xref: /freebsd/sys/kern/kern_exec.c (revision f95c930e6701aa675377bc5871ea490dd565eeba)
1 /*-
2  * Copyright (c) 1993, David Greenman
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_ktrace.h"
33 #include "opt_vm.h"
34 
35 #include <sys/param.h>
36 #include <sys/capsicum.h>
37 #include <sys/systm.h>
38 #include <sys/eventhandler.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/signalvar.h>
43 #include <sys/kernel.h>
44 #include <sys/mount.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
47 #include <sys/acct.h>
48 #include <sys/exec.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
51 #include <sys/wait.h>
52 #include <sys/malloc.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/pioctl.h>
56 #include <sys/namei.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/sf_buf.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysent.h>
64 #include <sys/shm.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/stat.h>
68 #ifdef KTRACE
69 #include <sys/ktrace.h>
70 #endif
71 
72 #include <vm/vm.h>
73 #include <vm/vm_param.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_pager.h>
81 
82 #ifdef	HWPMC_HOOKS
83 #include <sys/pmckern.h>
84 #endif
85 
86 #include <machine/reg.h>
87 
88 #include <security/audit/audit.h>
89 #include <security/mac/mac_framework.h>
90 
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t	dtrace_fasttrap_exec;
94 #endif
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
98 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
99 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
100 
101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
102 
103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106 static int do_execve(struct thread *td, struct image_args *args,
107     struct mac *mac_p);
108 
109 /* XXX This should be vm_size_t. */
110 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
111     NULL, 0, sysctl_kern_ps_strings, "LU", "");
112 
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
115     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
116 
117 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
118     NULL, 0, sysctl_kern_stackprot, "I", "");
119 
120 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
121 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
122     &ps_arg_cache_limit, 0, "");
123 
124 static int disallow_high_osrel;
125 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
126     &disallow_high_osrel, 0,
127     "Disallow execution of binaries built for higher version of the world");
128 
129 static int map_at_zero = 0;
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
131     "Permit processes to map an object at virtual address 0.");
132 
133 static int
134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
135 {
136 	struct proc *p;
137 	int error;
138 
139 	p = curproc;
140 #ifdef SCTL_MASK32
141 	if (req->flags & SCTL_MASK32) {
142 		unsigned int val;
143 		val = (unsigned int)p->p_sysent->sv_psstrings;
144 		error = SYSCTL_OUT(req, &val, sizeof(val));
145 	} else
146 #endif
147 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148 		   sizeof(p->p_sysent->sv_psstrings));
149 	return error;
150 }
151 
152 static int
153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
154 {
155 	struct proc *p;
156 	int error;
157 
158 	p = curproc;
159 #ifdef SCTL_MASK32
160 	if (req->flags & SCTL_MASK32) {
161 		unsigned int val;
162 		val = (unsigned int)p->p_sysent->sv_usrstack;
163 		error = SYSCTL_OUT(req, &val, sizeof(val));
164 	} else
165 #endif
166 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167 		    sizeof(p->p_sysent->sv_usrstack));
168 	return error;
169 }
170 
171 static int
172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
173 {
174 	struct proc *p;
175 
176 	p = curproc;
177 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
178 	    sizeof(p->p_sysent->sv_stackprot)));
179 }
180 
181 /*
182  * Each of the items is a pointer to a `const struct execsw', hence the
183  * double pointer here.
