xref: /netbsd-src/sys/kern/kern_exec.c (revision 796c32c94f6e154afc9de0f63da35c91bb739b45)
1 /*	$NetBSD: kern_exec.c,v 1.453 2017/11/13 22:01:45 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*-
30  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
31  * Copyright (C) 1992 Wolfgang Solfrank.
32  * Copyright (C) 1992 TooLs GmbH.
33  * All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. All advertising materials mentioning features or use of this software
44  *    must display the following acknowledgement:
45  *	This product includes software developed by TooLs GmbH.
46  * 4. The name of TooLs GmbH may not be used to endorse or promote products
47  *    derived from this software without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
50  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
51  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
52  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
53  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
54  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
55  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
56  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
57  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
58  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
59  */
60 
61 #include <sys/cdefs.h>
62 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.453 2017/11/13 22:01:45 christos Exp $");
63 
64 #include "opt_exec.h"
65 #include "opt_execfmt.h"
66 #include "opt_ktrace.h"
67 #include "opt_modular.h"
68 #include "opt_syscall_debug.h"
69 #include "veriexec.h"
70 #include "opt_pax.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/filedesc.h>
75 #include <sys/kernel.h>
76 #include <sys/proc.h>
77 #include <sys/mount.h>
78 #include <sys/kmem.h>
79 #include <sys/namei.h>
80 #include <sys/vnode.h>
81 #include <sys/file.h>
82 #include <sys/filedesc.h>
83 #include <sys/acct.h>
84 #include <sys/atomic.h>
85 #include <sys/exec.h>
86 #include <sys/ktrace.h>
87 #include <sys/uidinfo.h>
88 #include <sys/wait.h>
89 #include <sys/mman.h>
90 #include <sys/ras.h>
91 #include <sys/signalvar.h>
92 #include <sys/stat.h>
93 #include <sys/syscall.h>
94 #include <sys/kauth.h>
95 #include <sys/lwpctl.h>
96 #include <sys/pax.h>
97 #include <sys/cpu.h>
98 #include <sys/module.h>
99 #include <sys/syscallvar.h>
100 #include <sys/syscallargs.h>
101 #if NVERIEXEC > 0
102 #include <sys/verified_exec.h>
103 #endif /* NVERIEXEC > 0 */
104 #include <sys/sdt.h>
105 #include <sys/spawn.h>
106 #include <sys/prot.h>
107 #include <sys/cprng.h>
108 
109 #include <uvm/uvm_extern.h>
110 
111 #include <machine/reg.h>
112 
113 #include <compat/common/compat_util.h>
114 
115 #ifndef MD_TOPDOWN_INIT
116 #ifdef __USE_TOPDOWN_VM
117 #define	MD_TOPDOWN_INIT(epp)	(epp)->ep_flags |= EXEC_TOPDOWN_VM
118 #else
119 #define	MD_TOPDOWN_INIT(epp)
120 #endif
121 #endif
122 
123 struct execve_data;
124 
125 extern int user_va0_disable;
126 
127 static size_t calcargs(struct execve_data * restrict, const size_t);
128 static size_t calcstack(struct execve_data * restrict, const size_t);
129 static int copyoutargs(struct execve_data * restrict, struct lwp *,
130     char * const);
131 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
132 static int copyinargs(struct execve_data * restrict, char * const *,
133     char * const *, execve_fetch_element_t, char **);
134 static int copyinargstrs(struct execve_data * restrict, char * const *,
135     execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
136 static int exec_sigcode_map(struct proc *, const struct emul *);
137 
138 #if defined(DEBUG) && !defined(DEBUG_EXEC)
139 #define DEBUG_EXEC
140 #endif
141 #ifdef DEBUG_EXEC
142 #define DPRINTF(a) printf a
143 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
144     __LINE__, (s), (a), (b))
145 static void dump_vmcmds(const struct exec_package * const, size_t, int);
146 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
147 #else
148 #define DPRINTF(a)
149 #define COPYPRINTF(s, a, b)
150 #define DUMPVMCMDS(p, x, e) do {} while (0)
151 #endif /* DEBUG_EXEC */
152 
153 /*
154  * DTrace SDT provider definitions
155  */
156 SDT_PROVIDER_DECLARE(proc);
157 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
158 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
159 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
160 
161 /*
162  * Exec function switch:
163  *
164  * Note that each makecmds function is responsible for loading the
165  * exec package with the necessary functions for any exec-type-specific
166  * handling.
167  *
168  * Functions for specific exec types should be defined in their own
169  * header file.
170  */
171 static const struct execsw	**execsw = NULL;
172 static int			nexecs;
173 
174 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
175 
176 /* list of dynamically loaded execsw entries */
177 static LIST_HEAD(execlist_head, exec_entry) ex_head =
178     LIST_HEAD_INITIALIZER(ex_head);
179 struct exec_entry {
180 	LIST_ENTRY(exec_entry)	ex_list;
181 	SLIST_ENTRY(exec_entry)	ex_slist;
182 	const struct execsw	*ex_sw;
183 };
184 
185 #ifndef __HAVE_SYSCALL_INTERN
186 void	syscall(void);
187 #endif
188 
189 /* NetBSD autoloadable syscalls */
190 #ifdef MODULAR
191 #include <kern/syscalls_autoload.c>
192 #endif
193 
194 /* NetBSD emul struct */
195 struct emul emul_netbsd = {
196 	.e_name =		"netbsd",
197 #ifdef EMUL_NATIVEROOT
198 	.e_path =		EMUL_NATIVEROOT,
199 #else
200 	.e_path =		NULL,
201 #endif
202 #ifndef __HAVE_MINIMAL_EMUL
203 	.e_flags =		EMUL_HAS_SYS___syscall,
204 	.e_errno =		NULL,
205 	.e_nosys =		SYS_syscall,
206 	.e_nsysent =		SYS_NSYSENT,
207 #endif
208 #ifdef MODULAR
209 	.e_sc_autoload =	netbsd_syscalls_autoload,
210 #endif
211 	.e_sysent =		sysent,
212 #ifdef SYSCALL_DEBUG
213 	.e_syscallnames =	syscallnames,
214 #else
215 	.e_syscallnames =	NULL,
216 #endif
217 	.e_sendsig =		sendsig,
218 	.e_trapsignal =		trapsignal,
219 	.e_tracesig =		NULL,
220 	.e_sigcode =		NULL,
221 	.e_esigcode =		NULL,
222 	.e_sigobject =		NULL,
223 	.e_setregs =		setregs,
224 	.e_proc_exec =		NULL,
225 	.e_proc_fork =		NULL,
226 	.e_proc_exit =		NULL,
227 	.e_lwp_fork =		NULL,
228 	.e_lwp_exit =		NULL,
229 #ifdef __HAVE_SYSCALL_INTERN
230 	.e_syscall_intern =	syscall_intern,
231 #else
232 	.e_syscall =		syscall,
233 #endif
234 	.e_sysctlovly =		NULL,
235 	.e_fault =		NULL,
236 	.e_vm_default_addr =	uvm_default_mapaddr,
237 	.e_usertrap =		NULL,
238 	.e_ucsize =		sizeof(ucontext_t),
239 	.e_startlwp =		startlwp
240 };
241 
242 /*
243  * Exec lock. Used to control access to execsw[] structures.
244  * This must not be static so that netbsd32 can access it, too.
245  */
246 krwlock_t exec_lock;
247 
248 static kmutex_t sigobject_lock;
249 
250 /*
251  * Data used between a loadvm and execve part of an "exec" operation
252  */
253 struct execve_data {
254 	struct exec_package	ed_pack;
255 	struct pathbuf		*ed_pathbuf;
256 	struct vattr		ed_attr;
257 	struct ps_strings	ed_arginfo;
258 	char			*ed_argp;
259 	const char		*ed_pathstring;
260 	char			*ed_resolvedpathbuf;
261 	size_t			ed_ps_strings_sz;
262 	int			ed_szsigcode;
263 	size_t			ed_argslen;
264 	long			ed_argc;
265 	long			ed_envc;
266 };
267 
268 /*
269  * data passed from parent lwp to child during a posix_spawn()
270  */
271 struct spawn_exec_data {
272 	struct execve_data	sed_exec;
273 	struct posix_spawn_file_actions
274 				*sed_actions;
275 	struct posix_spawnattr	*sed_attrs;
276 	struct proc		*sed_parent;
277 	kcondvar_t		sed_cv_child_ready;
278 	kmutex_t		sed_mtx_child;
279 	int			sed_error;
280 	volatile uint32_t	sed_refcnt;
281 };
282 
283 static struct vm_map *exec_map;
284 static struct pool exec_pool;
285 
286 static void *
287 exec_pool_alloc(struct pool *pp, int flags)
288 {
289 
290 	return (void *)uvm_km_alloc(exec_map, NCARGS, 0,
291 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
292 }
293 
294 static void
295 exec_pool_free(struct pool *pp, void *addr)
296 {
297 
298 	uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
299 }
300 
301 static struct pool_allocator exec_palloc = {
302 	.pa_alloc = exec_pool_alloc,
303 	.pa_free = exec_pool_free,
304 	.pa_pagesz = NCARGS
305 };
306 
307 /*
308  * check exec:
309  * given an "executable" described in the exec package's namei info,
310  * see what we can do with it.
311  *
312  * ON ENTRY:
313  *	exec package with appropriate namei info
314  *	lwp pointer of exec'ing lwp
315  *	NO SELF-LOCKED VNODES
316  *
317  * ON EXIT:
318  *	error:	nothing held, etc.  exec header still allocated.
319  *	ok:	filled exec package, executable's vnode (unlocked).
320  *
321  * EXEC SWITCH ENTRY:
322  * 	Locked vnode to check, exec package, proc.
323  *
324  * EXEC SWITCH EXIT:
325  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
326  *	error:	destructive:
327  *			everything deallocated execept exec header.
328  *		non-destructive:
329  *			error code, executable's vnode (unlocked),
330  *			exec header unmodified.
331  */
332 int
333 /*ARGSUSED*/
334 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
335 {
336 	int		error, i;
337 	struct vnode	*vp;
338 	struct nameidata nd;
339 	size_t		resid;
340 
341 #if 1
342 	// grab the absolute pathbuf here before namei() trashes it.
343 	pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX);
344 #endif
345 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
346 
347 	/* first get the vnode */
348 	if ((error = namei(&nd)) != 0)
349 		return error;
350 	epp->ep_vp = vp = nd.ni_vp;
351 #if 0
352 	/*
353 	 * XXX: can't use nd.ni_pnbuf, because although pb contains an
354 	 * absolute path, nd.ni_pnbuf does not if the path contains symlinks.
