xref: /netbsd-src/sys/kern/kern_exec.c (revision 53b02e147d4ed531c0d2a5ca9b3e8026ba3e99b5)
1 /*	$NetBSD: kern_exec.c,v 1.514 2021/11/26 08:06:12 ryo 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 1.514 2021/11/26 08:06:12 ryo 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/futex.h>
91 #include <sys/ktrace.h>
92 #include <sys/uidinfo.h>
93 #include <sys/wait.h>
94 #include <sys/mman.h>
95 #include <sys/ras.h>
96 #include <sys/signalvar.h>
97 #include <sys/stat.h>
98 #include <sys/syscall.h>
99 #include <sys/kauth.h>
100 #include <sys/lwpctl.h>
101 #include <sys/pax.h>
102 #include <sys/cpu.h>
103 #include <sys/module.h>
104 #include <sys/syscallvar.h>
105 #include <sys/syscallargs.h>
106 #include <sys/vfs_syscalls.h>
107 #if NVERIEXEC > 0
108 #include <sys/verified_exec.h>
109 #endif /* NVERIEXEC > 0 */
110 #include <sys/sdt.h>
111 #include <sys/spawn.h>
112 #include <sys/prot.h>
113 #include <sys/cprng.h>
114 
115 #include <uvm/uvm_extern.h>
116 
117 #include <machine/reg.h>
118 
119 #include <compat/common/compat_util.h>
120 
121 #ifndef MD_TOPDOWN_INIT
122 #ifdef __USE_TOPDOWN_VM
123 #define	MD_TOPDOWN_INIT(epp)	(epp)->ep_flags |= EXEC_TOPDOWN_VM
124 #else
125 #define	MD_TOPDOWN_INIT(epp)
126 #endif
127 #endif
128 
129 struct execve_data;
130 
131 extern int user_va0_disable;
132 
133 static size_t calcargs(struct execve_data * restrict, const size_t);
134 static size_t calcstack(struct execve_data * restrict, const size_t);
135 static int copyoutargs(struct execve_data * restrict, struct lwp *,
136     char * const);
137 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
138 static int copyinargs(struct execve_data * restrict, char * const *,
139     char * const *, execve_fetch_element_t, char **);
140 static int copyinargstrs(struct execve_data * restrict, char * const *,
141     execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
142 static int exec_sigcode_map(struct proc *, const struct emul *);
143 
144 #if defined(DEBUG) && !defined(DEBUG_EXEC)
145 #define DEBUG_EXEC
146 #endif
147 #ifdef DEBUG_EXEC
148 #define DPRINTF(a) printf a
149 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
150     __LINE__, (s), (a), (b))
151 static void dump_vmcmds(const struct exec_package * const, size_t, int);
152 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
153 #else
154 #define DPRINTF(a)
155 #define COPYPRINTF(s, a, b)
156 #define DUMPVMCMDS(p, x, e) do {} while (0)
157 #endif /* DEBUG_EXEC */
158 
159 /*
160  * DTrace SDT provider definitions
161  */
162 SDT_PROVIDER_DECLARE(proc);
163 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
164 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
165 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
166 
167 /*
168  * Exec function switch:
169  *
170  * Note that each makecmds function is responsible for loading the
171  * exec package with the necessary functions for any exec-type-specific
172  * handling.
173  *
174  * Functions for specific exec types should be defined in their own
175  * header file.
176  */
177 static const struct execsw	**execsw = NULL;
178 static int			nexecs;
179 
180 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
181 
182 /* list of dynamically loaded execsw entries */
183 static LIST_HEAD(execlist_head, exec_entry) ex_head =
184     LIST_HEAD_INITIALIZER(ex_head);
185 struct exec_entry {
186 	LIST_ENTRY(exec_entry)	ex_list;
187 	SLIST_ENTRY(exec_entry)	ex_slist;
188 	const struct execsw	*ex_sw;
189 };
190 
191 #ifndef __HAVE_SYSCALL_INTERN
192 void	syscall(void);
193 #endif
194 
195 /* NetBSD autoloadable syscalls */
196 #ifdef MODULAR
197 #include <kern/syscalls_autoload.c>
198 #endif
199 
200 /* NetBSD emul struct */
201 struct emul emul_netbsd = {
202 	.e_name =		"netbsd",
203 #ifdef EMUL_NATIVEROOT
204 	.e_path =		EMUL_NATIVEROOT,
205 #else
206 	.e_path =		NULL,
207 #endif
208 #ifndef __HAVE_MINIMAL_EMUL
209 	.e_flags =		EMUL_HAS_SYS___syscall,
210 	.e_errno =		NULL,
211 	.e_nosys =		SYS_syscall,
212 	.e_nsysent =		SYS_NSYSENT,
213 #endif
214 #ifdef MODULAR
215 	.e_sc_autoload =	netbsd_syscalls_autoload,
216 #endif
217 	.e_sysent =		sysent,
218 	.e_nomodbits =		sysent_nomodbits,
219 #ifdef SYSCALL_DEBUG
220 	.e_syscallnames =	syscallnames,
221 #else
222 	.e_syscallnames =	NULL,
223 #endif
224 	.e_sendsig =		sendsig,
225 	.e_trapsignal =		trapsignal,
226 	.e_sigcode =		NULL,
227 	.e_esigcode =		NULL,
228 	.e_sigobject =		NULL,
229 	.e_setregs =		setregs,
230 	.e_proc_exec =		NULL,
231 	.e_proc_fork =		NULL,
232 	.e_proc_exit =		NULL,
233 	.e_lwp_fork =		NULL,
234 	.e_lwp_exit =		NULL,
235 #ifdef __HAVE_SYSCALL_INTERN
236 	.e_syscall_intern =	syscall_intern,
237 #else
238 	.e_syscall =		syscall,
239 #endif
240 	.e_sysctlovly =		NULL,
241 	.e_vm_default_addr =	uvm_default_mapaddr,
242 	.e_usertrap =		NULL,
243 	.e_ucsize =		sizeof(ucontext_t),
244 	.e_startlwp =		startlwp
245 };
246 
247 /*
248  * Exec lock. Used to control access to execsw[] structures.
249  * This must not be static so that netbsd32 can access it, too.
250  */
251 krwlock_t exec_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 DIAGNOSTIC
502 #define LMSG "%s: rejecting due to %s limit (%ju > %ju)\n"
503 #endif
504 #ifdef MAXTSIZ
505 			if (epp->ep_tsize > MAXTSIZ) {
506 #ifdef DIAGNOSTIC
507 				printf(LMSG, __func__, "text",
508 				    (uintmax_t)epp->ep_tsize,
509 				    (uintmax_t)MAXTSIZ);
510 #endif
511 				error = ENOMEM;
512 				break;
513 			}
514 #endif
515 			vsize_t dlimit =
516 			    (vsize_t)l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur;
517 			if (epp->ep_dsize > dlimit) {
518 #ifdef DIAGNOSTIC
519 				printf(LMSG, __func__, "data",
520 				    (uintmax_t)epp->ep_dsize,
521 				    (uintmax_t)dlimit);
522 #endif
523 				error = ENOMEM;
524 				break;
525 			}
526 			return 0;
527 		}
528 
529 		/*
530 		 * Reset all the fields that may have been modified by the
531 		 * loader.
532 		 */
533 		KASSERT(epp->ep_emul_arg == NULL);
534 		if (epp->ep_emul_root != NULL) {
535 			vrele(epp->ep_emul_root);
536 			epp->ep_emul_root = NULL;
537 		}
538 		if (epp->ep_interp != NULL) {
539 			vrele(epp->ep_interp);
540 			epp->ep_interp = NULL;
541 		}
542 		epp->ep_pax_flags = 0;
543 
544 		/* make sure the first "interesting" error code is saved. */
545 		if (error == ENOEXEC)
546 			error = newerror;
547 
548 		if (epp->ep_flags & EXEC_DESTR)
549 			/* Error from "#!" code, tidied up by recursive call */
550 			return error;
551 	}
552 
553 	/* not found, error */
554 
555 	/*
556 	 * free any vmspace-creation commands,
557 	 * and release their references
558 	 */
559 	kill_vmcmds(&epp->ep_vmcmds);
560 
561 #if NVERIEXEC > 0 || defined(PAX_SEGVGUARD)
562 bad2:
563 #endif
564 	/*
565 	 * close and release the vnode, restore the old one, free the
566 	 * pathname buf, and punt.
567 	 */
568 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
569 	VOP_CLOSE(vp, FREAD, l->l_cred);
570 	vput(vp);
571 	return error;
572 
573 bad1:
574 	/*
575 	 * free the namei pathname buffer, and put the vnode
576 	 * (which we don't yet have open).
577 	 */
578 	vput(vp);				/* was still locked */
579 	return error;
580 }
581 
582 #ifdef __MACHINE_STACK_GROWS_UP
583 #define STACK_PTHREADSPACE NBPG
584 #else
585 #define STACK_PTHREADSPACE 0
586 #endif
587 
588 static int
589 execve_fetch_element(char * const *array, size_t index, char **value)
590 {
591 	return copyin(array + index, value, sizeof(*value));
592 }
593 
594 /*
595  * exec system call
596  */
597 int
598 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
599 {
600 	/* {
601 		syscallarg(const char *)	path;
602 		syscallarg(char * const *)	argp;
603 		syscallarg(char * const *)	envp;
604 	} */
605 
606 	return execve1(l, true, SCARG(uap, path), -1, SCARG(uap, argp),
607 	    SCARG(uap, envp), execve_fetch_element);
608 }
609 
610 int
611 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
612     register_t *retval)
613 {
614 	/* {
615 		syscallarg(int)			fd;
616 		syscallarg(char * const *)	argp;
617 		syscallarg(char * const *)	envp;
618 	} */
619 
620 	return execve1(l, false, NULL, SCARG(uap, fd), SCARG(uap, argp),
621 	    SCARG(uap, envp), execve_fetch_element);
622 }
623 
624 /*
625  * Load modules to try and execute an image that we do not understand.
626  * If no execsw entries are present, we load those likely to be needed
627  * in order to run native images only.  Otherwise, we autoload all
628  * possible modules that could let us run the binary.  XXX lame
629  */
630 static void
631 exec_autoload(void)
632 {
633 #ifdef MODULAR
634 	static const char * const native[] = {
635 		"exec_elf32",
636 		"exec_elf64",
637 		"exec_script",
638 		NULL
639 	};
640 	static const char * const compat[] = {
641 		"exec_elf32",
642 		"exec_elf64",
643 		"exec_script",
644 		"exec_aout",
645 		"exec_coff",
646 		"exec_ecoff",
647 		"compat_aoutm68k",
648 		"compat_netbsd32",
649 #if 0
650 		"compat_linux",
651 		"compat_linux32",
652 #endif
653 		"compat_sunos",
654 		"compat_sunos32",
655 		"compat_ultrix",
656 		NULL
657 	};
658 	char const * const *list;
659 	int i;
660 
661 	list = nexecs == 0 ? native : compat;
662 	for (i = 0; list[i] != NULL; i++) {
663 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
664 			continue;
665 		}
666 		yield();
667 	}
668 #endif
669 }
670 
671 /*
672  * Copy the user or kernel supplied upath to the allocated pathbuffer pbp
673  * making it absolute in the process, by prepending the current working
674  * directory if it is not. If offs is supplied it will contain the offset
675  * where the original supplied copy of upath starts.