184  */
185 static const struct execsw **execsw;
186 
187 #ifndef _SYS_SYSPROTO_H_
188 struct execve_args {
189 	char    *fname;
190 	char    **argv;
191 	char    **envv;
192 };
193 #endif
194 
195 int
196 sys_execve(td, uap)
197 	struct thread *td;
198 	struct execve_args /* {
199 		char *fname;
200 		char **argv;
201 		char **envv;
202 	} */ *uap;
203 {
204 	int error;
205 	struct image_args args;
206 
207 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
208 	    uap->argv, uap->envv);
209 	if (error == 0)
210 		error = kern_execve(td, &args, NULL);
211 	return (error);
212 }
213 
214 #ifndef _SYS_SYSPROTO_H_
215 struct fexecve_args {
216 	int	fd;
217 	char	**argv;
218 	char	**envv;
219 }
220 #endif
221 int
222 sys_fexecve(struct thread *td, struct fexecve_args *uap)
223 {
224 	int error;
225 	struct image_args args;
226 
227 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
228 	    uap->argv, uap->envv);
229 	if (error == 0) {
230 		args.fd = uap->fd;
231 		error = kern_execve(td, &args, NULL);
232 	}
233 	return (error);
234 }
235 
236 #ifndef _SYS_SYSPROTO_H_
237 struct __mac_execve_args {
238 	char	*fname;
239 	char	**argv;
240 	char	**envv;
241 	struct mac	*mac_p;
242 };
243 #endif
244 
245 int
246 sys___mac_execve(td, uap)
247 	struct thread *td;
248 	struct __mac_execve_args /* {
249 		char *fname;
250 		char **argv;
251 		char **envv;
252 		struct mac *mac_p;
253 	} */ *uap;
254 {
255 #ifdef MAC
256 	int error;
257 	struct image_args args;
258 
259 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
260 	    uap->argv, uap->envv);
261 	if (error == 0)
262 		error = kern_execve(td, &args, uap->mac_p);
263 	return (error);
264 #else
265 	return (ENOSYS);
266 #endif
267 }
268 
269 /*
270  * XXX: kern_execve has the astonishing property of not always returning to
271  * the caller.  If sufficiently bad things happen during the call to
272  * do_execve(), it can end up calling exit1(); as a result, callers must
273  * avoid doing anything which they might need to undo (e.g., allocating
274  * memory).
275  */
276 int
277 kern_execve(td, args, mac_p)
278 	struct thread *td;
279 	struct image_args *args;
280 	struct mac *mac_p;
281 {
282 	struct proc *p = td->td_proc;
283 	struct vmspace *oldvmspace;
284 	int error;
285 
286 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
287 	    args->begin_envv - args->begin_argv);
288 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
289 	    args->endp - args->begin_envv);
290 	if (p->p_flag & P_HADTHREADS) {
291 		PROC_LOCK(p);
292 		if (thread_single(p, SINGLE_BOUNDARY)) {
293 			PROC_UNLOCK(p);
294 	       		exec_free_args(args);
295 			return (ERESTART);	/* Try again later. */
296 		}
297 		PROC_UNLOCK(p);
298 	}
299 
300 	KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
301 	oldvmspace = td->td_proc->p_vmspace;
302 	error = do_execve(td, args, mac_p);
303 
304 	if (p->p_flag & P_HADTHREADS) {
305 		PROC_LOCK(p);
306 		/*
307 		 * If success, we upgrade to SINGLE_EXIT state to
308 		 * force other threads to suicide.
309 		 */
310 		if (error == 0)
311 			thread_single(p, SINGLE_EXIT);
312 		else
313 			thread_single_end(p, SINGLE_BOUNDARY);
314 		PROC_UNLOCK(p);
315 	}
316 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
317 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
318 		    ("oldvmspace still used"));
319 		vmspace_free(oldvmspace);
320 		td->td_pflags &= ~TDP_EXECVMSPC;
321 	}
322 
323 	return (error);
324 }
325 
326 /*
327  * In-kernel implementation of execve().  All arguments are assumed to be
328  * userspace pointers from the passed thread.
