xref: /netbsd-src/sys/kern/kern_exec.c (revision c29d51755812ace2e87aeefdb06cb2b4dac7087a)
1 /*	$NetBSD: kern_exec.c,v 1.417 2015/09/26 16:12:24 maxv Exp $	*/
2 
3 /*-
4  * Copyright (c) 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*-
30  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
31  * Copyright (C) 1992 Wolfgang Solfrank.
32  * Copyright (C) 1992 TooLs GmbH.
33  * All rights reserved.
34  *
35  * Redistribution and use in source and binary forms, with or without
36  * modification, are permitted provided that the following conditions
37  * are met:
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. All advertising materials mentioning features or use of this software
44  *    must display the following acknowledgement:
45  *	This product includes software developed by TooLs GmbH.
46  * 4. The name of TooLs GmbH may not be used to endorse or promote products
47  *    derived from this software without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
50  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
51  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
52  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
53  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
54  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
55  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
56  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
57  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
58  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
59  */
60 
61 #include <sys/cdefs.h>
62 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.417 2015/09/26 16:12:24 maxv Exp $");
63 
64 #include "opt_exec.h"
65 #include "opt_execfmt.h"
66 #include "opt_ktrace.h"
67 #include "opt_modular.h"
68 #include "opt_syscall_debug.h"
69 #include "veriexec.h"
70 #include "opt_pax.h"
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/filedesc.h>
75 #include <sys/kernel.h>
76 #include <sys/proc.h>
77 #include <sys/mount.h>
78 #include <sys/kmem.h>
79 #include <sys/namei.h>
80 #include <sys/vnode.h>
81 #include <sys/file.h>
82 #include <sys/filedesc.h>
83 #include <sys/acct.h>
84 #include <sys/atomic.h>
85 #include <sys/exec.h>
86 #include <sys/ktrace.h>
87 #include <sys/uidinfo.h>
88 #include <sys/wait.h>
89 #include <sys/mman.h>
90 #include <sys/ras.h>
91 #include <sys/signalvar.h>
92 #include <sys/stat.h>
93 #include <sys/syscall.h>
94 #include <sys/kauth.h>
95 #include <sys/lwpctl.h>
96 #include <sys/pax.h>
97 #include <sys/cpu.h>
98 #include <sys/module.h>
99 #include <sys/syscallvar.h>
100 #include <sys/syscallargs.h>
101 #if NVERIEXEC > 0
102 #include <sys/verified_exec.h>
103 #endif /* NVERIEXEC > 0 */
104 #include <sys/sdt.h>
105 #include <sys/spawn.h>
106 #include <sys/prot.h>
107 #include <sys/cprng.h>
108 
109 #include <uvm/uvm_extern.h>
110 
111 #include <machine/reg.h>
112 
113 #include <compat/common/compat_util.h>
114 
115 #ifndef MD_TOPDOWN_INIT
116 #ifdef __USE_TOPDOWN_VM
117 #define	MD_TOPDOWN_INIT(epp)	(epp)->ep_flags |= EXEC_TOPDOWN_VM
118 #else
119 #define	MD_TOPDOWN_INIT(epp)
120 #endif
121 #endif
122 
123 struct execve_data;
124 
125 static size_t calcargs(struct execve_data * restrict, const size_t);
126 static size_t calcstack(struct execve_data * restrict, const size_t);
127 static int copyoutargs(struct execve_data * restrict, struct lwp *,
128     char * const);
129 static int copyoutpsstrs(struct execve_data * restrict, struct proc *);
130 static int copyinargs(struct execve_data * restrict, char * const *,
131     char * const *, execve_fetch_element_t, char **);
132 static int copyinargstrs(struct execve_data * restrict, char * const *,
133     execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t));
134 static int exec_sigcode_map(struct proc *, const struct emul *);
135 
136 #ifdef DEBUG_EXEC
137 #define DPRINTF(a) printf a
138 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \
139     __LINE__, (s), (a), (b))
140 static void dump_vmcmds(const struct exec_package * const, size_t, int);
141 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0)
142 #else
143 #define DPRINTF(a)
144 #define COPYPRINTF(s, a, b)
145 #define DUMPVMCMDS(p, x, e) do {} while (0)
146 #endif /* DEBUG_EXEC */
147 
148 /*
149  * DTrace SDT provider definitions
150  */
151 SDT_PROBE_DEFINE(proc,,,exec,exec,
152 	    "char *", NULL,
153 	    NULL, NULL, NULL, NULL,
154 	    NULL, NULL, NULL, NULL);
155 SDT_PROBE_DEFINE(proc,,,exec_success,exec-success,
156 	    "char *", NULL,
157 	    NULL, NULL, NULL, NULL,
158 	    NULL, NULL, NULL, NULL);
159 SDT_PROBE_DEFINE(proc,,,exec_failure,exec-failure,
160 	    "int", NULL,
161 	    NULL, NULL, NULL, NULL,
162 	    NULL, NULL, NULL, NULL);
163 
164 /*
165  * Exec function switch:
166  *
167  * Note that each makecmds function is responsible for loading the
168  * exec package with the necessary functions for any exec-type-specific
169  * handling.
170  *
171  * Functions for specific exec types should be defined in their own
172  * header file.
173  */
174 static const struct execsw	**execsw = NULL;
175 static int			nexecs;
176 
177 u_int	exec_maxhdrsz;	 /* must not be static - used by netbsd32 */
178 
179 /* list of dynamically loaded execsw entries */
180 static LIST_HEAD(execlist_head, exec_entry) ex_head =
181     LIST_HEAD_INITIALIZER(ex_head);
182 struct exec_entry {
183 	LIST_ENTRY(exec_entry)	ex_list;
184 	SLIST_ENTRY(exec_entry)	ex_slist;
185 	const struct execsw	*ex_sw;
186 };
187 
188 #ifndef __HAVE_SYSCALL_INTERN
189 void	syscall(void);
190 #endif
191 
192 /* NetBSD emul struct */
193 struct emul emul_netbsd = {
194 	.e_name =		"netbsd",
195 #ifdef EMUL_NATIVEROOT
196 	.e_path =		EMUL_NATIVEROOT,
197 #else
198 	.e_path =		NULL,
199 #endif
200 #ifndef __HAVE_MINIMAL_EMUL
201 	.e_flags =		EMUL_HAS_SYS___syscall,
202 	.e_errno =		NULL,
203 	.e_nosys =		SYS_syscall,
204 	.e_nsysent =		SYS_NSYSENT,
205 #endif
206 	.e_sysent =		sysent,
207 #ifdef SYSCALL_DEBUG
208 	.e_syscallnames =	syscallnames,
209 #else
210 	.e_syscallnames =	NULL,
211 #endif
212 	.e_sendsig =		sendsig,
213 	.e_trapsignal =		trapsignal,
214 	.e_tracesig =		NULL,
215 	.e_sigcode =		NULL,
216 	.e_esigcode =		NULL,
217 	.e_sigobject =		NULL,
218 	.e_setregs =		setregs,
219 	.e_proc_exec =		NULL,
220 	.e_proc_fork =		NULL,
221 	.e_proc_exit =		NULL,
222 	.e_lwp_fork =		NULL,
223 	.e_lwp_exit =		NULL,
224 #ifdef __HAVE_SYSCALL_INTERN
225 	.e_syscall_intern =	syscall_intern,
226 #else
227 	.e_syscall =		syscall,
228 #endif
229 	.e_sysctlovly =		NULL,
230 	.e_fault =		NULL,
231 	.e_vm_default_addr =	uvm_default_mapaddr,
232 	.e_usertrap =		NULL,
233 	.e_ucsize =		sizeof(ucontext_t),
234 	.e_startlwp =		startlwp
235 };
236 
237 /*
238  * Exec lock. Used to control access to execsw[] structures.
239  * This must not be static so that netbsd32 can access it, too.
240  */
241 krwlock_t exec_lock;
242 
243 static kmutex_t sigobject_lock;
244 
245 /*
246  * Data used between a loadvm and execve part of an "exec" operation
247  */
248 struct execve_data {
249 	struct exec_package	ed_pack;
250 	struct pathbuf		*ed_pathbuf;
251 	struct vattr		ed_attr;
252 	struct ps_strings	ed_arginfo;
253 	char			*ed_argp;
254 	const char		*ed_pathstring;
255 	char			*ed_resolvedpathbuf;
256 	size_t			ed_ps_strings_sz;
257 	int			ed_szsigcode;
258 	size_t			ed_argslen;
259 	long			ed_argc;
260 	long			ed_envc;
261 };
262 
263 /*
264  * data passed from parent lwp to child during a posix_spawn()
265  */
266 struct spawn_exec_data {
267 	struct execve_data	sed_exec;
268 	struct posix_spawn_file_actions
269 				*sed_actions;
270 	struct posix_spawnattr	*sed_attrs;
271 	struct proc		*sed_parent;
272 	kcondvar_t		sed_cv_child_ready;
273 	kmutex_t		sed_mtx_child;
274 	int			sed_error;
275 	volatile uint32_t	sed_refcnt;
276 };
277 
278 static void *
279 exec_pool_alloc(struct pool *pp, int flags)
280 {
281 
282 	return (void *)uvm_km_alloc(kernel_map, NCARGS, 0,
283 	    UVM_KMF_PAGEABLE | UVM_KMF_WAITVA);
284 }
285 
286 static void
287 exec_pool_free(struct pool *pp, void *addr)
288 {
289 
290 	uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE);
291 }
292 
293 static struct pool exec_pool;
294 
295 static struct pool_allocator exec_palloc = {
296 	.pa_alloc = exec_pool_alloc,
297 	.pa_free = exec_pool_free,
298 	.pa_pagesz = NCARGS
299 };
300 
301 /*
302  * check exec:
303  * given an "executable" described in the exec package's namei info,
304  * see what we can do with it.
305  *
306  * ON ENTRY:
307  *	exec package with appropriate namei info
308  *	lwp pointer of exec'ing lwp
309  *	NO SELF-LOCKED VNODES
310  *
311  * ON EXIT:
312  *	error:	nothing held, etc.  exec header still allocated.
313  *	ok:	filled exec package, executable's vnode (unlocked).
314  *
315  * EXEC SWITCH ENTRY:
316  * 	Locked vnode to check, exec package, proc.
317  *
318  * EXEC SWITCH EXIT:
319  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
320  *	error:	destructive:
321  *			everything deallocated execept exec header.
322  *		non-destructive:
323  *			error code, executable's vnode (unlocked),
324  *			exec header unmodified.
325  */
326 int
327 /*ARGSUSED*/
328 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb)
329 {
330 	int		error, i;
331 	struct vnode	*vp;
332 	struct nameidata nd;
333 	size_t		resid;
334 
335 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
336 
337 	/* first get the vnode */
338 	if ((error = namei(&nd)) != 0)
339 		return error;
340 	epp->ep_vp = vp = nd.ni_vp;
341 	/* normally this can't fail */
342 	error = copystr(nd.ni_pnbuf, epp->ep_resolvedname, PATH_MAX, NULL);
343 	KASSERT(error == 0);
344 
345 #ifdef DIAGNOSTIC
346 	/* paranoia (take this out once namei stuff stabilizes) */
347 	memset(nd.ni_pnbuf, '~', PATH_MAX);
348 #endif
349 
350 	/* check access and type */
351 	if (vp->v_type != VREG) {
352 		error = EACCES;
353 		goto bad1;
354 	}
355 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
356 		goto bad1;
357 
358 	/* get attributes */
359 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
360 		goto bad1;
361 
362 	/* Check mount point */
363 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
364 		error = EACCES;
365 		goto bad1;
366 	}
367 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
368 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
369 
370 	/* try to open it */
371 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
372 		goto bad1;
373 
374 	/* unlock vp, since we need it unlocked from here on out. */
375 	VOP_UNLOCK(vp);
376 
377 #if NVERIEXEC > 0
378 	error = veriexec_verify(l, vp, epp->ep_resolvedname,
379 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
380 	    NULL);
381 	if (error)
382 		goto bad2;
383 #endif /* NVERIEXEC > 0 */
384 
385 #ifdef PAX_SEGVGUARD
386 	error = pax_segvguard(l, vp, epp->ep_resolvedname, false);
387 	if (error)
388 		goto bad2;
389 #endif /* PAX_SEGVGUARD */
390 
391 	/* now we have the file, get the exec header */
392 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
393 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
394 	if (error)
395 		goto bad2;
396 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
397 
398 	/*
399 	 * Set up default address space limits.  Can be overridden
400 	 * by individual exec packages.