676  */
677 int
678 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg,
679     struct pathbuf **pbp, size_t *offs)
680 {
681 	char *path, *bp;
682 	size_t len, tlen;
683 	int error;
684 	struct cwdinfo *cwdi;
685 
686 	path = PNBUF_GET();
687 	if (seg == UIO_SYSSPACE) {
688 		error = copystr(upath, path, MAXPATHLEN, &len);
689 	} else {
690 		error = copyinstr(upath, path, MAXPATHLEN, &len);
691 	}
692 	if (error)
693 		goto err;
694 
695 	if (path[0] == '/') {
696 		if (offs)
697 			*offs = 0;
698 		goto out;
699 	}
700 
701 	len++;
702 	if (len + 1 >= MAXPATHLEN) {
703 		error = ENAMETOOLONG;
704 		goto err;
705 	}
706 	bp = path + MAXPATHLEN - len;
707 	memmove(bp, path, len);
708 	*(--bp) = '/';
709 
710 	cwdi = l->l_proc->p_cwdi;
711 	rw_enter(&cwdi->cwdi_lock, RW_READER);
712 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
713 	    GETCWD_CHECK_ACCESS, l);
714 	rw_exit(&cwdi->cwdi_lock);
715 
716 	if (error)
717 		goto err;
718 	tlen = path + MAXPATHLEN - bp;
719 
720 	memmove(path, bp, tlen);
721 	path[tlen - 1] = '\0';
722 	if (offs)
723 		*offs = tlen - len;
724 out:
725 	*pbp = pathbuf_assimilate(path);
726 	return 0;
727 err:
728 	PNBUF_PUT(path);
729 	return error;
730 }
731 
732 vaddr_t
733 exec_vm_minaddr(vaddr_t va_min)
734 {
735 	/*
736 	 * Increase va_min if we don't want NULL to be mappable by the
737 	 * process.
738 	 */
739 #define VM_MIN_GUARD	PAGE_SIZE
740 	if (user_va0_disable && (va_min < VM_MIN_GUARD))
741 		return VM_MIN_GUARD;
742 	return va_min;
743 }
744 
745 static int
746 execve_loadvm(struct lwp *l, bool has_path, const char *path, int fd,
747 	char * const *args, char * const *envs,
748 	execve_fetch_element_t fetch_element,
749 	struct execve_data * restrict data)
750 {
751 	struct exec_package	* const epp = &data->ed_pack;
752 	int			error;
753 	struct proc		*p;
754 	char			*dp;
755 	u_int			modgen;
756 
757 	KASSERT(data != NULL);
758 
759 	p = l->l_proc;
760 	modgen = 0;
761 
762 	SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0);
763 
764 	/*
765 	 * Check if we have exceeded our number of processes limit.
766 	 * This is so that we handle the case where a root daemon
767 	 * forked, ran setuid to become the desired user and is trying
768 	 * to exec. The obvious place to do the reference counting check
769 	 * is setuid(), but we don't do the reference counting check there
770 	 * like other OS's do because then all the programs that use setuid()
771 	 * must be modified to check the return code of setuid() and exit().
772 	 * It is dangerous to make setuid() fail, because it fails open and
773 	 * the program will continue to run as root. If we make it succeed
774 	 * and return an error code, again we are not enforcing the limit.
775 	 * The best place to enforce the limit is here, when the process tries
776 	 * to execute a new image, because eventually the process will need
777 	 * to call exec in order to do something useful.
778 	 */
779  retry:
780 	if (p->p_flag & PK_SUGID) {
781 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
782 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
783 		     &p->p_rlimit[RLIMIT_NPROC],
784 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
785 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
786 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
787 		return EAGAIN;
788 	}
789 
790 	/*
791 	 * Drain existing references and forbid new ones.  The process
792 	 * should be left alone until we're done here.  This is necessary
793 	 * to avoid race conditions - e.g. in ptrace() - that might allow
794 	 * a local user to illicitly obtain elevated privileges.
795 	 */
796 	rw_enter(&p->p_reflock, RW_WRITER);
797 
798 	if (has_path) {
799 		size_t	offs;
800 		/*
801 		 * Init the namei data to point the file user's program name.
802 		 * This is done here rather than in check_exec(), so that it's
803 		 * possible to override this settings if any of makecmd/probe
804 		 * functions call check_exec() recursively - for example,
805 		 * see exec_script_makecmds().
806 		 */
807 		if ((error = exec_makepathbuf(l, path, UIO_USERSPACE,
808 		    &data->ed_pathbuf, &offs)) != 0)
809 			goto clrflg;
810 		data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
811 		epp->ep_kname = data->ed_pathstring + offs;
812 		data->ed_resolvedname = PNBUF_GET();
813 		epp->ep_resolvedname = data->ed_resolvedname;
814 		epp->ep_xfd = -1;
815 	} else {
816 		data->ed_pathbuf = pathbuf_assimilate(strcpy(PNBUF_GET(), "/"));
817 		data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
818 		epp->ep_kname = "*fexecve*";
819 		data->ed_resolvedname = NULL;
820 		epp->ep_resolvedname = NULL;
821 		epp->ep_xfd = fd;
822 	}
823 
824 
825 	/*
826 	 * initialize the fields of the exec package.
827 	 */
828 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
829 	epp->ep_hdrlen = exec_maxhdrsz;
830 	epp->ep_hdrvalid = 0;
831 	epp->ep_emul_arg = NULL;
832 	epp->ep_emul_arg_free = NULL;
833 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
834 	epp->ep_vap = &data->ed_attr;
835 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
836 	MD_TOPDOWN_INIT(epp);
837 	epp->ep_emul_root = NULL;
838 	epp->ep_interp = NULL;
839 	epp->ep_esch = NULL;
840 	epp->ep_pax_flags = 0;
841 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
842 
843 	rw_enter(&exec_lock, RW_READER);
844 
845 	/* see if we can run it. */
846 	if ((error = check_exec(l, epp, data->ed_pathbuf,
847 	    &data->ed_resolvedname)) != 0) {
848 		if (error != ENOENT && error != EACCES && error != ENOEXEC) {
849 			DPRINTF(("%s: check exec failed for %s, error %d\n",
850 			    __func__, epp->ep_kname, error));
851 		}
852 		goto freehdr;
853 	}
854 
855 	/* allocate an argument buffer */
856 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
857 	KASSERT(data->ed_argp != NULL);
858 	dp = data->ed_argp;
859 
860 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
861 		goto bad;
862 	}
863 
864 	/*
865 	 * Calculate the new stack size.
866 	 */
867 
868 #ifdef __MACHINE_STACK_GROWS_UP
869 /*
870  * copyargs() fills argc/argv/envp from the lower address even on
871  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
872  * so that _rtld() use it.
873  */
874 #define	RTLD_GAP	32
875 #else
876 #define	RTLD_GAP	0
877 #endif
878 
879 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
880 
881 	data->ed_argslen = calcargs(data, argenvstrlen);
882 
883 	const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP);
884 
885 	if (len > epp->ep_ssize) {
886 		/* in effect, compare to initial limit */
887 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
888 		error = ENOMEM;
889 		goto bad;
890 	}
891 	/* adjust "active stack depth" for process VSZ */
892 	epp->ep_ssize = len;
893 
894 	return 0;
895 
896  bad:
897 	/* free the vmspace-creation commands, and release their references */
898 	kill_vmcmds(&epp->ep_vmcmds);
899 	/* kill any opened file descriptor, if necessary */
900 	if (epp->ep_flags & EXEC_HASFD) {
901 		epp->ep_flags &= ~EXEC_HASFD;
902 		fd_close(epp->ep_fd);
903 	}
904 	/* close and put the exec'd file */
905 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
906 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
907 	vput(epp->ep_vp);
908 	pool_put(&exec_pool, data->ed_argp);
909 
910  freehdr:
911 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
912 	if (epp->ep_emul_root != NULL)
913 		vrele(epp->ep_emul_root);
914 	if (epp->ep_interp != NULL)
915 		vrele(epp->ep_interp);
916 
917 	rw_exit(&exec_lock);
918 
919 	exec_path_free(data);
920 
921  clrflg:
922 	rw_exit(&p->p_reflock);
923 
924 	if (modgen != module_gen && error == ENOEXEC) {
925 		modgen = module_gen;
926 		exec_autoload();
927 		goto retry;
928 	}
929 
930 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
931 	return error;
932 }
933 
934 static int
935 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
936 {
937 	struct exec_package	* const epp = &data->ed_pack;
938 	struct proc		*p = l->l_proc;
939 	struct exec_vmcmd	*base_vcp;
940 	int			error = 0;
941 	size_t			i;
942 
943 	/* record proc's vnode, for use by procfs and others */
944 	if (p->p_textvp)
945 		vrele(p->p_textvp);
946 	vref(epp->ep_vp);
947 	p->p_textvp = epp->ep_vp;
948 
949 	/* create the new process's VM space by running the vmcmds */
950 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
951 
952 #ifdef TRACE_EXEC
953 	DUMPVMCMDS(epp, 0, 0);
954 #endif
955 
956 	base_vcp = NULL;
957 
958 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
959 		struct exec_vmcmd *vcp;
960 
961 		vcp = &epp->ep_vmcmds.evs_cmds[i];
962 		if (vcp->ev_flags & VMCMD_RELATIVE) {
963 			KASSERTMSG(base_vcp != NULL,
964 			    "%s: relative vmcmd with no base", __func__);
965 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
966 			    "%s: illegal base & relative vmcmd", __func__);
967 			vcp->ev_addr += base_vcp->ev_addr;
968 		}
969 		error = (*vcp->ev_proc)(l, vcp);
970 		if (error)
971 			DUMPVMCMDS(epp, i, error);
972 		if (vcp->ev_flags & VMCMD_BASE)
973 			base_vcp = vcp;
974 	}
975 
976 	/* free the vmspace-creation commands, and release their references */
977 	kill_vmcmds(&epp->ep_vmcmds);
978 
979 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
980 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
981 	vput(epp->ep_vp);
982 
983 	/* if an error happened, deallocate and punt */
984 	if (error != 0) {
985 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
986 	}
987 	return error;
988 }
989 
990 static void
991 execve_free_data(struct execve_data *data)
992 {
993 	struct exec_package	* const epp = &data->ed_pack;
994 
995 	/* free the vmspace-creation commands, and release their references */
996 	kill_vmcmds(&epp->ep_vmcmds);
997 	/* kill any opened file descriptor, if necessary */
998 	if (epp->ep_flags & EXEC_HASFD) {
999 		epp->ep_flags &= ~EXEC_HASFD;
1000 		fd_close(epp->ep_fd);
1001 	}
1002 
1003 	/* close and put the exec'd file */
1004 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
1005 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
1006 	vput(epp->ep_vp);
1007 	pool_put(&exec_pool, data->ed_argp);
1008 
1009 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1010 	if (epp->ep_emul_root != NULL)
1011 		vrele(epp->ep_emul_root);
1012 	if (epp->ep_interp != NULL)
1013 		vrele(epp->ep_interp);
1014 
1015 	exec_path_free(data);
1016 }
1017 
1018 static void
1019 pathexec(struct proc *p, const char *resolvedname)
1020 {
1021 	/* set command name & other accounting info */
1022 	const char *cmdname;
1023 
1024 	if (resolvedname == NULL) {
1025 		cmdname = "*fexecve*";
1026 		resolvedname = "/";
1027 	} else {
1028 		cmdname = strrchr(resolvedname, '/') + 1;
1029 	}
1030 	KASSERTMSG(resolvedname[0] == '/', "bad resolvedname `%s'",
1031 	    resolvedname);
1032 
1033 	strlcpy(p->p_comm, cmdname, sizeof(p->p_comm));
1034 
1035 	kmem_strfree(p->p_path);
1036 	p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP);
1037 }
1038 
1039 /* XXX elsewhere */
1040 static int
1041 credexec(struct lwp *l, struct vattr *attr)
1042 {
1043 	struct proc *p = l->l_proc;
1044 	int error;
1045 
1046 	/*
1047 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
1048 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
1049 	 * out additional references on the process for the moment.