329  */
330 static int
331 do_execve(td, args, mac_p)
332 	struct thread *td;
333 	struct image_args *args;
334 	struct mac *mac_p;
335 {
336 	struct proc *p = td->td_proc;
337 	struct nameidata nd;
338 	struct ucred *newcred = NULL, *oldcred;
339 	struct uidinfo *euip = NULL;
340 	register_t *stack_base;
341 	int error, i;
342 	struct image_params image_params, *imgp;
343 	struct vattr attr;
344 	int (*img_first)(struct image_params *);
345 	struct pargs *oldargs = NULL, *newargs = NULL;
346 	struct sigacts *oldsigacts, *newsigacts;
347 #ifdef KTRACE
348 	struct vnode *tracevp = NULL;
349 	struct ucred *tracecred = NULL;
350 #endif
351 	struct vnode *textvp = NULL, *binvp;
352 	cap_rights_t rights;
353 	int credential_changing;
354 	int textset;
355 #ifdef MAC
356 	struct label *interpvplabel = NULL;
357 	int will_transition;
358 #endif
359 #ifdef HWPMC_HOOKS
360 	struct pmckern_procexec pe;
361 #endif
362 	static const char fexecv_proc_title[] = "(fexecv)";
363 
364 	imgp = &image_params;
365 
366 	/*
367 	 * Lock the process and set the P_INEXEC flag to indicate that
368 	 * it should be left alone until we're done here.  This is
369 	 * necessary to avoid race conditions - e.g. in ptrace() -
370 	 * that might allow a local user to illicitly obtain elevated
371 	 * privileges.
372 	 */
373 	PROC_LOCK(p);
374 	KASSERT((p->p_flag & P_INEXEC) == 0,
375 	    ("%s(): process already has P_INEXEC flag", __func__));
376 	p->p_flag |= P_INEXEC;
377 	PROC_UNLOCK(p);
378 
379 	/*
380 	 * Initialize part of the common data
381 	 */
382 	bzero(imgp, sizeof(*imgp));
383 	imgp->proc = p;
384 	imgp->attr = &attr;
385 	imgp->args = args;
386 
387 #ifdef MAC
388 	error = mac_execve_enter(imgp, mac_p);
389 	if (error)
390 		goto exec_fail;
391 #endif
392 
393 	/*
394 	 * Translate the file name. namei() returns a vnode pointer
395 	 *	in ni_vp amoung other things.
396 	 *
397 	 * XXXAUDIT: It would be desirable to also audit the name of the
398 	 * interpreter if this is an interpreted binary.
399 	 */
400 	if (args->fname != NULL) {
401 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
402 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
403 	}
404 
405 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
406 
407 interpret:
408 	if (args->fname != NULL) {
409 #ifdef CAPABILITY_MODE
410 		/*
411 		 * While capability mode can't reach this point via direct
412 		 * path arguments to execve(), we also don't allow
413 		 * interpreters to be used in capability mode (for now).
414 		 * Catch indirect lookups and return a permissions error.
415 		 */
416 		if (IN_CAPABILITY_MODE(td)) {
417 			error = ECAPMODE;
418 			goto exec_fail;
419 		}
420 #endif
421 		error = namei(&nd);
422 		if (error)
423 			goto exec_fail;
424 
425 		binvp = nd.ni_vp;
426 		imgp->vp = binvp;
427 	} else {
428 		AUDIT_ARG_FD(args->fd);
429 		/*
430 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
431 		 */
432 		error = fgetvp_exec(td, args->fd,
433 		    cap_rights_init(&rights, CAP_FEXECVE), &binvp);
434 		if (error)
435 			goto exec_fail;
436 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
437 		AUDIT_ARG_VNODE1(binvp);
438 		imgp->vp = binvp;
439 	}
440 
441 	/*
442 	 * Check file permissions (also 'opens' file)
443 	 */
444 	error = exec_check_permissions(imgp);
445 	if (error)
446 		goto exec_fail_dealloc;
447 
448 	imgp->object = imgp->vp->v_object;
449 	if (imgp->object != NULL)
450 		vm_object_reference(imgp->object);
451 
452 	/*
453 	 * Set VV_TEXT now so no one can write to the executable while we're
454 	 * activating it.
455 	 *
456 	 * Remember if this was set before and unset it in case this is not
457 	 * actually an executable image.
458 	 */
459 	textset = VOP_IS_TEXT(imgp->vp);
460 	VOP_SET_TEXT(imgp->vp);
461 
462 	error = exec_map_first_page(imgp);
463 	if (error)
464 		goto exec_fail_dealloc;
465 
466 	imgp->proc->p_osrel = 0;
467 	/*
468 	 *	If the current process has a special image activator it
469 	 *	wants to try first, call it.   For example, emulating shell
470 	 *	scripts differently.