355 	 */
356 	/* normally this can't fail */
357 	error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
358 	KASSERT(error == 0);
359 #endif
360 
361 #ifdef DIAGNOSTIC
362 	/* paranoia (take this out once namei stuff stabilizes) */
363 	memset(nd.ni_pnbuf, '~', PATH_MAX);
364 #endif
365 
366 	/* check access and type */
367 	if (vp->v_type != VREG) {
368 		error = EACCES;
369 		goto bad1;
370 	}
371 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
372 		goto bad1;
373 
374 	/* get attributes */
375 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
376 		goto bad1;
377 
378 	/* Check mount point */
379 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
380 		error = EACCES;
381 		goto bad1;
382 	}
383 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
384 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
385 
386 	/* try to open it */
387 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
388 		goto bad1;
389 
390 	/* unlock vp, since we need it unlocked from here on out. */
391 	VOP_UNLOCK(vp);
392 
393 #if NVERIEXEC > 0
394 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
395 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
396 	    NULL);
397 	if (error)
398 		goto bad2;
399 #endif /* NVERIEXEC > 0 */
400 
401 #ifdef PAX_SEGVGUARD
402 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
403 	if (error)
404 		goto bad2;
405 #endif /* PAX_SEGVGUARD */
406 
407 	/* now we have the file, get the exec header */
408 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
409 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
410 	if (error)
411 		goto bad2;
412 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
413 
414 	/*
415 	 * Set up default address space limits.  Can be overridden
416 	 * by individual exec packages.
417 	 */
418 	epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
419 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
420 
421 	/*
422 	 * set up the vmcmds for creation of the process
423 	 * address space
424 	 */
425 	error = ENOEXEC;
426 	for (i = 0; i < nexecs; i++) {
427 		int newerror;
428 
429 		epp->ep_esch = execsw[i];
430 		newerror = (*execsw[i]->es_makecmds)(l, epp);
431 
432 		if (!newerror) {
433 			/* Seems ok: check that entry point is not too high */
434 			if (epp->ep_entry > epp->ep_vm_maxaddr) {
435 #ifdef DIAGNOSTIC
436 				printf("%s: rejecting %p due to "
437 				    "too high entry address (> %p)\n",
438 					 __func__, (void *)epp->ep_entry,
439 					 (void *)epp->ep_vm_maxaddr);
440 #endif
441 				error = ENOEXEC;
442 				break;
443 			}
444 			/* Seems ok: check that entry point is not too low */
445 			if (epp->ep_entry < epp->ep_vm_minaddr) {
446 #ifdef DIAGNOSTIC
447 				printf("%s: rejecting %p due to "
448 				    "too low entry address (< %p)\n",
449 				     __func__, (void *)epp->ep_entry,
450 				     (void *)epp->ep_vm_minaddr);
451 #endif
452 				error = ENOEXEC;
453 				break;
454 			}
455 
456 			/* check limits */
457 			if ((epp->ep_tsize > MAXTSIZ) ||
458 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
459 						    [RLIMIT_DATA].rlim_cur)) {
460 #ifdef DIAGNOSTIC
461 				printf("%s: rejecting due to "
462 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
463 				    __func__,
464 				    (unsigned long long)epp->ep_tsize,
465 				    (unsigned long long)MAXTSIZ,
466 				    (unsigned long long)epp->ep_dsize,
467 				    (unsigned long long)
468 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
469 #endif
470 				error = ENOMEM;
471 				break;
472 			}
473 			return 0;
474 		}
475 
476 		/*
477 		 * Reset all the fields that may have been modified by the
478 		 * loader.
479 		 */
480 		KASSERT(epp->ep_emul_arg == NULL);
481 		if (epp->ep_emul_root != NULL) {
482 			vrele(epp->ep_emul_root);
483 			epp->ep_emul_root = NULL;
484 		}
485 		if (epp->ep_interp != NULL) {
486 			vrele(epp->ep_interp);
487 			epp->ep_interp = NULL;
488 		}
489 		epp->ep_pax_flags = 0;
490 
491 		/* make sure the first "interesting" error code is saved. */
492 		if (error == ENOEXEC)
493 			error = newerror;
494 
495 		if (epp->ep_flags & EXEC_DESTR)
496 			/* Error from "#!" code, tidied up by recursive call */
497 			return error;
498 	}
499 
500 	/* not found, error */
501 
502 	/*
503 	 * free any vmspace-creation commands,
504 	 * and release their references
505 	 */
506 	kill_vmcmds(&epp->ep_vmcmds);
507 
508 bad2:
509 	/*
510 	 * close and release the vnode, restore the old one, free the
511 	 * pathname buf, and punt.
512 	 */
513 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
514 	VOP_CLOSE(vp, FREAD, l->l_cred);
515 	vput(vp);
516 	return error;
517 
518 bad1:
519 	/*
520 	 * free the namei pathname buffer, and put the vnode
521 	 * (which we don't yet have open).
522 	 */
523 	vput(vp);				/* was still locked */
524 	return error;
525 }
526 
527 #ifdef __MACHINE_STACK_GROWS_UP
528 #define STACK_PTHREADSPACE NBPG
529 #else
530 #define STACK_PTHREADSPACE 0
531 #endif
532 
533 static int
534 execve_fetch_element(char * const *array, size_t index, char **value)
535 {
536 	return copyin(array + index, value, sizeof(*value));
537 }
538 
539 /*
540  * exec system call
541  */
542 int
543 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
544 {
545 	/* {
546 		syscallarg(const char *)	path;
547 		syscallarg(char * const *)	argp;
548 		syscallarg(char * const *)	envp;
549 	} */
550 
551 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
552 	    SCARG(uap, envp), execve_fetch_element);
553 }
554 
555 int
556 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
557     register_t *retval)
558 {
559 	/* {
560 		syscallarg(int)			fd;
561 		syscallarg(char * const *)	argp;
562 		syscallarg(char * const *)	envp;
563 	} */
564 
565 	return ENOSYS;
566 }
567 
568 /*
569  * Load modules to try and execute an image that we do not understand.
570  * If no execsw entries are present, we load those likely to be needed
571  * in order to run native images only.  Otherwise, we autoload all
572  * possible modules that could let us run the binary.  XXX lame
573  */
574 static void
575 exec_autoload(void)
576 {
577 #ifdef MODULAR
578 	static const char * const native[] = {
579 		"exec_elf32",
580 		"exec_elf64",
581 		"exec_script",
582 		NULL
583 	};
584 	static const char * const compat[] = {
585 		"exec_elf32",
586 		"exec_elf64",
587 		"exec_script",
588 		"exec_aout",
589 		"exec_coff",
590 		"exec_ecoff",
591 		"compat_aoutm68k",
592 		"compat_netbsd32",
593 		"compat_sunos",
594 		"compat_sunos32",
595 		"compat_ultrix",
596 		NULL
597 	};
598 	char const * const *list;
599 	int i;
600 
601 	list = (nexecs == 0 ? native : compat);
602 	for (i = 0; list[i] != NULL; i++) {
603 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
604 			continue;
605 		}
606 		yield();
607 	}
608 #endif
609 }
610 
611 static int
612 makepathbuf(struct lwp *l, const char *upath, struct pathbuf **pbp,
613     size_t *offs)
614 {
615 	char *path, *bp;
616 	size_t len, tlen;
617 	int error;
618 	struct cwdinfo *cwdi;
619 
620 	path = PNBUF_GET();
621 	error = copyinstr(upath, path, MAXPATHLEN, &len);
622 	if (error) {
623 		PNBUF_PUT(path);
624 		DPRINTF(("%s: copyin path @%p %d\n", __func__, upath, error));
625 		return error;
626 	}
627 
628 	if (path[0] == '/') {
629 		*offs = 0;
630 		goto out;
631 	}
632 
633 	len++;
634 	if (len + 1 >= MAXPATHLEN)
635 		goto out;
636 	bp = path + MAXPATHLEN - len;
637 	memmove(bp, path, len);
638 	*(--bp) = '/';
639 
640 	cwdi = l->l_proc->p_cwdi;
641 	rw_enter(&cwdi->cwdi_lock, RW_READER);
642 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
643 	    GETCWD_CHECK_ACCESS, l);
644 	rw_exit(&cwdi->cwdi_lock);
645 
646 	if (error) {
647 		DPRINTF(("%s: getcwd_common path %s %d\n", __func__, path,
648 		    error));
649 		goto out;
650 	}
651 	tlen = path + MAXPATHLEN - bp;
652 
653 	memmove(path, bp, tlen);
654 	path[tlen] = '\0';
655 	*offs = tlen - len;
656 out:
657 	*pbp = pathbuf_assimilate(path);
658 	return 0;
659 }
660 
661 vaddr_t
662 exec_vm_minaddr(vaddr_t va_min)
663 {
664 	/*
665 	 * Increase va_min if we don't want NULL to be mappable by the
666 	 * process.
667 	 */
668 #define VM_MIN_GUARD	PAGE_SIZE
669 	if (user_va0_disable && (va_min < VM_MIN_GUARD))
670 		return VM_MIN_GUARD;
671 	return va_min;
672 }
673 
674 static int
675 execve_loadvm(struct lwp *l, const char *path, char * const *args,
676 	char * const *envs, execve_fetch_element_t fetch_element,
677 	struct execve_data * restrict data)
678 {
679 	struct exec_package	* const epp = &data->ed_pack;
680 	int			error;
681 	struct proc		*p;
682 	char			*dp;
683 	u_int			modgen;
684 	size_t			offs = 0;	// XXX: GCC
685 
686 	KASSERT(data != NULL);
687 
688 	p = l->l_proc;
689 	modgen = 0;
690 
691 	SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
692 
693 	/*
694 	 * Check if we have exceeded our number of processes limit.
695 	 * This is so that we handle the case where a root daemon
696 	 * forked, ran setuid to become the desired user and is trying
697 	 * to exec. The obvious place to do the reference counting check
698 	 * is setuid(), but we don't do the reference counting check there
699 	 * like other OS's do because then all the programs that use setuid()
700 	 * must be modified to check the return code of setuid() and exit().
701 	 * It is dangerous to make setuid() fail, because it fails open and
702 	 * the program will continue to run as root. If we make it succeed
703 	 * and return an error code, again we are not enforcing the limit.
704 	 * The best place to enforce the limit is here, when the process tries
705 	 * to execute a new image, because eventually the process will need
706 	 * to call exec in order to do something useful.
707 	 */
708  retry:
709 	if (p->p_flag & PK_SUGID) {
710 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
711 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
712 		     &p->p_rlimit[RLIMIT_NPROC],
713 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
714 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
715 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
716 		return EAGAIN;
717 	}
718 
719 	/*
720 	 * Drain existing references and forbid new ones.  The process
721 	 * should be left alone until we're done here.  This is necessary
722 	 * to avoid race conditions - e.g. in ptrace() - that might allow
723 	 * a local user to illicitly obtain elevated privileges.