401 	 *
402 	 * XXX probably should be all done in the exec packages.
403 	 */
404 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
405 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
406 	/*
407 	 * set up the vmcmds for creation of the process
408 	 * address space
409 	 */
410 	error = ENOEXEC;
411 	for (i = 0; i < nexecs; i++) {
412 		int newerror;
413 
414 		epp->ep_esch = execsw[i];
415 		newerror = (*execsw[i]->es_makecmds)(l, epp);
416 
417 		if (!newerror) {
418 			/* Seems ok: check that entry point is not too high */
419 			if (epp->ep_entry > epp->ep_vm_maxaddr) {
420 #ifdef DIAGNOSTIC
421 				printf("%s: rejecting %p due to "
422 				    "too high entry address (> %p)\n",
423 					 __func__, (void *)epp->ep_entry,
424 					 (void *)epp->ep_vm_maxaddr);
425 #endif
426 				error = ENOEXEC;
427 				break;
428 			}
429 			/* Seems ok: check that entry point is not too low */
430 			if (epp->ep_entry < epp->ep_vm_minaddr) {
431 #ifdef DIAGNOSTIC
432 				printf("%s: rejecting %p due to "
433 				    "too low entry address (< %p)\n",
434 				     __func__, (void *)epp->ep_entry,
435 				     (void *)epp->ep_vm_minaddr);
436 #endif
437 				error = ENOEXEC;
438 				break;
439 			}
440 
441 			/* check limits */
442 			if ((epp->ep_tsize > MAXTSIZ) ||
443 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
444 						    [RLIMIT_DATA].rlim_cur)) {
445 #ifdef DIAGNOSTIC
446 				printf("%s: rejecting due to "
447 				    "limits (t=%llu > %llu || d=%llu > %llu)\n",
448 				    __func__,
449 				    (unsigned long long)epp->ep_tsize,
450 				    (unsigned long long)MAXTSIZ,
451 				    (unsigned long long)epp->ep_dsize,
452 				    (unsigned long long)
453 				    l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur);
454 #endif
455 				error = ENOMEM;
456 				break;
457 			}
458 			return 0;
459 		}
460 
461 		if (epp->ep_emul_root != NULL) {
462 			vrele(epp->ep_emul_root);
463 			epp->ep_emul_root = NULL;
464 		}
465 		if (epp->ep_interp != NULL) {
466 			vrele(epp->ep_interp);
467 			epp->ep_interp = NULL;
468 		}
469 
470 		/* make sure the first "interesting" error code is saved. */
471 		if (error == ENOEXEC)
472 			error = newerror;
473 
474 		if (epp->ep_flags & EXEC_DESTR)
475 			/* Error from "#!" code, tidied up by recursive call */
476 			return error;
477 	}
478 
479 	/* not found, error */
480 
481 	/*
482 	 * free any vmspace-creation commands,
483 	 * and release their references
484 	 */
485 	kill_vmcmds(&epp->ep_vmcmds);
486 
487 bad2:
488 	/*
489 	 * close and release the vnode, restore the old one, free the
490 	 * pathname buf, and punt.
491 	 */
492 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
493 	VOP_CLOSE(vp, FREAD, l->l_cred);
494 	vput(vp);
495 	return error;
496 
497 bad1:
498 	/*
499 	 * free the namei pathname buffer, and put the vnode
500 	 * (which we don't yet have open).
501 	 */
502 	vput(vp);				/* was still locked */
503 	return error;
504 }
505 
506 #ifdef __MACHINE_STACK_GROWS_UP
507 #define STACK_PTHREADSPACE NBPG
508 #else
509 #define STACK_PTHREADSPACE 0
510 #endif
511 
512 static int
513 execve_fetch_element(char * const *array, size_t index, char **value)
514 {
515 	return copyin(array + index, value, sizeof(*value));
516 }
517 
518 /*
519  * exec system call
520  */
521 int
522 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
523 {
524 	/* {
525 		syscallarg(const char *)	path;
526 		syscallarg(char * const *)	argp;
527 		syscallarg(char * const *)	envp;
528 	} */
529 
530 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
531 	    SCARG(uap, envp), execve_fetch_element);
532 }
533 
534 int
535 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap,
536     register_t *retval)
537 {
538 	/* {
539 		syscallarg(int)			fd;
540 		syscallarg(char * const *)	argp;
541 		syscallarg(char * const *)	envp;
542 	} */
543 
544 	return ENOSYS;
545 }
546 
547 /*
548  * Load modules to try and execute an image that we do not understand.
549  * If no execsw entries are present, we load those likely to be needed
550  * in order to run native images only.  Otherwise, we autoload all
551  * possible modules that could let us run the binary.  XXX lame
552  */
553 static void
554 exec_autoload(void)
555 {
556 #ifdef MODULAR
557 	static const char * const native[] = {
558 		"exec_elf32",
559 		"exec_elf64",
560 		"exec_script",
561 		NULL
562 	};
563 	static const char * const compat[] = {
564 		"exec_elf32",
565 		"exec_elf64",
566 		"exec_script",
567 		"exec_aout",
568 		"exec_coff",
569 		"exec_ecoff",
570 		"compat_aoutm68k",
571 		"compat_freebsd",
572 		"compat_ibcs2",
573 		"compat_linux",
574 		"compat_linux32",
575 		"compat_netbsd32",
576 		"compat_sunos",
577 		"compat_sunos32",
578 		"compat_svr4",
579 		"compat_svr4_32",
580 		"compat_ultrix",
581 		NULL
582 	};
583 	char const * const *list;
584 	int i;
585 
586 	list = (nexecs == 0 ? native : compat);
587 	for (i = 0; list[i] != NULL; i++) {
588 		if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) {
589 			continue;
590 		}
591 		yield();
592 	}
593 #endif
594 }
595 
596 static int
597 makepathbuf(struct lwp *l, const char *upath, struct pathbuf **pbp,
598     size_t *offs)
599 {
600 	char *path, *bp;
601 	size_t len, tlen;
602 	int error;
603 	struct cwdinfo *cwdi;
604 
605 	path = PNBUF_GET();
606 	error = copyinstr(upath, path, MAXPATHLEN, &len);
607 	if (error) {
608 		PNBUF_PUT(path);
609 		DPRINTF(("%s: copyin path @%p %d\n", __func__, upath, error));
610 		return error;
611 	}
612 
613 	if (path[0] == '/') {
614 		*offs = 0;
615 		goto out;
616 	}
617 
618 	len++;
619 	if (len + 1 >= MAXPATHLEN)
620 		goto out;
621 	bp = path + MAXPATHLEN - len;
622 	memmove(bp, path, len);
623 	*(--bp) = '/';
624 
625 	cwdi = l->l_proc->p_cwdi;
626 	rw_enter(&cwdi->cwdi_lock, RW_READER);
627 	error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2,
628 	    GETCWD_CHECK_ACCESS, l);
629 	rw_exit(&cwdi->cwdi_lock);
630 
631 	if (error) {
632 		DPRINTF(("%s: getcwd_common path %s %d\n", __func__, path,
633 		    error));
634 		goto out;
635 	}
636 	tlen = path + MAXPATHLEN - bp;
637 
638 	memmove(path, bp, tlen);
639 	path[tlen] = '\0';
640 	*offs = tlen - len;
641 out:
642 	*pbp = pathbuf_assimilate(path);
643 	return 0;
644 }
645 
646 static int
647 execve_loadvm(struct lwp *l, const char *path, char * const *args,
648 	char * const *envs, execve_fetch_element_t fetch_element,
649 	struct execve_data * restrict data)
650 {
651 	struct exec_package	* const epp = &data->ed_pack;
652 	int			error;
653 	struct proc		*p;
654 	char			*dp;
655 	u_int			modgen;
656 	size_t			offs = 0;	// XXX: GCC
657 
658 	KASSERT(data != NULL);
659 
660 	p = l->l_proc;
661 	modgen = 0;
662 
663 	SDT_PROBE(proc,,,exec, path, 0, 0, 0, 0);
664 
665 	/*
666 	 * Check if we have exceeded our number of processes limit.
667 	 * This is so that we handle the case where a root daemon
668 	 * forked, ran setuid to become the desired user and is trying
669 	 * to exec. The obvious place to do the reference counting check
670 	 * is setuid(), but we don't do the reference counting check there
671 	 * like other OS's do because then all the programs that use setuid()
672 	 * must be modified to check the return code of setuid() and exit().
673 	 * It is dangerous to make setuid() fail, because it fails open and
674 	 * the program will continue to run as root. If we make it succeed
675 	 * and return an error code, again we are not enforcing the limit.
676 	 * The best place to enforce the limit is here, when the process tries
677 	 * to execute a new image, because eventually the process will need
678 	 * to call exec in order to do something useful.
679 	 */
680  retry:
681 	if (p->p_flag & PK_SUGID) {
682 		if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT,
683 		     p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
684 		     &p->p_rlimit[RLIMIT_NPROC],
685 		     KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
686 		    chgproccnt(kauth_cred_getuid(l->l_cred), 0) >
687 		     p->p_rlimit[RLIMIT_NPROC].rlim_cur)
688 		return EAGAIN;
689 	}
690 
691 	/*
692 	 * Drain existing references and forbid new ones.  The process
693 	 * should be left alone until we're done here.  This is necessary
694 	 * to avoid race conditions - e.g. in ptrace() - that might allow
695 	 * a local user to illicitly obtain elevated privileges.
696 	 */
697 	rw_enter(&p->p_reflock, RW_WRITER);
698 
699 	/*
700 	 * Init the namei data to point the file user's program name.
701 	 * This is done here rather than in check_exec(), so that it's
702 	 * possible to override this settings if any of makecmd/probe
703 	 * functions call check_exec() recursively - for example,
704 	 * see exec_script_makecmds().
705 	 */
706 	if ((error = makepathbuf(l, path, &data->ed_pathbuf, &offs)) != 0)
707 		goto clrflg;
708 	data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf);
709 	data->ed_resolvedpathbuf = PNBUF_GET();
710 
711 	/*
712 	 * initialize the fields of the exec package.
713 	 */
714 	epp->ep_kname = data->ed_pathstring + offs;
715 	epp->ep_resolvedname = data->ed_resolvedpathbuf;
716 	epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
717 	epp->ep_hdrlen = exec_maxhdrsz;
718 	epp->ep_hdrvalid = 0;
719 	epp->ep_emul_arg = NULL;
720 	epp->ep_emul_arg_free = NULL;
721 	memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds));
722 	epp->ep_vap = &data->ed_attr;
723 	epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0;
724 	MD_TOPDOWN_INIT(epp);
725 	epp->ep_emul_root = NULL;
726 	epp->ep_interp = NULL;
727 	epp->ep_esch = NULL;
728 	epp->ep_pax_flags = 0;
729 	memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch));
730 
731 	rw_enter(&exec_lock, RW_READER);
732 
733 	/* see if we can run it. */
734 	if ((error = check_exec(l, epp, data->ed_pathbuf)) != 0) {
735 		if (error != ENOENT) {
736 			DPRINTF(("%s: check exec failed %d\n",
737 			    __func__, error));
738 		}
739 		goto freehdr;
740 	}
741 
742 	/* allocate an argument buffer */
743 	data->ed_argp = pool_get(&exec_pool, PR_WAITOK);
744 	KASSERT(data->ed_argp != NULL);
745 	dp = data->ed_argp;
746 
747 	if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) {
748 		goto bad;
749 	}
750 
751 	/*
752 	 * Calculate the new stack size.
