xref: /netbsd-src/sys/kern/kern_exec.c (revision a8c74629f602faa0ccf8a463757d7baf858bbf3a)
1 /*	$NetBSD: kern_exec.c,v 1.502 2020/10/06 13:38:00 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008, 2019, 2020 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 1.502 2020/10/06 13:38:00 christos 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 __cacheline_aligned;
250 
251 static kmutex_t sigobject_lock __cacheline_aligned;
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 int
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 		DPRINTF(("%s: can't resolve name for %s, error %d\n",
332 		    __func__, epp->ep_kname, error));
333 		PNBUF_PUT(*rpath);
334 		*rpath = NULL;
335 		return error;
336 	}
337 	epp->ep_resolvedname = *rpath;
338 	if ((p = strrchr(*rpath, '/')) != NULL)
339 		epp->ep_kname = p + 1;
340 	return 0;
341 }
342 
343 
344 /*
345  * check exec:
346  * given an "executable" described in the exec package's namei info,
347  * see what we can do with it.
348  *
349  * ON ENTRY:
350  *	exec package with appropriate namei info
351  *	lwp pointer of exec'ing lwp
352  *	NO SELF-LOCKED VNODES
353  *
354  * ON EXIT:
355  *	error:	nothing held, etc.  exec header still allocated.
356  *	ok:	filled exec package, executable's vnode (unlocked).
357  *
358  * EXEC SWITCH ENTRY:
359  * 	Locked vnode to check, exec package, proc.
360  *
361  * EXEC SWITCH EXIT:
362  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
363  *	error:	destructive:
364  *			everything deallocated execept exec header.
365  *		non-destructive:
366  *			error code, executable's vnode (unlocked),
367  *			exec header unmodified.
368  */
369 int
370 /*ARGSUSED*/
371 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb,
372     char **rpath)
373 {
374 	int		error, i;
375 	struct vnode	*vp;
376 	size_t		resid;
377 
378 	if (epp->ep_resolvedname) {
379 		struct nameidata nd;
380 
381 		// grab the absolute pathbuf here before namei() trashes it.
382 		pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX);
383 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
384 
385 		/* first get the vnode */
386 		if ((error = namei(&nd)) != 0)
387 			return error;
388 
389 		epp->ep_vp = vp = nd.ni_vp;
390 #ifdef DIAGNOSTIC
391 		/* paranoia (take this out once namei stuff stabilizes) */
392 		memset(nd.ni_pnbuf, '~', PATH_MAX);
393 #endif
394 	} else {
395 		struct file *fp;
396 
397 		if ((error = fd_getvnode(epp->ep_xfd, &fp)) != 0)
398 			return error;
399 		epp->ep_vp = vp = fp->f_vnode;
400 		vref(vp);
401 		fd_putfile(epp->ep_xfd);
402 		if ((error = exec_resolvename(l, epp, vp, rpath)) != 0)
403 			return error;
404 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
405 	}
406 
407 	/* check access and type */
408 	if (vp->v_type != VREG) {
409 		error = EACCES;
410 		goto bad1;
411 	}
412 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
413 		goto bad1;
414 
415 	/* get attributes */
416 	/* XXX VOP_GETATTR is the only thing that needs LK_EXCLUSIVE here */
417 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
418 		goto bad1;
419 
420 	/* Check mount point */
421 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
422 		error = EACCES;
423 		goto bad1;
424 	}
425 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
426 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
427 
428 	/* try to open it */
429 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
430 		goto bad1;
431 
432 	/* now we have the file, get the exec header */
433 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
434 			UIO_SYSSPACE, IO_NODELOCKED, l->l_cred, &resid, NULL);
435 	if (error)
436 		goto bad1;
437 
438 	/* unlock vp, since we need it unlocked from here on out. */
439 	VOP_UNLOCK(vp);
440 
441 #if NVERIEXEC > 0
442 	error = veriexec_verify(l, vp,
443 	    epp->ep_resolvedname ? epp->ep_resolvedname : epp->ep_kname,
444 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
445 	    NULL);
446 	if (error)
447 		goto bad2;
448 #endif /* NVERIEXEC > 0 */
449 
450 #ifdef PAX_SEGVGUARD
451 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
452 	if (error)
453 		goto bad2;
454 #endif /* PAX_SEGVGUARD */
455 
456 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
457 
458 	/*
459 	 * Set up default address space limits.  Can be overridden
460 	 * by individual exec packages.
461 	 */
462 	epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS);
463 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
464 
465 	/*
466 	 * set up the vmcmds for creation of the process
467 	 * address space
468 	 */
469 	error = ENOEXEC;
470 	for (i = 0; i < nexecs; i++) {
471 		int newerror;
472 
473 		epp->ep_esch = execsw[i];
474 		newerror = (*execsw[i]->es_makecmds)(l, epp);
475 
476 		if (!newerror) {
477 			/* Seems ok: check that entry point is not too high */
478 			if (epp->ep_entry >= epp->ep_vm_maxaddr) {
479 #ifdef DIAGNOSTIC
480 				printf("%s: rejecting %p due to "
481 				    "too high entry address (>= %p)\n",
482 					 __func__, (void *)epp->ep_entry,
483 					 (void *)epp->ep_vm_maxaddr);
484 #endif
485 				error = ENOEXEC;
486 				break;
487 			}
488 			/* Seems ok: check that entry point is not too low */
489 			if (epp->ep_entry < epp->ep_vm_minaddr) {
490 #ifdef DIAGNOSTIC
491 				printf("%s: rejecting %p due to "
492 				    "too low entry address (< %p)\n",
493 				     __func__, (void *)epp->ep_entry,
494 				     (void *)epp->ep_vm_minaddr);
495 #endif
496 				error = ENOEXEC;
497 				break;
498 			}
499 
500 			/* check limits */
501 #ifdef MAXTSIZ
502 			if (epp->ep_tsize > MAXTSIZ) {
503 #ifdef DIAGNOSTIC
504 #define LMSG "%s: rejecting due to %s limit (%ju > %ju)\n"
505 				printf(LMSG, __func__, "text",
506 				    (uintmax_t)epp->ep_tsize,
507 				    (uintmax_t)MAXTSIZ);
508 #endif
509 				error = ENOMEM;
510 				break;
511 			}
512 #endif
513 			vsize_t dlimit =
514 			    (vsize_t)l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur;
515 			if (epp->ep_dsize > dlimit) {
516 #ifdef DIAGNOSTIC
517 				printf(LMSG, __func__, "data",
518 				    (uintmax_t)epp->ep_dsize,
519 				    (uintmax_t)dlimit);
520 #endif
521 				error = ENOMEM;
522 				break;
523 			}
524 			return 0;
525 		}
526 
527 		/*
528 		 * Reset all the fields that may have been modified by the
529 		 * loader.
530 		 */
531 		KASSERT(epp->ep_emul_arg == NULL);
532 		if (epp->ep_emul_root != NULL) {
533 			vrele(epp->ep_emul_root);
534 			epp->ep_emul_root = NULL;
535 		}
536 		if (epp->ep_interp != NULL) {
537 			vrele(epp->ep_interp);
538 			epp->ep_interp = NULL;
539 		}
540 		epp->ep_pax_flags = 0;
541 
542 		/* make sure the first "interesting" error code is saved. */
543 		if (error == ENOEXEC)
544 			error = newerror;
545 
546 		if (epp->ep_flags & EXEC_DESTR)
547 			/* Error from "#!" code, tidied up by recursive call */
548 			return error;
549 	}
550 
551 	/* not found, error */
552 
553 	/*
554 	 * free any vmspace-creation commands,
555 	 * and release their references
556 	 */
557 	kill_vmcmds(&epp->ep_vmcmds);
558 
559 #if NVERIEXEC > 0 || defined(PAX_SEGVGUARD)
560 bad2:
561 #endif
562 	/*
563 	 * close and release the vnode, restore the old one, free the
564 	 * pathname buf, and punt.
565 	 */
566 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
567 	VOP_CLOSE(vp, FREAD, l->l_cred);
568 	vput(vp);
569 	return error;
570 
571 bad1:
572 	/*
573 	 * free the namei pathname buffer, and put the vnode
574 	 * (which we don't yet have open).
575 	 */
576 	vput(vp);				/* was still locked */
577 	return error;
578 }
579 
580 #ifdef __MACHINE_STACK_GROWS_UP
581 #define STACK_PTHREADSPACE NBPG
582 #else
583 #define STACK_PTHREADSPACE 0
584 #endif
585 
586 static int
587 execve_fetch_element(char * const *array, size_t index, char **value)
588 {
589 	return copyin(array + index, value, sizeof(*value));
590 }
591 
592 /*
593  * exec system call
594  */
595 int
596 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
597 {
598 	/* {
599 		syscallarg(const char *)	path;
600 		syscallarg(char * const *)	argp;
601 		syscallarg(char * const *)	envp;
602 	} */
603 
604 	return execve1(l, true, SCARG(uap, path), -1, SCARG(uap, argp),
605 	    SCARG(uap, envp), execve_fetch_element);
606 }
607 
608 int
609 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
610     register_t *retval)
611 {
612 	/* {
613 		syscallarg(int)			fd;
614 		syscallarg(char * const *)	argp;
615 		syscallarg(char * const *)	envp;
616 	} */
617 
618 	return execve1(l, false, NULL, SCARG(uap, fd), SCARG(uap, argp),
619 	    SCARG(uap, envp), execve_fetch_element);
620 }
621 
622 /*
623  * Load modules to try and execute an image that we do not understand.
624  * If no execsw entries are present, we load those likely to be needed
625  * in order to run native images only.  Otherwise, we autoload all
626  * possible modules that could let us run the binary.  XXX lame
627  */
628 static void
629 exec_autoload(void)
630 {
631 #ifdef MODULAR
632 	static const char * const native[] = {
633 		"exec_elf32",
634 		"exec_elf64",
635 		"exec_script",
636 		NULL
637 	};
638 	static const char * const compat[] = {
639 		"exec_elf32",
640 		"exec_elf64",
641 		"exec_script",
642 		"exec_aout",
643 		"exec_coff",
644 		"exec_ecoff",
645 		"compat_aoutm68k",
646 		"compat_netbsd32",
647 #if 0
648 		"compat_linux",
649 		"compat_linux32",
650 #endif
651 		"compat_sunos",
652 		"compat_sunos32",
653 		"compat_ultrix",
654 		NULL
655 	};
656 	char const * const *list;
657 	int i;
658 
659 	list = nexecs == 0 ? native : compat;
660 	for (i = 0; list[i] != NULL; i++) {
661 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
662 			continue;
663 		}
664 		yield();
665 	}
666 #endif
667 }
668 
669 /*
670  * Copy the user or kernel supplied upath to the allocated pathbuffer pbp
671  * making it absolute in the process, by prepending the current working
672  * directory if it is not. If offs is supplied it will contain the offset
673  * where the original supplied copy of upath starts.