724 	 */
725 	rw_enter(&p->p_reflock, RW_WRITER);
726 
727 	/*
728 	 * Init the namei data to point the file user's program name.
729 	 * This is done here rather than in check_exec(), so that it's
730 	 * possible to override this settings if any of makecmd/probe
731 	 * functions call check_exec() recursively - for example,
732 	 * see exec_script_makecmds().
733 	 */
734 	if ((error = makepathbuf(l, path, &data->ed_pathbuf, &offs)) != 0)
735 		goto clrflg;
736 	data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
737 	data->ed_resolvedpathbuf = PNBUF_GET();
738 
739 	/*
740 	 * initialize the fields of the exec package.
741 	 */
742 	epp->ep_kname = data->ed_pathstring + offs;
743 	epp->ep_resolvedname = data->ed_resolvedpathbuf;
744 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
745 	epp->ep_hdrlen = exec_maxhdrsz;
746 	epp->ep_hdrvalid = 0;
747 	epp->ep_emul_arg = NULL;
748 	epp->ep_emul_arg_free = NULL;
749 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
750 	epp->ep_vap = &data->ed_attr;
751 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
752 	MD_TOPDOWN_INIT(epp);
753 	epp->ep_emul_root = NULL;
754 	epp->ep_interp = NULL;
755 	epp->ep_esch = NULL;
756 	epp->ep_pax_flags = 0;
757 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
758 
759 	rw_enter(&exec_lock, RW_READER);
760 
761 	/* see if we can run it. */
762 	if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
763 		if (error != ENOENT && error != EACCES) {
764 			DPRINTF(("%s: check exec failed for %s, error %d\n",
765 			    __func__, epp->ep_kname, error));
766 		}
767 		goto freehdr;
768 	}
769 
770 	/* allocate an argument buffer */
771 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
772 	KASSERT(data->ed_argp != NULL);
773 	dp = data->ed_argp;
774 
775 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
776 		goto bad;
777 	}
778 
779 	/*
780 	 * Calculate the new stack size.
781 	 */
782 
783 #ifdef __MACHINE_STACK_GROWS_UP
784 /*
785  * copyargs() fills argc/argv/envp from the lower address even on
786  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
787  * so that _rtld() use it.
788  */
789 #define	RTLD_GAP	32
790 #else
791 #define	RTLD_GAP	0
792 #endif
793 
794 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
795 
796 	data->ed_argslen = calcargs(data, argenvstrlen);
797 
798 	const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
799 
800 	if (len > epp->ep_ssize) {
801 		/* in effect, compare to initial limit */
802 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
803 		error = ENOMEM;
804 		goto bad;
805 	}
806 	/* adjust "active stack depth" for process VSZ */
807 	epp->ep_ssize = len;
808 
809 	return 0;
810 
811  bad:
812 	/* free the vmspace-creation commands, and release their references */
813 	kill_vmcmds(&epp->ep_vmcmds);
814 	/* kill any opened file descriptor, if necessary */
815 	if (epp->ep_flags & EXEC_HASFD) {
816 		epp->ep_flags &= ~EXEC_HASFD;
817 		fd_close(epp->ep_fd);
818 	}
819 	/* close and put the exec'd file */
820 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
821 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
822 	vput(epp->ep_vp);
823 	pool_put(&exec_pool, data->ed_argp);
824 
825  freehdr:
826 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
827 	if (epp->ep_emul_root != NULL)
828 		vrele(epp->ep_emul_root);
829 	if (epp->ep_interp != NULL)
830 		vrele(epp->ep_interp);
831 
832 	rw_exit(&exec_lock);
833 
834 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
835 	pathbuf_destroy(data->ed_pathbuf);
836 	PNBUF_PUT(data->ed_resolvedpathbuf);
837 
838  clrflg:
839 	rw_exit(&p->p_reflock);
840 
841 	if (modgen != module_gen && error == ENOEXEC) {
842 		modgen = module_gen;
843 		exec_autoload();
844 		goto retry;
845 	}
846 
847 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
848 	return error;
849 }
850 
851 static int
852 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
853 {
854 	struct exec_package	* const epp = &data->ed_pack;
855 	struct proc		*p = l->l_proc;
856 	struct exec_vmcmd	*base_vcp;
857 	int			error = 0;
858 	size_t			i;
859 
860 	/* record proc's vnode, for use by procfs and others */
861 	if (p->p_textvp)
862 		vrele(p->p_textvp);
863 	vref(epp->ep_vp);
864 	p->p_textvp = epp->ep_vp;
865 
866 	/* create the new process's VM space by running the vmcmds */
867 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
868 
869 #ifdef TRACE_EXEC
870 	DUMPVMCMDS(epp, 0, 0);
871 #endif
872 
873 	base_vcp = NULL;
874 
875 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
876 		struct exec_vmcmd *vcp;
877 
878 		vcp = &epp->ep_vmcmds.evs_cmds[i];
879 		if (vcp->ev_flags & VMCMD_RELATIVE) {
880 			KASSERTMSG(base_vcp != NULL,
881 			    "%s: relative vmcmd with no base", __func__);
882 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
883 			    "%s: illegal base & relative vmcmd", __func__);
884 			vcp->ev_addr += base_vcp->ev_addr;
885 		}
886 		error = (*vcp->ev_proc)(l, vcp);
887 		if (error)
888 			DUMPVMCMDS(epp, i, error);
889 		if (vcp->ev_flags & VMCMD_BASE)
890 			base_vcp = vcp;
891 	}
892 
893 	/* free the vmspace-creation commands, and release their references */
894 	kill_vmcmds(&epp->ep_vmcmds);
895 
896 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
897 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
898 	vput(epp->ep_vp);
899 
900 	/* if an error happened, deallocate and punt */
901 	if (error != 0) {
902 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
903 	}
904 	return error;
905 }
906 
907 static void
908 execve_free_data(struct execve_data *data)
909 {
910 	struct exec_package	* const epp = &data->ed_pack;
911 
912 	/* free the vmspace-creation commands, and release their references */
913 	kill_vmcmds(&epp->ep_vmcmds);
914 	/* kill any opened file descriptor, if necessary */
915 	if (epp->ep_flags & EXEC_HASFD) {
916 		epp->ep_flags &= ~EXEC_HASFD;
917 		fd_close(epp->ep_fd);
918 	}
919 
920 	/* close and put the exec'd file */
921 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
922 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
923 	vput(epp->ep_vp);
924 	pool_put(&exec_pool, data->ed_argp);
925 
926 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
927 	if (epp->ep_emul_root != NULL)
928 		vrele(epp->ep_emul_root);
929 	if (epp->ep_interp != NULL)
930 		vrele(epp->ep_interp);
931 
932 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
933 	pathbuf_destroy(data->ed_pathbuf);
934 	PNBUF_PUT(data->ed_resolvedpathbuf);
935 }
936 
937 static void
938 pathexec(struct proc *p, const char *resolvedname)
939 {
940 	KASSERT(resolvedname[0] == '/');
941 
942 	/* set command name & other accounting info */
943 	strlcpy(p->p_comm, strrchr(resolvedname, '/') + 1, sizeof(p->p_comm));
944 
945 	kmem_strfree(p->p_path);
946 	p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
947 }
948 
949 /* XXX elsewhere */
950 static int
951 credexec(struct lwp *l, struct vattr *attr)
952 {
953 	struct proc *p = l->l_proc;
954 	int error;
955 
956 	/*
957 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
958 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
959 	 * out additional references on the process for the moment.
960 	 */
961 	if ((p->p_slflag & PSL_TRACED) == 0 &&
962 
963 	    (((attr->va_mode & S_ISUID) != 0 &&
964 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
965 
966 	     ((attr->va_mode & S_ISGID) != 0 &&
967 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
968 		/*
969 		 * Mark the process as SUGID before we do
970 		 * anything that might block.
971 		 */
972 		proc_crmod_enter();
973 		proc_crmod_leave(NULL, NULL, true);
974 
975 		/* Make sure file descriptors 0..2 are in use. */
976 		if ((error = fd_checkstd()) != 0) {
977 			DPRINTF(("%s: fdcheckstd failed %d\n",
978 			    __func__, error));
979 			return error;
980 		}
981 
982 		/*
983 		 * Copy the credential so other references don't see our
984 		 * changes.
985 		 */
986 		l->l_cred = kauth_cred_copy(l->l_cred);
987 #ifdef KTRACE
988 		/*
989 		 * If the persistent trace flag isn't set, turn off.
990 		 */
991 		if (p->p_tracep) {
992 			mutex_enter(&ktrace_lock);
993 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
994 				ktrderef(p);
995 			mutex_exit(&ktrace_lock);
996 		}
997 #endif
998 		if (attr->va_mode & S_ISUID)
999 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
1000 		if (attr->va_mode & S_ISGID)
1001 			kauth_cred_setegid(l->l_cred, attr->va_gid);
1002 	} else {
1003 		if (kauth_cred_geteuid(l->l_cred) ==
1004 		    kauth_cred_getuid(l->l_cred) &&
1005 		    kauth_cred_getegid(l->l_cred) ==
1006 		    kauth_cred_getgid(l->l_cred))
1007 			p->p_flag &= ~PK_SUGID;
1008 	}
1009 
1010 	/*
1011 	 * Copy the credential so other references don't see our changes.
1012 	 * Test to see if this is necessary first, since in the common case
1013 	 * we won't need a private reference.
1014 	 */
1015 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1016 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1017 		l->l_cred = kauth_cred_copy(l->l_cred);
1018 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1019 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1020 	}
1021 
1022 	/* Update the master credentials. */
1023 	if (l->l_cred != p->p_cred) {
1024 		kauth_cred_t ocred;
1025 
1026 		kauth_cred_hold(l->l_cred);
1027 		mutex_enter(p->p_lock);
1028 		ocred = p->p_cred;
1029 		p->p_cred = l->l_cred;
1030 		mutex_exit(p->p_lock);
1031 		kauth_cred_free(ocred);
1032 	}
1033 
1034 	return 0;
1035 }
1036 
1037 static void
1038 emulexec(struct lwp *l, struct exec_package *epp)
1039 {
1040 	struct proc		*p = l->l_proc;
1041 
1042 	/* The emulation root will usually have been found when we looked
1043 	 * for the elf interpreter (or similar), if not look now. */
1044 	if (epp->ep_esch->es_emul->e_path != NULL &&
1045 	    epp->ep_emul_root == NULL)
1046 		emul_find_root(l, epp);
1047 
1048 	/* Any old emulation root got removed by fdcloseexec */
1049 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1050 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1051 	rw_exit(&p->p_cwdi->cwdi_lock);
1052 	epp->ep_emul_root = NULL;
1053 	if (epp->ep_interp != NULL)
1054 		vrele(epp->ep_interp);
1055 
1056 	/*
1057 	 * Call emulation specific exec hook. This can setup per-process
1058 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1059 	 *
1060 	 * If we are executing process of different emulation than the
1061 	 * original forked process, call e_proc_exit() of the old emulation
1062 	 * first, then e_proc_exec() of new emulation. If the emulation is
1063 	 * same, the exec hook code should deallocate any old emulation
1064 	 * resources held previously by this process.