753 	 */
754 
755 #ifdef PAX_ASLR
756 #define	ASLR_GAP(epp)	(pax_aslr_epp_active(epp) ? (cprng_fast32() % PAGE_SIZE) : 0)
757 #else
758 #define	ASLR_GAP(epp)	0
759 #endif
760 
761 #ifdef __MACHINE_STACK_GROWS_UP
762 /*
763  * copyargs() fills argc/argv/envp from the lower address even on
764  * __MACHINE_STACK_GROWS_UP machines.  Reserve a few words just below the SP
765  * so that _rtld() use it.
766  */
767 #define	RTLD_GAP	32
768 #else
769 #define	RTLD_GAP	0
770 #endif
771 
772 	const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp;
773 
774 	data->ed_argslen = calcargs(data, argenvstrlen);
775 
776 	const size_t len = calcstack(data, ASLR_GAP(epp) + RTLD_GAP);
777 
778 	if (len > epp->ep_ssize) {
779 		/* in effect, compare to initial limit */
780 		DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len));
781 		error = ENOMEM;
782 		goto bad;
783 	}
784 	/* adjust "active stack depth" for process VSZ */
785 	epp->ep_ssize = len;
786 
787 	return 0;
788 
789  bad:
790 	/* free the vmspace-creation commands, and release their references */
791 	kill_vmcmds(&epp->ep_vmcmds);
792 	/* kill any opened file descriptor, if necessary */
793 	if (epp->ep_flags & EXEC_HASFD) {
794 		epp->ep_flags &= ~EXEC_HASFD;
795 		fd_close(epp->ep_fd);
796 	}
797 	/* close and put the exec'd file */
798 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
799 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
800 	vput(epp->ep_vp);
801 	pool_put(&exec_pool, data->ed_argp);
802 
803  freehdr:
804 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
805 	if (epp->ep_emul_root != NULL)
806 		vrele(epp->ep_emul_root);
807 	if (epp->ep_interp != NULL)
808 		vrele(epp->ep_interp);
809 
810 	rw_exit(&exec_lock);
811 
812 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
813 	pathbuf_destroy(data->ed_pathbuf);
814 	PNBUF_PUT(data->ed_resolvedpathbuf);
815 
816  clrflg:
817 	rw_exit(&p->p_reflock);
818 
819 	if (modgen != module_gen && error == ENOEXEC) {
820 		modgen = module_gen;
821 		exec_autoload();
822 		goto retry;
823 	}
824 
825 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
826 	return error;
827 }
828 
829 static int
830 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data)
831 {
832 	struct exec_package	* const epp = &data->ed_pack;
833 	struct proc		*p = l->l_proc;
834 	struct exec_vmcmd	*base_vcp;
835 	int			error = 0;
836 	size_t			i;
837 
838 	/* record proc's vnode, for use by procfs and others */
839 	if (p->p_textvp)
840 		vrele(p->p_textvp);
841 	vref(epp->ep_vp);
842 	p->p_textvp = epp->ep_vp;
843 
844 	/* create the new process's VM space by running the vmcmds */
845 	KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__);
846 
847 	DUMPVMCMDS(epp, 0, 0);
848 
849 	base_vcp = NULL;
850 
851 	for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) {
852 		struct exec_vmcmd *vcp;
853 
854 		vcp = &epp->ep_vmcmds.evs_cmds[i];
855 		if (vcp->ev_flags & VMCMD_RELATIVE) {
856 			KASSERTMSG(base_vcp != NULL,
857 			    "%s: relative vmcmd with no base", __func__);
858 			KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0,
859 			    "%s: illegal base & relative vmcmd", __func__);
860 			vcp->ev_addr += base_vcp->ev_addr;
861 		}
862 		error = (*vcp->ev_proc)(l, vcp);
863 		if (error)
864 			DUMPVMCMDS(epp, i, error);
865 		if (vcp->ev_flags & VMCMD_BASE)
866 			base_vcp = vcp;
867 	}
868 
869 	/* free the vmspace-creation commands, and release their references */
870 	kill_vmcmds(&epp->ep_vmcmds);
871 
872 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
873 	VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred);
874 	vput(epp->ep_vp);
875 
876 	/* if an error happened, deallocate and punt */
877 	if (error != 0) {
878 		DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error));
879 	}
880 	return error;
881 }
882 
883 static void
884 execve_free_data(struct execve_data *data)
885 {
886 	struct exec_package	* const epp = &data->ed_pack;
887 
888 	/* free the vmspace-creation commands, and release their references */
889 	kill_vmcmds(&epp->ep_vmcmds);
890 	/* kill any opened file descriptor, if necessary */
891 	if (epp->ep_flags & EXEC_HASFD) {
892 		epp->ep_flags &= ~EXEC_HASFD;
893 		fd_close(epp->ep_fd);
894 	}
895 
896 	/* close and put the exec'd file */
897 	vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY);
898 	VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred);
899 	vput(epp->ep_vp);
900 	pool_put(&exec_pool, data->ed_argp);
901 
902 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
903 	if (epp->ep_emul_root != NULL)
904 		vrele(epp->ep_emul_root);
905 	if (epp->ep_interp != NULL)
906 		vrele(epp->ep_interp);
907 
908 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
909 	pathbuf_destroy(data->ed_pathbuf);
910 	PNBUF_PUT(data->ed_resolvedpathbuf);
911 }
912 
913 static void
914 pathexec(struct exec_package *epp, struct proc *p, const char *pathstring)
915 {
916 	const char		*commandname;
917 	size_t			commandlen;
918 	char			*path;
919 
920 	/* set command name & other accounting info */
921 	commandname = strrchr(epp->ep_resolvedname, '/');
922 	if (commandname != NULL) {
923 		commandname++;
924 	} else {
925 		commandname = epp->ep_resolvedname;
926 	}
927 	commandlen = min(strlen(commandname), MAXCOMLEN);
928 	(void)memcpy(p->p_comm, commandname, commandlen);
929 	p->p_comm[commandlen] = '\0';
930 
931 
932 	/*
933 	 * If the path starts with /, we don't need to do any work.
934 	 * This handles the majority of the cases.
935 	 * In the future perhaps we could canonicalize it?
936 	 */
937 	if (pathstring[0] == '/') {
938 		path = PNBUF_GET();
939 		(void)strlcpy(path, pathstring, MAXPATHLEN);
940 		epp->ep_path = path;
941 	} else
942 		epp->ep_path = NULL;
943 }
944 
945 /* XXX elsewhere */
946 static int
947 credexec(struct lwp *l, struct vattr *attr)
948 {
949 	struct proc *p = l->l_proc;
950 	int error;
951 
952 	/*
953 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
954 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
955 	 * out additional references on the process for the moment.
956 	 */
957 	if ((p->p_slflag & PSL_TRACED) == 0 &&
958 
959 	    (((attr->va_mode & S_ISUID) != 0 &&
960 	      kauth_cred_geteuid(l->l_cred) != attr->va_uid) ||
961 
962 	     ((attr->va_mode & S_ISGID) != 0 &&
963 	      kauth_cred_getegid(l->l_cred) != attr->va_gid))) {
964 		/*
965 		 * Mark the process as SUGID before we do
966 		 * anything that might block.
967 		 */
968 		proc_crmod_enter();
969 		proc_crmod_leave(NULL, NULL, true);
970 
971 		/* Make sure file descriptors 0..2 are in use. */
972 		if ((error = fd_checkstd()) != 0) {
973 			DPRINTF(("%s: fdcheckstd failed %d\n",
974 			    __func__, error));
975 			return error;
976 		}
977 
978 		/*
979 		 * Copy the credential so other references don't see our
980 		 * changes.
981 		 */
982 		l->l_cred = kauth_cred_copy(l->l_cred);
983 #ifdef KTRACE
984 		/*
985 		 * If the persistent trace flag isn't set, turn off.
986 		 */
987 		if (p->p_tracep) {
988 			mutex_enter(&ktrace_lock);
989 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
990 				ktrderef(p);
991 			mutex_exit(&ktrace_lock);
992 		}
993 #endif
994 		if (attr->va_mode & S_ISUID)
995 			kauth_cred_seteuid(l->l_cred, attr->va_uid);
996 		if (attr->va_mode & S_ISGID)
997 			kauth_cred_setegid(l->l_cred, attr->va_gid);
998 	} else {
999 		if (kauth_cred_geteuid(l->l_cred) ==
1000 		    kauth_cred_getuid(l->l_cred) &&
1001 		    kauth_cred_getegid(l->l_cred) ==
1002 		    kauth_cred_getgid(l->l_cred))
1003 			p->p_flag &= ~PK_SUGID;
1004 	}
1005 
1006 	/*
1007 	 * Copy the credential so other references don't see our changes.
1008 	 * Test to see if this is necessary first, since in the common case
1009 	 * we won't need a private reference.
1010 	 */
1011 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
1012 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
1013 		l->l_cred = kauth_cred_copy(l->l_cred);
1014 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
1015 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
1016 	}
1017 
1018 	/* Update the master credentials. */
1019 	if (l->l_cred != p->p_cred) {
1020 		kauth_cred_t ocred;
1021 
1022 		kauth_cred_hold(l->l_cred);
1023 		mutex_enter(p->p_lock);
1024 		ocred = p->p_cred;
1025 		p->p_cred = l->l_cred;
1026 		mutex_exit(p->p_lock);
1027 		kauth_cred_free(ocred);
1028 	}
1029 
1030 	return 0;
1031 }
1032 
1033 static void
1034 emulexec(struct lwp *l, struct exec_package *epp)
1035 {
1036 	struct proc		*p = l->l_proc;
1037 
1038 	/* The emulation root will usually have been found when we looked
1039 	 * for the elf interpreter (or similar), if not look now. */
1040 	if (epp->ep_esch->es_emul->e_path != NULL &&
1041 	    epp->ep_emul_root == NULL)
1042 		emul_find_root(l, epp);
1043 
1044 	/* Any old emulation root got removed by fdcloseexec */
1045 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
1046 	p->p_cwdi->cwdi_edir = epp->ep_emul_root;
1047 	rw_exit(&p->p_cwdi->cwdi_lock);
1048 	epp->ep_emul_root = NULL;
1049 	if (epp->ep_interp != NULL)
1050 		vrele(epp->ep_interp);
1051 
1052 	/*
1053 	 * Call emulation specific exec hook. This can setup per-process
1054 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
1055 	 *
1056 	 * If we are executing process of different emulation than the
1057 	 * original forked process, call e_proc_exit() of the old emulation
1058 	 * first, then e_proc_exec() of new emulation. If the emulation is
1059 	 * same, the exec hook code should deallocate any old emulation
1060 	 * resources held previously by this process.
1061 	 */
1062 	if (p->p_emul && p->p_emul->e_proc_exit
1063 	    && p->p_emul != epp->ep_esch->es_emul)
1064 		(*p->p_emul->e_proc_exit)(p);
1065 
1066 	/*
1067 	 * This is now LWP 1.