1050 	 */
1051 	if ((p->p_slflag & PSL_TRACED) == 0 &&
1052 
1053 	    (((attr->va_mode & S_ISUID) != 0 &&
1054 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
1055 
1056 	     ((attr->va_mode & S_ISGID) != 0 &&
1057 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
1058 		/*
1059 		 * Mark the process as SUGID before we do
1060 		 * anything that might block.
1061 		 */
1062 		proc_crmod_enter();
1063 		proc_crmod_leave(NULL, NULL, true);
1064 
1065 		/* Make sure file descriptors 0..2 are in use. */
1066 		if ((error = fd_checkstd()) != 0) {
1067 			DPRINTF(("%s: fdcheckstd failed %d\n",
1068 			    __func__, error));
1069 			return error;
1070 		}
1071 
1072 		/*
1073 		 * Copy the credential so other references don't see our
1074 		 * changes.
1075 		 */
1076 		l->l_cred = kauth_cred_copy(l->l_cred);
1077 #ifdef KTRACE
1078 		/*
1079 		 * If the persistent trace flag isn't set, turn off.
1080 		 */
1081 		if (p->p_tracep) {
1082 			mutex_enter(&ktrace_lock);
1083 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
1084 				ktrderef(p);
1085 			mutex_exit(&ktrace_lock);
1086 		}
1087 #endif
1088 		if (attr->va_mode & S_ISUID)
1089 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
1090 		if (attr->va_mode & S_ISGID)
1091 			kauth_cred_setegid(l->l_cred, attr->va_gid);
1092 	} else {
1093 		if (kauth_cred_geteuid(l->l_cred) ==
1094 		    kauth_cred_getuid(l->l_cred) &&
1095 		    kauth_cred_getegid(l->l_cred) ==
1096 		    kauth_cred_getgid(l->l_cred))
1097 			p->p_flag &= ~PK_SUGID;
1098 	}
1099 
1100 	/*
1101 	 * Copy the credential so other references don't see our changes.
1102 	 * Test to see if this is necessary first, since in the common case
1103 	 * we won't need a private reference.
1104 	 */
1105 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1106 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1107 		l->l_cred = kauth_cred_copy(l->l_cred);
1108 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1109 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1110 	}
1111 
1112 	/* Update the master credentials. */
1113 	if (l->l_cred != p->p_cred) {
1114 		kauth_cred_t ocred;
1115 
1116 		kauth_cred_hold(l->l_cred);
1117 		mutex_enter(p->p_lock);
1118 		ocred = p->p_cred;
1119 		p->p_cred = l->l_cred;
1120 		mutex_exit(p->p_lock);
1121 		kauth_cred_free(ocred);
1122 	}
1123 
1124 	return 0;
1125 }
1126 
1127 static void
1128 emulexec(struct lwp *l, struct exec_package *epp)
1129 {
1130 	struct proc		*p = l->l_proc;
1131 
1132 	/* The emulation root will usually have been found when we looked
1133 	 * for the elf interpreter (or similar), if not look now. */
1134 	if (epp->ep_esch->es_emul->e_path != NULL &&
1135 	    epp->ep_emul_root == NULL)
1136 		emul_find_root(l, epp);
1137 
1138 	/* Any old emulation root got removed by fdcloseexec */
1139 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1140 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1141 	rw_exit(&p->p_cwdi->cwdi_lock);
1142 	epp->ep_emul_root = NULL;
1143 	if (epp->ep_interp != NULL)
1144 		vrele(epp->ep_interp);
1145 
1146 	/*
1147 	 * Call emulation specific exec hook. This can setup per-process
1148 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1149 	 *
1150 	 * If we are executing process of different emulation than the
1151 	 * original forked process, call e_proc_exit() of the old emulation
1152 	 * first, then e_proc_exec() of new emulation. If the emulation is
1153 	 * same, the exec hook code should deallocate any old emulation
1154 	 * resources held previously by this process.
1155 	 */
1156 	if (p->p_emul && p->p_emul->e_proc_exit
1157 	    && p->p_emul != epp->ep_esch->es_emul)
1158 		(*p->p_emul->e_proc_exit)(p);
1159 
1160 	/*
1161 	 * Call exec hook. Emulation code may NOT store reference to anything
1162 	 * from &pack.
1163 	 */
1164 	if (epp->ep_esch->es_emul->e_proc_exec)
1165 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1166 
1167 	/* update p_emul, the old value is no longer needed */
1168 	p->p_emul = epp->ep_esch->es_emul;
1169 
1170 	/* ...and the same for p_execsw */
1171 	p->p_execsw = epp->ep_esch;
1172 
1173 #ifdef __HAVE_SYSCALL_INTERN
1174 	(*p->p_emul->e_syscall_intern)(p);
1175 #endif
1176 	ktremul();
1177 }
1178 
1179 static int
1180 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1181 	bool no_local_exec_lock, bool is_spawn)
1182 {
1183 	struct exec_package	* const epp = &data->ed_pack;
1184 	int error = 0;
1185 	struct proc		*p;
1186 	struct vmspace		*vm;
1187 
1188 	/*
1189 	 * In case of a posix_spawn operation, the child doing the exec
1190 	 * might not hold the reader lock on exec_lock, but the parent
1191 	 * will do this instead.
1192 	 */
1193 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1194 	KASSERT(!no_local_exec_lock || is_spawn);
1195 	KASSERT(data != NULL);
1196 
1197 	p = l->l_proc;
1198 
1199 	/* Get rid of other LWPs. */
1200 	if (p->p_nlwps > 1) {
1201 		mutex_enter(p->p_lock);
1202 		exit_lwps(l);
1203 		mutex_exit(p->p_lock);
1204 	}
1205 	KDASSERT(p->p_nlwps == 1);
1206 
1207 	/*
1208 	 * All of the other LWPs got rid of their robust futexes
1209 	 * when they exited above, but we might still have some
1210 	 * to dispose of.  Do that now.
1211 	 */
1212 	if (__predict_false(l->l_robust_head != 0)) {
1213 		futex_release_all_lwp(l);
1214 		/*
1215 		 * Since this LWP will live on with a different
1216 		 * program image, we need to clear the robust
1217 		 * futex list pointer here.
1218 		 */
1219 		l->l_robust_head = 0;
1220 	}
1221 
1222 	/* Destroy any lwpctl info. */
1223 	if (p->p_lwpctl != NULL)
1224 		lwp_ctl_exit();
1225 
1226 	/* Remove POSIX timers */
1227 	ptimers_free(p, TIMERS_POSIX);
1228 
1229 	/* Set the PaX flags. */
1230 	pax_set_flags(epp, p);
1231 
1232 	/*
1233 	 * Do whatever is necessary to prepare the address space
1234 	 * for remapping.  Note that this might replace the current
1235 	 * vmspace with another!
1236 	 *
1237 	 * vfork(): do not touch any user space data in the new child
1238 	 * until we have awoken the parent below, or it will defeat
1239 	 * lazy pmap switching (on x86).
1240 	 */
1241 	if (is_spawn)
1242 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1243 		    epp->ep_vm_maxaddr,
1244 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1245 	else
1246 		uvmspace_exec(l, epp->ep_vm_minaddr,
1247 		    epp->ep_vm_maxaddr,
1248 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1249 	vm = p->p_vmspace;
1250 
1251 	vm->vm_taddr = (void *)epp->ep_taddr;
1252 	vm->vm_tsize = btoc(epp->ep_tsize);
1253 	vm->vm_daddr = (void*)epp->ep_daddr;
1254 	vm->vm_dsize = btoc(epp->ep_dsize);
1255 	vm->vm_ssize = btoc(epp->ep_ssize);
1256 	vm->vm_issize = 0;
1257 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1258 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1259 
1260 	pax_aslr_init_vm(l, vm, epp);
1261 
1262 	cwdexec(p);
1263 	fd_closeexec();		/* handle close on exec */
1264 
1265 	if (__predict_false(ktrace_on))
1266 		fd_ktrexecfd();
1267 
1268 	execsigs(p);		/* reset caught signals */
1269 
1270 	mutex_enter(p->p_lock);
1271 	l->l_ctxlink = NULL;	/* reset ucontext link */
1272 	p->p_acflag &= ~AFORK;
1273 	p->p_flag |= PK_EXEC;
1274 	mutex_exit(p->p_lock);
1275 
1276 	error = credexec(l, &data->ed_attr);
1277 	if (error)
1278 		goto exec_abort;
1279 
1280 #if defined(__HAVE_RAS)
1281 	/*
1282 	 * Remove all RASs from the address space.
1283 	 */
1284 	ras_purgeall();
1285 #endif
1286 
1287 	/*
1288 	 * Stop profiling.