674  */
675 int
676 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg,
677     struct pathbuf **pbp, size_t *offs)
678 {
679 	char *path, *bp;
680 	size_t len, tlen;
681 	int error;
682 	struct cwdinfo *cwdi;
683 
684 	path = PNBUF_GET();
685 	if (seg == UIO_SYSSPACE) {
686 		error = copystr(upath, path, MAXPATHLEN, &len);
687 	} else {
688 		error = copyinstr(upath, path, MAXPATHLEN, &len);
689 	}
690 	if (error)
691 		goto err;
692 
693 	if (path[0] == '/') {
694 		if (offs)
695 			*offs = 0;
696 		goto out;
697 	}
698 
699 	len++;
700 	if (len + 1 >= MAXPATHLEN) {
701 		error = ENAMETOOLONG;
702 		goto err;
703 	}
704 	bp = path + MAXPATHLEN - len;
705 	memmove(bp, path, len);
706 	*(--bp) = '/';
707 
708 	cwdi = l->l_proc->p_cwdi;
709 	rw_enter(&cwdi->cwdi_lock, RW_READER);
710 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
711 	    GETCWD_CHECK_ACCESS, l);
712 	rw_exit(&cwdi->cwdi_lock);
713 
714 	if (error)
715 		goto err;
716 	tlen = path + MAXPATHLEN - bp;
717 
718 	memmove(path, bp, tlen);
719 	path[tlen - 1] = '\0';
720 	if (offs)
721 		*offs = tlen - len;
722 out:
723 	*pbp = pathbuf_assimilate(path);
724 	return 0;
725 err:
726 	PNBUF_PUT(path);
727 	return error;
728 }
729 
730 vaddr_t
731 exec_vm_minaddr(vaddr_t va_min)
732 {
733 	/*
734 	 * Increase va_min if we don't want NULL to be mappable by the
735 	 * process.
736 	 */
737 #define VM_MIN_GUARD	PAGE_SIZE
738 	if (user_va0_disable && (va_min < VM_MIN_GUARD))
739 		return VM_MIN_GUARD;
740 	return va_min;
741 }
742 
743 static int
744 execve_loadvm(struct lwp *l, bool has_path, const char *path, int fd,
745 	char * const *args, char * const *envs,
746 	execve_fetch_element_t fetch_element,
747 	struct execve_data * restrict data)
748 {
749 	struct exec_package	* const epp = &data->ed_pack;
750 	int			error;
751 	struct proc		*p;
752 	char			*dp;
753 	u_int			modgen;
754 
755 	KASSERT(data != NULL);
756 
757 	p = l->l_proc;
758 	modgen = 0;
759 
760 	SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
761 
762 	/*
763 	 * Check if we have exceeded our number of processes limit.
764 	 * This is so that we handle the case where a root daemon
765 	 * forked, ran setuid to become the desired user and is trying
766 	 * to exec. The obvious place to do the reference counting check
767 	 * is setuid(), but we don't do the reference counting check there
768 	 * like other OS's do because then all the programs that use setuid()
769 	 * must be modified to check the return code of setuid() and exit().
770 	 * It is dangerous to make setuid() fail, because it fails open and
771 	 * the program will continue to run as root. If we make it succeed
772 	 * and return an error code, again we are not enforcing the limit.
773 	 * The best place to enforce the limit is here, when the process tries
774 	 * to execute a new image, because eventually the process will need
775 	 * to call exec in order to do something useful.
776 	 */
777  retry:
778 	if (p->p_flag & PK_SUGID) {
779 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
780 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
781 		     &p->p_rlimit[RLIMIT_NPROC],
782 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
783 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
784 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
785 		return EAGAIN;
786 	}
787 
788 	/*
789 	 * Drain existing references and forbid new ones.  The process
790 	 * should be left alone until we're done here.  This is necessary
791 	 * to avoid race conditions - e.g. in ptrace() - that might allow
792 	 * a local user to illicitly obtain elevated privileges.
793 	 */
794 	rw_enter(&p->p_reflock, RW_WRITER);
795 
796 	if (has_path) {
797 		size_t	offs;
798 		/*
799 		 * Init the namei data to point the file user's program name.
800 		 * This is done here rather than in check_exec(), so that it's
801 		 * possible to override this settings if any of makecmd/probe
802 		 * functions call check_exec() recursively - for example,
803 		 * see exec_script_makecmds().
804 		 */
805 		if ((error = exec_makepathbuf(l, path, UIO_USERSPACE,
806 		    &data->ed_pathbuf, &offs)) != 0)
807 			goto clrflg;
808 		data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
809 		epp->ep_kname = data->ed_pathstring + offs;
810 		data->ed_resolvedname = PNBUF_GET();
811 		epp->ep_resolvedname = data->ed_resolvedname;
812 		epp->ep_xfd = -1;
813 	} else {
814 		data->ed_pathbuf = pathbuf_assimilate(strcpy(PNBUF_GET(), "/"));
815 		data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
816 		epp->ep_kname = "*fexecve*";
817 		data->ed_resolvedname = NULL;
818 		epp->ep_resolvedname = NULL;
819 		epp->ep_xfd = fd;
820 	}
821 
822 
823 	/*
824 	 * initialize the fields of the exec package.
825 	 */
826 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
827 	epp->ep_hdrlen = exec_maxhdrsz;
828 	epp->ep_hdrvalid = 0;
829 	epp->ep_emul_arg = NULL;
830 	epp->ep_emul_arg_free = NULL;
831 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
832 	epp->ep_vap = &data->ed_attr;
833 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
834 	MD_TOPDOWN_INIT(epp);
835 	epp->ep_emul_root = NULL;
836 	epp->ep_interp = NULL;
837 	epp->ep_esch = NULL;
838 	epp->ep_pax_flags = 0;
839 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
840 
841 	rw_enter(&exec_lock, RW_READER);
842 
843 	/* see if we can run it. */
844 	if ((error = check_exec(l, epp, data->ed_pathbuf,
845 	    &data->ed_resolvedname)) != 0) {
846 		if (error != ENOENT && error != EACCES && error != ENOEXEC) {
847 			DPRINTF(("%s: check exec failed for %s, error %d\n",
848 			    __func__, epp->ep_kname, error));
849 		}
850 		goto freehdr;
851 	}
852 
853 	/* allocate an argument buffer */
854 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
855 	KASSERT(data->ed_argp != NULL);
856 	dp = data->ed_argp;
857 
858 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
859 		goto bad;
860 	}
861 
862 	/*
863 	 * Calculate the new stack size.
864 	 */
865 
866 #ifdef __MACHINE_STACK_GROWS_UP
867 /*
868  * copyargs() fills argc/argv/envp from the lower address even on
869  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
870  * so that _rtld() use it.
871  */
872 #define	RTLD_GAP	32
873 #else
874 #define	RTLD_GAP	0
875 #endif
876 
877 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
878 
879 	data->ed_argslen = calcargs(data, argenvstrlen);
880 
881 	const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
882 
883 	if (len > epp->ep_ssize) {
884 		/* in effect, compare to initial limit */
885 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
886 		error = ENOMEM;
887 		goto bad;
888 	}
889 	/* adjust "active stack depth" for process VSZ */
890 	epp->ep_ssize = len;
891 
892 	return 0;
893 
894  bad:
895 	/* free the vmspace-creation commands, and release their references */
896 	kill_vmcmds(&epp->ep_vmcmds);
897 	/* kill any opened file descriptor, if necessary */
898 	if (epp->ep_flags & EXEC_HASFD) {
899 		epp->ep_flags &= ~EXEC_HASFD;
900 		fd_close(epp->ep_fd);
901 	}
902 	/* close and put the exec'd file */
903 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
904 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
905 	vput(epp->ep_vp);
906 	pool_put(&exec_pool, data->ed_argp);
907 
908  freehdr:
909 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
910 	if (epp->ep_emul_root != NULL)
911 		vrele(epp->ep_emul_root);
912 	if (epp->ep_interp != NULL)
913 		vrele(epp->ep_interp);
914 
915 	rw_exit(&exec_lock);
916 
917 	exec_path_free(data);
918 
919  clrflg:
920 	rw_exit(&p->p_reflock);
921 
922 	if (modgen != module_gen && error == ENOEXEC) {
923 		modgen = module_gen;
924 		exec_autoload();
925 		goto retry;
926 	}
927 
928 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
929 	return error;
930 }
931 
932 static int
933 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
934 {
935 	struct exec_package	* const epp = &data->ed_pack;
936 	struct proc		*p = l->l_proc;
937 	struct exec_vmcmd	*base_vcp;
938 	int			error = 0;
939 	size_t			i;
940 
941 	/* record proc's vnode, for use by procfs and others */
942 	if (p->p_textvp)
943 		vrele(p->p_textvp);
944 	vref(epp->ep_vp);
945 	p->p_textvp = epp->ep_vp;
946 
947 	/* create the new process's VM space by running the vmcmds */
948 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
949 
950 #ifdef TRACE_EXEC
951 	DUMPVMCMDS(epp, 0, 0);
952 #endif
953 
954 	base_vcp = NULL;
955 
956 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
957 		struct exec_vmcmd *vcp;
958 
959 		vcp = &epp->ep_vmcmds.evs_cmds[i];
960 		if (vcp->ev_flags & VMCMD_RELATIVE) {
961 			KASSERTMSG(base_vcp != NULL,
962 			    "%s: relative vmcmd with no base", __func__);
963 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
964 			    "%s: illegal base & relative vmcmd", __func__);
965 			vcp->ev_addr += base_vcp->ev_addr;
966 		}
967 		error = (*vcp->ev_proc)(l, vcp);
968 		if (error)
969 			DUMPVMCMDS(epp, i, error);
970 		if (vcp->ev_flags & VMCMD_BASE)
971 			base_vcp = vcp;
972 	}
973 
974 	/* free the vmspace-creation commands, and release their references */
975 	kill_vmcmds(&epp->ep_vmcmds);
976 
977 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
978 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
979 	vput(epp->ep_vp);
980 
981 	/* if an error happened, deallocate and punt */
982 	if (error != 0) {
983 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
984 	}
985 	return error;
986 }
987 
988 static void
989 execve_free_data(struct execve_data *data)
990 {
991 	struct exec_package	* const epp = &data->ed_pack;
992 
993 	/* free the vmspace-creation commands, and release their references */
994 	kill_vmcmds(&epp->ep_vmcmds);
995 	/* kill any opened file descriptor, if necessary */
996 	if (epp->ep_flags & EXEC_HASFD) {
997 		epp->ep_flags &= ~EXEC_HASFD;
998 		fd_close(epp->ep_fd);
999 	}
1000 
1001 	/* close and put the exec'd file */
1002 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
1003 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
1004 	vput(epp->ep_vp);
1005 	pool_put(&exec_pool, data->ed_argp);
1006 
1007 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1008 	if (epp->ep_emul_root != NULL)
1009 		vrele(epp->ep_emul_root);
1010 	if (epp->ep_interp != NULL)
1011 		vrele(epp->ep_interp);
1012 
1013 	exec_path_free(data);
1014 }
1015 
1016 static void
1017 pathexec(struct proc *p, const char *resolvedname)
1018 {
1019 	/* set command name & other accounting info */
1020 	const char *cmdname;
1021 
1022 	if (resolvedname == NULL) {
1023 		cmdname = "*fexecve*";
1024 		resolvedname = "/";
1025 	} else {
1026 		cmdname = strrchr(resolvedname, '/') + 1;
1027 	}
1028 	KASSERTMSG(resolvedname[0] == '/', "bad resolvedname `%s'",
1029 	    resolvedname);
1030 
1031 	strlcpy(p->p_comm, cmdname, sizeof(p->p_comm));
1032 
1033 	kmem_strfree(p->p_path);
1034 	p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
1035 }
1036 
1037 /* XXX elsewhere */
1038 static int
1039 credexec(struct lwp *l, struct vattr *attr)
1040 {
1041 	struct proc *p = l->l_proc;
1042 	int error;
1043 
1044 	/*
1045 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
1046 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
1047 	 * out additional references on the process for the moment.