1068 	 */
1069 	/* XXX elsewhere */
1070 	mutex_enter(p->p_lock);
1071 	p->p_nlwpid = 1;
1072 	l->l_lid = 1;
1073 	mutex_exit(p->p_lock);
1074 
1075 	/*
1076 	 * Call exec hook. Emulation code may NOT store reference to anything
1077 	 * from &pack.
1078 	 */
1079 	if (epp->ep_esch->es_emul->e_proc_exec)
1080 		(*epp->ep_esch->es_emul->e_proc_exec)(p, epp);
1081 
1082 	/* update p_emul, the old value is no longer needed */
1083 	p->p_emul = epp->ep_esch->es_emul;
1084 
1085 	/* ...and the same for p_execsw */
1086 	p->p_execsw = epp->ep_esch;
1087 
1088 #ifdef __HAVE_SYSCALL_INTERN
1089 	(*p->p_emul->e_syscall_intern)(p);
1090 #endif
1091 	ktremul();
1092 }
1093 
1094 static int
1095 execve_runproc(struct lwp *l, struct execve_data * restrict data,
1096 	bool no_local_exec_lock, bool is_spawn)
1097 {
1098 	struct exec_package	* const epp = &data->ed_pack;
1099 	int error = 0;
1100 	struct proc		*p;
1101 
1102 	/*
1103 	 * In case of a posix_spawn operation, the child doing the exec
1104 	 * might not hold the reader lock on exec_lock, but the parent
1105 	 * will do this instead.
1106 	 */
1107 	KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock));
1108 	KASSERT(!no_local_exec_lock || is_spawn);
1109 	KASSERT(data != NULL);
1110 
1111 	p = l->l_proc;
1112 
1113 	/* Get rid of other LWPs. */
1114 	if (p->p_nlwps > 1) {
1115 		mutex_enter(p->p_lock);
1116 		exit_lwps(l);
1117 		mutex_exit(p->p_lock);
1118 	}
1119 	KDASSERT(p->p_nlwps == 1);
1120 
1121 	/* Destroy any lwpctl info. */
1122 	if (p->p_lwpctl != NULL)
1123 		lwp_ctl_exit();
1124 
1125 	/* Remove POSIX timers */
1126 	timers_free(p, TIMERS_POSIX);
1127 
1128 	/* Set the PaX flags. */
1129 	p->p_pax = epp->ep_pax_flags;
1130 
1131 	/*
1132 	 * Do whatever is necessary to prepare the address space
1133 	 * for remapping.  Note that this might replace the current
1134 	 * vmspace with another!
1135 	 */
1136 	if (is_spawn)
1137 		uvmspace_spawn(l, epp->ep_vm_minaddr,
1138 		    epp->ep_vm_maxaddr,
1139 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1140 	else
1141 		uvmspace_exec(l, epp->ep_vm_minaddr,
1142 		    epp->ep_vm_maxaddr,
1143 		    epp->ep_flags & EXEC_TOPDOWN_VM);
1144 
1145 	struct vmspace		*vm;
1146 	vm = p->p_vmspace;
1147 	vm->vm_taddr = (void *)epp->ep_taddr;
1148 	vm->vm_tsize = btoc(epp->ep_tsize);
1149 	vm->vm_daddr = (void*)epp->ep_daddr;
1150 	vm->vm_dsize = btoc(epp->ep_dsize);
1151 	vm->vm_ssize = btoc(epp->ep_ssize);
1152 	vm->vm_issize = 0;
1153 	vm->vm_maxsaddr = (void *)epp->ep_maxsaddr;
1154 	vm->vm_minsaddr = (void *)epp->ep_minsaddr;
1155 
1156 #ifdef PAX_ASLR
1157 	pax_aslr_init_vm(l, vm);
1158 #endif /* PAX_ASLR */
1159 
1160 	/* Now map address space. */
1161 	error = execve_dovmcmds(l, data);
1162 	if (error != 0)
1163 		goto exec_abort;
1164 
1165 	pathexec(epp, p, data->ed_pathstring);
1166 
1167 	char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize);
1168 
1169 	error = copyoutargs(data, l, newstack);
1170 	if (error != 0)
1171 		goto exec_abort;
1172 
1173 	cwdexec(p);
1174 	fd_closeexec();		/* handle close on exec */
1175 
1176 	if (__predict_false(ktrace_on))
1177 		fd_ktrexecfd();
1178 
1179 	execsigs(p);		/* reset catched signals */
1180 
1181 	mutex_enter(p->p_lock);
1182 	l->l_ctxlink = NULL;	/* reset ucontext link */
1183 	p->p_acflag &= ~AFORK;
1184 	p->p_flag |= PK_EXEC;
1185 	mutex_exit(p->p_lock);
1186 
1187 	/*
1188 	 * Stop profiling.
1189 	 */
1190 	if ((p->p_stflag & PST_PROFIL) != 0) {
1191 		mutex_spin_enter(&p->p_stmutex);
1192 		stopprofclock(p);
1193 		mutex_spin_exit(&p->p_stmutex);
1194 	}
1195 
1196 	/*
1197 	 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have
1198 	 * exited and exec()/exit() are the only places it will be cleared.
1199 	 */
1200 	if ((p->p_lflag & PL_PPWAIT) != 0) {
1201 #if 0
1202 		lwp_t *lp;
1203 
1204 		mutex_enter(proc_lock);
1205 		lp = p->p_vforklwp;
1206 		p->p_vforklwp = NULL;
1207 
1208 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1209 		p->p_lflag &= ~PL_PPWAIT;
1210 
1211 		lp->l_pflag &= ~LP_VFORKWAIT; /* XXX */
1212 		cv_broadcast(&lp->l_waitcv);
1213 		mutex_exit(proc_lock);
1214 #else
1215 		mutex_enter(proc_lock);
1216 		l->l_lwpctl = NULL; /* was on loan from blocked parent */
1217 		p->p_lflag &= ~PL_PPWAIT;
1218 		cv_broadcast(&p->p_pptr->p_waitcv);
1219 		mutex_exit(proc_lock);
1220 #endif
1221 	}
1222 
1223 	error = credexec(l, &data->ed_attr);
1224 	if (error)
1225 		goto exec_abort;
1226 
1227 #if defined(__HAVE_RAS)
1228 	/*
1229 	 * Remove all RASs from the address space.
1230 	 */
1231 	ras_purgeall();
1232 #endif
1233 
1234 	doexechooks(p);
1235 
1236 	/*
1237 	 * Set initial SP at the top of the stack.
1238 	 *
1239 	 * Note that on machines where stack grows up (e.g. hppa), SP points to
1240 	 * the end of arg/env strings.  Userland guesses the address of argc
1241 	 * via ps_strings::ps_argvstr.
1242 	 */
1243 
1244 	/* Setup new registers and do misc. setup. */
1245 	(*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack);
1246 	if (epp->ep_esch->es_setregs)
1247 		(*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack);
1248 
1249 	/* Provide a consistent LWP private setting */
1250 	(void)lwp_setprivate(l, NULL);
1251 
1252 	/* Discard all PCU state; need to start fresh */
1253 	pcu_discard_all(l);
1254 
1255 	/* map the process's signal trampoline code */
1256 	if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) {
1257 		DPRINTF(("%s: map sigcode failed %d\n", __func__, error));
1258 		goto exec_abort;
1259 	}
1260 
1261 	pool_put(&exec_pool, data->ed_argp);
1262 
1263 	/* notify others that we exec'd */
1264 	KNOTE(&p->p_klist, NOTE_EXEC);
1265 
1266 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1267 
1268 	SDT_PROBE(proc,,,exec_success, epp->ep_kname, 0, 0, 0, 0);
1269 
1270 	emulexec(l, epp);
1271 
1272 	/* Allow new references from the debugger/procfs. */
1273 	rw_exit(&p->p_reflock);
1274 	if (!no_local_exec_lock)
1275 		rw_exit(&exec_lock);
1276 
1277 	mutex_enter(proc_lock);
1278 
1279 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1280 		ksiginfo_t ksi;
1281 
1282 		KSI_INIT_EMPTY(&ksi);
1283 		ksi.ksi_signo = SIGTRAP;
1284 		ksi.ksi_lid = l->l_lid;
1285 		kpsignal(p, &ksi, NULL);
1286 	}
1287 
1288 	if (p->p_sflag & PS_STOPEXEC) {
1289 		ksiginfoq_t kq;
1290 
1291 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1292 		p->p_pptr->p_nstopchild++;
1293 		p->p_pptr->p_waited = 0;
1294 		mutex_enter(p->p_lock);
1295 		ksiginfo_queue_init(&kq);
1296 		sigclearall(p, &contsigmask, &kq);
1297 		lwp_lock(l);
1298 		l->l_stat = LSSTOP;
1299 		p->p_stat = SSTOP;
1300 		p->p_nrlwps--;
1301 		lwp_unlock(l);
1302 		mutex_exit(p->p_lock);
1303 		mutex_exit(proc_lock);
1304 		lwp_lock(l);
1305 		mi_switch(l);
1306 		ksiginfo_queue_drain(&kq);
1307 		KERNEL_LOCK(l->l_biglocks, l);
1308 	} else {
1309 		mutex_exit(proc_lock);
1310 	}
1311 
1312 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1313 	pathbuf_destroy(data->ed_pathbuf);
1314 	PNBUF_PUT(data->ed_resolvedpathbuf);
1315 	DPRINTF(("%s finished\n", __func__));
1316 	return EJUSTRETURN;
1317 
1318  exec_abort:
1319 	SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0);
1320 	rw_exit(&p->p_reflock);
1321 	if (!no_local_exec_lock)
1322 		rw_exit(&exec_lock);
1323 
1324 	pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring);
1325 	pathbuf_destroy(data->ed_pathbuf);
1326 	PNBUF_PUT(data->ed_resolvedpathbuf);
1327 
1328 	/*
1329 	 * the old process doesn't exist anymore.  exit gracefully.
1330 	 * get rid of the (new) address space we have created, if any, get rid
1331 	 * of our namei data and vnode, and exit noting failure
1332 	 */
1333 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1334 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1335 
1336 	exec_free_emul_arg(epp);
1337 	pool_put(&exec_pool, data->ed_argp);
1338 	kmem_free(epp->ep_hdr, epp->ep_hdrlen);
1339 	if (epp->ep_emul_root != NULL)
1340 		vrele(epp->ep_emul_root);
1341 	if (epp->ep_interp != NULL)
1342 		vrele(epp->ep_interp);
1343 
1344 	/* Acquire the sched-state mutex (exit1() will release it). */
1345 	if (!is_spawn) {
1346 		mutex_enter(p->p_lock);
1347 		exit1(l, W_EXITCODE(error, SIGABRT));
1348 	}
1349 
1350 	return error;
1351 }
1352 
1353 int
1354 execve1(struct lwp *l, const char *path, char * const *args,
1355     char * const *envs, execve_fetch_element_t fetch_element)
1356 {
1357 	struct execve_data data;
1358 	int error;
1359 
1360 	error = execve_loadvm(l, path, args, envs, fetch_element, &data);
1361 	if (error)
1362 		return error;
1363 	error = execve_runproc(l, &data, false, false);
1364 	return error;
1365 }
1366 
1367 static size_t
1368 fromptrsz(const struct exec_package *epp)
1369 {
1370 	return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *);
1371 }
1372 
1373 static size_t
1374 ptrsz(const struct exec_package *epp)
1375 {
1376 	return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *);
1377 }
1378 
1379 static size_t
1380 calcargs(struct execve_data * restrict data, const size_t argenvstrlen)
1381 {
1382 	struct exec_package	* const epp = &data->ed_pack;
1383 
1384 	const size_t nargenvptrs =
1385 	    1 +				/* long argc */
1386 	    data->ed_argc +		/* char *argv[] */
1387 	    1 +				/* \0 */
1388 	    data->ed_envc +		/* char *env[] */
1389 	    1 +				/* \0 */
1390 	    epp->ep_esch->es_arglen;	/* auxinfo */
1391 
1392 	return (nargenvptrs * ptrsz(epp)) + argenvstrlen;
1393 }
1394 
1395 static size_t
1396 calcstack(struct execve_data * restrict data, const size_t gaplen)
1397 {
1398 	struct exec_package	* const epp = &data->ed_pack;
1399 
1400 	data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode -
1401 	    epp->ep_esch->es_emul->e_sigcode;
1402 
1403 	data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ?