1289 	 */
1290 	if ((p->p_stflag & PST_PROFIL) != 0) {
1291 		mutex_spin_enter(&p->p_stmutex);
1292 		stopprofclock(p);
1293 		mutex_spin_exit(&p->p_stmutex);
1294 	}
1295 
1296 	/*
1297 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1298 	 * exited and exec()/exit() are the only places it will be cleared.
1299 	 *
1300 	 * Once the parent has been awoken, curlwp may teleport to a new CPU
1301 	 * in sched_vforkexec(), and it's then OK to start messing with user
1302 	 * data.  See comment above.
1303 	 */
1304 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1305 		bool samecpu;
1306 		lwp_t *lp;
1307 
1308 		mutex_enter(&proc_lock);
1309 		lp = p->p_vforklwp;
1310 		p->p_vforklwp = NULL;
1311 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1312 		cv_broadcast(&lp->l_waitcv);
1313 
1314 		/* Clear flags after cv_broadcast() (scheduler needs them). */
1315 		p->p_lflag &= ~PL_PPWAIT;
1316 		lp->l_vforkwaiting = false;
1317 
1318 		/* If parent is still on same CPU, teleport curlwp elsewhere. */
1319 		samecpu = (lp->l_cpu == curlwp->l_cpu);
1320 		mutex_exit(&proc_lock);
1321 
1322 		/* Give the parent its CPU back - find a new home. */
1323 		KASSERT(!is_spawn);
1324 		sched_vforkexec(l, samecpu);
1325 	}
1326 
1327 	/* Now map address space. */
1328 	error = execve_dovmcmds(l, data);
1329 	if (error != 0)
1330 		goto exec_abort;
1331 
1332 	pathexec(p, epp->ep_resolvedname);
1333 
1334 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1335 
1336 	error = copyoutargs(data, l, newstack);
1337 	if (error != 0)
1338 		goto exec_abort;
1339 
1340 	doexechooks(p);
1341 
1342 	/*
1343 	 * Set initial SP at the top of the stack.
1344 	 *
1345 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1346 	 * the end of arg/env strings.  Userland guesses the address of argc
1347 	 * via ps_strings::ps_argvstr.
1348 	 */
1349 
1350 	/* Setup new registers and do misc. setup. */
1351 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1352 	if (epp->ep_esch->es_setregs)
1353 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1354 
1355 	/* Provide a consistent LWP private setting */
1356 	(void)lwp_setprivate(l, NULL);
1357 
1358 	/* Discard all PCU state; need to start fresh */
1359 	pcu_discard_all(l);
1360 
1361 	/* map the process's signal trampoline code */
1362 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1363 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1364 		goto exec_abort;
1365 	}
1366 
1367 	pool_put(&exec_pool, data->ed_argp);
1368 
1369 	/*
1370 	 * Notify anyone who might care that we've exec'd.
1371 	 *
1372 	 * This is slightly racy; someone could sneak in and
1373 	 * attach a knote after we've decided not to notify,
1374 	 * or vice-versa, but that's not particularly bothersome.
1375 	 * knote_proc_exec() will acquire p->p_lock as needed.
1376 	 */
1377 	if (!SLIST_EMPTY(&p->p_klist)) {
1378 		knote_proc_exec(p);
1379 	}
1380 
1381 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1382 
1383 	SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0);
1384 
1385 	emulexec(l, epp);
1386 
1387 	/* Allow new references from the debugger/procfs. */
1388 	rw_exit(&p->p_reflock);
1389 	if (!no_local_exec_lock)
1390 		rw_exit(&exec_lock);
1391 
1392 	mutex_enter(&proc_lock);
1393 
1394 	/* posix_spawn(3) reports a single event with implied exec(3) */
1395 	if ((p->p_slflag & PSL_TRACED) && !is_spawn) {
1396 		mutex_enter(p->p_lock);
1397 		eventswitch(TRAP_EXEC, 0, 0);
1398 		mutex_enter(&proc_lock);
1399 	}
1400 
1401 	if (p->p_sflag & PS_STOPEXEC) {
1402 		ksiginfoq_t kq;
1403 
1404 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1405 		p->p_pptr->p_nstopchild++;
1406 		p->p_waited = 0;
1407 		mutex_enter(p->p_lock);
1408 		ksiginfo_queue_init(&kq);
1409 		sigclearall(p, &contsigmask, &kq);
1410 		lwp_lock(l);
1411 		l->l_stat = LSSTOP;
1412 		p->p_stat = SSTOP;
1413 		p->p_nrlwps--;
1414 		lwp_unlock(l);
1415 		mutex_exit(p->p_lock);
1416 		mutex_exit(&proc_lock);
1417 		lwp_lock(l);
1418 		spc_lock(l->l_cpu);
1419 		mi_switch(l);
1420 		ksiginfo_queue_drain(&kq);
1421 	} else {
1422 		mutex_exit(&proc_lock);
1423 	}
1424 
1425 	exec_path_free(data);
1426 #ifdef TRACE_EXEC
1427 	DPRINTF(("%s finished\n", __func__));
1428 #endif
1429 	return EJUSTRETURN;
1430 
1431  exec_abort:
1432 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
1433 	rw_exit(&p->p_reflock);
1434 	if (!no_local_exec_lock)
1435 		rw_exit(&exec_lock);
1436 
1437 	exec_path_free(data);
1438 
1439 	/*
1440 	 * the old process doesn't exist anymore.  exit gracefully.
1441 	 * get rid of the (new) address space we have created, if any, get rid
1442 	 * of our namei data and vnode, and exit noting failure
1443 	 */
1444 	if (vm != NULL) {
1445 		uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1446 			VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1447 	}
1448 
1449 	exec_free_emul_arg(epp);
1450 	pool_put(&exec_pool, data->ed_argp);
1451 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1452 	if (epp->ep_emul_root != NULL)
1453 		vrele(epp->ep_emul_root);
1454 	if (epp->ep_interp != NULL)
1455 		vrele(epp->ep_interp);
1456 
1457 	/* Acquire the sched-state mutex (exit1() will release it). */
1458 	if (!is_spawn) {
1459 		mutex_enter(p->p_lock);
1460 		exit1(l, error, SIGABRT);
1461 	}
1462 
1463 	return error;
1464 }
1465 
1466 int
1467 execve1(struct lwp *l, bool has_path, const char *path, int fd,
1468     char * const *args, char * const *envs,
1469     execve_fetch_element_t fetch_element)
1470 {
1471 	struct execve_data data;
1472 	int error;
1473 
1474 	error = execve_loadvm(l, has_path, path, fd, args, envs, fetch_element,
1475 	    &data);
1476 	if (error)
1477 		return error;
1478 	error = execve_runproc(l, &data, false, false);
1479 	return error;
1480 }
1481 
1482 static size_t
1483 fromptrsz(const struct exec_package *epp)
1484 {
1485 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1486 }
1487 
1488 static size_t
1489 ptrsz(const struct exec_package *epp)
1490 {
1491 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1492 }
1493 
1494 static size_t
1495 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1496 {
1497 	struct exec_package	* const epp = &data->ed_pack;
1498 
1499 	const size_t nargenvptrs =
1500 	    1 +				/* long argc */
1501 	    data->ed_argc +		/* char *argv[] */
1502 	    1 +				/* \0 */
1503 	    data->ed_envc +		/* char *env[] */
1504 	    1;				/* \0 */
1505 
1506 	return (nargenvptrs * ptrsz(epp))	/* pointers */
1507 	    + argenvstrlen			/* strings */
1508 	    + epp->ep_esch->es_arglen;		/* auxinfo */
1509 }
1510 
1511 static size_t
1512 calcstack(struct execve_data * restrict data, const size_t gaplen)
1513 {
1514 	struct exec_package	* const epp = &data->ed_pack;
1515 
1516 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1517 	    epp->ep_esch->es_emul->e_sigcode;
1518 
1519 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1520 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1521 
1522 	const size_t sigcode_psstr_sz =
1523 	    data->ed_szsigcode +	/* sigcode */
1524 	    data->ed_ps_strings_sz +	/* ps_strings */
1525 	    STACK_PTHREADSPACE;		/* pthread space */
1526 
1527 	const size_t stacklen =
1528 	    data->ed_argslen +
1529 	    gaplen +
1530 	    sigcode_psstr_sz;
1531 
1532 	/* make the stack "safely" aligned */
1533 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1534 }
1535 
1536 static int
1537 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1538     char * const newstack)
1539 {
1540 	struct exec_package	* const epp = &data->ed_pack;
1541 	struct proc		*p = l->l_proc;
1542 	int			error;
1543 
1544 	memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo));
1545 
1546 	/* remember information about the process */
1547 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1548 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1549 
1550 	/*
1551 	 * Allocate the stack address passed to the newly execve()'ed process.
1552 	 *
1553 	 * The new stack address will be set to the SP (stack pointer) register
1554 	 * in setregs().
1555 	 */
1556 
1557 	char *newargs = STACK_ALLOC(
1558 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1559 
1560 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1561 	    &data->ed_arginfo, &newargs, data->ed_argp);
1562 
1563 	if (error) {
1564 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1565 		return error;
1566 	}
1567 
1568 	error = copyoutpsstrs(data, p);
1569 	if (error != 0)
1570 		return error;
1571 
1572 	return 0;
1573 }
1574 
1575 static int
1576 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1577 {
1578 	struct exec_package	* const epp = &data->ed_pack;
1579 	struct ps_strings32	arginfo32;
1580 	void			*aip;
1581 	int			error;
1582 
1583 	/* fill process ps_strings info */
1584 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1585 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1586 
1587 	if (epp->ep_flags & EXEC_32) {
1588 		aip = &arginfo32;
1589 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1590 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1591 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1592 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1593 	} else
1594 		aip = &data->ed_arginfo;
1595 
1596 	/* copy out the process's ps_strings structure */
1597 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1598 	    != 0) {
1599 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1600 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1601 		return error;
1602 	}
1603 
1604 	return 0;
1605 }
1606 
1607 static int
1608 copyinargs(struct execve_data * restrict data, char * const *args,
1609     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1610 {
1611 	struct exec_package	* const epp = &data->ed_pack;
1612 	char			*dp;
1613 	size_t			i;
1614 	int			error;
1615 
1616 	dp = *dpp;
1617 
1618 	data->ed_argc = 0;
1619 
1620 	/* copy the fake args list, if there's one, freeing it as we go */
1621 	if (epp->ep_flags & EXEC_HASARGL) {
1622 		struct exec_fakearg	*fa = epp->ep_fa;
1623 
1624 		while (fa->fa_arg != NULL) {
1625 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1626 			size_t len;
1627 
1628 			len = strlcpy(dp, fa->fa_arg, maxlen);
1629 			/* Count NUL into len. */
1630 			if (len < maxlen)
1631 				len++;
1632 			else {
1633 				while (fa->fa_arg != NULL) {
1634 					kmem_free(fa->fa_arg, fa->fa_len);
1635 					fa++;
1636 				}
1637 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1638 				epp->ep_flags &= ~EXEC_HASARGL;
1639 				return E2BIG;
1640 			}
1641 			ktrexecarg(fa->fa_arg, len - 1);
1642 			dp += len;
1643 
1644 			kmem_free(fa->fa_arg, fa->fa_len);
1645 			fa++;
1646 			data->ed_argc++;
1647 		}
1648 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1649 		epp->ep_flags &= ~EXEC_HASARGL;
1650 	}
1651 
1652 	/*
1653 	 * Read and count argument strings from user.