1048 	 */
1049 	if ((p->p_slflag & PSL_TRACED) == 0 &&
1050 
1051 	    (((attr->va_mode & S_ISUID) != 0 &&
1052 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
1053 
1054 	     ((attr->va_mode & S_ISGID) != 0 &&
1055 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
1056 		/*
1057 		 * Mark the process as SUGID before we do
1058 		 * anything that might block.
1059 		 */
1060 		proc_crmod_enter();
1061 		proc_crmod_leave(NULL, NULL, true);
1062 
1063 		/* Make sure file descriptors 0..2 are in use. */
1064 		if ((error = fd_checkstd()) != 0) {
1065 			DPRINTF(("%s: fdcheckstd failed %d\n",
1066 			    __func__, error));
1067 			return error;
1068 		}
1069 
1070 		/*
1071 		 * Copy the credential so other references don't see our
1072 		 * changes.
1073 		 */
1074 		l->l_cred = kauth_cred_copy(l->l_cred);
1075 #ifdef KTRACE
1076 		/*
1077 		 * If the persistent trace flag isn't set, turn off.
1078 		 */
1079 		if (p->p_tracep) {
1080 			mutex_enter(&ktrace_lock);
1081 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1082 				ktrderef(p);
1083 			mutex_exit(&ktrace_lock);
1084 		}
1085 #endif
1086 		if (attr->va_mode & S_ISUID)
1087 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
1088 		if (attr->va_mode & S_ISGID)
1089 			kauth_cred_setegid(l->l_cred, attr->va_gid);
1090 	} else {
1091 		if (kauth_cred_geteuid(l->l_cred) ==
1092 		    kauth_cred_getuid(l->l_cred) &&
1093 		    kauth_cred_getegid(l->l_cred) ==
1094 		    kauth_cred_getgid(l->l_cred))
1095 			p->p_flag &= ~PK_SUGID;
1096 	}
1097 
1098 	/*
1099 	 * Copy the credential so other references don't see our changes.
1100 	 * Test to see if this is necessary first, since in the common case
1101 	 * we won't need a private reference.
1102 	 */
1103 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1104 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1105 		l->l_cred = kauth_cred_copy(l->l_cred);
1106 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1107 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1108 	}
1109 
1110 	/* Update the master credentials. */
1111 	if (l->l_cred != p->p_cred) {
1112 		kauth_cred_t ocred;
1113 
1114 		kauth_cred_hold(l->l_cred);
1115 		mutex_enter(p->p_lock);
1116 		ocred = p->p_cred;
1117 		p->p_cred = l->l_cred;
1118 		mutex_exit(p->p_lock);
1119 		kauth_cred_free(ocred);
1120 	}
1121 
1122 	return 0;
1123 }
1124 
1125 static void
1126 emulexec(struct lwp *l, struct exec_package *epp)
1127 {
1128 	struct proc		*p = l->l_proc;
1129 
1130 	/* The emulation root will usually have been found when we looked
1131 	 * for the elf interpreter (or similar), if not look now. */
1132 	if (epp->ep_esch->es_emul->e_path != NULL &&
1133 	    epp->ep_emul_root == NULL)
1134 		emul_find_root(l, epp);
1135 
1136 	/* Any old emulation root got removed by fdcloseexec */
1137 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1138 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1139 	rw_exit(&p->p_cwdi->cwdi_lock);
1140 	epp->ep_emul_root = NULL;
1141 	if (epp->ep_interp != NULL)
1142 		vrele(epp->ep_interp);
1143 
1144 	/*
1145 	 * Call emulation specific exec hook. This can setup per-process
1146 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1147 	 *
1148 	 * If we are executing process of different emulation than the
1149 	 * original forked process, call e_proc_exit() of the old emulation
1150 	 * first, then e_proc_exec() of new emulation. If the emulation is
1151 	 * same, the exec hook code should deallocate any old emulation
1152 	 * resources held previously by this process.
1153 	 */
1154 	if (p->p_emul && p->p_emul->e_proc_exit
1155 	    && p->p_emul != epp->ep_esch->es_emul)
1156 		(*p->p_emul->e_proc_exit)(p);
1157 
1158 	/*
1159 	 * Call exec hook. Emulation code may NOT store reference to anything
1160 	 * from &pack.
1161 	 */
1162 	if (epp->ep_esch->es_emul->e_proc_exec)
1163 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1164 
1165 	/* update p_emul, the old value is no longer needed */
1166 	p->p_emul = epp->ep_esch->es_emul;
1167 
1168 	/* ...and the same for p_execsw */
1169 	p->p_execsw = epp->ep_esch;
1170 
1171 #ifdef __HAVE_SYSCALL_INTERN
1172 	(*p->p_emul->e_syscall_intern)(p);
1173 #endif
1174 	ktremul();
1175 }
1176 
1177 static int
1178 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1179 	bool no_local_exec_lock, bool is_spawn)
1180 {
1181 	struct exec_package	* const epp = &data->ed_pack;
1182 	int error = 0;
1183 	struct proc		*p;
1184 	struct vmspace		*vm;
1185 
1186 	/*
1187 	 * In case of a posix_spawn operation, the child doing the exec
1188 	 * might not hold the reader lock on exec_lock, but the parent
1189 	 * will do this instead.
1190 	 */
1191 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1192 	KASSERT(!no_local_exec_lock || is_spawn);
1193 	KASSERT(data != NULL);
1194 
1195 	p = l->l_proc;
1196 
1197 	/* Get rid of other LWPs. */
1198 	if (p->p_nlwps > 1) {
1199 		mutex_enter(p->p_lock);
1200 		exit_lwps(l);
1201 		mutex_exit(p->p_lock);
1202 	}
1203 	KDASSERT(p->p_nlwps == 1);
1204 
1205 	/* Destroy any lwpctl info. */
1206 	if (p->p_lwpctl != NULL)
1207 		lwp_ctl_exit();
1208 
1209 	/* Remove POSIX timers */
1210 	timers_free(p, TIMERS_POSIX);
1211 
1212 	/* Set the PaX flags. */
1213 	pax_set_flags(epp, p);
1214 
1215 	/*
1216 	 * Do whatever is necessary to prepare the address space
1217 	 * for remapping.  Note that this might replace the current
1218 	 * vmspace with another!
1219 	 *
1220 	 * vfork(): do not touch any user space data in the new child
1221 	 * until we have awoken the parent below, or it will defeat
1222 	 * lazy pmap switching (on x86).
1223 	 */
1224 	if (is_spawn)
1225 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1226 		    epp->ep_vm_maxaddr,
1227 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1228 	else
1229 		uvmspace_exec(l, epp->ep_vm_minaddr,
1230 		    epp->ep_vm_maxaddr,
1231 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1232 	vm = p->p_vmspace;
1233 
1234 	vm->vm_taddr = (void *)epp->ep_taddr;
1235 	vm->vm_tsize = btoc(epp->ep_tsize);
1236 	vm->vm_daddr = (void*)epp->ep_daddr;
1237 	vm->vm_dsize = btoc(epp->ep_dsize);
1238 	vm->vm_ssize = btoc(epp->ep_ssize);
1239 	vm->vm_issize = 0;
1240 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1241 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1242 
1243 	pax_aslr_init_vm(l, vm, epp);
1244 
1245 	cwdexec(p);
1246 	fd_closeexec();		/* handle close on exec */
1247 
1248 	if (__predict_false(ktrace_on))
1249 		fd_ktrexecfd();
1250 
1251 	execsigs(p);		/* reset caught signals */
1252 
1253 	mutex_enter(p->p_lock);
1254 	l->l_ctxlink = NULL;	/* reset ucontext link */
1255 	p->p_acflag &= ~AFORK;
1256 	p->p_flag |= PK_EXEC;
1257 	mutex_exit(p->p_lock);
1258 
1259 	error = credexec(l, &data->ed_attr);
1260 	if (error)
1261 		goto exec_abort;
1262 
1263 #if defined(__HAVE_RAS)
1264 	/*
1265 	 * Remove all RASs from the address space.
