 
2111 	error = 0;
2112 	/* handle posix_spawn_file_actions */
2113 	if (spawn_data->sed_actions != NULL) {
2114 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
2115 			fae = &spawn_data->sed_actions->fae[i];
2116 			switch (fae->fae_action) {
2117 			case FAE_OPEN:
2118 				if (fd_getfile(fae->fae_fildes) != NULL) {
2119 					error = fd_close(fae->fae_fildes);
2120 					if (error)
2121 						break;
2122 				}
2123 				error = fd_open(fae->fae_path, fae->fae_oflag,
2124 				    fae->fae_mode, &newfd);
2125 				if (error)
2126 					break;
2127 				if (newfd != fae->fae_fildes) {
2128 					error = dodup(l, newfd,
2129 					    fae->fae_fildes, 0, &retval);
2130 					if (fd_getfile(newfd) != NULL)
2131 						fd_close(newfd);
2132 				}
2133 				break;
2134 			case FAE_DUP2:
2135 				error = dodup(l, fae->fae_fildes,
2136 				    fae->fae_newfildes, 0, &retval);
2137 				break;
2138 			case FAE_CLOSE:
2139 				if (fd_getfile(fae->fae_fildes) == NULL) {
2140 					error = EBADF;
2141 					break;
2142 				}
2143 				error = fd_close(fae->fae_fildes);
2144 				break;
2145 			}
2146 			if (error)
2147 				goto report_error;
2148 		}
2149 	}
2150 
2151 	/* handle posix_spawnattr */
2152 	if (spawn_data->sed_attrs != NULL) {
2153 		struct sigaction sigact;
2154 		memset(&sigact, 0, sizeof(sigact));
2155 		sigact._sa_u._sa_handler = SIG_DFL;
2156 		sigact.sa_flags = 0;
2157 
2158 		/*
2159 		 * set state to SSTOP so that this proc can be found by pid.
2160 		 * see proc_enterprp, do_sched_setparam below
2161 		 */
2162 		mutex_enter(proc_lock);
2163 		/*
2164 		 * p_stat should be SACTIVE, so we need to adjust the
2165 		 * parent's p_nstopchild here.  For safety, just make
2166 		 * we're on the good side of SDEAD before we adjust.
2167 		 */
2168 		ostat = p->p_stat;
2169 		KASSERT(ostat < SSTOP);
2170 		p->p_stat = SSTOP;
2171 		p->p_waited = 0;
2172 		p->p_pptr->p_nstopchild++;
2173 		mutex_exit(proc_lock);
2174 
2175 		/* Set process group */
2176 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2177 			pid_t mypid = p->p_pid,
2178 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
2179 
2180 			if (pgrp == 0)
2181 				pgrp = mypid;
2182 
2183 			error = proc_enterpgrp(spawn_data->sed_parent,
2184 			    mypid, pgrp, false);
2185 			if (error)
2186 				goto report_error_stopped;
2187 		}
2188 
2189 		/* Set scheduler policy */
2190 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2191 			error = do_sched_setparam(p->p_pid, 0,
2192 			    spawn_data->sed_attrs->sa_schedpolicy,
2193 			    &spawn_data->sed_attrs->sa_schedparam);
2194 		else if (spawn_data->sed_attrs->sa_flags
2195 		    & POSIX_SPAWN_SETSCHEDPARAM) {
2196 			error = do_sched_setparam(ppid, 0,
2197 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2198 		}
2199 		if (error)
2200 			goto report_error_stopped;
2201 
2202 		/* Reset user ID's */
2203 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2204 			error = do_setresuid(l, -1,
2205 			     kauth_cred_getgid(l->l_cred), -1,
2206 			     ID_E_EQ_R | ID_E_EQ_S);
2207 			if (error)
2208 				goto report_error_stopped;
2209 			error = do_setresuid(l, -1,
2210 			    kauth_cred_getuid(l->l_cred), -1,
2211 			    ID_E_EQ_R | ID_E_EQ_S);
2212 			if (error)
2213 				goto report_error_stopped;
2214 		}
2215 
2216 		/* Set signal masks/defaults */
2217 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2218 			mutex_enter(p->p_lock);
2219 			error = sigprocmask1(l, SIG_SETMASK,
2220 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
2221 			mutex_exit(p->p_lock);
2222 			if (error)
2223 				goto report_error_stopped;
2224 		}
2225 
2226 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2227 			/*
2228 			 * The following sigaction call is using a sigaction
2229 			 * version 0 trampoline which is in the compatibility
2230 			 * code only. This is not a problem because for SIG_DFL
2231 			 * and SIG_IGN, the trampolines are now ignored. If they
2232 			 * were not, this would be a problem because we are
2233 			 * holding the exec_lock, and the compat code needs
2234 			 * to do the same in order to replace the trampoline
2235 			 * code of the process.