1654 	 */
1655 
1656 	if (args == NULL) {
1657 		DPRINTF(("%s: null args\n", __func__));
1658 		return EINVAL;
1659 	}
1660 	if (epp->ep_flags & EXEC_SKIPARG)
1661 		args = (const void *)((const char *)args + fromptrsz(epp));
1662 	i = 0;
1663 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1664 	if (error != 0) {
1665 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1666 		return error;
1667 	}
1668 	data->ed_argc += i;
1669 
1670 	/*
1671 	 * Read and count environment strings from user.
1672 	 */
1673 
1674 	data->ed_envc = 0;
1675 	/* environment need not be there */
1676 	if (envs == NULL)
1677 		goto done;
1678 	i = 0;
1679 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1680 	if (error != 0) {
1681 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1682 		return error;
1683 	}
1684 	data->ed_envc += i;
1685 
1686 done:
1687 	*dpp = dp;
1688 
1689 	return 0;
1690 }
1691 
1692 static int
1693 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1694     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1695     void (*ktr)(const void *, size_t))
1696 {
1697 	char			*dp, *sp;
1698 	size_t			i;
1699 	int			error;
1700 
1701 	dp = *dpp;
1702 
1703 	i = 0;
1704 	while (1) {
1705 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1706 		size_t len;
1707 
1708 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1709 			return error;
1710 		}
1711 		if (!sp)
1712 			break;
1713 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1714 			if (error == ENAMETOOLONG)
1715 				error = E2BIG;
1716 			return error;
1717 		}
1718 		if (__predict_false(ktrace_on))
1719 			(*ktr)(dp, len - 1);
1720 		dp += len;
1721 		i++;
1722 	}
1723 
1724 	*dpp = dp;
1725 	*ip = i;
1726 
1727 	return 0;
1728 }
1729 
1730 /*
1731  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1732  * Those strings are located just after auxinfo.
1733  */
1734 int
1735 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1736     char **stackp, void *argp)
1737 {
1738 	char	**cpp, *dp, *sp;
1739 	size_t	len;
1740 	void	*nullp;
1741 	long	argc, envc;
1742 	int	error;
1743 
1744 	cpp = (char **)*stackp;
1745 	nullp = NULL;
1746 	argc = arginfo->ps_nargvstr;
1747 	envc = arginfo->ps_nenvstr;
1748 
1749 	/* argc on stack is long */
1750 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1751 
1752 	dp = (char *)(cpp +
1753 	    1 +				/* long argc */
1754 	    argc +			/* char *argv[] */
1755 	    1 +				/* \0 */
1756 	    envc +			/* char *env[] */
1757 	    1) +			/* \0 */
1758 	    pack->ep_esch->es_arglen;	/* auxinfo */
1759 	sp = argp;
1760 
1761 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1762 		COPYPRINTF("", cpp - 1, sizeof(argc));
1763 		return error;
1764 	}
1765 
1766 	/* XXX don't copy them out, remap them! */
1767 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1768 
1769 	for (; --argc >= 0; sp += len, dp += len) {
1770 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1771 			COPYPRINTF("", cpp - 1, sizeof(dp));
1772 			return error;
1773 		}
1774 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1775 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1776 			return error;
1777 		}
1778 	}
1779 
1780 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1781 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1782 		return error;
1783 	}
1784 
1785 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1786 
1787 	for (; --envc >= 0; sp += len, dp += len) {
1788 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1789 			COPYPRINTF("", cpp - 1, sizeof(dp));
1790 			return error;
1791 		}
1792 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1793 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1794 			return error;
1795 		}
1796 
1797 	}
1798 
1799 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1800 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1801 		return error;
1802 	}
1803 
1804 	*stackp = (char *)cpp;
1805 	return 0;
1806 }
1807 
1808 
1809 /*
1810  * Add execsw[] entries.
1811  */
1812 int
1813 exec_add(struct execsw *esp, int count)
1814 {
1815 	struct exec_entry	*it;
1816 	int			i, error = 0;
1817 
1818 	if (count == 0) {
1819 		return 0;
1820 	}
1821 
1822 	/* Check for duplicates. */
1823 	rw_enter(&exec_lock, RW_WRITER);
1824 	for (i = 0; i < count; i++) {
1825 		LIST_FOREACH(it, &ex_head, ex_list) {
1826 			/* assume unique (makecmds, probe_func, emulation) */
1827 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1828 			    it->ex_sw->u.elf_probe_func ==
1829 			    esp[i].u.elf_probe_func &&
1830 			    it->ex_sw->es_emul == esp[i].es_emul) {
1831 				rw_exit(&exec_lock);
1832 				return EEXIST;
1833 			}
1834 		}
1835 	}
1836 
1837 	/* Allocate new entries. */
1838 	for (i = 0; i < count; i++) {
1839 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1840 		it->ex_sw = &esp[i];
1841 		error = exec_sigcode_alloc(it->ex_sw->es_emul);
1842 		if (error != 0) {
1843 			kmem_free(it, sizeof(*it));
1844 			break;
1845 		}
1846 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1847 	}
1848 	/* If even one fails, remove them all back. */
1849 	if (error != 0) {
1850 		for (i--; i >= 0; i--) {
1851 			it = LIST_FIRST(&ex_head);
1852 			LIST_REMOVE(it, ex_list);
1853 			exec_sigcode_free(it->ex_sw->es_emul);
1854 			kmem_free(it, sizeof(*it));
1855 		}
1856 		return error;
1857 	}
1858 
1859 	/* update execsw[] */
1860 	exec_init(0);
1861 	rw_exit(&exec_lock);
1862 	return 0;
1863 }
1864 
1865 /*
1866  * Remove execsw[] entry.
1867  */
1868 int
1869 exec_remove(struct execsw *esp, int count)
1870 {
1871 	struct exec_entry	*it, *next;
1872 	int			i;
1873 	const struct proclist_desc *pd;
1874 	proc_t			*p;
1875 
1876 	if (count == 0) {
1877 		return 0;
1878 	}
1879 
1880 	/* Abort if any are busy. */
1881 	rw_enter(&exec_lock, RW_WRITER);
1882 	for (i = 0; i < count; i++) {
1883 		mutex_enter(&proc_lock);
1884 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1885 			PROCLIST_FOREACH(p, pd->pd_list) {
1886 				if (p->p_execsw == &esp[i]) {
1887 					mutex_exit(&proc_lock);
1888 					rw_exit(&exec_lock);
1889 					return EBUSY;
1890 				}
1891 			}
1892 		}
1893 		mutex_exit(&proc_lock);
1894 	}
1895 
1896 	/* None are busy, so remove them all. */
1897 	for (i = 0; i < count; i++) {
1898 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1899 			next = LIST_NEXT(it, ex_list);
1900 			if (it->ex_sw == &esp[i]) {
1901 				LIST_REMOVE(it, ex_list);
1902 				exec_sigcode_free(it->ex_sw->es_emul);
1903 				kmem_free(it, sizeof(*it));
1904 				break;
1905 			}
1906 		}
1907 	}
1908 
1909 	/* update execsw[] */
1910 	exec_init(0);
1911 	rw_exit(&exec_lock);
1912 	return 0;
1913 }
1914 
1915 /*
1916  * Initialize exec structures. If init_boot is true, also does necessary
1917  * one-time initialization (it's called from main() that way).
1918  * Once system is multiuser, this should be called with exec_lock held,
1919  * i.e. via exec_{add|remove}().
1920  */
1921 int
1922 exec_init(int init_boot)
1923 {
1924 	const struct execsw 	**sw;
1925 	struct exec_entry	*ex;
1926 	SLIST_HEAD(,exec_entry)	first;
1927 	SLIST_HEAD(,exec_entry)	any;
1928 	SLIST_HEAD(,exec_entry)	last;
1929 	int			i, sz;
1930 
1931 	if (init_boot) {
1932 		/* do one-time initializations */
1933 		vaddr_t vmin = 0, vmax;
1934 
1935 		rw_init(&exec_lock);
1936 		exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax,
1937 		    maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL);
1938 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1939 		    "execargs", &exec_palloc, IPL_NONE);
1940 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1941 	} else {
1942 		KASSERT(rw_write_held(&exec_lock));
1943 	}
1944 
1945 	/* Sort each entry onto the appropriate queue. */
1946 	SLIST_INIT(&first);
1947 	SLIST_INIT(&any);
1948 	SLIST_INIT(&last);
1949 	sz = 0;
1950 	LIST_FOREACH(ex, &ex_head, ex_list) {
1951 		switch(ex->ex_sw->es_prio) {
1952 		case EXECSW_PRIO_FIRST:
1953 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1954 			break;
1955 		case EXECSW_PRIO_ANY:
1956 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1957 			break;
1958 		case EXECSW_PRIO_LAST:
1959 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1960 			break;
1961 		default:
1962 			panic("%s", __func__);
1963 			break;
1964 		}
1965 		sz++;
1966 	}
1967 
1968 	/*
1969 	 * Create new execsw[].  Ensure we do not try a zero-sized
1970 	 * allocation.
1971 	 */
1972 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1973 	i = 0;
1974 	SLIST_FOREACH(ex, &first, ex_slist) {
1975 		sw[i++] = ex->ex_sw;
1976 	}
1977 	SLIST_FOREACH(ex, &any, ex_slist) {
1978 		sw[i++] = ex->ex_sw;
1979 	}
1980 	SLIST_FOREACH(ex, &last, ex_slist) {
1981 		sw[i++] = ex->ex_sw;
1982 	}
1983 
1984 	/* Replace old execsw[] and free used memory. */
1985 	if (execsw != NULL) {
1986 		kmem_free(__UNCONST(execsw),
1987 		    nexecs * sizeof(struct execsw *) + 1);
1988 	}
1989 	execsw = sw;
1990 	nexecs = sz;
1991 
1992 	/* Figure out the maximum size of an exec header. */
1993 	exec_maxhdrsz = sizeof(int);
1994 	for (i = 0; i < nexecs; i++) {
1995 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1996 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1997 	}
1998 
1999 	return 0;
2000 }
2001 
2002 int
2003 exec_sigcode_alloc(const struct emul *e)
2004 {
2005 	vaddr_t va;
2006 	vsize_t sz;
2007 	int error;
2008 	struct uvm_object *uobj;
2009 
2010 	KASSERT(rw_lock_held(&exec_lock));
2011 
2012 	if (e == NULL || e->e_sigobject == NULL)
2013 		return 0;
2014 
2015 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
2016 	if (sz == 0)
2017 		return 0;
2018 
2019 	/*
2020 	 * Create a sigobject for this emulation.