471 	 */
472 	error = -1;
473 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
474 		error = img_first(imgp);
475 
476 	/*
477 	 *	Loop through the list of image activators, calling each one.
478 	 *	An activator returns -1 if there is no match, 0 on success,
479 	 *	and an error otherwise.
480 	 */
481 	for (i = 0; error == -1 && execsw[i]; ++i) {
482 		if (execsw[i]->ex_imgact == NULL ||
483 		    execsw[i]->ex_imgact == img_first) {
484 			continue;
485 		}
486 		error = (*execsw[i]->ex_imgact)(imgp);
487 	}
488 
489 	if (error) {
490 		if (error == -1) {
491 			if (textset == 0)
492 				VOP_UNSET_TEXT(imgp->vp);
493 			error = ENOEXEC;
494 		}
495 		goto exec_fail_dealloc;
496 	}
497 
498 	/*
499 	 * Special interpreter operation, cleanup and loop up to try to
500 	 * activate the interpreter.
501 	 */
502 	if (imgp->interpreted) {
503 		exec_unmap_first_page(imgp);
504 		/*
505 		 * VV_TEXT needs to be unset for scripts.  There is a short
506 		 * period before we determine that something is a script where
507 		 * VV_TEXT will be set. The vnode lock is held over this
508 		 * entire period so nothing should illegitimately be blocked.
509 		 */
510 		VOP_UNSET_TEXT(imgp->vp);
511 		/* free name buffer and old vnode */
512 		if (args->fname != NULL)
513 			NDFREE(&nd, NDF_ONLY_PNBUF);
514 #ifdef MAC
515 		mac_execve_interpreter_enter(binvp, &interpvplabel);
516 #endif
517 		if (imgp->opened) {
518 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
519 			imgp->opened = 0;
520 		}
521 		vput(binvp);
522 		vm_object_deallocate(imgp->object);
523 		imgp->object = NULL;
524 		/* set new name to that of the interpreter */
525 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
526 		    UIO_SYSSPACE, imgp->interpreter_name, td);
527 		args->fname = imgp->interpreter_name;
528 		goto interpret;
529 	}
530 
531 	/*
532 	 * NB: We unlock the vnode here because it is believed that none
533 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
534 	 */
535 	VOP_UNLOCK(imgp->vp, 0);
536 
537 	/*
538 	 * Do the best to calculate the full path to the image file.
539 	 */
540 	if (imgp->auxargs != NULL &&
541 	    ((args->fname != NULL && args->fname[0] == '/') ||
542 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
543 		imgp->execpath = args->fname;
544 
545 	if (disallow_high_osrel &&
546 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
547 		error = ENOEXEC;
548 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
549 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
550 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
551 		goto exec_fail_dealloc;
552 	}
553 
554 	/*
555 	 * Copy out strings (args and env) and initialize stack base
556 	 */
557 	if (p->p_sysent->sv_copyout_strings)
558 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
559 	else
560 		stack_base = exec_copyout_strings(imgp);
561 
562 	/*
563 	 * If custom stack fixup routine present for this process
564 	 * let it do the stack setup.
565 	 * Else stuff argument count as first item on stack
566 	 */
567 	if (p->p_sysent->sv_fixup != NULL)
568 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
569 	else
570 		suword(--stack_base, imgp->args->argc);
571 
572 	/*
573 	 * For security and other reasons, the file descriptor table cannot
574 	 * be shared after an exec.
575 	 */
576 	fdunshare(td);
577 	/* close files on exec */
578 	fdcloseexec(td);
579 
580 	/*
581 	 * Malloc things before we need locks.
582 	 */
583 	i = imgp->args->begin_envv - imgp->args->begin_argv;
584 	/* Cache arguments if they fit inside our allowance */
585 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
586 		newargs = pargs_alloc(i);
587 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
588 	}
589 
590 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
591 
592 	/* Get a reference to the vnode prior to locking the proc */
593 	VREF(binvp);
594 
595 	/*
596 	 * For security and other reasons, signal handlers cannot
597 	 * be shared after an exec. The new process gets a copy of the old
598 	 * handlers. In execsigs(), the new process will have its signals
599 	 * reset.