1065 	 */
1066 	if (p->p_emul && p->p_emul->e_proc_exit
1067 	    && p->p_emul != epp->ep_esch->es_emul)
1068 		(*p->p_emul->e_proc_exit)(p);
1069 
1070 	/*
1071 	 * This is now LWP 1.
1072 	 */
1073 	/* XXX elsewhere */
1074 	mutex_enter(p->p_lock);
1075 	p->p_nlwpid = 1;
1076 	l->l_lid = 1;
1077 	mutex_exit(p->p_lock);
1078 
1079 	/*
1080 	 * Call exec hook. Emulation code may NOT store reference to anything
1081 	 * from &pack.
1082 	 */
1083 	if (epp->ep_esch->es_emul->e_proc_exec)
1084 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1085 
1086 	/* update p_emul, the old value is no longer needed */
1087 	p->p_emul = epp->ep_esch->es_emul;
1088 
1089 	/* ...and the same for p_execsw */
1090 	p->p_execsw = epp->ep_esch;
1091 
1092 #ifdef __HAVE_SYSCALL_INTERN
1093 	(*p->p_emul->e_syscall_intern)(p);
1094 #endif
1095 	ktremul();
1096 }
1097 
1098 static int
1099 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1100 	bool no_local_exec_lock, bool is_spawn)
1101 {
1102 	struct exec_package	* const epp = &data->ed_pack;
1103 	int error = 0;
1104 	struct proc		*p;
1105 
1106 	/*
1107 	 * In case of a posix_spawn operation, the child doing the exec
1108 	 * might not hold the reader lock on exec_lock, but the parent
1109 	 * will do this instead.
1110 	 */
1111 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1112 	KASSERT(!no_local_exec_lock || is_spawn);
1113 	KASSERT(data != NULL);
1114 
1115 	p = l->l_proc;
1116 
1117 	/* Get rid of other LWPs. */
1118 	if (p->p_nlwps > 1) {
1119 		mutex_enter(p->p_lock);
1120 		exit_lwps(l);
1121 		mutex_exit(p->p_lock);
1122 	}
1123 	KDASSERT(p->p_nlwps == 1);
1124 
1125 	/* Destroy any lwpctl info. */
1126 	if (p->p_lwpctl != NULL)
1127 		lwp_ctl_exit();
1128 
1129 	/* Remove POSIX timers */
1130 	timers_free(p, TIMERS_POSIX);
1131 
1132 	/* Set the PaX flags. */
1133 	pax_set_flags(epp, p);
1134 
1135 	/*
1136 	 * Do whatever is necessary to prepare the address space
1137 	 * for remapping.  Note that this might replace the current
1138 	 * vmspace with another!
1139 	 */
1140 	if (is_spawn)
1141 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1142 		    epp->ep_vm_maxaddr,
1143 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1144 	else
1145 		uvmspace_exec(l, epp->ep_vm_minaddr,
1146 		    epp->ep_vm_maxaddr,
1147 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1148 
1149 	struct vmspace		*vm;
1150 	vm = p->p_vmspace;
1151 	vm->vm_taddr = (void *)epp->ep_taddr;
1152 	vm->vm_tsize = btoc(epp->ep_tsize);
1153 	vm->vm_daddr = (void*)epp->ep_daddr;
1154 	vm->vm_dsize = btoc(epp->ep_dsize);
1155 	vm->vm_ssize = btoc(epp->ep_ssize);
1156 	vm->vm_issize = 0;
1157 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1158 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1159 
1160 	pax_aslr_init_vm(l, vm, epp);
1161 
1162 	/* Now map address space. */
1163 	error = execve_dovmcmds(l, data);
1164 	if (error != 0)
1165 		goto exec_abort;
1166 
1167 	pathexec(p, epp->ep_resolvedname);
1168 
1169 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1170 
1171 	error = copyoutargs(data, l, newstack);
1172 	if (error != 0)
1173 		goto exec_abort;
1174 
1175 	cwdexec(p);
1176 	fd_closeexec();		/* handle close on exec */
1177 
1178 	if (__predict_false(ktrace_on))
1179 		fd_ktrexecfd();
1180 
1181 	execsigs(p);		/* reset caught signals */
1182 
1183 	mutex_enter(p->p_lock);
1184 	l->l_ctxlink = NULL;	/* reset ucontext link */
1185 	p->p_acflag &= ~AFORK;
1186 	p->p_flag |= PK_EXEC;
1187 	mutex_exit(p->p_lock);
1188 
1189 	/*
1190 	 * Stop profiling.
1191 	 */
1192 	if ((p->p_stflag & PST_PROFIL) != 0) {
1193 		mutex_spin_enter(&p->p_stmutex);
1194 		stopprofclock(p);
1195 		mutex_spin_exit(&p->p_stmutex);
1196 	}
1197 
1198 	/*
1199 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1200 	 * exited and exec()/exit() are the only places it will be cleared.
1201 	 */
1202 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1203 		mutex_enter(proc_lock);
1204 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1205 		p->p_lflag &= ~PL_PPWAIT;
1206 		cv_broadcast(&p->p_pptr->p_waitcv);
1207 		mutex_exit(proc_lock);
1208 	}
1209 
1210 	error = credexec(l, &data->ed_attr);
1211 	if (error)
1212 		goto exec_abort;
1213 
1214 #if defined(__HAVE_RAS)
1215 	/*
1216 	 * Remove all RASs from the address space.
1217 	 */
1218 	ras_purgeall();
1219 #endif
1220 
1221 	doexechooks(p);
1222 
1223 	/*
1224 	 * Set initial SP at the top of the stack.
1225 	 *
1226 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1227 	 * the end of arg/env strings.  Userland guesses the address of argc
1228 	 * via ps_strings::ps_argvstr.
1229 	 */
1230 
1231 	/* Setup new registers and do misc. setup. */
1232 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1233 	if (epp->ep_esch->es_setregs)
1234 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1235 
1236 	/* Provide a consistent LWP private setting */
1237 	(void)lwp_setprivate(l, NULL);
1238 
1239 	/* Discard all PCU state; need to start fresh */
1240 	pcu_discard_all(l);
1241 
1242 	/* map the process's signal trampoline code */
1243 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1244 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1245 		goto exec_abort;
1246 	}
1247 
1248 	pool_put(&exec_pool, data->ed_argp);
1249 
1250 	/* notify others that we exec'd */
1251 	KNOTE(&p->p_klist, NOTE_EXEC);
1252 
1253 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1254 
1255 	SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1256 
1257 	emulexec(l, epp);
1258 
1259 	/* Allow new references from the debugger/procfs. */
1260 	rw_exit(&p->p_reflock);
1261 	if (!no_local_exec_lock)
1262 		rw_exit(&exec_lock);
1263 
1264 	mutex_enter(proc_lock);
1265 
1266 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1267 		ksiginfo_t ksi;
1268 
1269 		KSI_INIT_EMPTY(&ksi);
1270 		ksi.ksi_signo = SIGTRAP;
1271 		ksi.ksi_code = TRAP_EXEC;
1272 		ksi.ksi_lid = l->l_lid;
1273 		kpsignal(p, &ksi, NULL);
1274 	}
1275 
1276 	if (p->p_sflag & PS_STOPEXEC) {
1277 		ksiginfoq_t kq;
1278 
1279 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1280 		p->p_pptr->p_nstopchild++;
1281 		p->p_waited = 0;
1282 		mutex_enter(p->p_lock);
1283 		ksiginfo_queue_init(&kq);
1284 		sigclearall(p, &contsigmask, &kq);
1285 		lwp_lock(l);
1286 		l->l_stat = LSSTOP;
1287 		p->p_stat = SSTOP;
1288 		p->p_nrlwps--;
1289 		lwp_unlock(l);
1290 		mutex_exit(p->p_lock);
1291 		mutex_exit(proc_lock);
1292 		lwp_lock(l);
1293 		mi_switch(l);
1294 		ksiginfo_queue_drain(&kq);
1295 		KERNEL_LOCK(l->l_biglocks, l);
1296 	} else {
1297 		mutex_exit(proc_lock);
1298 	}
1299 
1300 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1301 	pathbuf_destroy(data->ed_pathbuf);
1302 	PNBUF_PUT(data->ed_resolvedpathbuf);
1303 #ifdef TRACE_EXEC
1304 	DPRINTF(("%s finished\n", __func__));
1305 #endif
1306 	return EJUSTRETURN;
1307 
1308  exec_abort:
1309 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1310 	rw_exit(&p->p_reflock);
1311 	if (!no_local_exec_lock)
1312 		rw_exit(&exec_lock);
1313 
1314 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1315 	pathbuf_destroy(data->ed_pathbuf);
1316 	PNBUF_PUT(data->ed_resolvedpathbuf);
1317 
1318 	/*
1319 	 * the old process doesn't exist anymore.  exit gracefully.