2236 			 */
2237 			for (i = 1; i <= NSIG; i++) {
2238 				if (sigismember(
2239 				    &spawn_data->sed_attrs->sa_sigdefault, i))
2240 					sigaction1(l, i, &sigact, NULL, NULL,
2241 					    0);
2242 			}
2243 		}
2244 		mutex_enter(proc_lock);
2245 		p->p_stat = ostat;
2246 		p->p_pptr->p_nstopchild--;
2247 		mutex_exit(proc_lock);
2248 	}
2249 
2250 	/* now do the real exec */
2251 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2252 	    true);
2253 	have_reflock = false;
2254 	if (error == EJUSTRETURN)
2255 		error = 0;
2256 	else if (error)
2257 		goto report_error;
2258 
2259 	if (parent_is_waiting) {
2260 		mutex_enter(&spawn_data->sed_mtx_child);
2261 		cv_signal(&spawn_data->sed_cv_child_ready);
2262 		mutex_exit(&spawn_data->sed_mtx_child);
2263 	}
2264 
2265 	/* release our refcount on the data */
2266 	spawn_exec_data_release(spawn_data);
2267 
2268 	if (p->p_slflag & PSL_TRACED) {
2269 		/* Paranoid check */
2270 		mutex_enter(proc_lock);
2271 		if (!(p->p_slflag & PSL_TRACED)) {
2272 			mutex_exit(proc_lock);
2273 			goto cpu_return;
2274 		}
2275 
2276 		mutex_enter(p->p_lock);
2277 		eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, p->p_opptr->p_pid);
2278 	}
2279 
2280  cpu_return:
2281 	/* and finally: leave to userland for the first time */
2282 	cpu_spawn_return(l);
2283 
2284 	/* NOTREACHED */
2285 	return;
2286 
2287  report_error_stopped:
2288 	mutex_enter(proc_lock);
2289 	p->p_stat = ostat;
2290 	p->p_pptr->p_nstopchild--;
2291 	mutex_exit(proc_lock);
2292  report_error:
2293 	if (have_reflock) {
2294 		/*
2295 		 * We have not passed through execve_runproc(),
2296 		 * which would have released the p_reflock and also
2297 		 * taken ownership of the sed_exec part of spawn_data,
2298 		 * so release/free both here.
2299 		 */
2300 		rw_exit(&p->p_reflock);
2301 		execve_free_data(&spawn_data->sed_exec);
2302 	}
2303 
2304 	if (parent_is_waiting) {
2305 		/* pass error to parent */
2306 		mutex_enter(&spawn_data->sed_mtx_child);
2307 		spawn_data->sed_error = error;
2308 		cv_signal(&spawn_data->sed_cv_child_ready);
2309 		mutex_exit(&spawn_data->sed_mtx_child);
2310 	} else {
2311 		rw_exit(&exec_lock);
2312 	}
2313 
2314 	/* release our refcount on the data */
2315 	spawn_exec_data_release(spawn_data);
2316 
2317 	/* done, exit */
2318 	mutex_enter(p->p_lock);
2319 	/*
2320 	 * Posix explicitly asks for an exit code of 127 if we report
2321 	 * errors from the child process - so, unfortunately, there
2322 	 * is no way to report a more exact error code.