1404 	    sizeof(struct ps_strings32) : sizeof(struct ps_strings);
1405 
1406 	const size_t sigcode_psstr_sz =
1407 	    data->ed_szsigcode +	/* sigcode */
1408 	    data->ed_ps_strings_sz +	/* ps_strings */
1409 	    STACK_PTHREADSPACE;		/* pthread space */
1410 
1411 	const size_t stacklen =
1412 	    data->ed_argslen +
1413 	    gaplen +
1414 	    sigcode_psstr_sz;
1415 
1416 	/* make the stack "safely" aligned */
1417 	return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES);
1418 }
1419 
1420 static int
1421 copyoutargs(struct execve_data * restrict data, struct lwp *l,
1422     char * const newstack)
1423 {
1424 	struct exec_package	* const epp = &data->ed_pack;
1425 	struct proc		*p = l->l_proc;
1426 	int			error;
1427 
1428 	/* remember information about the process */
1429 	data->ed_arginfo.ps_nargvstr = data->ed_argc;
1430 	data->ed_arginfo.ps_nenvstr = data->ed_envc;
1431 
1432 	/*
1433 	 * Allocate the stack address passed to the newly execve()'ed process.
1434 	 *
1435 	 * The new stack address will be set to the SP (stack pointer) register
1436 	 * in setregs().
1437 	 */
1438 
1439 	char *newargs = STACK_ALLOC(
1440 	    STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen);
1441 
1442 	error = (*epp->ep_esch->es_copyargs)(l, epp,
1443 	    &data->ed_arginfo, &newargs, data->ed_argp);
1444 
1445 	if (epp->ep_path) {
1446 		PNBUF_PUT(epp->ep_path);
1447 		epp->ep_path = NULL;
1448 	}
1449 	if (error) {
1450 		DPRINTF(("%s: copyargs failed %d\n", __func__, error));
1451 		return error;
1452 	}
1453 
1454 	error = copyoutpsstrs(data, p);
1455 	if (error != 0)
1456 		return error;
1457 
1458 	return 0;
1459 }
1460 
1461 static int
1462 copyoutpsstrs(struct execve_data * restrict data, struct proc *p)
1463 {
1464 	struct exec_package	* const epp = &data->ed_pack;
1465 	struct ps_strings32	arginfo32;
1466 	void			*aip;
1467 	int			error;
1468 
1469 	/* fill process ps_strings info */
1470 	p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr,
1471 	    STACK_PTHREADSPACE), data->ed_ps_strings_sz);
1472 
1473 	if (epp->ep_flags & EXEC_32) {
1474 		aip = &arginfo32;
1475 		arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr;
1476 		arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr;
1477 		arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr;
1478 		arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr;
1479 	} else
1480 		aip = &data->ed_arginfo;
1481 
1482 	/* copy out the process's ps_strings structure */
1483 	if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz))
1484 	    != 0) {
1485 		DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n",
1486 		    __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz));
1487 		return error;
1488 	}
1489 
1490 	return 0;
1491 }
1492 
1493 static int
1494 copyinargs(struct execve_data * restrict data, char * const *args,
1495     char * const *envs, execve_fetch_element_t fetch_element, char **dpp)
1496 {
1497 	struct exec_package	* const epp = &data->ed_pack;
1498 	char			*dp;
1499 	size_t			i;
1500 	int			error;
1501 
1502 	dp = *dpp;
1503 
1504 	data->ed_argc = 0;
1505 
1506 	/* copy the fake args list, if there's one, freeing it as we go */
1507 	if (epp->ep_flags & EXEC_HASARGL) {
1508 		struct exec_fakearg	*fa = epp->ep_fa;
1509 
1510 		while (fa->fa_arg != NULL) {
1511 			const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1512 			size_t len;
1513 
1514 			len = strlcpy(dp, fa->fa_arg, maxlen);
1515 			/* Count NUL into len. */
1516 			if (len < maxlen)
1517 				len++;
1518 			else {
1519 				while (fa->fa_arg != NULL) {
1520 					kmem_free(fa->fa_arg, fa->fa_len);
1521 					fa++;
1522 				}
1523 				kmem_free(epp->ep_fa, epp->ep_fa_len);
1524 				epp->ep_flags &= ~EXEC_HASARGL;
1525 				return E2BIG;
1526 			}
1527 			ktrexecarg(fa->fa_arg, len - 1);
1528 			dp += len;
1529 
1530 			kmem_free(fa->fa_arg, fa->fa_len);
1531 			fa++;
1532 			data->ed_argc++;
1533 		}
1534 		kmem_free(epp->ep_fa, epp->ep_fa_len);
1535 		epp->ep_flags &= ~EXEC_HASARGL;
1536 	}
1537 
1538 	/*
1539 	 * Read and count argument strings from user.
1540 	 */
1541 
1542 	if (args == NULL) {
1543 		DPRINTF(("%s: null args\n", __func__));
1544 		return EINVAL;
1545 	}
1546 	if (epp->ep_flags & EXEC_SKIPARG)
1547 		args = (const void *)((const char *)args + fromptrsz(epp));
1548 	i = 0;
1549 	error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg);
1550 	if (error != 0) {
1551 		DPRINTF(("%s: copyin arg %d\n", __func__, error));
1552 		return error;
1553 	}
1554 	data->ed_argc += i;
1555 
1556 	/*
1557 	 * Read and count environment strings from user.
1558 	 */
1559 
1560 	data->ed_envc = 0;
1561 	/* environment need not be there */
1562 	if (envs == NULL)
1563 		goto done;
1564 	i = 0;
1565 	error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv);
1566 	if (error != 0) {
1567 		DPRINTF(("%s: copyin env %d\n", __func__, error));
1568 		return error;
1569 	}
1570 	data->ed_envc += i;
1571 
1572 done:
1573 	*dpp = dp;
1574 
1575 	return 0;
1576 }
1577 
1578 static int
1579 copyinargstrs(struct execve_data * restrict data, char * const *strs,
1580     execve_fetch_element_t fetch_element, char **dpp, size_t *ip,
1581     void (*ktr)(const void *, size_t))
1582 {
1583 	char			*dp, *sp;
1584 	size_t			i;
1585 	int			error;
1586 
1587 	dp = *dpp;
1588 
1589 	i = 0;
1590 	while (1) {
1591 		const size_t maxlen = ARG_MAX - (dp - data->ed_argp);
1592 		size_t len;
1593 
1594 		if ((error = (*fetch_element)(strs, i, &sp)) != 0) {
1595 			return error;
1596 		}
1597 		if (!sp)
1598 			break;
1599 		if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) {
1600 			if (error == ENAMETOOLONG)
1601 				error = E2BIG;
1602 			return error;
1603 		}
1604 		if (__predict_false(ktrace_on))
1605 			(*ktr)(dp, len - 1);
1606 		dp += len;
1607 		i++;
1608 	}
1609 
1610 	*dpp = dp;
1611 	*ip = i;
1612 
1613 	return 0;
1614 }
1615 
1616 /*
1617  * Copy argv and env strings from kernel buffer (argp) to the new stack.
1618  * Those strings are located just after auxinfo.
1619  */
1620 int
1621 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1622     char **stackp, void *argp)
1623 {
1624 	char	**cpp, *dp, *sp;
1625 	size_t	len;
1626 	void	*nullp;
1627 	long	argc, envc;
1628 	int	error;
1629 
1630 	cpp = (char **)*stackp;
1631 	nullp = NULL;
1632 	argc = arginfo->ps_nargvstr;
1633 	envc = arginfo->ps_nenvstr;
1634 
1635 	/* argc on stack is long */
1636 	CTASSERT(sizeof(*cpp) == sizeof(argc));
1637 
1638 	dp = (char *)(cpp +
1639 	    1 +				/* long argc */
1640 	    argc +			/* char *argv[] */
1641 	    1 +				/* \0 */
1642 	    envc +			/* char *env[] */
1643 	    1 +				/* \0 */
1644 	    /* XXX auxinfo multiplied by ptr size? */
1645 	    pack->ep_esch->es_arglen);	/* auxinfo */
1646 	sp = argp;
1647 
1648 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) {
1649 		COPYPRINTF("", cpp - 1, sizeof(argc));
1650 		return error;
1651 	}
1652 
1653 	/* XXX don't copy them out, remap them! */
1654 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1655 
1656 	for (; --argc >= 0; sp += len, dp += len) {
1657 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1658 			COPYPRINTF("", cpp - 1, sizeof(dp));
1659 			return error;
1660 		}
1661 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1662 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1663 			return error;
1664 		}
1665 	}
1666 
1667 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1668 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1669 		return error;
1670 	}
1671 
1672 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1673 
1674 	for (; --envc >= 0; sp += len, dp += len) {
1675 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) {
1676 			COPYPRINTF("", cpp - 1, sizeof(dp));
1677 			return error;
1678 		}
1679 		if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) {
1680 			COPYPRINTF("str", dp, (size_t)ARG_MAX);
1681 			return error;
1682 		}
1683 
1684 	}
1685 
1686 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) {
1687 		COPYPRINTF("", cpp - 1, sizeof(nullp));
1688 		return error;
1689 	}
1690 
1691 	*stackp = (char *)cpp;
1692 	return 0;
1693 }
1694 
1695 
1696 /*
1697  * Add execsw[] entries.
1698  */
1699 int
1700 exec_add(struct execsw *esp, int count)
1701 {
1702 	struct exec_entry	*it;
1703 	int			i;
1704 
1705 	if (count == 0) {
1706 		return 0;
1707 	}
1708 
1709 	/* Check for duplicates. */
1710 	rw_enter(&exec_lock, RW_WRITER);
1711 	for (i = 0; i < count; i++) {
1712 		LIST_FOREACH(it, &ex_head, ex_list) {
1713 			/* assume unique (makecmds, probe_func, emulation) */
1714 			if (it->ex_sw->es_makecmds == esp[i].es_makecmds &&
1715 			    it->ex_sw->u.elf_probe_func ==
1716 			    esp[i].u.elf_probe_func &&
1717 			    it->ex_sw->es_emul == esp[i].es_emul) {
1718 				rw_exit(&exec_lock);
1719 				return EEXIST;
1720 			}
1721 		}
1722 	}
1723 
1724 	/* Allocate new entries. */
1725 	for (i = 0; i < count; i++) {
1726 		it = kmem_alloc(sizeof(*it), KM_SLEEP);
1727 		it->ex_sw = &esp[i];
1728 		LIST_INSERT_HEAD(&ex_head, it, ex_list);
1729 	}
1730 
1731 	/* update execsw[] */
1732 	exec_init(0);
1733 	rw_exit(&exec_lock);
1734 	return 0;
1735 }
1736 
1737 /*
1738  * Remove execsw[] entry.