1320 	 * get rid of the (new) address space we have created, if any, get rid
1321 	 * of our namei data and vnode, and exit noting failure
1322 	 */
1323 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1324 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1325 
1326 	exec_free_emul_arg(epp);
1327 	pool_put(&exec_pool, data->ed_argp);
1328 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1329 	if (epp->ep_emul_root != NULL)
1330 		vrele(epp->ep_emul_root);
1331 	if (epp->ep_interp != NULL)
1332 		vrele(epp->ep_interp);
1333 
1334 	/* Acquire the sched-state mutex (exit1() will release it). */
1335 	if (!is_spawn) {
1336 		mutex_enter(p->p_lock);
1337 		exit1(l, error, SIGABRT);
1338 	}
1339 
1340 	return error;
1341 }
1342 
1343 int
1344 execve1(struct lwp *l, const char *path, char * const *args,
1345     char * const *envs, execve_fetch_element_t fetch_element)
1346 {
1347 	struct execve_data data;
1348 	int error;
1349 
1350 	error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1351 	if (error)
1352 		return error;
1353 	error = execve_runproc(l, &data, false, false);
1354 	return error;
1355 }
1356 
1357 static size_t
1358 fromptrsz(const struct exec_package *epp)
1359 {
1360 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1361 }
1362 
1363 static size_t
1364 ptrsz(const struct exec_package *epp)
1365 {
1366 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1367 }
1368 
1369 static size_t
1370 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1371 {
1372 	struct exec_package	* const epp = &data->ed_pack;
1373 
1374 	const size_t nargenvptrs =
1375 	    1 +				/* long argc */
1376 	    data->ed_argc +		/* char *argv[] */
1377 	    1 +				/* \0 */
1378 	    data->ed_envc +		/* char *env[] */
1379 	    1;				/* \0 */
1380 
1381 	return (nargenvptrs * ptrsz(epp))	/* pointers */
1382 	    + argenvstrlen			/* strings */
1383 	    + epp->ep_esch->es_arglen;		/* auxinfo */
1384 }
1385 
1386 static size_t
1387 calcstack(struct execve_data * restrict data, const size_t gaplen)
1388 {
1389 	struct exec_package	* const epp = &data->ed_pack;
1390 
1391 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1392 	    epp->ep_esch->es_emul->e_sigcode;
1393 
1394 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1395 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1396 
1397 	const size_t sigcode_psstr_sz =
1398 	    data->ed_szsigcode +	/* sigcode */
1399 	    data->ed_ps_strings_sz +	/* ps_strings */
1400 	    STACK_PTHREADSPACE;		/* pthread space */
1401 
1402 	const size_t stacklen =
1403 	    data->ed_argslen +
1404 	    gaplen +
1405 	    sigcode_psstr_sz;
1406 
1407 	/* make the stack "safely" aligned */
1408 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1409 }
1410 
1411 static int
1412 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1413     char * const newstack)
1414 {
1415 	struct exec_package	* const epp = &data->ed_pack;
1416 	struct proc		*p = l->l_proc;
1417 	int			error;
1418 
1419 	/* remember information about the process */
1420 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1421 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1422 
1423 	/*
1424 	 * Allocate the stack address passed to the newly execve()'ed process.
1425 	 *
1426 	 * The new stack address will be set to the SP (stack pointer) register
1427 	 * in setregs().
1428 	 */
1429 
1430 	char *newargs = STACK_ALLOC(
1431 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1432 
1433 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1434 	    &data->ed_arginfo, &newargs, data->ed_argp);
1435 
1436 	if (error) {
1437 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1438 		return error;
1439 	}
1440 
1441 	error = copyoutpsstrs(data, p);
1442 	if (error != 0)
1443 		return error;
1444 
1445 	return 0;
1446 }
1447 
1448 static int
1449 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1450 {
1451 	struct exec_package	* const epp = &data->ed_pack;
1452 	struct ps_strings32	arginfo32;
1453 	void			*aip;
1454 	int			error;
1455 
1456 	/* fill process ps_strings info */
1457 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1458 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1459 
1460 	if (epp->ep_flags & EXEC_32) {
1461 		aip = &arginfo32;
1462 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1463 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1464 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1465 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1466 	} else
1467 		aip = &data->ed_arginfo;
1468 
1469 	/* copy out the process's ps_strings structure */
1470 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1471 	    != 0) {
1472 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1473 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1474 		return error;
1475 	}
1476 
1477 	return 0;
1478 }
1479 
1480 static int
1481 copyinargs(struct execve_data * restrict data, char * const *args,
1482     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1483 {
1484 	struct exec_package	* const epp = &data->ed_pack;
1485 	char			*dp;
1486 	size_t			i;
1487 	int			error;
1488 
1489 	dp = *dpp;
1490 
1491 	data->ed_argc = 0;
1492 
1493 	/* copy the fake args list, if there's one, freeing it as we go */
1494 	if (epp->ep_flags & EXEC_HASARGL) {
1495 		struct exec_fakearg	*fa = epp->ep_fa;
1496 
1497 		while (fa->fa_arg != NULL) {
1498 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1499 			size_t len;
1500 
1501 			len = strlcpy(dp, fa->fa_arg, maxlen);
1502 			/* Count NUL into len. */
1503 			if (len < maxlen)
1504 				len++;
1505 			else {
1506 				while (fa->fa_arg != NULL) {
1507 					kmem_free(fa->fa_arg, fa->fa_len);
1508 					fa++;
1509 				}
1510 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1511 				epp->ep_flags &= ~EXEC_HASARGL;
1512 				return E2BIG;
1513 			}
1514 			ktrexecarg(fa->fa_arg, len - 1);
1515 			dp += len;
1516 
1517 			kmem_free(fa->fa_arg, fa->fa_len);
1518 			fa++;
1519 			data->ed_argc++;
1520 		}
1521 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1522 		epp->ep_flags &= ~EXEC_HASARGL;
1523 	}
1524 
1525 	/*
1526 	 * Read and count argument strings from user.
1527 	 */
1528 
1529 	if (args == NULL) {
1530 		DPRINTF(("%s: null args\n", __func__));
1531 		return EINVAL;
1532 	}
1533 	if (epp->ep_flags & EXEC_SKIPARG)
1534 		args = (const void *)((const char *)args + fromptrsz(epp));
1535 	i = 0;
1536 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1537 	if (error != 0) {
1538 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1539 		return error;
1540 	}
1541 	data->ed_argc += i;
1542 
1543 	/*
1544 	 * Read and count environment strings from user.
1545 	 */
1546 
1547 	data->ed_envc = 0;
1548 	/* environment need not be there */
1549 	if (envs == NULL)
1550 		goto done;
1551 	i = 0;
1552 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1553 	if (error != 0) {
1554 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1555 		return error;
1556 	}
1557 	data->ed_envc += i;
1558 
1559 done:
1560 	*dpp = dp;
1561 
1562 	return 0;
1563 }
1564 
1565 static int
1566 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1567     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1568     void (*ktr)(const void *, size_t))
1569 {
1570 	char			*dp, *sp;
1571 	size_t			i;
1572 	int			error;
1573 
1574 	dp = *dpp;
1575 
1576 	i = 0;
1577 	while (1) {
1578 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1579 		size_t len;
1580 
1581 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1582 			return error;
1583 		}
1584 		if (!sp)
1585 			break;
1586 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1587 			if (error == ENAMETOOLONG)
1588 				error = E2BIG;
1589 			return error;
1590 		}
1591 		if (__predict_false(ktrace_on))
1592 			(*ktr)(dp, len - 1);
1593 		dp += len;
1594 		i++;
1595 	}
1596 
1597 	*dpp = dp;
1598 	*ip = i;
1599 
1600 	return 0;
1601 }
1602 
1603 /*
1604  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1605  * Those strings are located just after auxinfo.
1606  */
1607 int
1608 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1609     char **stackp, void *argp)
1610 {
1611 	char	**cpp, *dp, *sp;
1612 	size_t	len;
1613 	void	*nullp;
1614 	long	argc, envc;
1615 	int	error;
1616 
1617 	cpp = (char **)*stackp;
1618 	nullp = NULL;
1619 	argc = arginfo->ps_nargvstr;
1620 	envc = arginfo->ps_nenvstr;
1621 
1622 	/* argc on stack is long */
1623 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1624 
1625 	dp = (char *)(cpp +
1626 	    1 +				/* long argc */
1627 	    argc +			/* char *argv[] */
1628 	    1 +				/* \0 */
1629 	    envc +			/* char *env[] */
1630 	    1) +			/* \0 */
1631 	    pack->ep_esch->es_arglen;	/* auxinfo */
1632 	sp = argp;
1633 
1634 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1635 		COPYPRINTF("", cpp - 1, sizeof(argc));
1636 		return error;
1637 	}
1638 
1639 	/* XXX don't copy them out, remap them! */
1640 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1641 
1642 	for (; --argc >= 0; sp += len, dp += len) {
1643 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1644 			COPYPRINTF("", cpp - 1, sizeof(dp));
1645 			return error;
1646 		}
1647 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1648 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1649 			return error;
1650 		}
1651 	}
1652 
1653 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1654 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1655 		return error;
1656 	}
1657 
1658 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1659 
1660 	for (; --envc >= 0; sp += len, dp += len) {
1661 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1662 			COPYPRINTF("", cpp - 1, sizeof(dp));
1663 			return error;
1664 		}
1665 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1666 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1667 			return error;
1668 		}
1669 
1670 	}
1671 
1672 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1673 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1674 		return error;
1675 	}
1676 
1677 	*stackp = (char *)cpp;
1678 	return 0;
1679 }
1680 
1681 
1682 /*
1683  * Add execsw[] entries.
1684  */
1685 int
1686 exec_add(struct execsw *esp, int count)
1687 {
1688 	struct exec_entry	*it;
1689 	int			i;
1690 
1691 	if (count == 0) {
1692 		return 0;
1693 	}
1694 
1695 	/* Check for duplicates. */
1696 	rw_enter(&exec_lock, RW_WRITER);
1697 	for (i = 0; i < count; i++) {
1698 		LIST_FOREACH(it, &ex_head, ex_list) {
1699 			/* assume unique (makecmds, probe_func, emulation) */
1700 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1701 			    it->ex_sw->u.elf_probe_func ==
1702 			    esp[i].u.elf_probe_func &&
1703 			    it->ex_sw->es_emul == esp[i].es_emul) {
1704 				rw_exit(&exec_lock);
1705 				return EEXIST;
1706 			}
1707 		}
1708 	}
1709 
1710 	/* Allocate new entries. */
1711 	for (i = 0; i < count; i++) {
1712 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1713 		it->ex_sw = &esp[i];
1714 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1715 	}
1716 
1717 	/* update execsw[] */
1718 	exec_init(0);
1719 	rw_exit(&exec_lock);
1720 	return 0;
1721 }
1722 
1723 /*
1724  * Remove execsw[] entry.
1725  */
1726 int
1727 exec_remove(struct execsw *esp, int count)
1728 {
1729 	struct exec_entry	*it, *next;
1730 	int			i;
1731 	const struct proclist_desc *pd;
1732 	proc_t			*p;
1733 
1734 	if (count == 0) {
1735 		return 0;
1736 	}
1737 
1738 	/* Abort if any are busy. */
1739 	rw_enter(&exec_lock, RW_WRITER);
1740 	for (i = 0; i < count; i++) {
1741 		mutex_enter(proc_lock);
1742 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1743 			PROCLIST_FOREACH(p, pd->pd_list) {
1744 				if (p->p_execsw == &esp[i]) {
1745 					mutex_exit(proc_lock);
1746 					rw_exit(&exec_lock);
1747 					return EBUSY;
1748 				}
1749 			}
1750 		}
1751 		mutex_exit(proc_lock);
1752 	}
1753 
1754 	/* None are busy, so remove them all. */
1755 	for (i = 0; i < count; i++) {
1756 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1757 			next = LIST_NEXT(it, ex_list);
1758 			if (it->ex_sw == &esp[i]) {
1759 				LIST_REMOVE(it, ex_list);
1760 				kmem_free(it, sizeof(*it));
1761 				break;
1762 			}
1763 		}
1764 	}
1765 
1766 	/* update execsw[] */
1767 	exec_init(0);
1768 	rw_exit(&exec_lock);
1769 	return 0;
1770 }
1771 
1772 /*
1773  * Initialize exec structures. If init_boot is true, also does necessary
1774  * one-time initialization (it's called from main() that way).