2021 	 *
2022 	 * sigobject is an anonymous memory object (just like SYSV shared
2023 	 * memory) that we keep a permanent reference to and that we map
2024 	 * in all processes that need this sigcode. The creation is simple,
2025 	 * we create an object, add a permanent reference to it, map it in
2026 	 * kernel space, copy out the sigcode to it and unmap it.
2027 	 * We map it with PROT_READ|PROT_EXEC into the process just
2028 	 * the way sys_mmap() would map it.
2029 	 */
2030 	if (*e->e_sigobject == NULL) {
2031 		uobj = uao_create(sz, 0);
2032 		(*uobj->pgops->pgo_reference)(uobj);
2033 		va = vm_map_min(kernel_map);
2034 		if ((error = uvm_map(kernel_map, &va, round_page(sz),
2035 		    uobj, 0, 0,
2036 		    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
2037 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
2038 			printf("sigcode kernel mapping failed %d\n", error);
2039 			(*uobj->pgops->pgo_detach)(uobj);
2040 			return error;
2041 		}
2042 		memcpy((void *)va, e->e_sigcode, sz);
2043 #ifdef PMAP_NEED_PROCWR
2044 		pmap_procwr(&proc0, va, sz);
2045 #endif
2046 		uvm_unmap(kernel_map, va, va + round_page(sz));
2047 		*e->e_sigobject = uobj;
2048 		KASSERT(uobj->uo_refs == 1);
2049 	} else {
2050 		/* if already created, reference++ */
2051 		uobj = *e->e_sigobject;
2052 		(*uobj->pgops->pgo_reference)(uobj);
2053 	}
2054 
2055 	return 0;
2056 }
2057 
2058 void
2059 exec_sigcode_free(const struct emul *e)
2060 {
2061 	struct uvm_object *uobj;
2062 
2063 	KASSERT(rw_lock_held(&exec_lock));
2064 
2065 	if (e == NULL || e->e_sigobject == NULL)
2066 		return;
2067 
2068 	uobj = *e->e_sigobject;
2069 	if (uobj == NULL)
2070 		return;
2071 
2072 	if (uobj->uo_refs == 1)
2073 		*e->e_sigobject = NULL;	/* I'm the last person to reference. */
2074 	(*uobj->pgops->pgo_detach)(uobj);
2075 }
2076 
2077 static int
2078 exec_sigcode_map(struct proc *p, const struct emul *e)
2079 {
2080 	vaddr_t va;
2081 	vsize_t sz;
2082 	int error;
2083 	struct uvm_object *uobj;
2084 
2085 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
2086 	if (e->e_sigobject == NULL || sz == 0)
2087 		return 0;
2088 
2089 	uobj = *e->e_sigobject;
2090 	if (uobj == NULL)
2091 		return 0;
2092 
2093 	/* Just a hint to uvm_map where to put it. */
2094 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
2095 	    round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN);
2096 
2097 #ifdef __alpha__
2098 	/*
2099 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
2100 	 * which causes the above calculation to put the sigcode at
2101 	 * an invalid address.  Put it just below the text instead.
2102 	 */
2103 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
2104 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
2105 	}
2106 #endif
2107 
2108 	(*uobj->pgops->pgo_reference)(uobj);
2109 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
2110 			uobj, 0, 0,
2111 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
2112 				    UVM_ADV_RANDOM, 0));
2113 	if (error) {
2114 		DPRINTF(("%s, %d: map %p "
2115 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
2116 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
2117 		    va, error));
2118 		(*uobj->pgops->pgo_detach)(uobj);
2119 		return error;
2120 	}
2121 	p->p_sigctx.ps_sigcode = (void *)va;
2122 	return 0;
2123 }
2124 
2125 /*
2126  * Release a refcount on spawn_exec_data and destroy memory, if this
2127  * was the last one.
2128  */
2129 static void
2130 spawn_exec_data_release(struct spawn_exec_data *data)
2131 {
2132 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
2133 		return;
2134 
2135 	cv_destroy(&data->sed_cv_child_ready);
2136 	mutex_destroy(&data->sed_mtx_child);
2137 
2138 	if (data->sed_actions)
2139 		posix_spawn_fa_free(data->sed_actions,
2140 		    data->sed_actions->len);
2141 	if (data->sed_attrs)
2142 		kmem_free(data->sed_attrs,
2143 		    sizeof(*data->sed_attrs));
2144 	kmem_free(data, sizeof(*data));
2145 }
2146 
2147 static int
2148 handle_posix_spawn_file_actions(struct posix_spawn_file_actions *actions)
2149 {
2150 	struct lwp *l = curlwp;
2151 	register_t retval;
2152 	int error, newfd;
2153 
2154 	if (actions == NULL)
2155 		return 0;
2156 
2157 	for (size_t i = 0; i < actions->len; i++) {
2158 		const struct posix_spawn_file_actions_entry *fae =
2159 		    &actions->fae[i];
2160 		switch (fae->fae_action) {
2161 		case FAE_OPEN:
2162 			if (fd_getfile(fae->fae_fildes) != NULL) {
2163 				error = fd_close(fae->fae_fildes);
2164 				if (error)
2165 					return error;
2166 			}
2167 			error = fd_open(fae->fae_path, fae->fae_oflag,
2168 			    fae->fae_mode, &newfd);
2169 			if (error)
2170 				return error;
2171 			if (newfd != fae->fae_fildes) {
2172 				error = dodup(l, newfd,
2173 				    fae->fae_fildes, 0, &retval);
2174 				if (fd_getfile(newfd) != NULL)
2175 					fd_close(newfd);
2176 			}
2177 			break;
2178 		case FAE_DUP2:
2179 			error = dodup(l, fae->fae_fildes,
2180 			    fae->fae_newfildes, 0, &retval);
2181 			break;
2182 		case FAE_CLOSE:
2183 			if (fd_getfile(fae->fae_fildes) == NULL) {
2184 				return EBADF;
2185 			}
2186 			error = fd_close(fae->fae_fildes);
2187 			break;
2188 		case FAE_CHDIR:
2189 			error = do_sys_chdir(l, fae->fae_chdir_path,
2190 			    UIO_SYSSPACE, &retval);
2191 			break;
2192 		case FAE_FCHDIR:
2193 			error = do_sys_fchdir(l, fae->fae_fildes, &retval);
2194 			break;
2195 		}
2196 		if (error)
2197 			return error;
2198 	}
2199 	return 0;
2200 }
2201 
2202 static int
2203 handle_posix_spawn_attrs(struct posix_spawnattr *attrs, struct proc *parent)
2204 {
2205 	struct sigaction sigact;
2206 	int error;
2207 	struct proc *p = curproc;
2208 	struct lwp *l = curlwp;
2209 
2210 	if (attrs == NULL)
2211 		return 0;
2212 
2213 	memset(&sigact, 0, sizeof(sigact));
2214 	sigact._sa_u._sa_handler = SIG_DFL;
2215 	sigact.sa_flags = 0;
2216 
2217 	/*
2218 	 * set state to SSTOP so that this proc can be found by pid.
2219 	 * see proc_enterprp, do_sched_setparam below
2220 	 */
2221 	mutex_enter(&proc_lock);
2222 	/*
2223 	 * p_stat should be SACTIVE, so we need to adjust the
2224 	 * parent's p_nstopchild here.  For safety, just make
2225 	 * we're on the good side of SDEAD before we adjust.
2226 	 */
2227 	int ostat = p->p_stat;
2228 	KASSERT(ostat < SSTOP);
2229 	p->p_stat = SSTOP;
2230 	p->p_waited = 0;
2231 	p->p_pptr->p_nstopchild++;
2232 	mutex_exit(&proc_lock);
2233 
2234 	/* Set process group */
2235 	if (attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2236 		pid_t mypid = p->p_pid;
2237 		pid_t pgrp = attrs->sa_pgroup;
2238 
2239 		if (pgrp == 0)
2240 			pgrp = mypid;
2241 
2242 		error = proc_enterpgrp(parent, mypid, pgrp, false);
2243 		if (error)
2244 			goto out;
2245 	}
2246 
2247 	/* Set scheduler policy */
2248 	if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2249 		error = do_sched_setparam(p->p_pid, 0, attrs->sa_schedpolicy,
2250 		    &attrs->sa_schedparam);
2251 	else if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDPARAM) {
2252 		error = do_sched_setparam(parent->p_pid, 0,
2253 		    SCHED_NONE, &attrs->sa_schedparam);
2254 	}
2255 	if (error)
2256 		goto out;
2257 
2258 	/* Reset user ID's */
2259 	if (attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2260 		error = do_setresgid(l, -1, kauth_cred_getgid(l->l_cred), -1,
2261 		     ID_E_EQ_R | ID_E_EQ_S);
2262 		if (error)
2263 			return error;
2264 		error = do_setresuid(l, -1, kauth_cred_getuid(l->l_cred), -1,
2265 		    ID_E_EQ_R | ID_E_EQ_S);
2266 		if (error)
2267 			goto out;
2268 	}
2269 
2270 	/* Set signal masks/defaults */
2271 	if (attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2272 		mutex_enter(p->p_lock);
2273 		error = sigprocmask1(l, SIG_SETMASK, &attrs->sa_sigmask, NULL);
2274 		mutex_exit(p->p_lock);
2275 		if (error)
2276 			goto out;
2277 	}
2278 
2279 	if (attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2280 		/*
2281 		 * The following sigaction call is using a sigaction
2282 		 * version 0 trampoline which is in the compatibility
2283 		 * code only. This is not a problem because for SIG_DFL
2284 		 * and SIG_IGN, the trampolines are now ignored. If they
2285 		 * were not, this would be a problem because we are
2286 		 * holding the exec_lock, and the compat code needs
2287 		 * to do the same in order to replace the trampoline
2288 		 * code of the process.
2289 		 */
2290 		for (int i = 1; i <= NSIG; i++) {
2291 			if (sigismember(&attrs->sa_sigdefault, i))
2292 				sigaction1(l, i, &sigact, NULL, NULL, 0);
2293 		}
2294 	}
2295 	error = 0;
2296 out:
2297 	mutex_enter(&proc_lock);
2298 	p->p_stat = ostat;
2299 	p->p_pptr->p_nstopchild--;
2300 	mutex_exit(&proc_lock);
2301 	return error;
2302 }
2303 
2304 /*
2305  * A child lwp of a posix_spawn operation starts here and ends up in
2306  * cpu_spawn_return, dealing with all filedescriptor and scheduler
2307  * manipulations in between.