1739  */
1740 int
1741 exec_remove(struct execsw *esp, int count)
1742 {
1743 	struct exec_entry	*it, *next;
1744 	int			i;
1745 	const struct proclist_desc *pd;
1746 	proc_t			*p;
1747 
1748 	if (count == 0) {
1749 		return 0;
1750 	}
1751 
1752 	/* Abort if any are busy. */
1753 	rw_enter(&exec_lock, RW_WRITER);
1754 	for (i = 0; i < count; i++) {
1755 		mutex_enter(proc_lock);
1756 		for (pd = proclists; pd->pd_list != NULL; pd++) {
1757 			PROCLIST_FOREACH(p, pd->pd_list) {
1758 				if (p->p_execsw == &esp[i]) {
1759 					mutex_exit(proc_lock);
1760 					rw_exit(&exec_lock);
1761 					return EBUSY;
1762 				}
1763 			}
1764 		}
1765 		mutex_exit(proc_lock);
1766 	}
1767 
1768 	/* None are busy, so remove them all. */
1769 	for (i = 0; i < count; i++) {
1770 		for (it = LIST_FIRST(&ex_head); it != NULL; it = next) {
1771 			next = LIST_NEXT(it, ex_list);
1772 			if (it->ex_sw == &esp[i]) {
1773 				LIST_REMOVE(it, ex_list);
1774 				kmem_free(it, sizeof(*it));
1775 				break;
1776 			}
1777 		}
1778 	}
1779 
1780 	/* update execsw[] */
1781 	exec_init(0);
1782 	rw_exit(&exec_lock);
1783 	return 0;
1784 }
1785 
1786 /*
1787  * Initialize exec structures. If init_boot is true, also does necessary
1788  * one-time initialization (it's called from main() that way).
1789  * Once system is multiuser, this should be called with exec_lock held,
1790  * i.e. via exec_{add|remove}().
1791  */
1792 int
1793 exec_init(int init_boot)
1794 {
1795 	const struct execsw 	**sw;
1796 	struct exec_entry	*ex;
1797 	SLIST_HEAD(,exec_entry)	first;
1798 	SLIST_HEAD(,exec_entry)	any;
1799 	SLIST_HEAD(,exec_entry)	last;
1800 	int			i, sz;
1801 
1802 	if (init_boot) {
1803 		/* do one-time initializations */
1804 		rw_init(&exec_lock);
1805 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1806 		pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH,
1807 		    "execargs", &exec_palloc, IPL_NONE);
1808 		pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0);
1809 	} else {
1810 		KASSERT(rw_write_held(&exec_lock));
1811 	}
1812 
1813 	/* Sort each entry onto the appropriate queue. */
1814 	SLIST_INIT(&first);
1815 	SLIST_INIT(&any);
1816 	SLIST_INIT(&last);
1817 	sz = 0;
1818 	LIST_FOREACH(ex, &ex_head, ex_list) {
1819 		switch(ex->ex_sw->es_prio) {
1820 		case EXECSW_PRIO_FIRST:
1821 			SLIST_INSERT_HEAD(&first, ex, ex_slist);
1822 			break;
1823 		case EXECSW_PRIO_ANY:
1824 			SLIST_INSERT_HEAD(&any, ex, ex_slist);
1825 			break;
1826 		case EXECSW_PRIO_LAST:
1827 			SLIST_INSERT_HEAD(&last, ex, ex_slist);
1828 			break;
1829 		default:
1830 			panic("%s", __func__);
1831 			break;
1832 		}
1833 		sz++;
1834 	}
1835 
1836 	/*
1837 	 * Create new execsw[].  Ensure we do not try a zero-sized
1838 	 * allocation.
1839 	 */
1840 	sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP);
1841 	i = 0;
1842 	SLIST_FOREACH(ex, &first, ex_slist) {
1843 		sw[i++] = ex->ex_sw;
1844 	}
1845 	SLIST_FOREACH(ex, &any, ex_slist) {
1846 		sw[i++] = ex->ex_sw;
1847 	}
1848 	SLIST_FOREACH(ex, &last, ex_slist) {
1849 		sw[i++] = ex->ex_sw;
1850 	}
1851 
1852 	/* Replace old execsw[] and free used memory. */
1853 	if (execsw != NULL) {
1854 		kmem_free(__UNCONST(execsw),
1855 		    nexecs * sizeof(struct execsw *) + 1);
1856 	}
1857 	execsw = sw;
1858 	nexecs = sz;
1859 
1860 	/* Figure out the maximum size of an exec header. */
1861 	exec_maxhdrsz = sizeof(int);
1862 	for (i = 0; i < nexecs; i++) {
1863 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1864 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1865 	}
1866 
1867 	return 0;
1868 }
1869 
1870 static int
1871 exec_sigcode_map(struct proc *p, const struct emul *e)
1872 {
1873 	vaddr_t va;
1874 	vsize_t sz;
1875 	int error;
1876 	struct uvm_object *uobj;
1877 
1878 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1879 
1880 	if (e->e_sigobject == NULL || sz == 0) {
1881 		return 0;
1882 	}
1883 
1884 	/*
1885 	 * If we don't have a sigobject for this emulation, create one.
1886 	 *
1887 	 * sigobject is an anonymous memory object (just like SYSV shared
1888 	 * memory) that we keep a permanent reference to and that we map
1889 	 * in all processes that need this sigcode. The creation is simple,
1890 	 * we create an object, add a permanent reference to it, map it in
1891 	 * kernel space, copy out the sigcode to it and unmap it.
1892 	 * We map it with PROT_READ|PROT_EXEC into the process just
1893 	 * the way sys_mmap() would map it.
1894 	 */
1895 
1896 	uobj = *e->e_sigobject;
1897 	if (uobj == NULL) {
1898 		mutex_enter(&sigobject_lock);
1899 		if ((uobj = *e->e_sigobject) == NULL) {
1900 			uobj = uao_create(sz, 0);
1901 			(*uobj->pgops->pgo_reference)(uobj);
1902 			va = vm_map_min(kernel_map);
1903 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1904 			    uobj, 0, 0,
1905 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1906 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1907 				printf("kernel mapping failed %d\n", error);
1908 				(*uobj->pgops->pgo_detach)(uobj);
1909 				mutex_exit(&sigobject_lock);
1910 				return error;
1911 			}
1912 			memcpy((void *)va, e->e_sigcode, sz);
1913 #ifdef PMAP_NEED_PROCWR
1914 			pmap_procwr(&proc0, va, sz);
1915 #endif
1916 			uvm_unmap(kernel_map, va, va + round_page(sz));
1917 			*e->e_sigobject = uobj;
1918 		}
1919 		mutex_exit(&sigobject_lock);
1920 	}
1921 
1922 	/* Just a hint to uvm_map where to put it. */
1923 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1924 	    round_page(sz));
1925 
1926 #ifdef __alpha__
1927 	/*
1928 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1929 	 * which causes the above calculation to put the sigcode at
1930 	 * an invalid address.  Put it just below the text instead.
1931 	 */
1932 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1933 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1934 	}
1935 #endif
1936 
1937 	(*uobj->pgops->pgo_reference)(uobj);
1938 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1939 			uobj, 0, 0,
1940 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1941 				    UVM_ADV_RANDOM, 0));
1942 	if (error) {
1943 		DPRINTF(("%s, %d: map %p "
1944 		    "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n",
1945 		    __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz),
1946 		    va, error));
1947 		(*uobj->pgops->pgo_detach)(uobj);
1948 		return error;
1949 	}
1950 	p->p_sigctx.ps_sigcode = (void *)va;
1951 	return 0;
1952 }
1953 
1954 /*
1955  * Release a refcount on spawn_exec_data and destroy memory, if this
1956  * was the last one.
1957  */
1958 static void
1959 spawn_exec_data_release(struct spawn_exec_data *data)
1960 {
1961 	if (atomic_dec_32_nv(&data->sed_refcnt) != 0)
1962 		return;
1963 
1964 	cv_destroy(&data->sed_cv_child_ready);
1965 	mutex_destroy(&data->sed_mtx_child);
1966 
1967 	if (data->sed_actions)
1968 		posix_spawn_fa_free(data->sed_actions,
1969 		    data->sed_actions->len);
1970 	if (data->sed_attrs)
1971 		kmem_free(data->sed_attrs,
1972 		    sizeof(*data->sed_attrs));
1973 	kmem_free(data, sizeof(*data));
1974 }
1975 
1976 /*
1977  * A child lwp of a posix_spawn operation starts here and ends up in
1978  * cpu_spawn_return, dealing with all filedescriptor and scheduler
1979  * manipulations in between.
1980  * The parent waits for the child, as it is not clear whether the child
1981  * will be able to acquire its own exec_lock. If it can, the parent can
1982  * be released early and continue running in parallel. If not (or if the
1983  * magic debug flag is passed in the scheduler attribute struct), the
1984  * child rides on the parent's exec lock until it is ready to return to
1985  * to userland - and only then releases the parent. This method loses
1986  * concurrency, but improves error reporting.
1987  */
1988 static void
1989 spawn_return(void *arg)
1990 {
1991 	struct spawn_exec_data *spawn_data = arg;
1992 	struct lwp *l = curlwp;
1993 	int error, newfd;
1994 	size_t i;
1995 	const struct posix_spawn_file_actions_entry *fae;
1996 	pid_t ppid;
1997 	register_t retval;
1998 	bool have_reflock;
1999 	bool parent_is_waiting = true;
2000 
2001 	/*
2002 	 * Check if we can release parent early.
2003 	 * We either need to have no sed_attrs, or sed_attrs does not
2004 	 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require
2005 	 * safe access to the parent proc (passed in sed_parent).
2006 	 * We then try to get the exec_lock, and only if that works, we can
2007 	 * release the parent here already.
2008 	 */
2009 	ppid = spawn_data->sed_parent->p_pid;
2010 	if ((!spawn_data->sed_attrs
2011 	    || (spawn_data->sed_attrs->sa_flags
2012 	        & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0)
2013 	    && rw_tryenter(&exec_lock, RW_READER)) {
2014 		parent_is_waiting = false;
2015 		mutex_enter(&spawn_data->sed_mtx_child);
2016 		cv_signal(&spawn_data->sed_cv_child_ready);
2017 		mutex_exit(&spawn_data->sed_mtx_child);
2018 	}
2019 
2020 	/* don't allow debugger access yet */
2021 	rw_enter(&l->l_proc->p_reflock, RW_WRITER);
2022 	have_reflock = true;
2023 
2024 	error = 0;
2025 	/* handle posix_spawn_file_actions */
2026 	if (spawn_data->sed_actions != NULL) {
2027 		for (i = 0; i < spawn_data->sed_actions->len; i++) {
2028 			fae = &spawn_data->sed_actions->fae[i];
2029 			switch (fae->fae_action) {
2030 			case FAE_OPEN:
2031 				if (fd_getfile(fae->fae_fildes) != NULL) {
2032 					error = fd_close(fae->fae_fildes);
2033 					if (error)
2034 						break;
2035 				}
2036 				error = fd_open(fae->fae_path, fae->fae_oflag,
2037 				    fae->fae_mode, &newfd);
2038 				if (error)
2039 					break;
2040 				if (newfd != fae->fae_fildes) {
2041 					error = dodup(l, newfd,
2042 					    fae->fae_fildes, 0, &retval);
2043 					if (fd_getfile(newfd) != NULL)
2044 						fd_close(newfd);
2045 				}
2046 				break;
2047 			case FAE_DUP2:
2048 				error = dodup(l, fae->fae_fildes,
2049 				    fae->fae_newfildes, 0, &retval);
2050 				break;
2051 			case FAE_CLOSE:
2052 				if (fd_getfile(fae->fae_fildes) == NULL) {
2053 					error = EBADF;
2054 					break;
2055 				}
2056 				error = fd_close(fae->fae_fildes);
2057 				break;
2058 			}
2059 			if (error)
2060 				goto report_error;
2061 		}
2062 	}
2063 
2064 	/* handle posix_spawnattr */
2065 	if (spawn_data->sed_attrs != NULL) {
2066 		int ostat;
2067 		struct sigaction sigact;
2068 		sigact._sa_u._sa_handler = SIG_DFL;
2069 		sigact.sa_flags = 0;
2070 
2071 		/*
2072 		 * set state to SSTOP so that this proc can be found by pid.