1266 	 */
1267 	ras_purgeall();
1268 #endif
1269 
1270 	/*
1271 	 * Stop profiling.
1272 	 */
1273 	if ((p->p_stflag & PST_PROFIL) != 0) {
1274 		mutex_spin_enter(&p->p_stmutex);
1275 		stopprofclock(p);
1276 		mutex_spin_exit(&p->p_stmutex);
1277 	}
1278 
1279 	/*
1280 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1281 	 * exited and exec()/exit() are the only places it will be cleared.
1282 	 *
1283 	 * Once the parent has been awoken, curlwp may teleport to a new CPU
1284 	 * in sched_vforkexec(), and it's then OK to start messing with user
1285 	 * data.  See comment above.
1286 	 */
1287 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1288 		bool samecpu;
1289 		lwp_t *lp;
1290 
1291 		mutex_enter(&proc_lock);
1292 		lp = p->p_vforklwp;
1293 		p->p_vforklwp = NULL;
1294 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1295 		cv_broadcast(&lp->l_waitcv);
1296 
1297 		/* Clear flags after cv_broadcast() (scheduler needs them). */
1298 		p->p_lflag &= ~PL_PPWAIT;
1299 		lp->l_vforkwaiting = false;
1300 
1301 		/* If parent is still on same CPU, teleport curlwp elsewhere. */
1302 		samecpu = (lp->l_cpu == curlwp->l_cpu);
1303 		mutex_exit(&proc_lock);
1304 
1305 		/* Give the parent its CPU back - find a new home. */
1306 		KASSERT(!is_spawn);
1307 		sched_vforkexec(l, samecpu);
1308 	}
1309 
1310 	/* Now map address space. */
1311 	error = execve_dovmcmds(l, data);
1312 	if (error != 0)
1313 		goto exec_abort;
1314 
1315 	pathexec(p, epp->ep_resolvedname);
1316 
1317 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1318 
1319 	error = copyoutargs(data, l, newstack);
1320 	if (error != 0)
1321 		goto exec_abort;
1322 
1323 	doexechooks(p);
1324 
1325 	/*
1326 	 * Set initial SP at the top of the stack.
1327 	 *
1328 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1329 	 * the end of arg/env strings.  Userland guesses the address of argc
1330 	 * via ps_strings::ps_argvstr.
1331 	 */
1332 
1333 	/* Setup new registers and do misc. setup. */
1334 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1335 	if (epp->ep_esch->es_setregs)
1336 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1337 
1338 	/* Provide a consistent LWP private setting */
1339 	(void)lwp_setprivate(l, NULL);
1340 
1341 	/* Discard all PCU state; need to start fresh */
1342 	pcu_discard_all(l);
1343 
1344 	/* map the process's signal trampoline code */
1345 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1346 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1347 		goto exec_abort;
1348 	}
1349 
1350 	pool_put(&exec_pool, data->ed_argp);
1351 
1352 	/* notify others that we exec'd */
1353 	KNOTE(&p->p_klist, NOTE_EXEC);
1354 
1355 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1356 
1357 	SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1358 
1359 	emulexec(l, epp);
1360 
1361 	/* Allow new references from the debugger/procfs. */
1362 	rw_exit(&p->p_reflock);
1363 	if (!no_local_exec_lock)
1364 		rw_exit(&exec_lock);
1365 
1366 	mutex_enter(&proc_lock);
1367 
1368 	/* posix_spawn(3) reports a single event with implied exec(3) */
1369 	if ((p->p_slflag & PSL_TRACED) && !is_spawn) {
1370 		mutex_enter(p->p_lock);
1371 		eventswitch(TRAP_EXEC, 0, 0);
1372 		mutex_enter(&proc_lock);
1373 	}
1374 
1375 	if (p->p_sflag & PS_STOPEXEC) {
1376 		ksiginfoq_t kq;
1377 
1378 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1379 		p->p_pptr->p_nstopchild++;
1380 		p->p_waited = 0;
1381 		mutex_enter(p->p_lock);
1382 		ksiginfo_queue_init(&kq);
1383 		sigclearall(p, &contsigmask, &kq);
1384 		lwp_lock(l);
1385 		l->l_stat = LSSTOP;
1386 		p->p_stat = SSTOP;
1387 		p->p_nrlwps--;
1388 		lwp_unlock(l);
1389 		mutex_exit(p->p_lock);
1390 		mutex_exit(&proc_lock);
1391 		lwp_lock(l);
1392 		spc_lock(l->l_cpu);
1393 		mi_switch(l);
1394 		ksiginfo_queue_drain(&kq);
1395 	} else {
1396 		mutex_exit(&proc_lock);
1397 	}
1398 
1399 	exec_path_free(data);
1400 #ifdef TRACE_EXEC
1401 	DPRINTF(("%s finished\n", __func__));
1402 #endif
1403 	return EJUSTRETURN;
1404 
1405  exec_abort:
1406 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1407 	rw_exit(&p->p_reflock);
1408 	if (!no_local_exec_lock)
1409 		rw_exit(&exec_lock);
1410 
1411 	exec_path_free(data);
1412 
1413 	/*
1414 	 * the old process doesn't exist anymore.  exit gracefully.
1415 	 * get rid of the (new) address space we have created, if any, get rid
1416 	 * of our namei data and vnode, and exit noting failure
1417 	 */
1418 	if (vm != NULL) {
1419 		uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1420 			VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1421 	}
1422 
1423 	exec_free_emul_arg(epp);
1424 	pool_put(&exec_pool, data->ed_argp);
1425 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1426 	if (epp->ep_emul_root != NULL)
1427 		vrele(epp->ep_emul_root);
1428 	if (epp->ep_interp != NULL)
1429 		vrele(epp->ep_interp);
1430 
1431 	/* Acquire the sched-state mutex (exit1() will release it). */
1432 	if (!is_spawn) {
1433 		mutex_enter(p->p_lock);
1434 		exit1(l, error, SIGABRT);
1435 	}
1436 
1437 	return error;
1438 }
1439 
1440 int
1441 execve1(struct lwp *l, bool has_path, const char *path, int fd,
1442     char * const *args, char * const *envs,
1443     execve_fetch_element_t fetch_element)
1444 {
1445 	struct execve_data data;
1446 	int error;
1447 
1448 	error = execve_loadvm(l, has_path, path, fd, args, envs, fetch_element,
1449 	    &data);
1450 	if (error)
1451 		return error;
1452 	error = execve_runproc(l, &data, false, false);
1453 	return error;
1454 }
1455 
1456 static size_t
1457 fromptrsz(const struct exec_package *epp)
1458 {
1459 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1460 }
1461 
1462 static size_t
1463 ptrsz(const struct exec_package *epp)
1464 {
1465 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1466 }
1467 
1468 static size_t
1469 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1470 {
1471 	struct exec_package	* const epp = &data->ed_pack;
1472 
1473 	const size_t nargenvptrs =
1474 	    1 +				/* long argc */
1475 	    data->ed_argc +		/* char *argv[] */
1476 	    1 +				/* \0 */
1477 	    data->ed_envc +		/* char *env[] */
1478 	    1;				/* \0 */
1479 
1480 	return (nargenvptrs * ptrsz(epp))	/* pointers */
1481 	    + argenvstrlen			/* strings */
1482 	    + epp->ep_esch->es_arglen;		/* auxinfo */
1483 }
1484 
1485 static size_t
1486 calcstack(struct execve_data * restrict data, const size_t gaplen)
1487 {
1488 	struct exec_package	* const epp = &data->ed_pack;
1489 
1490 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1491 	    epp->ep_esch->es_emul->e_sigcode;
1492 
1493 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1494 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1495 
1496 	const size_t sigcode_psstr_sz =
1497 	    data->ed_szsigcode +	/* sigcode */
1498 	    data->ed_ps_strings_sz +	/* ps_strings */
1499 	    STACK_PTHREADSPACE;		/* pthread space */
1500 
1501 	const size_t stacklen =
1502 	    data->ed_argslen +
1503 	    gaplen +
1504 	    sigcode_psstr_sz;
1505 
1506 	/* make the stack "safely" aligned */
1507 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1508 }
1509 
1510 static int
1511 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1512     char * const newstack)
1513 {
1514 	struct exec_package	* const epp = &data->ed_pack;
1515 	struct proc		*p = l->l_proc;
1516 	int			error;
1517 
1518 	memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo));
1519 
1520 	/* remember information about the process */
1521 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1522 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1523 
1524 	/*
1525 	 * Allocate the stack address passed to the newly execve()'ed process.
1526 	 *
1527 	 * The new stack address will be set to the SP (stack pointer) register
1528 	 * in setregs().