2323 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2324 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2325 	 */
2326 	exit1(l, 127, 0);
2327 }
2328 
2329 void
2330 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2331 {
2332 
2333 	for (size_t i = 0; i < len; i++) {
2334 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2335 		if (fae->fae_action != FAE_OPEN)
2336 			continue;
2337 		kmem_strfree(fae->fae_path);
2338 	}
2339 	if (fa->len > 0)
2340 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2341 	kmem_free(fa, sizeof(*fa));
2342 }
2343 
2344 static int
2345 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2346     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2347 {
2348 	struct posix_spawn_file_actions *fa;
2349 	struct posix_spawn_file_actions_entry *fae;
2350 	char *pbuf = NULL;
2351 	int error;
2352 	size_t i = 0;
2353 
2354 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2355 	error = copyin(ufa, fa, sizeof(*fa));
2356 	if (error || fa->len == 0) {
2357 		kmem_free(fa, sizeof(*fa));
2358 		return error;	/* 0 if not an error, and len == 0 */
2359 	}
2360 
2361 	if (fa->len > lim) {
2362 		kmem_free(fa, sizeof(*fa));
2363 		return EINVAL;
2364 	}
2365 
2366 	fa->size = fa->len;
2367 	size_t fal = fa->len * sizeof(*fae);
2368 	fae = fa->fae;
2369 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2370 	error = copyin(fae, fa->fae, fal);
2371 	if (error)
2372 		goto out;
2373 
2374 	pbuf = PNBUF_GET();
2375 	for (; i < fa->len; i++) {
2376 		fae = &fa->fae[i];
2377 		if (fae->fae_action != FAE_OPEN)
2378 			continue;
2379 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2380 		if (error)
2381 			goto out;
2382 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2383 		memcpy(fae->fae_path, pbuf, fal);
2384 	}
2385 	PNBUF_PUT(pbuf);
2386 
2387 	*fap = fa;
2388 	return 0;
2389 out:
2390 	if (pbuf)
2391 		PNBUF_PUT(pbuf);
2392 	posix_spawn_fa_free(fa, i);
2393 	return error;
2394 }
2395 
2396 int
2397 check_posix_spawn(struct lwp *l1)
2398 {
2399 	int error, tnprocs, count;
2400 	uid_t uid;
2401 	struct proc *p1;
2402 
2403 	p1 = l1->l_proc;
2404 	uid = kauth_cred_getuid(l1->l_cred);
2405 	tnprocs = atomic_inc_uint_nv(&nprocs);
2406 
2407 	/*
2408 	 * Although process entries are dynamically created, we still keep
2409 	 * a global limit on the maximum number we will create.
2410 	 */
2411 	if (__predict_false(tnprocs >= maxproc))
2412 		error = -1;
2413 	else
2414 		error = kauth_authorize_process(l1->l_cred,
2415 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2416 
2417 	if (error) {
2418 		atomic_dec_uint(&nprocs);
2419 		return EAGAIN;
2420 	}
2421 
2422 	/*
2423 	 * Enforce limits.
2424 	 */
2425 	count = chgproccnt(uid, 1);
2426 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2427 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2428 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2429 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2430 		(void)chgproccnt(uid, -1);
2431 		atomic_dec_uint(&nprocs);
2432 		return EAGAIN;
2433 	}
2434 
2435 	return 0;
2436 }
2437 
2438 int
2439 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2440 	struct posix_spawn_file_actions *fa,
2441 	struct posix_spawnattr *sa,
2442 	char *const *argv, char *const *envp,
2443 	execve_fetch_element_t fetch)
2444 {
2445 
2446 	struct proc *p1, *p2;
2447 	struct lwp *l2;
2448 	int error;
2449 	struct spawn_exec_data *spawn_data;
2450 	vaddr_t uaddr;
2451 	pid_t pid;
2452 	bool have_exec_lock = false;
2453 
2454 	p1 = l1->l_proc;
2455 
2456 	/* Allocate and init spawn_data */
2457 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2458 	spawn_data->sed_refcnt = 1; /* only parent so far */
2459 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2460 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2461 	mutex_enter(&spawn_data->sed_mtx_child);
2462 
2463 	/*
2464 	 * Do the first part of the exec now, collect state
2465 	 * in spawn_data.