2073 		 * see proc_enterprp, do_sched_setparam below
2074 		 */
2075 		ostat = l->l_proc->p_stat;
2076 		l->l_proc->p_stat = SSTOP;
2077 
2078 		/* Set process group */
2079 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETPGROUP) {
2080 			pid_t mypid = l->l_proc->p_pid,
2081 			     pgrp = spawn_data->sed_attrs->sa_pgroup;
2082 
2083 			if (pgrp == 0)
2084 				pgrp = mypid;
2085 
2086 			error = proc_enterpgrp(spawn_data->sed_parent,
2087 			    mypid, pgrp, false);
2088 			if (error)
2089 				goto report_error;
2090 		}
2091 
2092 		/* Set scheduler policy */
2093 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER)
2094 			error = do_sched_setparam(l->l_proc->p_pid, 0,
2095 			    spawn_data->sed_attrs->sa_schedpolicy,
2096 			    &spawn_data->sed_attrs->sa_schedparam);
2097 		else if (spawn_data->sed_attrs->sa_flags
2098 		    & POSIX_SPAWN_SETSCHEDPARAM) {
2099 			error = do_sched_setparam(ppid, 0,
2100 			    SCHED_NONE, &spawn_data->sed_attrs->sa_schedparam);
2101 		}
2102 		if (error)
2103 			goto report_error;
2104 
2105 		/* Reset user ID's */
2106 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_RESETIDS) {
2107 			error = do_setresuid(l, -1,
2108 			     kauth_cred_getgid(l->l_cred), -1,
2109 			     ID_E_EQ_R | ID_E_EQ_S);
2110 			if (error)
2111 				goto report_error;
2112 			error = do_setresuid(l, -1,
2113 			    kauth_cred_getuid(l->l_cred), -1,
2114 			    ID_E_EQ_R | ID_E_EQ_S);
2115 			if (error)
2116 				goto report_error;
2117 		}
2118 
2119 		/* Set signal masks/defaults */
2120 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) {
2121 			mutex_enter(l->l_proc->p_lock);
2122 			error = sigprocmask1(l, SIG_SETMASK,
2123 			    &spawn_data->sed_attrs->sa_sigmask, NULL);
2124 			mutex_exit(l->l_proc->p_lock);
2125 			if (error)
2126 				goto report_error;
2127 		}
2128 
2129 		if (spawn_data->sed_attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) {
2130 			/*
2131 			 * The following sigaction call is using a sigaction
2132 			 * version 0 trampoline which is in the compatibility
2133 			 * code only. This is not a problem because for SIG_DFL
2134 			 * and SIG_IGN, the trampolines are now ignored. If they
2135 			 * were not, this would be a problem because we are
2136 			 * holding the exec_lock, and the compat code needs
2137 			 * to do the same in order to replace the trampoline
2138 			 * code of the process.
2139 			 */
2140 			for (i = 1; i <= NSIG; i++) {
2141 				if (sigismember(
2142 				    &spawn_data->sed_attrs->sa_sigdefault, i))
2143 					sigaction1(l, i, &sigact, NULL, NULL,
2144 					    0);
2145 			}
2146 		}
2147 		l->l_proc->p_stat = ostat;
2148 	}
2149 
2150 	/* now do the real exec */
2151 	error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting,
2152 	    true);
2153 	have_reflock = false;
2154 	if (error == EJUSTRETURN)
2155 		error = 0;
2156 	else if (error)
2157 		goto report_error;
2158 
2159 	if (parent_is_waiting) {
2160 		mutex_enter(&spawn_data->sed_mtx_child);
2161 		cv_signal(&spawn_data->sed_cv_child_ready);
2162 		mutex_exit(&spawn_data->sed_mtx_child);
2163 	}
2164 
2165 	/* release our refcount on the data */
2166 	spawn_exec_data_release(spawn_data);
2167 
2168 	/* and finally: leave to userland for the first time */
2169 	cpu_spawn_return(l);
2170 
2171 	/* NOTREACHED */
2172 	return;
2173 
2174  report_error:
2175 	if (have_reflock) {
2176 		/*
2177 		 * We have not passed through execve_runproc(),
2178 		 * which would have released the p_reflock and also
2179 		 * taken ownership of the sed_exec part of spawn_data,
2180 		 * so release/free both here.
2181 		 */
2182 		rw_exit(&l->l_proc->p_reflock);
2183 		execve_free_data(&spawn_data->sed_exec);
2184 	}
2185 
2186 	if (parent_is_waiting) {
2187 		/* pass error to parent */
2188 		mutex_enter(&spawn_data->sed_mtx_child);
2189 		spawn_data->sed_error = error;
2190 		cv_signal(&spawn_data->sed_cv_child_ready);
2191 		mutex_exit(&spawn_data->sed_mtx_child);
2192 	} else {
2193 		rw_exit(&exec_lock);
2194 	}
2195 
2196 	/* release our refcount on the data */
2197 	spawn_exec_data_release(spawn_data);
2198 
2199 	/* done, exit */
2200 	mutex_enter(l->l_proc->p_lock);
2201 	/*
2202 	 * Posix explicitly asks for an exit code of 127 if we report
2203 	 * errors from the child process - so, unfortunately, there
2204 	 * is no way to report a more exact error code.
2205 	 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as
2206 	 * flag bit in the attrp argument to posix_spawn(2), see above.
2207 	 */
2208 	exit1(l, W_EXITCODE(127, 0));
2209 }
2210 
2211 void
2212 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len)
2213 {
2214 
2215 	for (size_t i = 0; i < len; i++) {
2216 		struct posix_spawn_file_actions_entry *fae = &fa->fae[i];
2217 		if (fae->fae_action != FAE_OPEN)
2218 			continue;
2219 		kmem_free(fae->fae_path, strlen(fae->fae_path) + 1);
2220 	}
2221 	if (fa->len > 0)
2222 		kmem_free(fa->fae, sizeof(*fa->fae) * fa->len);
2223 	kmem_free(fa, sizeof(*fa));
2224 }
2225 
2226 static int
2227 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap,
2228     const struct posix_spawn_file_actions *ufa, rlim_t lim)
2229 {
2230 	struct posix_spawn_file_actions *fa;
2231 	struct posix_spawn_file_actions_entry *fae;
2232 	char *pbuf = NULL;
2233 	int error;
2234 	size_t i = 0;
2235 
2236 	fa = kmem_alloc(sizeof(*fa), KM_SLEEP);
2237 	error = copyin(ufa, fa, sizeof(*fa));
2238 	if (error || fa->len == 0) {
2239 		kmem_free(fa, sizeof(*fa));
2240 		return error;	/* 0 if not an error, and len == 0 */
2241 	}
2242 
2243 	if (fa->len > lim) {
2244 		kmem_free(fa, sizeof(*fa));
2245 		return EINVAL;
2246 	}
2247 
2248 	fa->size = fa->len;
2249 	size_t fal = fa->len * sizeof(*fae);
2250 	fae = fa->fae;
2251 	fa->fae = kmem_alloc(fal, KM_SLEEP);
2252 	error = copyin(fae, fa->fae, fal);
2253 	if (error)
2254 		goto out;
2255 
2256 	pbuf = PNBUF_GET();
2257 	for (; i < fa->len; i++) {
2258 		fae = &fa->fae[i];
2259 		if (fae->fae_action != FAE_OPEN)
2260 			continue;
2261 		error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal);
2262 		if (error)
2263 			goto out;
2264 		fae->fae_path = kmem_alloc(fal, KM_SLEEP);
2265 		memcpy(fae->fae_path, pbuf, fal);
2266 	}
2267 	PNBUF_PUT(pbuf);
2268 
2269 	*fap = fa;
2270 	return 0;
2271 out:
2272 	if (pbuf)
2273 		PNBUF_PUT(pbuf);
2274 	posix_spawn_fa_free(fa, i);
2275 	return error;
2276 }
2277 
2278 int
2279 check_posix_spawn(struct lwp *l1)
2280 {
2281 	int error, tnprocs, count;
2282 	uid_t uid;
2283 	struct proc *p1;
2284 
2285 	p1 = l1->l_proc;
2286 	uid = kauth_cred_getuid(l1->l_cred);
2287 	tnprocs = atomic_inc_uint_nv(&nprocs);
2288 
2289 	/*
2290 	 * Although process entries are dynamically created, we still keep
2291 	 * a global limit on the maximum number we will create.
2292 	 */
2293 	if (__predict_false(tnprocs >= maxproc))
2294 		error = -1;
2295 	else
2296 		error = kauth_authorize_process(l1->l_cred,
2297 		    KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL);
2298 
2299 	if (error) {
2300 		atomic_dec_uint(&nprocs);
2301 		return EAGAIN;
2302 	}
2303 
2304 	/*
2305 	 * Enforce limits.
2306 	 */
2307 	count = chgproccnt(uid, 1);
2308 	if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT,
2309 	     p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS),
2310 	     &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 &&
2311 	    __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) {
2312 		(void)chgproccnt(uid, -1);
2313 		atomic_dec_uint(&nprocs);
2314 		return EAGAIN;
2315 	}
2316 
2317 	return 0;
2318 }
2319 
2320 int
2321 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path,
2322 	struct posix_spawn_file_actions *fa,
2323 	struct posix_spawnattr *sa,
2324 	char *const *argv, char *const *envp,
2325 	execve_fetch_element_t fetch)
2326 {
2327 
2328 	struct proc *p1, *p2;
2329 	struct lwp *l2;
2330 	int error;
2331 	struct spawn_exec_data *spawn_data;
2332 	vaddr_t uaddr;
2333 	pid_t pid;
2334 	bool have_exec_lock = false;
2335 
2336 	p1 = l1->l_proc;
2337 
2338 	/* Allocate and init spawn_data */
2339 	spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP);
2340 	spawn_data->sed_refcnt = 1; /* only parent so far */
2341 	cv_init(&spawn_data->sed_cv_child_ready, "pspawn");
2342 	mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE);
2343 	mutex_enter(&spawn_data->sed_mtx_child);
2344 
2345 	/*
2346 	 * Do the first part of the exec now, collect state
2347 	 * in spawn_data.
2348 	 */
2349 	error = execve_loadvm(l1, path, argv,
2350 	    envp, fetch, &spawn_data->sed_exec);
2351 	if (error == EJUSTRETURN)
2352 		error = 0;
2353 	else if (error)
2354 		goto error_exit;
2355 
2356 	have_exec_lock = true;
2357 
2358 	/*
2359 	 * Allocate virtual address space for the U-area now, while it
2360 	 * is still easy to abort the fork operation if we're out of
2361 	 * kernel virtual address space.
2362 	 */
2363 	uaddr = uvm_uarea_alloc();
2364 	if (__predict_false(uaddr == 0)) {
2365 		error = ENOMEM;
2366 		goto error_exit;
2367 	}
2368 
2369 	/*
2370 	 * Allocate new proc. Borrow proc0 vmspace for it, we will
2371 	 * replace it with its own before returning to userland
2372 	 * in the child.
2373 	 * This is a point of no return, we will have to go through
2374 	 * the child proc to properly clean it up past this point.
2375 	 */
2376 	p2 = proc_alloc();
2377 	pid = p2->p_pid;
2378 
2379 	/*
2380 	 * Make a proc table entry for the new process.
2381 	 * Start by zeroing the section of proc that is zero-initialized,
2382 	 * then copy the section that is copied directly from the parent.