2308  * The parent waits for the child, as it is not clear whether the child
2309  * will be able to acquire its own exec_lock. If it can, the parent can
2310  * be released early and continue running in parallel. If not (or if the
2311  * magic debug flag is passed in the scheduler attribute struct), the
2312  * child rides on the parent's exec lock until it is ready to return to
2313  * to userland - and only then releases the parent. This method loses
2314  * concurrency, but improves error reporting.
2315  */
2316 static void
2317 spawn_return(void *arg)
2318 {
2319 	struct spawn_exec_data *spawn_data = arg;
2320 	struct lwp *l = curlwp;
2321 	struct proc *p = l->l_proc;
2322 	int error;
2323 	bool have_reflock;
2324 	bool parent_is_waiting = true;
2325 
2326 	/*
2327 	 * Check if we can release parent early.
2328 	 * We either need to have no sed_attrs, or sed_attrs does not
2329 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2330 	 * safe access to the parent proc (passed in sed_parent).
2331 	 * We then try to get the exec_lock, and only if that works, we can
2332 	 * release the parent here already.
2333 	 */
2334 	struct posix_spawnattr *attrs = spawn_data->sed_attrs;
2335 	if ((!attrs || (attrs->sa_flags
2336 		& (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2337 	    && rw_tryenter(&exec_lock, RW_READER)) {
2338 		parent_is_waiting = false;
2339 		mutex_enter(&spawn_data->sed_mtx_child);
2340 		cv_signal(&spawn_data->sed_cv_child_ready);
2341 		mutex_exit(&spawn_data->sed_mtx_child);
2342 	}
2343 
2344 	/* don't allow debugger access yet */
2345 	rw_enter(&p->p_reflock, RW_WRITER);
2346 	have_reflock = true;
2347 
2348 	/* handle posix_spawnattr */
2349 	error = handle_posix_spawn_attrs(attrs, spawn_data->sed_parent);
2350 	if (error)
2351 		goto report_error;
2352 
2353 	/* handle posix_spawn_file_actions */
2354 	error = handle_posix_spawn_file_actions(spawn_data->sed_actions);
2355 	if (error)
2356 		goto report_error;
2357 
2358 	/* now do the real exec */
2359 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2360 	    true);
2361 	have_reflock = false;
2362 	if (error == EJUSTRETURN)
2363 		error = 0;
2364 	else if (error)
2365 		goto report_error;
2366 
2367 	if (parent_is_waiting) {
2368 		mutex_enter(&spawn_data->sed_mtx_child);
2369 		cv_signal(&spawn_data->sed_cv_child_ready);
2370 		mutex_exit(&spawn_data->sed_mtx_child);
2371 	}
2372 
2373 	/* release our refcount on the data */
2374 	spawn_exec_data_release(spawn_data);
2375 
2376 	if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) ==
2377 	    (PSL_TRACED|PSL_TRACEDCHILD)) {
2378 		eventswitchchild(p, TRAP_CHLD, PTRACE_POSIX_SPAWN);
2379 	}
2380 
2381 	/* and finally: leave to userland for the first time */
2382 	cpu_spawn_return(l);
2383 
2384 	/* NOTREACHED */
2385 	return;
2386 
2387  report_error:
2388 	if (have_reflock) {
2389 		/*
2390 		 * We have not passed through execve_runproc(),
2391 		 * which would have released the p_reflock and also
2392 		 * taken ownership of the sed_exec part of spawn_data,
2393 		 * so release/free both here.
2394 		 */
2395 		rw_exit(&p->p_reflock);
2396 		execve_free_data(&spawn_data->sed_exec);
2397 	}
2398 
2399 	if (parent_is_waiting) {
2400 		/* pass error to parent */
2401 		mutex_enter(&spawn_data->sed_mtx_child);
2402 		spawn_data->sed_error = error;
2403 		cv_signal(&spawn_data->sed_cv_child_ready);
2404 		mutex_exit(&spawn_data->sed_mtx_child);
2405 	} else {
2406 		rw_exit(&exec_lock);
2407 	}
2408 
2409 	/* release our refcount on the data */
2410 	spawn_exec_data_release(spawn_data);
2411 
2412 	/* done, exit */
2413 	mutex_enter(p->p_lock);
2414 	/*
2415 	 * Posix explicitly asks for an exit code of 127 if we report
2416 	 * errors from the child process - so, unfortunately, there
2417 	 * is no way to report a more exact error code.
2418 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2419 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2420 	 */
2421 	exit1(l, 127, 0);
2422 }
2423 
2424 static __inline char **
2425 posix_spawn_fae_path(struct posix_spawn_file_actions_entry *fae)
2426 {
2427 	switch (fae->fae_action) {
2428 	case FAE_OPEN:
2429 		return &fae->fae_path;
2430 	case FAE_CHDIR:
2431 		return &fae->fae_chdir_path;
2432 	default:
2433 		return NULL;
2434 	}
2435 }
2436 
2437 void
2438 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2439 {
2440 
2441 	for (size_t i = 0; i < len; i++) {
2442 		char **pathp = posix_spawn_fae_path(&fa->fae[i]);
2443 		if (pathp)
2444 			kmem_strfree(*pathp);
2445 	}
2446 	if (fa->len > 0)
2447 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2448 	kmem_free(fa, sizeof(*fa));
2449 }
2450 
2451 static int
2452 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2453     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2454 {
2455 	struct posix_spawn_file_actions *fa;
2456 	struct posix_spawn_file_actions_entry *fae;
2457 	char *pbuf = NULL;
2458 	int error;
2459 	size_t i = 0;
2460 
2461 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2462 	error = copyin(ufa, fa, sizeof(*fa));
2463 	if (error || fa->len == 0) {
2464 		kmem_free(fa, sizeof(*fa));
2465 		return error;	/* 0 if not an error, and len == 0 */
2466 	}
2467 
2468 	if (fa->len > lim) {
2469 		kmem_free(fa, sizeof(*fa));
2470 		return EINVAL;
2471 	}
2472 
2473 	fa->size = fa->len;
2474 	size_t fal = fa->len * sizeof(*fae);
2475 	fae = fa->fae;
2476 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2477 	error = copyin(fae, fa->fae, fal);
2478 	if (error)
2479 		goto out;
2480 
2481 	pbuf = PNBUF_GET();
2482 	for (; i < fa->len; i++) {
2483 		char **pathp = posix_spawn_fae_path(&fa->fae[i]);
2484 		if (pathp == NULL)
2485 			continue;
2486 		error = copyinstr(*pathp, pbuf, MAXPATHLEN, &fal);
2487 		if (error)
2488 			goto out;
2489 		*pathp = kmem_alloc(fal, KM_SLEEP);
2490 		memcpy(*pathp, pbuf, fal);
2491 	}
2492 	PNBUF_PUT(pbuf);
2493 
2494 	*fap = fa;
2495 	return 0;
2496 out:
2497 	if (pbuf)
2498 		PNBUF_PUT(pbuf);
2499 	posix_spawn_fa_free(fa, i);
2500 	return error;
2501 }
2502 
2503 /*
2504  * N.B. increments nprocs upon success.  Callers need to drop nprocs if
2505  * they fail for some other reason.
2506  */
2507 int
2508 check_posix_spawn(struct lwp *l1)
2509 {
2510 	int error, tnprocs, count;
2511 	uid_t uid;
2512 	struct proc *p1;
2513 
2514 	p1 = l1->l_proc;
2515 	uid = kauth_cred_getuid(l1->l_cred);
2516 	tnprocs = atomic_inc_uint_nv(&nprocs);
2517 
2518 	/*
2519 	 * Although process entries are dynamically created, we still keep
2520 	 * a global limit on the maximum number we will create.
2521 	 */
2522 	if (__predict_false(tnprocs >= maxproc))
2523 		error = -1;
2524 	else
2525 		error = kauth_authorize_process(l1->l_cred,
2526 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2527 
2528 	if (error) {
2529 		atomic_dec_uint(&nprocs);
2530 		return EAGAIN;
2531 	}
2532 
2533 	/*
2534 	 * Enforce limits.
2535 	 */
2536 	count = chgproccnt(uid, 1);
2537 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2538 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2539 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2540 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2541 		(void)chgproccnt(uid, -1);
2542 		atomic_dec_uint(&nprocs);
2543 		return EAGAIN;
2544 	}
2545 
2546 	return 0;
2547 }
2548 
2549 int
2550 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2551 	struct posix_spawn_file_actions *fa,
2552 	struct posix_spawnattr *sa,
2553 	char *const *argv, char *const *envp,
2554 	execve_fetch_element_t fetch)
2555 {
2556 
2557 	struct proc *p1, *p2;
2558 	struct lwp *l2;
2559 	int error;
2560 	struct spawn_exec_data *spawn_data;
2561 	vaddr_t uaddr;
2562 	pid_t pid;
2563 	bool have_exec_lock = false;
2564 
2565 	p1 = l1->l_proc;
2566 
2567 	/* Allocate and init spawn_data */
2568 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2569 	spawn_data->sed_refcnt = 1; /* only parent so far */
2570 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2571 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2572 	mutex_enter(&spawn_data->sed_mtx_child);
2573 
2574 	/*
2575 	 * Do the first part of the exec now, collect state
2576 	 * in spawn_data.
2577 	 */
2578 	error = execve_loadvm(l1, true, path, -1, argv,
2579 	    envp, fetch, &spawn_data->sed_exec);
2580 	if (error == EJUSTRETURN)
2581 		error = 0;
2582 	else if (error)
2583 		goto error_exit;
2584 
2585 	have_exec_lock = true;
2586 
2587 	/*
2588 	 * Allocate virtual address space for the U-area now, while it
2589 	 * is still easy to abort the fork operation if we're out of
2590 	 * kernel virtual address space.
2591 	 */
2592 	uaddr = uvm_uarea_alloc();
2593 	if (__predict_false(uaddr == 0)) {
2594 		error = ENOMEM;
2595 		goto error_exit;
2596 	}
2597 
2598 	/*
2599 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2600 	 * replace it with its own before returning to userland
2601 	 * in the child.
2602 	 */
2603 	p2 = proc_alloc();
2604 	if (p2 == NULL) {
2605 		/* We were unable to allocate a process ID. */
2606 		error = EAGAIN;
2607 		goto error_exit;
2608 	}
2609 
2610 	/*
2611 	 * This is a point of no return, we will have to go through
2612 	 * the child proc to properly clean it up past this point.
2613 	 */
2614 	pid = p2->p_pid;
2615 
2616 	/*
2617 	 * Make a proc table entry for the new process.