1529 	 */
1530 
1531 	char *newargs = STACK_ALLOC(
1532 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1533 
1534 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1535 	    &data->ed_arginfo, &newargs, data->ed_argp);
1536 
1537 	if (error) {
1538 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1539 		return error;
1540 	}
1541 
1542 	error = copyoutpsstrs(data, p);
1543 	if (error != 0)
1544 		return error;
1545 
1546 	return 0;
1547 }
1548 
1549 static int
1550 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1551 {
1552 	struct exec_package	* const epp = &data->ed_pack;
1553 	struct ps_strings32	arginfo32;
1554 	void			*aip;
1555 	int			error;
1556 
1557 	/* fill process ps_strings info */
1558 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1559 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1560 
1561 	if (epp->ep_flags & EXEC_32) {
1562 		aip = &arginfo32;
1563 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1564 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1565 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1566 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1567 	} else
1568 		aip = &data->ed_arginfo;
1569 
1570 	/* copy out the process's ps_strings structure */
1571 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1572 	    != 0) {
1573 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1574 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1575 		return error;
1576 	}
1577 
1578 	return 0;
1579 }
1580 
1581 static int
1582 copyinargs(struct execve_data * restrict data, char * const *args,
1583     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1584 {
1585 	struct exec_package	* const epp = &data->ed_pack;
1586 	char			*dp;
1587 	size_t			i;
1588 	int			error;
1589 
1590 	dp = *dpp;
1591 
1592 	data->ed_argc = 0;
1593 
1594 	/* copy the fake args list, if there's one, freeing it as we go */
1595 	if (epp->ep_flags & EXEC_HASARGL) {
1596 		struct exec_fakearg	*fa = epp->ep_fa;
1597 
1598 		while (fa->fa_arg != NULL) {
1599 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1600 			size_t len;
1601 
1602 			len = strlcpy(dp, fa->fa_arg, maxlen);
1603 			/* Count NUL into len. */
1604 			if (len < maxlen)
1605 				len++;
1606 			else {
1607 				while (fa->fa_arg != NULL) {
1608 					kmem_free(fa->fa_arg, fa->fa_len);
1609 					fa++;
1610 				}
1611 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1612 				epp->ep_flags &= ~EXEC_HASARGL;
1613 				return E2BIG;
1614 			}
1615 			ktrexecarg(fa->fa_arg, len - 1);
1616 			dp += len;
1617 
1618 			kmem_free(fa->fa_arg, fa->fa_len);
1619 			fa++;
1620 			data->ed_argc++;
1621 		}
1622 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1623 		epp->ep_flags &= ~EXEC_HASARGL;
1624 	}
1625 
1626 	/*
1627 	 * Read and count argument strings from user.
1628 	 */
1629 
1630 	if (args == NULL) {
1631 		DPRINTF(("%s: null args\n", __func__));
1632 		return EINVAL;
1633 	}
1634 	if (epp->ep_flags & EXEC_SKIPARG)
1635 		args = (const void *)((const char *)args + fromptrsz(epp));
1636 	i = 0;
1637 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1638 	if (error != 0) {
1639 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1640 		return error;
1641 	}
1642 	data->ed_argc += i;
1643 
1644 	/*
1645 	 * Read and count environment strings from user.
1646 	 */
1647 
1648 	data->ed_envc = 0;
1649 	/* environment need not be there */
1650 	if (envs == NULL)
1651 		goto done;
1652 	i = 0;
1653 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1654 	if (error != 0) {
1655 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1656 		return error;
1657 	}
1658 	data->ed_envc += i;
1659 
1660 done:
1661 	*dpp = dp;
1662 
1663 	return 0;
1664 }
1665 
1666 static int
1667 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1668     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1669     void (*ktr)(const void *, size_t))
1670 {
1671 	char			*dp, *sp;
1672 	size_t			i;
1673 	int			error;
1674 
1675 	dp = *dpp;
1676 
1677 	i = 0;
1678 	while (1) {
1679 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1680 		size_t len;
1681 
1682 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1683 			return error;
1684 		}
1685 		if (!sp)
1686 			break;
1687 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1688 			if (error == ENAMETOOLONG)
1689 				error = E2BIG;
1690 			return error;
1691 		}
1692 		if (__predict_false(ktrace_on))
1693 			(*ktr)(dp, len - 1);
1694 		dp += len;
1695 		i++;
1696 	}
1697 
1698 	*dpp = dp;
1699 	*ip = i;
1700 
1701 	return 0;
1702 }
1703 
1704 /*
1705  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1706  * Those strings are located just after auxinfo.
1707  */
1708 int
1709 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1710     char **stackp, void *argp)
1711 {
1712 	char	**cpp, *dp, *sp;
1713 	size_t	len;
1714 	void	*nullp;
1715 	long	argc, envc;
1716 	int	error;
1717 
1718 	cpp = (char **)*stackp;
1719 	nullp = NULL;
1720 	argc = arginfo->ps_nargvstr;
1721 	envc = arginfo->ps_nenvstr;
1722 
1723 	/* argc on stack is long */
1724 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1725 
1726 	dp = (char *)(cpp +
1727 	    1 +				/* long argc */
1728 	    argc +			/* char *argv[] */
1729 	    1 +				/* \0 */
1730 	    envc +			/* char *env[] */
1731 	    1) +			/* \0 */
1732 	    pack->ep_esch->es_arglen;	/* auxinfo */
1733 	sp = argp;
1734 
1735 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1736 		COPYPRINTF("", cpp - 1, sizeof(argc));
1737 		return error;
1738 	}
1739 
1740 	/* XXX don't copy them out, remap them! */
1741 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1742 
1743 	for (; --argc >= 0; sp += len, dp += len) {
1744 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1745 			COPYPRINTF("", cpp - 1, sizeof(dp));
1746 			return error;
1747 		}
1748 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1749 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1750 			return error;
1751 		}
1752 	}
1753 
1754 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1755 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1756 		return error;
1757 	}
1758 
1759 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1760 
1761 	for (; --envc >= 0; sp += len, dp += len) {
1762 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1763 			COPYPRINTF("", cpp - 1, sizeof(dp));
1764 			return error;
1765 		}
1766 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1767 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1768 			return error;
1769 		}
1770 
1771 	}
1772 
1773 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1774 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1775 		return error;
1776 	}
1777 
1778 	*stackp = (char *)cpp;
1779 	return 0;
1780 }
1781 
1782 
1783 /*
1784  * Add execsw[] entries.
1785  */
1786 int
1787 exec_add(struct execsw *esp, int count)
1788 {
1789 	struct exec_entry	*it;
1790 	int			i;
1791 
1792 	if (count == 0) {
1793 		return 0;
1794 	}
1795 
1796 	/* Check for duplicates. */
1797 	rw_enter(&exec_lock, RW_WRITER);
1798 	for (i = 0; i < count; i++) {
1799 		LIST_FOREACH(it, &ex_head, ex_list) {
1800 			/* assume unique (makecmds, probe_func, emulation) */
1801 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1802 			    it->ex_sw->u.elf_probe_func ==
1803 			    esp[i].u.elf_probe_func &&
1804 			    it->ex_sw->es_emul == esp[i].es_emul) {
1805 				rw_exit(&exec_lock);
1806 				return EEXIST;
1807 			}
1808 		}
1809 	}
1810 
1811 	/* Allocate new entries. */
1812 	for (i = 0; i < count; i++) {
1813 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1814 		it->ex_sw = &esp[i];
1815 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1816 	}
1817 
1818 	/* update execsw[] */
1819 	exec_init(0);
1820 	rw_exit(&exec_lock);
1821 	return 0;
1822 }
1823 
1824 /*
1825  * Remove execsw[] entry.
1826  */
1827 int
1828 exec_remove(struct execsw *esp, int count)
1829 {
1830 	struct exec_entry	*it, *next;
1831 	int			i;
1832 	const struct proclist_desc *pd;
1833 	proc_t			*p;
1834 
1835 	if (count == 0) {
1836 		return 0;
1837 	}
1838 
1839 	/* Abort if any are busy. */
1840 	rw_enter(&exec_lock, RW_WRITER);
1841 	for (i = 0; i < count; i++) {
1842 		mutex_enter(&proc_lock);
1843 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1844 			PROCLIST_FOREACH(p, pd->pd_list) {
1845 				if (p->p_execsw == &esp[i]) {
1846 					mutex_exit(&proc_lock);
1847 					rw_exit(&exec_lock);
1848 					return EBUSY;
1849 				}
1850 			}
1851 		}
1852 		mutex_exit(&proc_lock);
1853 	}
1854 
1855 	/* None are busy, so remove them all. */
1856 	for (i = 0; i < count; i++) {
1857 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1858 			next = LIST_NEXT(it, ex_list);
1859 			if (it->ex_sw == &esp[i]) {
1860 				LIST_REMOVE(it, ex_list);
1861 				kmem_free(it, sizeof(*it));
1862 				break;
1863 			}
1864 		}
1865 	}
1866 
1867 	/* update execsw[] */
1868 	exec_init(0);
1869 	rw_exit(&exec_lock);
1870 	return 0;
1871 }
1872 
1873 /*
1874  * Initialize exec structures. If init_boot is true, also does necessary
1875  * one-time initialization (it's called from main() that way).
1876  * Once system is multiuser, this should be called with exec_lock held,
1877  * i.e. via exec_{add|remove}().
1878  */
1879 int
1880 exec_init(int init_boot)
1881 {
1882 	const struct execsw 	**sw;
1883 	struct exec_entry	*ex;
1884 	SLIST_HEAD(,exec_entry)	first;
1885 	SLIST_HEAD(,exec_entry)	any;
1886 	SLIST_HEAD(,exec_entry)	last;
1887 	int			i, sz;
1888 
1889 	if (init_boot) {
1890 		/* do one-time initializations */
1891 		vaddr_t vmin = 0, vmax;
1892 
1893 		rw_init(&exec_lock);
1894 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1895 		exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1896 		    maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1897 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1898 		    "execargs", &exec_palloc, IPL_NONE);
1899 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1900 	} else {
1901 		KASSERT(rw_write_held(&exec_lock));
1902 	}
1903 
1904 	/* Sort each entry onto the appropriate queue. */
1905 	SLIST_INIT(&first);
1906 	SLIST_INIT(&any);
1907 	SLIST_INIT(&last);
1908 	sz = 0;
1909 	LIST_FOREACH(ex, &ex_head, ex_list) {
1910 		switch(ex->ex_sw->es_prio) {
1911 		case EXECSW_PRIO_FIRST:
1912 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1913 			break;
1914 		case EXECSW_PRIO_ANY:
1915 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1916 			break;
1917 		case EXECSW_PRIO_LAST:
1918 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1919 			break;
1920 		default:
1921 			panic("%s", __func__);
1922 			break;
1923 		}
1924 		sz++;
1925 	}
1926 
1927 	/*
1928 	 * Create new execsw[].  Ensure we do not try a zero-sized
1929 	 * allocation.