2383 	 */
2384 	memset(&p2->p_startzero, 0,
2385 	    (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero));
2386 	memcpy(&p2->p_startcopy, &p1->p_startcopy,
2387 	    (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy));
2388 	p2->p_vmspace = proc0.p_vmspace;
2389 
2390 	TAILQ_INIT(&p2->p_sigpend.sp_info);
2391 
2392 	LIST_INIT(&p2->p_lwps);
2393 	LIST_INIT(&p2->p_sigwaiters);
2394 
2395 	/*
2396 	 * Duplicate sub-structures as needed.
2397 	 * Increase reference counts on shared objects.
2398 	 * Inherit flags we want to keep.  The flags related to SIGCHLD
2399 	 * handling are important in order to keep a consistent behaviour
2400 	 * for the child after the fork.  If we are a 32-bit process, the
2401 	 * child will be too.
2402 	 */
2403 	p2->p_flag =
2404 	    p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32);
2405 	p2->p_emul = p1->p_emul;
2406 	p2->p_execsw = p1->p_execsw;
2407 
2408 	mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
2409 	mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
2410 	rw_init(&p2->p_reflock);
2411 	cv_init(&p2->p_waitcv, "wait");
2412 	cv_init(&p2->p_lwpcv, "lwpwait");
2413 
2414 	p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
2415 
2416 	kauth_proc_fork(p1, p2);
2417 
2418 	p2->p_raslist = NULL;
2419 	p2->p_fd = fd_copy();
2420 
2421 	/* XXX racy */
2422 	p2->p_mqueue_cnt = p1->p_mqueue_cnt;
2423 
2424 	p2->p_cwdi = cwdinit();
2425 
2426 	/*
2427 	 * Note: p_limit (rlimit stuff) is copy-on-write, so normally
2428 	 * we just need increase pl_refcnt.
2429 	 */
2430 	if (!p1->p_limit->pl_writeable) {
2431 		lim_addref(p1->p_limit);
2432 		p2->p_limit = p1->p_limit;
2433 	} else {
2434 		p2->p_limit = lim_copy(p1->p_limit);
2435 	}
2436 
2437 	p2->p_lflag = 0;
2438 	p2->p_sflag = 0;
2439 	p2->p_slflag = 0;
2440 	p2->p_pptr = p1;
2441 	p2->p_ppid = p1->p_pid;
2442 	LIST_INIT(&p2->p_children);
2443 
2444 	p2->p_aio = NULL;
2445 
2446 #ifdef KTRACE
2447 	/*
2448 	 * Copy traceflag and tracefile if enabled.
2449 	 * If not inherited, these were zeroed above.
2450 	 */
2451 	if (p1->p_traceflag & KTRFAC_INHERIT) {
2452 		mutex_enter(&ktrace_lock);
2453 		p2->p_traceflag = p1->p_traceflag;
2454 		if ((p2->p_tracep = p1->p_tracep) != NULL)
2455 			ktradref(p2);
2456 		mutex_exit(&ktrace_lock);
2457 	}
2458 #endif
2459 
2460 	/*
2461 	 * Create signal actions for the child process.
2462 	 */
2463 	p2->p_sigacts = sigactsinit(p1, 0);
2464 	mutex_enter(p1->p_lock);
2465 	p2->p_sflag |=
2466 	    (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP));
2467 	sched_proc_fork(p1, p2);
2468 	mutex_exit(p1->p_lock);
2469 
2470 	p2->p_stflag = p1->p_stflag;
2471 
2472 	/*
2473 	 * p_stats.
2474 	 * Copy parts of p_stats, and zero out the rest.
2475 	 */
2476 	p2->p_stats = pstatscopy(p1->p_stats);
2477 
2478 	/* copy over machdep flags to the new proc */
2479 	cpu_proc_fork(p1, p2);
2480 
2481 	/*
2482 	 * Prepare remaining parts of spawn data
2483 	 */
2484 	spawn_data->sed_actions = fa;
2485 	spawn_data->sed_attrs = sa;
2486 
2487 	spawn_data->sed_parent = p1;
2488 
2489 	/* create LWP */
2490 	lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data,
2491 	    &l2, l1->l_class);
2492 	l2->l_ctxlink = NULL;	/* reset ucontext link */
2493 
2494 	/*
2495 	 * Copy the credential so other references don't see our changes.
2496 	 * Test to see if this is necessary first, since in the common case
2497 	 * we won't need a private reference.
2498 	 */
2499 	if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) ||
2500 	    kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) {
2501 		l2->l_cred = kauth_cred_copy(l2->l_cred);
2502 		kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred));
2503 		kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred));
2504 	}
2505 
2506 	/* Update the master credentials. */
2507 	if (l2->l_cred != p2->p_cred) {
2508 		kauth_cred_t ocred;
2509 
2510 		kauth_cred_hold(l2->l_cred);
2511 		mutex_enter(p2->p_lock);
2512 		ocred = p2->p_cred;
2513 		p2->p_cred = l2->l_cred;
2514 		mutex_exit(p2->p_lock);
2515 		kauth_cred_free(ocred);
2516 	}
2517 
2518 	*child_ok = true;
2519 	spawn_data->sed_refcnt = 2;	/* child gets it as well */
2520 #if 0
2521 	l2->l_nopreempt = 1; /* start it non-preemptable */
2522 #endif
2523 
2524 	/*
2525 	 * It's now safe for the scheduler and other processes to see the
2526 	 * child process.
2527 	 */
2528 	mutex_enter(proc_lock);
2529 
2530 	if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT)
2531 		p2->p_lflag |= PL_CONTROLT;
2532 
2533 	LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling);
2534 	p2->p_exitsig = SIGCHLD;	/* signal for parent on exit */
2535 
2536 	LIST_INSERT_AFTER(p1, p2, p_pglist);
2537 	LIST_INSERT_HEAD(&allproc, p2, p_list);
2538 
2539 	p2->p_trace_enabled = trace_is_enabled(p2);
2540 #ifdef __HAVE_SYSCALL_INTERN
2541 	(*p2->p_emul->e_syscall_intern)(p2);
2542 #endif
2543 
2544 	/*
2545 	 * Make child runnable, set start time, and add to run queue except
2546 	 * if the parent requested the child to start in SSTOP state.
2547 	 */
2548 	mutex_enter(p2->p_lock);
2549 
2550 	getmicrotime(&p2->p_stats->p_start);
2551 
2552 	lwp_lock(l2);
2553 	KASSERT(p2->p_nrlwps == 1);
2554 	p2->p_nrlwps = 1;
2555 	p2->p_stat = SACTIVE;
2556 	l2->l_stat = LSRUN;
2557 	sched_enqueue(l2, false);
2558 	lwp_unlock(l2);
2559 
2560 	mutex_exit(p2->p_lock);
2561 	mutex_exit(proc_lock);
2562 
2563 	cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child);
2564 	error = spawn_data->sed_error;
2565 	mutex_exit(&spawn_data->sed_mtx_child);
2566 	spawn_exec_data_release(spawn_data);
2567 
2568 	rw_exit(&p1->p_reflock);
2569 	rw_exit(&exec_lock);
2570 	have_exec_lock = false;
2571 
2572 	*pid_res = pid;
2573 	return error;
2574 
2575  error_exit:
2576 	if (have_exec_lock) {
2577 		execve_free_data(&spawn_data->sed_exec);
2578 		rw_exit(&p1->p_reflock);
2579 		rw_exit(&exec_lock);
2580 	}
2581 	mutex_exit(&spawn_data->sed_mtx_child);
2582 	spawn_exec_data_release(spawn_data);
2583 
2584 	return error;
2585 }
2586 
2587 int
2588 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap,
2589     register_t *retval)
2590 {
2591 	/* {
2592 		syscallarg(pid_t *) pid;
2593 		syscallarg(const char *) path;
2594 		syscallarg(const struct posix_spawn_file_actions *) file_actions;
2595 		syscallarg(const struct posix_spawnattr *) attrp;
2596 		syscallarg(char *const *) argv;
2597 		syscallarg(char *const *) envp;
2598 	} */
2599 
2600 	int error;
2601 	struct posix_spawn_file_actions *fa = NULL;
2602 	struct posix_spawnattr *sa = NULL;
2603 	pid_t pid;
2604 	bool child_ok = false;
2605 	rlim_t max_fileactions;
2606 	proc_t *p = l1->l_proc;
2607 
2608 	error = check_posix_spawn(l1);
2609 	if (error) {
2610 		*retval = error;
2611 		return 0;
2612 	}
2613 
2614 	/* copy in file_actions struct */
2615 	if (SCARG(uap, file_actions) != NULL) {
2616 		max_fileactions = 2 * min(p->p_rlimit[RLIMIT_NOFILE].rlim_cur,
2617 		    maxfiles);
2618 		error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions),
2619 		    max_fileactions);
2620 		if (error)
2621 			goto error_exit;
2622 	}
2623 
2624 	/* copyin posix_spawnattr struct */
2625 	if (SCARG(uap, attrp) != NULL) {
2626 		sa = kmem_alloc(sizeof(*sa), KM_SLEEP);
2627 		error = copyin(SCARG(uap, attrp), sa, sizeof(*sa));
2628 		if (error)
2629 			goto error_exit;
2630 	}
2631 
2632 	/*
2633 	 * Do the spawn
2634 	 */
2635 	error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa,
2636 	    SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element);
2637 	if (error)
2638 		goto error_exit;
2639 
2640 	if (error == 0 && SCARG(uap, pid) != NULL)
2641 		error = copyout(&pid, SCARG(uap, pid), sizeof(pid));
2642 
2643 	*retval = error;
2644 	return 0;
2645 
2646  error_exit:
2647 	if (!child_ok) {
2648 		(void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1);
2649 		atomic_dec_uint(&nprocs);
2650 
2651 		if (sa)
2652 			kmem_free(sa, sizeof(*sa));
2653 		if (fa)
2654 			posix_spawn_fa_free(fa, fa->len);
2655 	}
2656 
2657 	*retval = error;
2658 	return 0;
2659 }
2660 
2661 void
2662 exec_free_emul_arg(struct exec_package *epp)
2663 {
2664 	if (epp->ep_emul_arg_free != NULL) {
2665 		KASSERT(epp->ep_emul_arg != NULL);
2666 		(*epp->ep_emul_arg_free)(epp->ep_emul_arg);
2667 		epp->ep_emul_arg_free = NULL;
2668 		epp->ep_emul_arg = NULL;
2669 	} else {
2670 		KASSERT(epp->ep_emul_arg == NULL);
2671 	}
2672 }
2673 
2674 #ifdef DEBUG_EXEC
2675 static void
2676 dump_vmcmds(const struct exec_package * const epp, size_t x, int error)
2677 {
2678 	struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0];
2679 	size_t j;
2680 
2681 	if (error == 0)
2682 		DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used));
2683 	else
2684 		DPRINTF(("vmcmds %zu/%u, error %d\n", x,
2685 		    epp->ep_vmcmds.evs_used, error));
2686 
2687 	for (j = 0; j < epp->ep_vmcmds.evs_used; j++) {
2688 		DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#"
2689 		    PRIxVADDR"/%#"PRIxVSIZE" fd@%#"
2690 		    PRIxVSIZE" prot=0%o flags=%d\n", j,
2691 		    vp[j].ev_proc == vmcmd_map_pagedvn ?
2692 		    "pagedvn" :
2693 		    vp[j].ev_proc == vmcmd_map_readvn ?
2694 		    "readvn" :
2695 		    vp[j].ev_proc == vmcmd_map_zero ?
2696 		    "zero" : "*unknown*",
2697 		    vp[j].ev_addr, vp[j].ev_len,
2698 		    vp[j].ev_offset, vp[j].ev_prot,
2699 		    vp[j].ev_flags));
2700 		if (error != 0 && j == x)
2701 			DPRINTF(("     ^--- failed\n"));
2702 	}
2703 }
2704 #endif
2705