2466 	 */
2467 	error = execve_loadvm(l1, true, path, -1, argv,
2468 	    envp, fetch, &spawn_data->sed_exec);
2469 	if (error == EJUSTRETURN)
2470 		error = 0;
2471 	else if (error)
2472 		goto error_exit;
2473 
2474 	have_exec_lock = true;
2475 
2476 	/*
2477 	 * Allocate virtual address space for the U-area now, while it
2478 	 * is still easy to abort the fork operation if we're out of
2479 	 * kernel virtual address space.
2480 	 */
2481 	uaddr = uvm_uarea_alloc();
2482 	if (__predict_false(uaddr == 0)) {
2483 		error = ENOMEM;
2484 		goto error_exit;
2485 	}
2486 
2487 	/*
2488 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2489 	 * replace it with its own before returning to userland
2490 	 * in the child.
2491 	 * This is a point of no return, we will have to go through
2492 	 * the child proc to properly clean it up past this point.
2493 	 */
2494 	p2 = proc_alloc();
2495 	pid = p2->p_pid;
2496 
2497 	/*
2498 	 * Make a proc table entry for the new process.
2499 	 * Start by zeroing the section of proc that is zero-initialized,
2500 	 * then copy the section that is copied directly from the parent.
2501 	 */
2502 	memset(&p2->p_startzero, 0,
2503 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2504 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2505 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2506 	p2->p_vmspace = proc0.p_vmspace;
2507 
2508 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2509 
2510 	LIST_INIT(&p2->p_lwps);
2511 	LIST_INIT(&p2->p_sigwaiters);
2512 
2513 	/*
2514 	 * Duplicate sub-structures as needed.
2515 	 * Increase reference counts on shared objects.
2516 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2517 	 * handling are important in order to keep a consistent behaviour
2518 	 * for the child after the fork.  If we are a 32-bit process, the
2519 	 * child will be too.
2520 	 */
2521 	p2->p_flag =
2522 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2523 	p2->p_emul = p1->p_emul;
2524 	p2->p_execsw = p1->p_execsw;
2525 
2526 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2527 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2528 	rw_init(&p2->p_reflock);
2529 	cv_init(&p2->p_waitcv, "wait");
2530 	cv_init(&p2->p_lwpcv, "lwpwait");
2531 
2532 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2533 
2534 	kauth_proc_fork(p1, p2);
2535 
2536 	p2->p_raslist = NULL;
2537 	p2->p_fd = fd_copy();
2538 
2539 	/* XXX racy */
2540 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2541 
2542 	p2->p_cwdi = cwdinit();
2543 
2544 	/*
2545 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2546 	 * we just need increase pl_refcnt.
2547 	 */
2548 	if (!p1->p_limit->pl_writeable) {
2549 		lim_addref(p1->p_limit);
2550 		p2->p_limit = p1->p_limit;
2551 	} else {
2552 		p2->p_limit = lim_copy(p1->p_limit);
2553 	}
2554 
2555 	p2->p_lflag = 0;
2556 	l1->l_vforkwaiting = false;
2557 	p2->p_sflag = 0;
2558 	p2->p_slflag = 0;
2559 	p2->p_pptr = p1;
2560 	p2->p_ppid = p1->p_pid;
2561 	LIST_INIT(&p2->p_children);
2562 
2563 	p2->p_aio = NULL;
2564 
2565 #ifdef KTRACE
2566 	/*
2567 	 * Copy traceflag and tracefile if enabled.
2568 	 * If not inherited, these were zeroed above.
2569 	 */
2570 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2571 		mutex_enter(&ktrace_lock);
2572 		p2->p_traceflag = p1->p_traceflag;
2573 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2574 			ktradref(p2);
2575 		mutex_exit(&ktrace_lock);
2576 	}
2577 #endif
2578 
2579 	/*
2580 	 * Create signal actions for the child process.