1930 	 */
1931 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1932 	i = 0;
1933 	SLIST_FOREACH(ex, &first, ex_slist) {
1934 		sw[i++] = ex->ex_sw;
1935 	}
1936 	SLIST_FOREACH(ex, &any, ex_slist) {
1937 		sw[i++] = ex->ex_sw;
1938 	}
1939 	SLIST_FOREACH(ex, &last, ex_slist) {
1940 		sw[i++] = ex->ex_sw;
1941 	}
1942 
1943 	/* Replace old execsw[] and free used memory. */
1944 	if (execsw != NULL) {
1945 		kmem_free(__UNCONST(execsw),
1946 		    nexecs * sizeof(struct execsw *) + 1);
1947 	}
1948 	execsw = sw;
1949 	nexecs = sz;
1950 
1951 	/* Figure out the maximum size of an exec header. */
1952 	exec_maxhdrsz = sizeof(int);
1953 	for (i = 0; i < nexecs; i++) {
1954 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1955 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1956 	}
1957 
1958 	return 0;
1959 }
1960 
1961 static int
1962 exec_sigcode_map(struct proc *p, const struct emul *e)
1963 {
1964 	vaddr_t va;
1965 	vsize_t sz;
1966 	int error;
1967 	struct uvm_object *uobj;
1968 
1969 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1970 
1971 	if (e->e_sigobject == NULL || sz == 0) {
1972 		return 0;
1973 	}
1974 
1975 	/*
1976 	 * If we don't have a sigobject for this emulation, create one.
1977 	 *
1978 	 * sigobject is an anonymous memory object (just like SYSV shared
1979 	 * memory) that we keep a permanent reference to and that we map
1980 	 * in all processes that need this sigcode. The creation is simple,
1981 	 * we create an object, add a permanent reference to it, map it in
1982 	 * kernel space, copy out the sigcode to it and unmap it.
1983 	 * We map it with PROT_READ|PROT_EXEC into the process just
1984 	 * the way sys_mmap() would map it.
1985 	 */
1986 
1987 	uobj = *e->e_sigobject;
1988 	if (uobj == NULL) {
1989 		mutex_enter(&sigobject_lock);
1990 		if ((uobj = *e->e_sigobject) == NULL) {
1991 			uobj = uao_create(sz, 0);
1992 			(*uobj->pgops->pgo_reference)(uobj);
1993 			va = vm_map_min(kernel_map);
1994 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1995 			    uobj, 0, 0,
1996 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1997 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1998 				printf("kernel mapping failed %d\n", error);
1999 				(*uobj->pgops->pgo_detach)(uobj);
2000 				mutex_exit(&sigobject_lock);
2001 				return error;
2002 			}
2003 			memcpy((void *)va, e->e_sigcode, sz);
2004 #ifdef PMAP_NEED_PROCWR
2005 			pmap_procwr(&proc0, va, sz);
2006 #endif
2007 			uvm_unmap(kernel_map, va, va + round_page(sz));
2008 			*e->e_sigobject = uobj;
2009 		}
2010 		mutex_exit(&sigobject_lock);
2011 	}
2012 
2013 	/* Just a hint to uvm_map where to put it. */
2014 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
2015 	    round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
2016 
2017 #ifdef __alpha__
2018 	/*
2019 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
2020 	 * which causes the above calculation to put the sigcode at
2021 	 * an invalid address.  Put it just below the text instead.
2022 	 */
2023 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
2024 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
2025 	}
2026 #endif
2027 
2028 	(*uobj->pgops->pgo_reference)(uobj);
2029 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
2030 			uobj, 0, 0,
2031 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
2032 				    UVM_ADV_RANDOM, 0));
2033 	if (error) {
2034 		DPRINTF(("%s, %d: map %p "
2035 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
2036 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
2037 		    va, error));
2038 		(*uobj->pgops->pgo_detach)(uobj);
2039 		return error;
2040 	}
2041 	p->p_sigctx.ps_sigcode = (void *)va;
2042 	return 0;
2043 }
2044 
2045 /*
2046  * Release a refcount on spawn_exec_data and destroy memory, if this
2047  * was the last one.
2048  */
2049 static void
2050 spawn_exec_data_release(struct spawn_exec_data *data)
2051 {
2052 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
2053 		return;
2054 
2055 	cv_destroy(&data->sed_cv_child_ready);
2056 	mutex_destroy(&data->sed_mtx_child);
2057 
2058 	if (data->sed_actions)
2059 		posix_spawn_fa_free(data->sed_actions,
2060 		    data->sed_actions->len);
2061 	if (data->sed_attrs)
2062 		kmem_free(data->sed_attrs,
2063 		    sizeof(*data->sed_attrs));
2064 	kmem_free(data, sizeof(*data));
2065 }
2066 
2067 static int
2068 handle_posix_spawn_file_actions(struct posix_spawn_file_actions *actions)
2069 {
2070 	struct lwp *l = curlwp;
2071 	register_t retval;
2072 	int error, newfd;
2073 
2074 	if (actions == NULL)
2075 		return 0;
2076 
2077 	for (size_t i = 0; i < actions->len; i++) {
2078 		const struct posix_spawn_file_actions_entry *fae =
2079 		    &actions->fae[i];
2080 		switch (fae->fae_action) {
2081 		case FAE_OPEN:
2082 			if (fd_getfile(fae->fae_fildes) != NULL) {
2083 				error = fd_close(fae->fae_fildes);
2084 				if (error)
2085 					return error;
2086 			}
2087 			error = fd_open(fae->fae_path, fae->fae_oflag,
2088 			    fae->fae_mode, &newfd);
2089 			if (error)
2090 				return error;
2091 			if (newfd != fae->fae_fildes) {
2092 				error = dodup(l, newfd,
2093 				    fae->fae_fildes, 0, &retval);
2094 				if (fd_getfile(newfd) != NULL)
2095 					fd_close(newfd);
2096 			}
2097 			break;
2098 		case FAE_DUP2:
2099 			error = dodup(l, fae->fae_fildes,
2100 			    fae->fae_newfildes, 0, &retval);
2101 			break;
2102 		case FAE_CLOSE:
2103 			if (fd_getfile(fae->fae_fildes) == NULL) {
2104 				return EBADF;
2105 			}
2106 			error = fd_close(fae->fae_fildes);
2107 			break;
2108 		}
2109 		if (error)
2110 			return error;
2111 	}
2112 	return 0;
2113 }
2114 
2115 static int
2116 handle_posix_spawn_attrs(struct posix_spawnattr *attrs, struct proc *parent)
2117 {
2118 	struct sigaction sigact;
2119 	int error;
2120 	struct proc *p = curproc;
2121 	struct lwp *l = curlwp;
2122 
2123 	if (attrs == NULL)
2124 		return 0;
2125 
2126 	memset(&sigact, 0, sizeof(sigact));
2127 	sigact._sa_u._sa_handler = SIG_DFL;
2128 	sigact.sa_flags = 0;
2129 
2130 	/*
2131 	 * set state to SSTOP so that this proc can be found by pid.
2132 	 * see proc_enterprp, do_sched_setparam below
2133 	 */
2134 	mutex_enter(&proc_lock);
2135 	/*
2136 	 * p_stat should be SACTIVE, so we need to adjust the
2137 	 * parent's p_nstopchild here.  For safety, just make
2138 	 * we're on the good side of SDEAD before we adjust.
2139 	 */
2140 	int ostat = p->p_stat;
2141 	KASSERT(ostat < SSTOP);
2142 	p->p_stat = SSTOP;
2143 	p->p_waited = 0;
2144 	p->p_pptr->p_nstopchild++;
2145 	mutex_exit(&proc_lock);
2146 
2147 	/* Set process group */
2148 	if (attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2149 		pid_t mypid = p->p_pid;
2150 		pid_t pgrp = attrs->sa_pgroup;
2151 
2152 		if (pgrp == 0)
2153 			pgrp = mypid;
2154 
2155 		error = proc_enterpgrp(parent, mypid, pgrp, false);
2156 		if (error)
2157 			goto out;
2158 	}
2159 
2160 	/* Set scheduler policy */
2161 	if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2162 		error = do_sched_setparam(p->p_pid, 0, attrs->sa_schedpolicy,
2163 		    &attrs->sa_schedparam);
2164 	else if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDPARAM) {
2165 		error = do_sched_setparam(parent->p_pid, 0,
2166 		    SCHED_NONE, &attrs->sa_schedparam);
2167 	}
2168 	if (error)
2169 		goto out;
2170 
2171 	/* Reset user ID's */
2172 	if (attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2173 		error = do_setresuid(l, -1, kauth_cred_getgid(l->l_cred), -1,
2174 		     ID_E_EQ_R | ID_E_EQ_S);
2175 		if (error)
2176 			return error;
2177 		error = do_setresuid(l, -1, kauth_cred_getuid(l->l_cred), -1,
2178 		    ID_E_EQ_R | ID_E_EQ_S);
2179 		if (error)
2180 			goto out;
2181 	}
2182 
2183 	/* Set signal masks/defaults */
2184 	if (attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2185 		mutex_enter(p->p_lock);
2186 		error = sigprocmask1(l, SIG_SETMASK, &attrs->sa_sigmask, NULL);
2187 		mutex_exit(p->p_lock);
2188 		if (error)
2189 			goto out;
2190 	}
2191 
2192 	if (attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2193 		/*
2194 		 * The following sigaction call is using a sigaction
2195 		 * version 0 trampoline which is in the compatibility
2196 		 * code only. This is not a problem because for SIG_DFL
2197 		 * and SIG_IGN, the trampolines are now ignored. If they
2198 		 * were not, this would be a problem because we are
2199 		 * holding the exec_lock, and the compat code needs
2200 		 * to do the same in order to replace the trampoline
2201 		 * code of the process.
2202 		 */
2203 		for (int i = 1; i <= NSIG; i++) {
2204 			if (sigismember(&attrs->sa_sigdefault, i))
2205 				sigaction1(l, i, &sigact, NULL, NULL, 0);
2206 		}
2207 	}
2208 	error = 0;
2209 out:
2210 	mutex_enter(&proc_lock);
2211 	p->p_stat = ostat;
2212 	p->p_pptr->p_nstopchild--;
2213 	mutex_exit(&proc_lock);
2214 	return error;
2215 }
2216 
2217 /*
2218  * A child lwp of a posix_spawn operation starts here and ends up in
2219  * cpu_spawn_return, dealing with all filedescriptor and scheduler
2220  * manipulations in between.
2221  * The parent waits for the child, as it is not clear whether the child
2222  * will be able to acquire its own exec_lock. If it can, the parent can
2223  * be released early and continue running in parallel. If not (or if the
2224  * magic debug flag is passed in the scheduler attribute struct), the
2225  * child rides on the parent's exec lock until it is ready to return to
2226  * to userland - and only then releases the parent. This method loses
2227  * concurrency, but improves error reporting.