600 	 */
601 	if (sigacts_shared(p->p_sigacts)) {
602 		oldsigacts = p->p_sigacts;
603 		newsigacts = sigacts_alloc();
604 		sigacts_copy(newsigacts, oldsigacts);
605 	} else {
606 		oldsigacts = NULL;
607 		newsigacts = NULL; /* satisfy gcc */
608 	}
609 
610 	PROC_LOCK(p);
611 	if (oldsigacts)
612 		p->p_sigacts = newsigacts;
613 	oldcred = p->p_ucred;
614 	/* Stop profiling */
615 	stopprofclock(p);
616 
617 	/* reset caught signals */
618 	execsigs(p);
619 
620 	/* name this process - nameiexec(p, ndp) */
621 	bzero(p->p_comm, sizeof(p->p_comm));
622 	if (args->fname)
623 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
624 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
625 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
626 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
627 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
628 #ifdef KTR
629 	sched_clear_tdname(td);
630 #endif
631 
632 	/*
633 	 * mark as execed, wakeup the process that vforked (if any) and tell
634 	 * it that it now has its own resources back
635 	 */
636 	p->p_flag |= P_EXEC;
637 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
638 		p->p_flag2 &= ~P2_NOTRACE;
639 	if (p->p_flag & P_PPWAIT) {
640 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
641 		cv_broadcast(&p->p_pwait);
642 	}
643 
644 	/*
645 	 * Implement image setuid/setgid.
646 	 *
647 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
648 	 * the process is being traced.
649 	 *
650 	 * We disable setuid/setgid/etc in compatibility mode on the basis
651 	 * that most setugid applications are not written with that
652 	 * environment in mind, and will therefore almost certainly operate
653 	 * incorrectly. In principle there's no reason that setugid
654 	 * applications might not be useful in capability mode, so we may want
655 	 * to reconsider this conservative design choice in the future.
656 	 *
657 	 * XXXMAC: For the time being, use NOSUID to also prohibit
658 	 * transitions on the file system.
659 	 */
660 	credential_changing = 0;
661 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
662 	    attr.va_uid;
663 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
664 	    attr.va_gid;
665 #ifdef MAC
666 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
667 	    interpvplabel, imgp);
668 	credential_changing |= will_transition;
669 #endif
670 
671 	if (credential_changing &&
672 #ifdef CAPABILITY_MODE
673 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
674 #endif
675 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
676 	    (p->p_flag & P_TRACED) == 0) {
677 		/*
678 		 * Turn off syscall tracing for set-id programs, except for
679 		 * root.  Record any set-id flags first to make sure that
680 		 * we do not regain any tracing during a possible block.
681 		 */
682 		setsugid(p);
683 
684 #ifdef KTRACE
685 		if (p->p_tracecred != NULL &&
686 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
687 			ktrprocexec(p, &tracecred, &tracevp);
688 #endif
689 		/*
690 		 * Close any file descriptors 0..2 that reference procfs,
691 		 * then make sure file descriptors 0..2 are in use.
692 		 *
693 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
694 		 * taking sleepable locks, so temporarily drop our locks.
695 		 */
696 		PROC_UNLOCK(p);
697 		VOP_UNLOCK(imgp->vp, 0);
698 		fdsetugidsafety(td);
699 		error = fdcheckstd(td);
700 		if (error != 0)
701 			goto done1;
702 		newcred = crdup(oldcred);
703 		euip = uifind(attr.va_uid);
704 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
705 		PROC_LOCK(p);
706 		/*
707 		 * Set the new credentials.
708 		 */
709 		if (attr.va_mode & S_ISUID)
710 			change_euid(newcred, euip);
711 		if (attr.va_mode & S_ISGID)
712 			change_egid(newcred, attr.va_gid);
713 #ifdef MAC
714 		if (will_transition) {
715 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
716 			    interpvplabel, imgp);
717 		}
718 #endif
719 		/*
720 		 * Implement correct POSIX saved-id behavior.
721 		 *
722 		 * XXXMAC: Note that the current logic will save the
723 		 * uid and gid if a MAC domain transition occurs, even
724 		 * though maybe it shouldn't.