2581 	 */
2582 	p2->p_sigacts = sigactsinit(p1, 0);
2583 	mutex_enter(p1->p_lock);
2584 	p2->p_sflag |=
2585 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2586 	sched_proc_fork(p1, p2);
2587 	mutex_exit(p1->p_lock);
2588 
2589 	p2->p_stflag = p1->p_stflag;
2590 
2591 	/*
2592 	 * p_stats.
2593 	 * Copy parts of p_stats, and zero out the rest.
2594 	 */
2595 	p2->p_stats = pstatscopy(p1->p_stats);
2596 
2597 	/* copy over machdep flags to the new proc */
2598 	cpu_proc_fork(p1, p2);
2599 
2600 	/*
2601 	 * Prepare remaining parts of spawn data
2602 	 */
2603 	spawn_data->sed_actions = fa;
2604 	spawn_data->sed_attrs = sa;
2605 
2606 	spawn_data->sed_parent = p1;
2607 
2608 	/* create LWP */
2609 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2610 	    &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2611 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2612 
2613 	/*
2614 	 * Copy the credential so other references don't see our changes.
2615 	 * Test to see if this is necessary first, since in the common case
2616 	 * we won't need a private reference.
2617 	 */
2618 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2619 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2620 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2621 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2622 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2623 	}
2624 
2625 	/* Update the master credentials. */
2626 	if (l2->l_cred != p2->p_cred) {
2627 		kauth_cred_t ocred;
2628 
2629 		kauth_cred_hold(l2->l_cred);
2630 		mutex_enter(p2->p_lock);
2631 		ocred = p2->p_cred;
2632 		p2->p_cred = l2->l_cred;
2633 		mutex_exit(p2->p_lock);
2634 		kauth_cred_free(ocred);
2635 	}
2636 
2637 	*child_ok = true;
2638 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2639 #if 0
2640 	l2->l_nopreempt = 1; /* start it non-preemptable */
2641 #endif
2642 
2643 	/*
2644 	 * It's now safe for the scheduler and other processes to see the
2645 	 * child process.
2646 	 */
2647 	mutex_enter(proc_lock);
2648 
2649 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2650 		p2->p_lflag |= PL_CONTROLT;
2651 
2652 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2653 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2654 
2655 	if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) ==
2656 	    (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2657 		proc_changeparent(p2, p1->p_pptr);
2658 	}
2659 
2660 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2661 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2662 
2663 	p2->p_trace_enabled = trace_is_enabled(p2);
2664 #ifdef __HAVE_SYSCALL_INTERN
2665 	(*p2->p_emul->e_syscall_intern)(p2);
2666 #endif
2667 
2668 	/*
2669 	 * Make child runnable, set start time, and add to run queue except
2670 	 * if the parent requested the child to start in SSTOP state.
2671 	 */
2672 	mutex_enter(p2->p_lock);
2673 
2674 	getmicrotime(&p2->p_stats->p_start);
2675 
2676 	lwp_lock(l2);
2677 	KASSERT(p2->p_nrlwps == 1);
2678 	KASSERT(l2->l_stat == LSIDL);
2679 	p2->p_nrlwps = 1;
2680 	p2->p_stat = SACTIVE;
2681 	setrunnable(l2);
2682 	/* LWP now unlocked */
2683 
2684 	mutex_exit(p2->p_lock);
2685 	mutex_exit(proc_lock);
2686 
2687 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2688 	error = spawn_data->sed_error;
2689 	mutex_exit(&spawn_data->sed_mtx_child);
2690 	spawn_exec_data_release(spawn_data);
2691 
2692 	rw_exit(&p1->p_reflock);
2693 	rw_exit(&exec_lock);
2694 	have_exec_lock = false;
2695 
2696 	*pid_res = pid;
2697 
2698 	if (error)
2699 		return error;
2700 
2701 	if (p1->p_slflag & PSL_TRACED) {
2702 		/* Paranoid check */
2703 		mutex_enter(proc_lock);
2704 		if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) !=
2705 		    (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2706 			mutex_exit(proc_lock);