2228  */
2229 static void
2230 spawn_return(void *arg)
2231 {
2232 	struct spawn_exec_data *spawn_data = arg;
2233 	struct lwp *l = curlwp;
2234 	struct proc *p = l->l_proc;
2235 	int error;
2236 	bool have_reflock;
2237 	bool parent_is_waiting = true;
2238 
2239 	/*
2240 	 * Check if we can release parent early.
2241 	 * We either need to have no sed_attrs, or sed_attrs does not
2242 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2243 	 * safe access to the parent proc (passed in sed_parent).
2244 	 * We then try to get the exec_lock, and only if that works, we can
2245 	 * release the parent here already.
2246 	 */
2247 	struct posix_spawnattr *attrs = spawn_data->sed_attrs;
2248 	if ((!attrs || (attrs->sa_flags
2249 		& (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2250 	    && rw_tryenter(&exec_lock, RW_READER)) {
2251 		parent_is_waiting = false;
2252 		mutex_enter(&spawn_data->sed_mtx_child);
2253 		cv_signal(&spawn_data->sed_cv_child_ready);
2254 		mutex_exit(&spawn_data->sed_mtx_child);
2255 	}
2256 
2257 	/* don't allow debugger access yet */
2258 	rw_enter(&p->p_reflock, RW_WRITER);
2259 	have_reflock = true;
2260 
2261 	/* handle posix_spawn_file_actions */
2262 	error = handle_posix_spawn_file_actions(spawn_data->sed_actions);
2263 	if (error)
2264 		goto report_error;
2265 
2266 	/* handle posix_spawnattr */
2267 	error = handle_posix_spawn_attrs(attrs, spawn_data->sed_parent);
2268 	if (error)
2269 		goto report_error;
2270 
2271 	/* now do the real exec */
2272 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2273 	    true);
2274 	have_reflock = false;
2275 	if (error == EJUSTRETURN)
2276 		error = 0;
2277 	else if (error)
2278 		goto report_error;
2279 
2280 	if (parent_is_waiting) {
2281 		mutex_enter(&spawn_data->sed_mtx_child);
2282 		cv_signal(&spawn_data->sed_cv_child_ready);
2283 		mutex_exit(&spawn_data->sed_mtx_child);
2284 	}
2285 
2286 	/* release our refcount on the data */
2287 	spawn_exec_data_release(spawn_data);
2288 
2289 	if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) ==
2290 	    (PSL_TRACED|PSL_TRACEDCHILD)) {
2291 		eventswitchchild(p, TRAP_CHLD, PTRACE_POSIX_SPAWN);
2292 	}
2293 
2294 	/* and finally: leave to userland for the first time */
2295 	cpu_spawn_return(l);
2296 
2297 	/* NOTREACHED */
2298 	return;
2299 
2300  report_error:
2301 	if (have_reflock) {
2302 		/*
2303 		 * We have not passed through execve_runproc(),
2304 		 * which would have released the p_reflock and also
2305 		 * taken ownership of the sed_exec part of spawn_data,
2306 		 * so release/free both here.
2307 		 */
2308 		rw_exit(&p->p_reflock);
2309 		execve_free_data(&spawn_data->sed_exec);
2310 	}
2311 
2312 	if (parent_is_waiting) {
2313 		/* pass error to parent */
2314 		mutex_enter(&spawn_data->sed_mtx_child);
2315 		spawn_data->sed_error = error;
2316 		cv_signal(&spawn_data->sed_cv_child_ready);
2317 		mutex_exit(&spawn_data->sed_mtx_child);
2318 	} else {
2319 		rw_exit(&exec_lock);
2320 	}
2321 
2322 	/* release our refcount on the data */
2323 	spawn_exec_data_release(spawn_data);
2324 
2325 	/* done, exit */
2326 	mutex_enter(p->p_lock);
2327 	/*
2328 	 * Posix explicitly asks for an exit code of 127 if we report
2329 	 * errors from the child process - so, unfortunately, there
2330 	 * is no way to report a more exact error code.
2331 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2332 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2333 	 */
2334 	exit1(l, 127, 0);
2335 }
2336 
2337 void
2338 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2339 {
2340 
2341 	for (size_t i = 0; i < len; i++) {
2342 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2343 		if (fae->fae_action != FAE_OPEN)
2344 			continue;
2345 		kmem_strfree(fae->fae_path);
2346 	}
2347 	if (fa->len > 0)
2348 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2349 	kmem_free(fa, sizeof(*fa));
2350 }
2351 
2352 static int
2353 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2354     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2355 {
2356 	struct posix_spawn_file_actions *fa;
2357 	struct posix_spawn_file_actions_entry *fae;
2358 	char *pbuf = NULL;
2359 	int error;
2360 	size_t i = 0;
2361 
2362 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2363 	error = copyin(ufa, fa, sizeof(*fa));
2364 	if (error || fa->len == 0) {
2365 		kmem_free(fa, sizeof(*fa));
2366 		return error;	/* 0 if not an error, and len == 0 */
2367 	}
2368 
2369 	if (fa->len > lim) {
2370 		kmem_free(fa, sizeof(*fa));
2371 		return EINVAL;
2372 	}
2373 
2374 	fa->size = fa->len;
2375 	size_t fal = fa->len * sizeof(*fae);
2376 	fae = fa->fae;
2377 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2378 	error = copyin(fae, fa->fae, fal);
2379 	if (error)
2380 		goto out;
2381 
2382 	pbuf = PNBUF_GET();
2383 	for (; i < fa->len; i++) {
2384 		fae = &fa->fae[i];
2385 		if (fae->fae_action != FAE_OPEN)
2386 			continue;
2387 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2388 		if (error)
2389 			goto out;
2390 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2391 		memcpy(fae->fae_path, pbuf, fal);
2392 	}
2393 	PNBUF_PUT(pbuf);
2394 
2395 	*fap = fa;
2396 	return 0;
2397 out:
2398 	if (pbuf)
2399 		PNBUF_PUT(pbuf);
2400 	posix_spawn_fa_free(fa, i);
2401 	return error;
2402 }
2403 
2404 /*
2405  * N.B. increments nprocs upon success.  Callers need to drop nprocs if
2406  * they fail for some other reason.
2407  */
2408 int
2409 check_posix_spawn(struct lwp *l1)
2410 {
2411 	int error, tnprocs, count;
2412 	uid_t uid;
2413 	struct proc *p1;
2414 
2415 	p1 = l1->l_proc;
2416 	uid = kauth_cred_getuid(l1->l_cred);
2417 	tnprocs = atomic_inc_uint_nv(&nprocs);
2418 
2419 	/*
2420 	 * Although process entries are dynamically created, we still keep
2421 	 * a global limit on the maximum number we will create.
2422 	 */
2423 	if (__predict_false(tnprocs >= maxproc))
2424 		error = -1;
2425 	else
2426 		error = kauth_authorize_process(l1->l_cred,
2427 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2428 
2429 	if (error) {
2430 		atomic_dec_uint(&nprocs);
2431 		return EAGAIN;
2432 	}
2433 
2434 	/*
2435 	 * Enforce limits.
2436 	 */
2437 	count = chgproccnt(uid, 1);
2438 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2439 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2440 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2441 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2442 		(void)chgproccnt(uid, -1);
2443 		atomic_dec_uint(&nprocs);
2444 		return EAGAIN;
2445 	}
2446 
2447 	return 0;
2448 }
2449 
2450 int
2451 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2452 	struct posix_spawn_file_actions *fa,
2453 	struct posix_spawnattr *sa,
2454 	char *const *argv, char *const *envp,
2455 	execve_fetch_element_t fetch)
2456 {
2457 
2458 	struct proc *p1, *p2;
2459 	struct lwp *l2;
2460 	int error;
2461 	struct spawn_exec_data *spawn_data;
2462 	vaddr_t uaddr;
2463 	pid_t pid;
2464 	bool have_exec_lock = false;
2465 
2466 	p1 = l1->l_proc;
2467 
2468 	/* Allocate and init spawn_data */
2469 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2470 	spawn_data->sed_refcnt = 1; /* only parent so far */
2471 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2472 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2473 	mutex_enter(&spawn_data->sed_mtx_child);
2474 
2475 	/*
2476 	 * Do the first part of the exec now, collect state
2477 	 * in spawn_data.
2478 	 */
2479 	error = execve_loadvm(l1, true, path, -1, argv,
2480 	    envp, fetch, &spawn_data->sed_exec);
2481 	if (error == EJUSTRETURN)
2482 		error = 0;
2483 	else if (error)
2484 		goto error_exit;
2485 
2486 	have_exec_lock = true;
2487 
2488 	/*
2489 	 * Allocate virtual address space for the U-area now, while it
2490 	 * is still easy to abort the fork operation if we're out of
2491 	 * kernel virtual address space.
2492 	 */
2493 	uaddr = uvm_uarea_alloc();
2494 	if (__predict_false(uaddr == 0)) {
2495 		error = ENOMEM;
2496 		goto error_exit;
2497 	}
2498 
2499 	/*
2500 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2501 	 * replace it with its own before returning to userland
2502 	 * in the child.
2503 	 */
2504 	p2 = proc_alloc();
2505 	if (p2 == NULL) {
2506 		/* We were unable to allocate a process ID. */
2507 		error = EAGAIN;
2508 		goto error_exit;
2509 	}
2510 
2511 	/*
2512 	 * This is a point of no return, we will have to go through
2513 	 * the child proc to properly clean it up past this point.
2514 	 */
2515 	pid = p2->p_pid;
2516 
2517 	/*
2518 	 * Make a proc table entry for the new process.
2519 	 * Start by zeroing the section of proc that is zero-initialized,
2520 	 * then copy the section that is copied directly from the parent.
2521 	 */
2522 	memset(&p2->p_startzero, 0,
2523 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2524 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2525 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2526 	p2->p_vmspace = proc0.p_vmspace;
2527 
2528 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2529 
2530 	LIST_INIT(&p2->p_lwps);
2531 	LIST_INIT(&p2->p_sigwaiters);
2532 
2533 	/*
2534 	 * Duplicate sub-structures as needed.
2535 	 * Increase reference counts on shared objects.