1775  * Once system is multiuser, this should be called with exec_lock held,
1776  * i.e. via exec_{add|remove}().
1777  */
1778 int
1779 exec_init(int init_boot)
1780 {
1781 	const struct execsw 	**sw;
1782 	struct exec_entry	*ex;
1783 	SLIST_HEAD(,exec_entry)	first;
1784 	SLIST_HEAD(,exec_entry)	any;
1785 	SLIST_HEAD(,exec_entry)	last;
1786 	int			i, sz;
1787 
1788 	if (init_boot) {
1789 		/* do one-time initializations */
1790 		vaddr_t vmin = 0, vmax;
1791 
1792 		rw_init(&exec_lock);
1793 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1794 		exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1795 		    maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1796 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1797 		    "execargs", &exec_palloc, IPL_NONE);
1798 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1799 	} else {
1800 		KASSERT(rw_write_held(&exec_lock));
1801 	}
1802 
1803 	/* Sort each entry onto the appropriate queue. */
1804 	SLIST_INIT(&first);
1805 	SLIST_INIT(&any);
1806 	SLIST_INIT(&last);
1807 	sz = 0;
1808 	LIST_FOREACH(ex, &ex_head, ex_list) {
1809 		switch(ex->ex_sw->es_prio) {
1810 		case EXECSW_PRIO_FIRST:
1811 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1812 			break;
1813 		case EXECSW_PRIO_ANY:
1814 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1815 			break;
1816 		case EXECSW_PRIO_LAST:
1817 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1818 			break;
1819 		default:
1820 			panic("%s", __func__);
1821 			break;
1822 		}
1823 		sz++;
1824 	}
1825 
1826 	/*
1827 	 * Create new execsw[].  Ensure we do not try a zero-sized
1828 	 * allocation.
1829 	 */
1830 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1831 	i = 0;
1832 	SLIST_FOREACH(ex, &first, ex_slist) {
1833 		sw[i++] = ex->ex_sw;
1834 	}
1835 	SLIST_FOREACH(ex, &any, ex_slist) {
1836 		sw[i++] = ex->ex_sw;
1837 	}
1838 	SLIST_FOREACH(ex, &last, ex_slist) {
1839 		sw[i++] = ex->ex_sw;
1840 	}
1841 
1842 	/* Replace old execsw[] and free used memory. */
1843 	if (execsw != NULL) {
1844 		kmem_free(__UNCONST(execsw),
1845 		    nexecs * sizeof(struct execsw *) + 1);
1846 	}
1847 	execsw = sw;
1848 	nexecs = sz;
1849 
1850 	/* Figure out the maximum size of an exec header. */
1851 	exec_maxhdrsz = sizeof(int);
1852 	for (i = 0; i < nexecs; i++) {
1853 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1854 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1855 	}
1856 
1857 	return 0;
1858 }
1859 
1860 static int
1861 exec_sigcode_map(struct proc *p, const struct emul *e)
1862 {
1863 	vaddr_t va;
1864 	vsize_t sz;
1865 	int error;
1866 	struct uvm_object *uobj;
1867 
1868 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1869 
1870 	if (e->e_sigobject == NULL || sz == 0) {
1871 		return 0;
1872 	}
1873 
1874 	/*
1875 	 * If we don't have a sigobject for this emulation, create one.
1876 	 *
1877 	 * sigobject is an anonymous memory object (just like SYSV shared
1878 	 * memory) that we keep a permanent reference to and that we map
1879 	 * in all processes that need this sigcode. The creation is simple,
1880 	 * we create an object, add a permanent reference to it, map it in
1881 	 * kernel space, copy out the sigcode to it and unmap it.
1882 	 * We map it with PROT_READ|PROT_EXEC into the process just
1883 	 * the way sys_mmap() would map it.
1884 	 */
1885 
1886 	uobj = *e->e_sigobject;
1887 	if (uobj == NULL) {
1888 		mutex_enter(&sigobject_lock);
1889 		if ((uobj = *e->e_sigobject) == NULL) {
1890 			uobj = uao_create(sz, 0);
1891 			(*uobj->pgops->pgo_reference)(uobj);
1892 			va = vm_map_min(kernel_map);
1893 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1894 			    uobj, 0, 0,
1895 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1896 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1897 				printf("kernel mapping failed %d\n", error);
1898 				(*uobj->pgops->pgo_detach)(uobj);
1899 				mutex_exit(&sigobject_lock);
1900 				return error;
1901 			}
1902 			memcpy((void *)va, e->e_sigcode, sz);
1903 #ifdef PMAP_NEED_PROCWR
1904 			pmap_procwr(&proc0, va, sz);
1905 #endif
1906 			uvm_unmap(kernel_map, va, va + round_page(sz));
1907 			*e->e_sigobject = uobj;
1908 		}
1909 		mutex_exit(&sigobject_lock);
1910 	}
1911 
1912 	/* Just a hint to uvm_map where to put it. */
1913 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1914 	    round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
1915 
1916 #ifdef __alpha__
1917 	/*
1918 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1919 	 * which causes the above calculation to put the sigcode at
1920 	 * an invalid address.  Put it just below the text instead.
1921 	 */
1922 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1923 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1924 	}
1925 #endif
1926 
1927 	(*uobj->pgops->pgo_reference)(uobj);
1928 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1929 			uobj, 0, 0,
1930 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1931 				    UVM_ADV_RANDOM, 0));
1932 	if (error) {
1933 		DPRINTF(("%s, %d: map %p "
1934 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1935 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1936 		    va, error));
1937 		(*uobj->pgops->pgo_detach)(uobj);
1938 		return error;
1939 	}
1940 	p->p_sigctx.ps_sigcode = (void *)va;
1941 	return 0;
1942 }
1943 
1944 /*
1945  * Release a refcount on spawn_exec_data and destroy memory, if this
1946  * was the last one.
1947  */
1948 static void
1949 spawn_exec_data_release(struct spawn_exec_data *data)
1950 {
1951 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1952 		return;
1953 
1954 	cv_destroy(&data->sed_cv_child_ready);
1955 	mutex_destroy(&data->sed_mtx_child);
1956 
1957 	if (data->sed_actions)
1958 		posix_spawn_fa_free(data->sed_actions,
1959 		    data->sed_actions->len);
1960 	if (data->sed_attrs)
1961 		kmem_free(data->sed_attrs,
1962 		    sizeof(*data->sed_attrs));
1963 	kmem_free(data, sizeof(*data));
1964 }
1965 
1966 /*
1967  * A child lwp of a posix_spawn operation starts here and ends up in
1968  * cpu_spawn_return, dealing with all filedescriptor and scheduler
1969  * manipulations in between.
1970  * The parent waits for the child, as it is not clear whether the child
1971  * will be able to acquire its own exec_lock. If it can, the parent can
1972  * be released early and continue running in parallel. If not (or if the
1973  * magic debug flag is passed in the scheduler attribute struct), the
1974  * child rides on the parent's exec lock until it is ready to return to
1975  * to userland - and only then releases the parent. This method loses
1976  * concurrency, but improves error reporting.
1977  */
1978 static void
1979 spawn_return(void *arg)
1980 {
1981 	struct spawn_exec_data *spawn_data = arg;
1982 	struct lwp *l = curlwp;
1983 	int error, newfd;
1984 	int ostat;
1985 	size_t i;
1986 	const struct posix_spawn_file_actions_entry *fae;
1987 	pid_t ppid;
1988 	register_t retval;
1989 	bool have_reflock;
1990 	bool parent_is_waiting = true;
1991 
1992 	/*
1993 	 * Check if we can release parent early.
1994 	 * We either need to have no sed_attrs, or sed_attrs does not
1995 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
1996 	 * safe access to the parent proc (passed in sed_parent).
1997 	 * We then try to get the exec_lock, and only if that works, we can
1998 	 * release the parent here already.
1999 	 */
2000 	ppid = spawn_data->sed_parent->p_pid;
2001 	if ((!spawn_data->sed_attrs
2002 	    || (spawn_data->sed_attrs->sa_flags
2003 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2004 	    && rw_tryenter(&exec_lock, RW_READER)) {
2005 		parent_is_waiting = false;
2006 		mutex_enter(&spawn_data->sed_mtx_child);
2007 		cv_signal(&spawn_data->sed_cv_child_ready);
2008 		mutex_exit(&spawn_data->sed_mtx_child);
2009 	}
2010 
2011 	/* don't allow debugger access yet */
2012 	rw_enter(&l->l_proc->p_reflock, RW_WRITER);
2013 	have_reflock = true;
2014 
2015 	error = 0;
2016 	/* handle posix_spawn_file_actions */
2017 	if (spawn_data->sed_actions != NULL) {
2018 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
2019 			fae = &spawn_data->sed_actions->fae[i];
2020 			switch (fae->fae_action) {
2021 			case FAE_OPEN:
2022 				if (fd_getfile(fae->fae_fildes) != NULL) {
2023 					error = fd_close(fae->fae_fildes);
2024 					if (error)
2025 						break;
2026 				}
2027 				error = fd_open(fae->fae_path, fae->fae_oflag,
2028 				    fae->fae_mode, &newfd);
2029 				if (error)
2030 					break;
2031 				if (newfd != fae->fae_fildes) {
2032 					error = dodup(l, newfd,
2033 					    fae->fae_fildes, 0, &retval);
2034 					if (fd_getfile(newfd) != NULL)
2035 						fd_close(newfd);
2036 				}
2037 				break;
2038 			case FAE_DUP2:
2039 				error = dodup(l, fae->fae_fildes,
2040 				    fae->fae_newfildes, 0, &retval);
2041 				break;
2042 			case FAE_CLOSE:
2043 				if (fd_getfile(fae->fae_fildes) == NULL) {
2044 					error = EBADF;
2045 					break;
2046 				}
2047 				error = fd_close(fae->fae_fildes);
2048 				break;
2049 			}
2050 			if (error)
2051 				goto report_error;
2052 		}
2053 	}
2054 
2055 	/* handle posix_spawnattr */
2056 	if (spawn_data->sed_attrs != NULL) {
2057 		struct sigaction sigact;
2058 		sigact._sa_u._sa_handler = SIG_DFL;
2059 		sigact.sa_flags = 0;
2060 
2061 		/*
2062 		 * set state to SSTOP so that this proc can be found by pid.