2707 			return 0;
2708 		}
2709 
2710 		mutex_enter(p1->p_lock);
2711 		eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, pid);
2712 	}
2713 	return 0;
2714 
2715  error_exit:
2716 	if (have_exec_lock) {
2717 		execve_free_data(&spawn_data->sed_exec);
2718 		rw_exit(&p1->p_reflock);
2719 		rw_exit(&exec_lock);
2720 	}
2721 	mutex_exit(&spawn_data->sed_mtx_child);
2722 	spawn_exec_data_release(spawn_data);
2723 
2724 	return error;
2725 }
2726 
2727 int
2728 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2729     register_t *retval)
2730 {
2731 	/* {
2732 		syscallarg(pid_t *) pid;
2733 		syscallarg(const char *) path;
2734 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2735 		syscallarg(const struct posix_spawnattr *) attrp;
2736 		syscallarg(char *const *) argv;
2737 		syscallarg(char *const *) envp;
2738 	} */
2739 
2740 	int error;
2741 	struct posix_spawn_file_actions *fa = NULL;
2742 	struct posix_spawnattr *sa = NULL;
2743 	pid_t pid;
2744 	bool child_ok = false;
2745 	rlim_t max_fileactions;
2746 	proc_t *p = l1->l_proc;
2747 
2748 	error = check_posix_spawn(l1);
2749 	if (error) {
2750 		*retval = error;
2751 		return 0;
2752 	}
2753 
2754 	/* copy in file_actions struct */
2755 	if (SCARG(uap, file_actions) != NULL) {
2756 		max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2757 		    maxfiles);
2758 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2759 		    max_fileactions);
2760 		if (error)
2761 			goto error_exit;
2762 	}
2763 
2764 	/* copyin posix_spawnattr struct */
2765 	if (SCARG(uap, attrp) != NULL) {
2766 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2767 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2768 		if (error)
2769 			goto error_exit;
2770 	}
2771 
2772 	/*
2773 	 * Do the spawn
2774 	 */
2775 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2776 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2777 	if (error)
2778 		goto error_exit;
2779 
2780 	if (error == 0 && SCARG(uap, pid) != NULL)
2781 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2782 
2783 	*retval = error;
2784 	return 0;
2785 
2786  error_exit:
2787 	if (!child_ok) {
2788 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2789 		atomic_dec_uint(&nprocs);
2790 
2791 		if (sa)
2792 			kmem_free(sa, sizeof(*sa));
2793 		if (fa)
2794 			posix_spawn_fa_free(fa, fa->len);
2795 	}
2796 
2797 	*retval = error;
2798 	return 0;
2799 }
2800 
2801 void
2802 exec_free_emul_arg(struct exec_package *epp)
2803 {
2804 	if (epp->ep_emul_arg_free != NULL) {
2805 		KASSERT(epp->ep_emul_arg != NULL);
2806 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2807 		epp->ep_emul_arg_free = NULL;
2808 		epp->ep_emul_arg = NULL;
2809 	} else {
2810 		KASSERT(epp->ep_emul_arg == NULL);
2811 	}
2812 }
2813 
2814 #ifdef DEBUG_EXEC
2815 static void
2816 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2817 {
2818 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2819 	size_t j;
2820 
2821 	if (error == 0)
2822 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2823 	else
2824 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2825 		    epp->ep_vmcmds.evs_used, error));
2826 
2827 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2828 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2829 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2830 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2831 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2832 		    "pagedvn" :
2833 		    vp[j].ev_proc == vmcmd_map_readvn ?
2834 		    "readvn" :
2835 		    vp[j].ev_proc == vmcmd_map_zero ?
2836 		    "zero" : "*unknown*",
2837 		    vp[j].ev_addr, vp[j].ev_len,
2838 		    vp[j].ev_offset, vp[j].ev_prot,
2839 		    vp[j].ev_flags));
2840 		if (error != 0 && j == x)
2841 			DPRINTF(("     ^--- failed\n"));
2842 	}
2843 }
2844 #endif
2845