2536 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2537 	 * handling are important in order to keep a consistent behaviour
2538 	 * for the child after the fork.  If we are a 32-bit process, the
2539 	 * child will be too.
2540 	 */
2541 	p2->p_flag =
2542 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2543 	p2->p_emul = p1->p_emul;
2544 	p2->p_execsw = p1->p_execsw;
2545 
2546 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2547 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2548 	rw_init(&p2->p_reflock);
2549 	cv_init(&p2->p_waitcv, "wait");
2550 	cv_init(&p2->p_lwpcv, "lwpwait");
2551 
2552 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2553 
2554 	kauth_proc_fork(p1, p2);
2555 
2556 	p2->p_raslist = NULL;
2557 	p2->p_fd = fd_copy();
2558 
2559 	/* XXX racy */
2560 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2561 
2562 	p2->p_cwdi = cwdinit();
2563 
2564 	/*
2565 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2566 	 * we just need increase pl_refcnt.
2567 	 */
2568 	if (!p1->p_limit->pl_writeable) {
2569 		lim_addref(p1->p_limit);
2570 		p2->p_limit = p1->p_limit;
2571 	} else {
2572 		p2->p_limit = lim_copy(p1->p_limit);
2573 	}
2574 
2575 	p2->p_lflag = 0;
2576 	l1->l_vforkwaiting = false;
2577 	p2->p_sflag = 0;
2578 	p2->p_slflag = 0;
2579 	p2->p_pptr = p1;
2580 	p2->p_ppid = p1->p_pid;
2581 	LIST_INIT(&p2->p_children);
2582 
2583 	p2->p_aio = NULL;
2584 
2585 #ifdef KTRACE
2586 	/*
2587 	 * Copy traceflag and tracefile if enabled.
2588 	 * If not inherited, these were zeroed above.
2589 	 */
2590 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2591 		mutex_enter(&ktrace_lock);
2592 		p2->p_traceflag = p1->p_traceflag;
2593 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2594 			ktradref(p2);
2595 		mutex_exit(&ktrace_lock);
2596 	}
2597 #endif
2598 
2599 	/*
2600 	 * Create signal actions for the child process.
2601 	 */
2602 	p2->p_sigacts = sigactsinit(p1, 0);
2603 	mutex_enter(p1->p_lock);
2604 	p2->p_sflag |=
2605 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2606 	sched_proc_fork(p1, p2);
2607 	mutex_exit(p1->p_lock);
2608 
2609 	p2->p_stflag = p1->p_stflag;
2610 
2611 	/*
2612 	 * p_stats.
2613 	 * Copy parts of p_stats, and zero out the rest.
2614 	 */
2615 	p2->p_stats = pstatscopy(p1->p_stats);
2616 
2617 	/* copy over machdep flags to the new proc */
2618 	cpu_proc_fork(p1, p2);
2619 
2620 	/*
2621 	 * Prepare remaining parts of spawn data
2622 	 */
2623 	spawn_data->sed_actions = fa;
2624 	spawn_data->sed_attrs = sa;
2625 
2626 	spawn_data->sed_parent = p1;
2627 
2628 	/* create LWP */
2629 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2630 	    &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2631 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2632 
2633 	/*
2634 	 * Copy the credential so other references don't see our changes.
2635 	 * Test to see if this is necessary first, since in the common case
2636 	 * we won't need a private reference.
2637 	 */
2638 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2639 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2640 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2641 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2642 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2643 	}
2644 
2645 	/* Update the master credentials. */
2646 	if (l2->l_cred != p2->p_cred) {
2647 		kauth_cred_t ocred;
2648 
2649 		kauth_cred_hold(l2->l_cred);
2650 		mutex_enter(p2->p_lock);
2651 		ocred = p2->p_cred;
2652 		p2->p_cred = l2->l_cred;
2653 		mutex_exit(p2->p_lock);
2654 		kauth_cred_free(ocred);
2655 	}
2656 
2657 	*child_ok = true;
2658 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2659 #if 0
2660 	l2->l_nopreempt = 1; /* start it non-preemptable */
2661 #endif
2662 
2663 	/*
2664 	 * It's now safe for the scheduler and other processes to see the
2665 	 * child process.
2666 	 */
2667 	mutex_enter(&proc_lock);
2668 
2669 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2670 		p2->p_lflag |= PL_CONTROLT;
2671 
2672 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2673 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2674 
2675 	if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) ==
2676 	    (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2677 		proc_changeparent(p2, p1->p_pptr);
2678 		SET(p2->p_slflag, PSL_TRACEDCHILD);
2679 	}
2680 
2681 	p2->p_oppid = p1->p_pid;  /* Remember the original parent id. */
2682 
2683 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2684 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2685 
2686 	p2->p_trace_enabled = trace_is_enabled(p2);
2687 #ifdef __HAVE_SYSCALL_INTERN
2688 	(*p2->p_emul->e_syscall_intern)(p2);
2689 #endif
2690 
2691 	/*
2692 	 * Make child runnable, set start time, and add to run queue except
2693 	 * if the parent requested the child to start in SSTOP state.
2694 	 */
2695 	mutex_enter(p2->p_lock);
2696 
2697 	getmicrotime(&p2->p_stats->p_start);
2698 
2699 	lwp_lock(l2);
2700 	KASSERT(p2->p_nrlwps == 1);
2701 	KASSERT(l2->l_stat == LSIDL);
2702 	p2->p_nrlwps = 1;
2703 	p2->p_stat = SACTIVE;
2704 	setrunnable(l2);
2705 	/* LWP now unlocked */
2706 
2707 	mutex_exit(p2->p_lock);
2708 	mutex_exit(&proc_lock);
2709 
2710 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2711 	error = spawn_data->sed_error;
2712 	mutex_exit(&spawn_data->sed_mtx_child);
2713 	spawn_exec_data_release(spawn_data);
2714 
2715 	rw_exit(&p1->p_reflock);
2716 	rw_exit(&exec_lock);
2717 	have_exec_lock = false;
2718 
2719 	*pid_res = pid;
2720 
2721 	if (error)
2722 		return error;
2723 
2724 	if (p1->p_slflag & PSL_TRACED) {
2725 		/* Paranoid check */
2726 		mutex_enter(&proc_lock);
2727 		if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) !=
2728 		    (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2729 			mutex_exit(&proc_lock);
2730 			return 0;
2731 		}
2732 
2733 		mutex_enter(p1->p_lock);
2734 		eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, pid);
2735 	}
2736 	return 0;
2737 
2738  error_exit:
2739 	if (have_exec_lock) {
2740 		execve_free_data(&spawn_data->sed_exec);
2741 		rw_exit(&p1->p_reflock);
2742 		rw_exit(&exec_lock);
2743 	}
2744 	mutex_exit(&spawn_data->sed_mtx_child);
2745 	spawn_exec_data_release(spawn_data);
2746 
2747 	return error;
2748 }
2749 
2750 int
2751 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2752     register_t *retval)
2753 {
2754 	/* {
2755 		syscallarg(pid_t *) pid;
2756 		syscallarg(const char *) path;
2757 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2758 		syscallarg(const struct posix_spawnattr *) attrp;
2759 		syscallarg(char *const *) argv;
2760 		syscallarg(char *const *) envp;
2761 	} */
2762 
2763 	int error;
2764 	struct posix_spawn_file_actions *fa = NULL;
2765 	struct posix_spawnattr *sa = NULL;
2766 	pid_t pid;
2767 	bool child_ok = false;
2768 	rlim_t max_fileactions;
2769 	proc_t *p = l1->l_proc;
2770 
2771 	/* check_posix_spawn() increments nprocs for us. */
2772 	error = check_posix_spawn(l1);
2773 	if (error) {
2774 		*retval = error;
2775 		return 0;
2776 	}
2777 
2778 	/* copy in file_actions struct */
2779 	if (SCARG(uap, file_actions) != NULL) {
2780 		max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2781 		    maxfiles);
2782 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2783 		    max_fileactions);
2784 		if (error)
2785 			goto error_exit;
2786 	}
2787 
2788 	/* copyin posix_spawnattr struct */
2789 	if (SCARG(uap, attrp) != NULL) {
2790 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2791 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2792 		if (error)
2793 			goto error_exit;
2794 	}
2795 
2796 	/*
2797 	 * Do the spawn
2798 	 */
2799 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2800 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2801 	if (error)
2802 		goto error_exit;
2803 
2804 	if (error == 0 && SCARG(uap, pid) != NULL)
2805 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2806 
2807 	*retval = error;
2808 	return 0;
2809 
2810  error_exit:
2811 	if (!child_ok) {
2812 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2813 		atomic_dec_uint(&nprocs);
2814 
2815 		if (sa)
2816 			kmem_free(sa, sizeof(*sa));
2817 		if (fa)
2818 			posix_spawn_fa_free(fa, fa->len);
2819 	}
2820 
2821 	*retval = error;
2822 	return 0;
2823 }
2824 
2825 void
2826 exec_free_emul_arg(struct exec_package *epp)
2827 {
2828 	if (epp->ep_emul_arg_free != NULL) {
2829 		KASSERT(epp->ep_emul_arg != NULL);
2830 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2831 		epp->ep_emul_arg_free = NULL;
2832 		epp->ep_emul_arg = NULL;
2833 	} else {
2834 		KASSERT(epp->ep_emul_arg == NULL);
2835 	}
2836 }
2837 
2838 #ifdef DEBUG_EXEC
2839 static void
2840 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2841 {
2842 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2843 	size_t j;
2844 
2845 	if (error == 0)
2846 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2847 	else
2848 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2849 		    epp->ep_vmcmds.evs_used, error));
2850 
2851 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2852 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2853 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2854 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2855 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2856 		    "pagedvn" :
2857 		    vp[j].ev_proc == vmcmd_map_readvn ?
2858 		    "readvn" :
2859 		    vp[j].ev_proc == vmcmd_map_zero ?
2860 		    "zero" : "*unknown*",
2861 		    vp[j].ev_addr, vp[j].ev_len,
2862 		    vp[j].ev_offset, vp[j].ev_prot,
2863 		    vp[j].ev_flags));
2864 		if (error != 0 && j == x)
2865 			DPRINTF(("     ^--- failed\n"));
2866 	}
2867 }
2868 #endif
2869