2063 		 * see proc_enterprp, do_sched_setparam below
2064 		 */
2065 		mutex_enter(proc_lock);
2066 		/*
2067 		 * p_stat should be SACTIVE, so we need to adjust the
2068 		 * parent's p_nstopchild here.  For safety, just make
2069 		 * we're on the good side of SDEAD before we adjust.
2070 		 */
2071 		ostat = l->l_proc->p_stat;
2072 		KASSERT(ostat < SSTOP);
2073 		l->l_proc->p_stat = SSTOP;
2074 		l->l_proc->p_waited = 0;
2075 		l->l_proc->p_pptr->p_nstopchild++;
2076 		mutex_exit(proc_lock);
2077 
2078 		/* Set process group */
2079 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2080 			pid_t mypid = l->l_proc->p_pid,
2081 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
2082 
2083 			if (pgrp == 0)
2084 				pgrp = mypid;
2085 
2086 			error = proc_enterpgrp(spawn_data->sed_parent,
2087 			    mypid, pgrp, false);
2088 			if (error)
2089 				goto report_error_stopped;
2090 		}
2091 
2092 		/* Set scheduler policy */
2093 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2094 			error = do_sched_setparam(l->l_proc->p_pid, 0,
2095 			    spawn_data->sed_attrs->sa_schedpolicy,
2096 			    &spawn_data->sed_attrs->sa_schedparam);
2097 		else if (spawn_data->sed_attrs->sa_flags
2098 		    & POSIX_SPAWN_SETSCHEDPARAM) {
2099 			error = do_sched_setparam(ppid, 0,
2100 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2101 		}
2102 		if (error)
2103 			goto report_error_stopped;
2104 
2105 		/* Reset user ID's */
2106 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2107 			error = do_setresuid(l, -1,
2108 			     kauth_cred_getgid(l->l_cred), -1,
2109 			     ID_E_EQ_R | ID_E_EQ_S);
2110 			if (error)
2111 				goto report_error_stopped;
2112 			error = do_setresuid(l, -1,
2113 			    kauth_cred_getuid(l->l_cred), -1,
2114 			    ID_E_EQ_R | ID_E_EQ_S);
2115 			if (error)
2116 				goto report_error_stopped;
2117 		}
2118 
2119 		/* Set signal masks/defaults */
2120 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2121 			mutex_enter(l->l_proc->p_lock);
2122 			error = sigprocmask1(l, SIG_SETMASK,
2123 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
2124 			mutex_exit(l->l_proc->p_lock);
2125 			if (error)
2126 				goto report_error_stopped;
2127 		}
2128 
2129 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2130 			/*
2131 			 * The following sigaction call is using a sigaction
2132 			 * version 0 trampoline which is in the compatibility
2133 			 * code only. This is not a problem because for SIG_DFL
2134 			 * and SIG_IGN, the trampolines are now ignored. If they
2135 			 * were not, this would be a problem because we are
2136 			 * holding the exec_lock, and the compat code needs
2137 			 * to do the same in order to replace the trampoline
2138 			 * code of the process.
2139 			 */
2140 			for (i = 1; i <= NSIG; i++) {
2141 				if (sigismember(
2142 				    &spawn_data->sed_attrs->sa_sigdefault, i))
2143 					sigaction1(l, i, &sigact, NULL, NULL,
2144 					    0);
2145 			}
2146 		}
2147 		mutex_enter(proc_lock);
2148 		l->l_proc->p_stat = ostat;
2149 		l->l_proc->p_pptr->p_nstopchild--;
2150 		mutex_exit(proc_lock);
2151 	}
2152 
2153 	/* now do the real exec */
2154 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2155 	    true);
2156 	have_reflock = false;
2157 	if (error == EJUSTRETURN)
2158 		error = 0;
2159 	else if (error)
2160 		goto report_error;
2161 
2162 	if (parent_is_waiting) {
2163 		mutex_enter(&spawn_data->sed_mtx_child);
2164 		cv_signal(&spawn_data->sed_cv_child_ready);
2165 		mutex_exit(&spawn_data->sed_mtx_child);
2166 	}
2167 
2168 	/* release our refcount on the data */
2169 	spawn_exec_data_release(spawn_data);
2170 
2171 	/* and finally: leave to userland for the first time */
2172 	cpu_spawn_return(l);
2173 
2174 	/* NOTREACHED */
2175 	return;
2176 
2177  report_error_stopped:
2178 	mutex_enter(proc_lock);
2179 	l->l_proc->p_stat = ostat;
2180 	l->l_proc->p_pptr->p_nstopchild--;
2181 	mutex_exit(proc_lock);
2182  report_error:
2183 	if (have_reflock) {
2184 		/*
2185 		 * We have not passed through execve_runproc(),
2186 		 * which would have released the p_reflock and also
2187 		 * taken ownership of the sed_exec part of spawn_data,
2188 		 * so release/free both here.
2189 		 */
2190 		rw_exit(&l->l_proc->p_reflock);
2191 		execve_free_data(&spawn_data->sed_exec);
2192 	}
2193 
2194 	if (parent_is_waiting) {
2195 		/* pass error to parent */
2196 		mutex_enter(&spawn_data->sed_mtx_child);
2197 		spawn_data->sed_error = error;
2198 		cv_signal(&spawn_data->sed_cv_child_ready);
2199 		mutex_exit(&spawn_data->sed_mtx_child);
2200 	} else {
2201 		rw_exit(&exec_lock);
2202 	}
2203 
2204 	/* release our refcount on the data */
2205 	spawn_exec_data_release(spawn_data);
2206 
2207 	/* done, exit */
2208 	mutex_enter(l->l_proc->p_lock);
2209 	/*
2210 	 * Posix explicitly asks for an exit code of 127 if we report
2211 	 * errors from the child process - so, unfortunately, there
2212 	 * is no way to report a more exact error code.
2213 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2214 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2215 	 */
2216 	exit1(l, 127, 0);
2217 }
2218 
2219 void
2220 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2221 {
2222 
2223 	for (size_t i = 0; i < len; i++) {
2224 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2225 		if (fae->fae_action != FAE_OPEN)
2226 			continue;
2227 		kmem_strfree(fae->fae_path);
2228 	}
2229 	if (fa->len > 0)
2230 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2231 	kmem_free(fa, sizeof(*fa));
2232 }
2233 
2234 static int
2235 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2236     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2237 {
2238 	struct posix_spawn_file_actions *fa;
2239 	struct posix_spawn_file_actions_entry *fae;
2240 	char *pbuf = NULL;
2241 	int error;
2242 	size_t i = 0;
2243 
2244 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2245 	error = copyin(ufa, fa, sizeof(*fa));
2246 	if (error || fa->len == 0) {
2247 		kmem_free(fa, sizeof(*fa));
2248 		return error;	/* 0 if not an error, and len == 0 */
2249 	}
2250 
2251 	if (fa->len > lim) {
2252 		kmem_free(fa, sizeof(*fa));
2253 		return EINVAL;
2254 	}
2255 
2256 	fa->size = fa->len;
2257 	size_t fal = fa->len * sizeof(*fae);
2258 	fae = fa->fae;
2259 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2260 	error = copyin(fae, fa->fae, fal);
2261 	if (error)
2262 		goto out;
2263 
2264 	pbuf = PNBUF_GET();
2265 	for (; i < fa->len; i++) {
2266 		fae = &fa->fae[i];
2267 		if (fae->fae_action != FAE_OPEN)
2268 			continue;
2269 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2270 		if (error)
2271 			goto out;
2272 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2273 		memcpy(fae->fae_path, pbuf, fal);
2274 	}
2275 	PNBUF_PUT(pbuf);
2276 
2277 	*fap = fa;
2278 	return 0;
2279 out:
2280 	if (pbuf)
2281 		PNBUF_PUT(pbuf);
2282 	posix_spawn_fa_free(fa, i);
2283 	return error;
2284 }
2285 
2286 int
2287 check_posix_spawn(struct lwp *l1)
2288 {
2289 	int error, tnprocs, count;
2290 	uid_t uid;
2291 	struct proc *p1;
2292 
2293 	p1 = l1->l_proc;
2294 	uid = kauth_cred_getuid(l1->l_cred);
2295 	tnprocs = atomic_inc_uint_nv(&nprocs);
2296 
2297 	/*
2298 	 * Although process entries are dynamically created, we still keep
2299 	 * a global limit on the maximum number we will create.
2300 	 */
2301 	if (__predict_false(tnprocs >= maxproc))
2302 		error = -1;
2303 	else
2304 		error = kauth_authorize_process(l1->l_cred,
2305 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2306 
2307 	if (error) {
2308 		atomic_dec_uint(&nprocs);
2309 		return EAGAIN;
2310 	}
2311 
2312 	/*
2313 	 * Enforce limits.
2314 	 */
2315 	count = chgproccnt(uid, 1);
2316 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2317 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2318 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2319 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2320 		(void)chgproccnt(uid, -1);
2321 		atomic_dec_uint(&nprocs);
2322 		return EAGAIN;
2323 	}
2324 
2325 	return 0;
2326 }
2327 
2328 int
2329 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2330 	struct posix_spawn_file_actions *fa,
2331 	struct posix_spawnattr *sa,
2332 	char *const *argv, char *const *envp,
2333 	execve_fetch_element_t fetch)
2334 {
2335 
2336 	struct proc *p1, *p2;
2337 	struct lwp *l2;
2338 	int error;
2339 	struct spawn_exec_data *spawn_data;
2340 	vaddr_t uaddr;
2341 	pid_t pid;
2342 	bool have_exec_lock = false;
2343 
2344 	p1 = l1->l_proc;
2345 
2346 	/* Allocate and init spawn_data */
2347 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2348 	spawn_data->sed_refcnt = 1; /* only parent so far */
2349 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2350 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2351 	mutex_enter(&spawn_data->sed_mtx_child);
2352 
2353 	/*
2354 	 * Do the first part of the exec now, collect state
2355 	 * in spawn_data.
2356 	 */
2357 	error = execve_loadvm(l1, path, argv,
2358 	    envp, fetch, &spawn_data->sed_exec);
2359 	if (error == EJUSTRETURN)
2360 		error = 0;
2361 	else if (error)
2362 		goto error_exit;
2363 
2364 	have_exec_lock = true;
2365 
2366 	/*
2367 	 * Allocate virtual address space for the U-area now, while it
2368 	 * is still easy to abort the fork operation if we're out of
2369 	 * kernel virtual address space.