2618 	 * Start by zeroing the section of proc that is zero-initialized,
2619 	 * then copy the section that is copied directly from the parent.
2620 	 */
2621 	memset(&p2->p_startzero, 0,
2622 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2623 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2624 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2625 	p2->p_vmspace = proc0.p_vmspace;
2626 
2627 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2628 
2629 	LIST_INIT(&p2->p_lwps);
2630 	LIST_INIT(&p2->p_sigwaiters);
2631 
2632 	/*
2633 	 * Duplicate sub-structures as needed.
2634 	 * Increase reference counts on shared objects.
2635 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2636 	 * handling are important in order to keep a consistent behaviour
2637 	 * for the child after the fork.  If we are a 32-bit process, the
2638 	 * child will be too.
2639 	 */
2640 	p2->p_flag =
2641 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2642 	p2->p_emul = p1->p_emul;
2643 	p2->p_execsw = p1->p_execsw;
2644 
2645 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2646 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2647 	rw_init(&p2->p_reflock);
2648 	cv_init(&p2->p_waitcv, "wait");
2649 	cv_init(&p2->p_lwpcv, "lwpwait");
2650 
2651 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2652 
2653 	kauth_proc_fork(p1, p2);
2654 
2655 	p2->p_raslist = NULL;
2656 	p2->p_fd = fd_copy();
2657 
2658 	/* XXX racy */
2659 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2660 
2661 	p2->p_cwdi = cwdinit();
2662 
2663 	/*
2664 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2665 	 * we just need increase pl_refcnt.
2666 	 */
2667 	if (!p1->p_limit->pl_writeable) {
2668 		lim_addref(p1->p_limit);
2669 		p2->p_limit = p1->p_limit;
2670 	} else {
2671 		p2->p_limit = lim_copy(p1->p_limit);
2672 	}
2673 
2674 	p2->p_lflag = 0;
2675 	l1->l_vforkwaiting = false;
2676 	p2->p_sflag = 0;
2677 	p2->p_slflag = 0;
2678 	p2->p_pptr = p1;
2679 	p2->p_ppid = p1->p_pid;
2680 	LIST_INIT(&p2->p_children);
2681 
2682 	p2->p_aio = NULL;
2683 
2684 #ifdef KTRACE
2685 	/*
2686 	 * Copy traceflag and tracefile if enabled.
2687 	 * If not inherited, these were zeroed above.
2688 	 */
2689 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2690 		mutex_enter(&ktrace_lock);
2691 		p2->p_traceflag = p1->p_traceflag;
2692 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2693 			ktradref(p2);
2694 		mutex_exit(&ktrace_lock);
2695 	}
2696 #endif
2697 
2698 	/*
2699 	 * Create signal actions for the child process.
2700 	 */
2701 	p2->p_sigacts = sigactsinit(p1, 0);
2702 	mutex_enter(p1->p_lock);
2703 	p2->p_sflag |=
2704 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2705 	sched_proc_fork(p1, p2);
2706 	mutex_exit(p1->p_lock);
2707 
2708 	p2->p_stflag = p1->p_stflag;
2709 
2710 	/*
2711 	 * p_stats.
2712 	 * Copy parts of p_stats, and zero out the rest.
2713 	 */
2714 	p2->p_stats = pstatscopy(p1->p_stats);
2715 
2716 	/* copy over machdep flags to the new proc */
2717 	cpu_proc_fork(p1, p2);
2718 
2719 	/*
2720 	 * Prepare remaining parts of spawn data
2721 	 */
2722 	spawn_data->sed_actions = fa;
2723 	spawn_data->sed_attrs = sa;
2724 
2725 	spawn_data->sed_parent = p1;
2726 
2727 	/* create LWP */
2728 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2729 	    &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk);
2730 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2731 
2732 	/*
2733 	 * Copy the credential so other references don't see our changes.
2734 	 * Test to see if this is necessary first, since in the common case
2735 	 * we won't need a private reference.
2736 	 */
2737 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2738 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2739 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2740 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2741 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2742 	}
2743 
2744 	/* Update the master credentials. */
2745 	if (l2->l_cred != p2->p_cred) {
2746 		kauth_cred_t ocred;
2747 
2748 		kauth_cred_hold(l2->l_cred);
2749 		mutex_enter(p2->p_lock);
2750 		ocred = p2->p_cred;
2751 		p2->p_cred = l2->l_cred;
2752 		mutex_exit(p2->p_lock);
2753 		kauth_cred_free(ocred);
2754 	}
2755 
2756 	*child_ok = true;
2757 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2758 #if 0
2759 	l2->l_nopreempt = 1; /* start it non-preemptable */
2760 #endif
2761 
2762 	/*
2763 	 * It's now safe for the scheduler and other processes to see the
2764 	 * child process.
2765 	 */
2766 	mutex_enter(&proc_lock);
2767 
2768 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2769 		p2->p_lflag |= PL_CONTROLT;
2770 
2771 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2772 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2773 
2774 	if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) ==
2775 	    (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2776 		proc_changeparent(p2, p1->p_pptr);
2777 		SET(p2->p_slflag, PSL_TRACEDCHILD);
2778 	}
2779 
2780 	p2->p_oppid = p1->p_pid;  /* Remember the original parent id. */
2781 
2782 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2783 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2784 
2785 	p2->p_trace_enabled = trace_is_enabled(p2);
2786 #ifdef __HAVE_SYSCALL_INTERN
2787 	(*p2->p_emul->e_syscall_intern)(p2);
2788 #endif
2789 
2790 	/*
2791 	 * Make child runnable, set start time, and add to run queue except
2792 	 * if the parent requested the child to start in SSTOP state.
2793 	 */
2794 	mutex_enter(p2->p_lock);
2795 
2796 	getmicrotime(&p2->p_stats->p_start);
2797 
2798 	lwp_lock(l2);
2799 	KASSERT(p2->p_nrlwps == 1);
2800 	KASSERT(l2->l_stat == LSIDL);
2801 	p2->p_nrlwps = 1;
2802 	p2->p_stat = SACTIVE;
2803 	setrunnable(l2);
2804 	/* LWP now unlocked */
2805 
2806 	mutex_exit(p2->p_lock);
2807 	mutex_exit(&proc_lock);
2808 
2809 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2810 	error = spawn_data->sed_error;
2811 	mutex_exit(&spawn_data->sed_mtx_child);
2812 	spawn_exec_data_release(spawn_data);
2813 
2814 	rw_exit(&p1->p_reflock);
2815 	rw_exit(&exec_lock);
2816 	have_exec_lock = false;
2817 
2818 	*pid_res = pid;
2819 
2820 	if (error)
2821 		return error;
2822 
2823 	if (p1->p_slflag & PSL_TRACED) {
2824 		/* Paranoid check */
2825 		mutex_enter(&proc_lock);
2826 		if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) !=
2827 		    (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) {
2828 			mutex_exit(&proc_lock);
2829 			return 0;
2830 		}
2831 
2832 		mutex_enter(p1->p_lock);
2833 		eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, pid);
2834 	}
2835 	return 0;
2836 
2837  error_exit:
2838 	if (have_exec_lock) {
2839 		execve_free_data(&spawn_data->sed_exec);
2840 		rw_exit(&p1->p_reflock);
2841 		rw_exit(&exec_lock);
2842 	}
2843 	mutex_exit(&spawn_data->sed_mtx_child);
2844 	spawn_exec_data_release(spawn_data);
2845 
2846 	return error;
2847 }
2848 
2849 int
2850 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2851     register_t *retval)
2852 {
2853 	/* {
2854 		syscallarg(pid_t *) pid;
2855 		syscallarg(const char *) path;
2856 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2857 		syscallarg(const struct posix_spawnattr *) attrp;
2858 		syscallarg(char *const *) argv;
2859 		syscallarg(char *const *) envp;
2860 	} */
2861 
2862 	int error;
2863 	struct posix_spawn_file_actions *fa = NULL;
2864 	struct posix_spawnattr *sa = NULL;
2865 	pid_t pid;
2866 	bool child_ok = false;
2867 	rlim_t max_fileactions;
2868 	proc_t *p = l1->l_proc;
2869 
2870 	/* check_posix_spawn() increments nprocs for us. */
2871 	error = check_posix_spawn(l1);
2872 	if (error) {
2873 		*retval = error;
2874 		return 0;
2875 	}
2876 
2877 	/* copy in file_actions struct */
2878 	if (SCARG(uap, file_actions) != NULL) {
2879 		max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2880 		    maxfiles);
2881 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2882 		    max_fileactions);
2883 		if (error)
2884 			goto error_exit;
2885 	}
2886 
2887 	/* copyin posix_spawnattr struct */
2888 	if (SCARG(uap, attrp) != NULL) {
2889 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2890 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2891 		if (error)
2892 			goto error_exit;
2893 	}
2894 
2895 	/*
2896 	 * Do the spawn
2897 	 */
2898 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2899 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2900 	if (error)
2901 		goto error_exit;
2902 
2903 	if (error == 0 && SCARG(uap, pid) != NULL)
2904 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2905 
2906 	*retval = error;
2907 	return 0;
2908 
2909  error_exit:
2910 	if (!child_ok) {
2911 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2912 		atomic_dec_uint(&nprocs);
2913 
2914 		if (sa)
2915 			kmem_free(sa, sizeof(*sa));
2916 		if (fa)
2917 			posix_spawn_fa_free(fa, fa->len);
2918 	}
2919 
2920 	*retval = error;
2921 	return 0;
2922 }
2923 
2924 void
2925 exec_free_emul_arg(struct exec_package *epp)
2926 {
2927 	if (epp->ep_emul_arg_free != NULL) {
2928 		KASSERT(epp->ep_emul_arg != NULL);
2929 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2930 		epp->ep_emul_arg_free = NULL;
2931 		epp->ep_emul_arg = NULL;
2932 	} else {
2933 		KASSERT(epp->ep_emul_arg == NULL);
2934 	}
2935 }
2936 
2937 #ifdef DEBUG_EXEC
2938 static void
2939 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2940 {
2941 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2942 	size_t j;
2943 
2944 	if (error == 0)
2945 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2946 	else
2947 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2948 		    epp->ep_vmcmds.evs_used, error));
2949 
2950 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2951 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2952 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2953 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2954 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2955 		    "pagedvn" :
2956 		    vp[j].ev_proc == vmcmd_map_readvn ?
2957 		    "readvn" :
2958 		    vp[j].ev_proc == vmcmd_map_zero ?
2959 		    "zero" : "*unknown*",
2960 		    vp[j].ev_addr, vp[j].ev_len,
2961 		    vp[j].ev_offset, vp[j].ev_prot,
2962 		    vp[j].ev_flags));
2963 		if (error != 0 && j == x)
2964 			DPRINTF(("     ^--- failed\n"));
2965 	}
2966 }
2967 #endif
2968