2370 	 */
2371 	uaddr = uvm_uarea_alloc();
2372 	if (__predict_false(uaddr == 0)) {
2373 		error = ENOMEM;
2374 		goto error_exit;
2375 	}
2376 
2377 	/*
2378 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2379 	 * replace it with its own before returning to userland
2380 	 * in the child.
2381 	 * This is a point of no return, we will have to go through
2382 	 * the child proc to properly clean it up past this point.
2383 	 */
2384 	p2 = proc_alloc();
2385 	pid = p2->p_pid;
2386 
2387 	/*
2388 	 * Make a proc table entry for the new process.
2389 	 * Start by zeroing the section of proc that is zero-initialized,
2390 	 * then copy the section that is copied directly from the parent.
2391 	 */
2392 	memset(&p2->p_startzero, 0,
2393 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2394 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2395 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2396 	p2->p_vmspace = proc0.p_vmspace;
2397 
2398 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2399 
2400 	LIST_INIT(&p2->p_lwps);
2401 	LIST_INIT(&p2->p_sigwaiters);
2402 
2403 	/*
2404 	 * Duplicate sub-structures as needed.
2405 	 * Increase reference counts on shared objects.
2406 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2407 	 * handling are important in order to keep a consistent behaviour
2408 	 * for the child after the fork.  If we are a 32-bit process, the
2409 	 * child will be too.
2410 	 */
2411 	p2->p_flag =
2412 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2413 	p2->p_emul = p1->p_emul;
2414 	p2->p_execsw = p1->p_execsw;
2415 
2416 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2417 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2418 	rw_init(&p2->p_reflock);
2419 	cv_init(&p2->p_waitcv, "wait");
2420 	cv_init(&p2->p_lwpcv, "lwpwait");
2421 
2422 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2423 
2424 	kauth_proc_fork(p1, p2);
2425 
2426 	p2->p_raslist = NULL;
2427 	p2->p_fd = fd_copy();
2428 
2429 	/* XXX racy */
2430 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2431 
2432 	p2->p_cwdi = cwdinit();
2433 
2434 	/*
2435 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2436 	 * we just need increase pl_refcnt.
2437 	 */
2438 	if (!p1->p_limit->pl_writeable) {
2439 		lim_addref(p1->p_limit);
2440 		p2->p_limit = p1->p_limit;
2441 	} else {
2442 		p2->p_limit = lim_copy(p1->p_limit);
2443 	}
2444 
2445 	p2->p_lflag = 0;
2446 	p2->p_sflag = 0;
2447 	p2->p_slflag = 0;
2448 	p2->p_pptr = p1;
2449 	p2->p_ppid = p1->p_pid;
2450 	LIST_INIT(&p2->p_children);
2451 
2452 	p2->p_aio = NULL;
2453 
2454 #ifdef KTRACE
2455 	/*
2456 	 * Copy traceflag and tracefile if enabled.
2457 	 * If not inherited, these were zeroed above.
2458 	 */
2459 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2460 		mutex_enter(&ktrace_lock);
2461 		p2->p_traceflag = p1->p_traceflag;
2462 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2463 			ktradref(p2);
2464 		mutex_exit(&ktrace_lock);
2465 	}
2466 #endif
2467 
2468 	/*
2469 	 * Create signal actions for the child process.
2470 	 */
2471 	p2->p_sigacts = sigactsinit(p1, 0);
2472 	mutex_enter(p1->p_lock);
2473 	p2->p_sflag |=
2474 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2475 	sched_proc_fork(p1, p2);
2476 	mutex_exit(p1->p_lock);
2477 
2478 	p2->p_stflag = p1->p_stflag;
2479 
2480 	/*
2481 	 * p_stats.
2482 	 * Copy parts of p_stats, and zero out the rest.
2483 	 */
2484 	p2->p_stats = pstatscopy(p1->p_stats);
2485 
2486 	/* copy over machdep flags to the new proc */
2487 	cpu_proc_fork(p1, p2);
2488 
2489 	/*
2490 	 * Prepare remaining parts of spawn data
2491 	 */
2492 	spawn_data->sed_actions = fa;
2493 	spawn_data->sed_attrs = sa;
2494 
2495 	spawn_data->sed_parent = p1;
2496 
2497 	/* create LWP */
2498 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2499 	    &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2500 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2501 
2502 	/*
2503 	 * Copy the credential so other references don't see our changes.
2504 	 * Test to see if this is necessary first, since in the common case
2505 	 * we won't need a private reference.
2506 	 */
2507 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2508 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2509 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2510 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2511 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2512 	}
2513 
2514 	/* Update the master credentials. */
2515 	if (l2->l_cred != p2->p_cred) {
2516 		kauth_cred_t ocred;
2517 
2518 		kauth_cred_hold(l2->l_cred);
2519 		mutex_enter(p2->p_lock);
2520 		ocred = p2->p_cred;
2521 		p2->p_cred = l2->l_cred;
2522 		mutex_exit(p2->p_lock);
2523 		kauth_cred_free(ocred);
2524 	}
2525 
2526 	*child_ok = true;
2527 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2528 #if 0
2529 	l2->l_nopreempt = 1; /* start it non-preemptable */
2530 #endif
2531 
2532 	/*
2533 	 * It's now safe for the scheduler and other processes to see the
2534 	 * child process.
2535 	 */
2536 	mutex_enter(proc_lock);
2537 
2538 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2539 		p2->p_lflag |= PL_CONTROLT;
2540 
2541 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2542 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2543 
2544 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2545 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2546 
2547 	p2->p_trace_enabled = trace_is_enabled(p2);
2548 #ifdef __HAVE_SYSCALL_INTERN
2549 	(*p2->p_emul->e_syscall_intern)(p2);
2550 #endif
2551 
2552 	/*
2553 	 * Make child runnable, set start time, and add to run queue except
2554 	 * if the parent requested the child to start in SSTOP state.
2555 	 */
2556 	mutex_enter(p2->p_lock);
2557 
2558 	getmicrotime(&p2->p_stats->p_start);
2559 
2560 	lwp_lock(l2);
2561 	KASSERT(p2->p_nrlwps == 1);
2562 	p2->p_nrlwps = 1;
2563 	p2->p_stat = SACTIVE;
2564 	l2->l_stat = LSRUN;
2565 	sched_enqueue(l2, false);
2566 	lwp_unlock(l2);
2567 
2568 	mutex_exit(p2->p_lock);
2569 	mutex_exit(proc_lock);
2570 
2571 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2572 	error = spawn_data->sed_error;
2573 	mutex_exit(&spawn_data->sed_mtx_child);
2574 	spawn_exec_data_release(spawn_data);
2575 
2576 	rw_exit(&p1->p_reflock);
2577 	rw_exit(&exec_lock);
2578 	have_exec_lock = false;
2579 
2580 	*pid_res = pid;
2581 	return error;
2582 
2583  error_exit:
2584 	if (have_exec_lock) {
2585 		execve_free_data(&spawn_data->sed_exec);
2586 		rw_exit(&p1->p_reflock);
2587 		rw_exit(&exec_lock);
2588 	}
2589 	mutex_exit(&spawn_data->sed_mtx_child);
2590 	spawn_exec_data_release(spawn_data);
2591 
2592 	return error;
2593 }
2594 
2595 int
2596 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2597     register_t *retval)
2598 {
2599 	/* {
2600 		syscallarg(pid_t *) pid;
2601 		syscallarg(const char *) path;
2602 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2603 		syscallarg(const struct posix_spawnattr *) attrp;
2604 		syscallarg(char *const *) argv;
2605 		syscallarg(char *const *) envp;
2606 	} */
2607 
2608 	int error;
2609 	struct posix_spawn_file_actions *fa = NULL;
2610 	struct posix_spawnattr *sa = NULL;
2611 	pid_t pid;
2612 	bool child_ok = false;
2613 	rlim_t max_fileactions;
2614 	proc_t *p = l1->l_proc;
2615 
2616 	error = check_posix_spawn(l1);
2617 	if (error) {
2618 		*retval = error;
2619 		return 0;
2620 	}
2621 
2622 	/* copy in file_actions struct */
2623 	if (SCARG(uap, file_actions) != NULL) {
2624 		max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2625 		    maxfiles);
2626 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2627 		    max_fileactions);
2628 		if (error)
2629 			goto error_exit;
2630 	}
2631 
2632 	/* copyin posix_spawnattr struct */
2633 	if (SCARG(uap, attrp) != NULL) {
2634 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2635 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2636 		if (error)
2637 			goto error_exit;
2638 	}
2639 
2640 	/*
2641 	 * Do the spawn
2642 	 */
2643 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2644 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2645 	if (error)
2646 		goto error_exit;
2647 
2648 	if (error == 0 && SCARG(uap, pid) != NULL)
2649 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2650 
2651 	*retval = error;
2652 	return 0;
2653 
2654  error_exit:
2655 	if (!child_ok) {
2656 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2657 		atomic_dec_uint(&nprocs);
2658 
2659 		if (sa)
2660 			kmem_free(sa, sizeof(*sa));
2661 		if (fa)
2662 			posix_spawn_fa_free(fa, fa->len);
2663 	}
2664 
2665 	*retval = error;
2666 	return 0;
2667 }
2668 
2669 void
2670 exec_free_emul_arg(struct exec_package *epp)
2671 {
2672 	if (epp->ep_emul_arg_free != NULL) {
2673 		KASSERT(epp->ep_emul_arg != NULL);
2674 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2675 		epp->ep_emul_arg_free = NULL;
2676 		epp->ep_emul_arg = NULL;
2677 	} else {
2678 		KASSERT(epp->ep_emul_arg == NULL);
2679 	}
2680 }
2681 
2682 #ifdef DEBUG_EXEC
2683 static void
2684 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2685 {
2686 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2687 	size_t j;
2688 
2689 	if (error == 0)
2690 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2691 	else
2692 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2693 		    epp->ep_vmcmds.evs_used, error));
2694 
2695 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2696 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2697 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2698 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2699 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2700 		    "pagedvn" :
2701 		    vp[j].ev_proc == vmcmd_map_readvn ?
2702 		    "readvn" :
2703 		    vp[j].ev_proc == vmcmd_map_zero ?
2704 		    "zero" : "*unknown*",
2705 		    vp[j].ev_addr, vp[j].ev_len,
2706 		    vp[j].ev_offset, vp[j].ev_prot,
2707 		    vp[j].ev_flags));
2708 		if (error != 0 && j == x)
2709 			DPRINTF(("     ^--- failed\n"));
2710 	}
2711 }
2712 #endif
2713