725 		 */
726 		change_svuid(newcred, newcred->cr_uid);
727 		change_svgid(newcred, newcred->cr_gid);
728 		proc_set_cred(p, newcred);
729 	} else {
730 		if (oldcred->cr_uid == oldcred->cr_ruid &&
731 		    oldcred->cr_gid == oldcred->cr_rgid)
732 			p->p_flag &= ~P_SUGID;
733 		/*
734 		 * Implement correct POSIX saved-id behavior.
735 		 *
736 		 * XXX: It's not clear that the existing behavior is
737 		 * POSIX-compliant.  A number of sources indicate that the
738 		 * saved uid/gid should only be updated if the new ruid is
739 		 * not equal to the old ruid, or the new euid is not equal
740 		 * to the old euid and the new euid is not equal to the old
741 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
742 		 * Also, this code uses the new (replaced) euid and egid as
743 		 * the source, which may or may not be the right ones to use.
744 		 */
745 		if (oldcred->cr_svuid != oldcred->cr_uid ||
746 		    oldcred->cr_svgid != oldcred->cr_gid) {
747 			PROC_UNLOCK(p);
748 			VOP_UNLOCK(imgp->vp, 0);
749 			newcred = crdup(oldcred);
750 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
751 			PROC_LOCK(p);
752 			change_svuid(newcred, newcred->cr_uid);
753 			change_svgid(newcred, newcred->cr_gid);
754 			proc_set_cred(p, newcred);
755 		}
756 	}
757 
758 	/*
759 	 * Store the vp for use in procfs.  This vnode was referenced prior
760 	 * to locking the proc lock.
761 	 */
762 	textvp = p->p_textvp;
763 	p->p_textvp = binvp;
764 
765 #ifdef KDTRACE_HOOKS
766 	/*
767 	 * Tell the DTrace fasttrap provider about the exec if it
768 	 * has declared an interest.
769 	 */
770 	if (dtrace_fasttrap_exec)
771 		dtrace_fasttrap_exec(p);
772 #endif
773 
774 	/*
775 	 * Notify others that we exec'd, and clear the P_INEXEC flag
776 	 * as we're now a bona fide freshly-execed process.
777 	 */
778 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
779 	p->p_flag &= ~P_INEXEC;
780 
781 	/* clear "fork but no exec" flag, as we _are_ execing */
782 	p->p_acflag &= ~AFORK;
783 
784 	/*
785 	 * Free any previous argument cache and replace it with
786 	 * the new argument cache, if any.
787 	 */
788 	oldargs = p->p_args;
789 	p->p_args = newargs;
790 	newargs = NULL;
791 
792 #ifdef	HWPMC_HOOKS
793 	/*
794 	 * Check if system-wide sampling is in effect or if the
795 	 * current process is using PMCs.  If so, do exec() time
796 	 * processing.  This processing needs to happen AFTER the
797 	 * P_INEXEC flag is cleared.
798 	 *
799 	 * The proc lock needs to be released before taking the PMC
800 	 * SX.
801 	 */
802 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
803 		PROC_UNLOCK(p);
804 		VOP_UNLOCK(imgp->vp, 0);
805 		pe.pm_credentialschanged = credential_changing;
806 		pe.pm_entryaddr = imgp->entry_addr;
807 
808 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
809 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
810 	} else
811 		PROC_UNLOCK(p);
812 #else  /* !HWPMC_HOOKS */
813 	PROC_UNLOCK(p);
814 #endif
815 
816 	/* Set values passed into the program in registers. */
817 	if (p->p_sysent->sv_setregs)
818 		(*p->p_sysent->sv_setregs)(td, imgp,
819 		    (u_long)(uintptr_t)stack_base);
820 	else
821 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
822 
823 	vfs_mark_atime(imgp->vp, td->td_ucred);
824 
825 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
826 
827 	VOP_UNLOCK(imgp->vp, 0);
828 done1:
829 	/*
830 	 * Free any resources malloc'd earlier that we didn't use.
831 	 */
832 	if (euip != NULL)
833 		uifree(euip);
834 	if (newcred != NULL)
835 		crfree(oldcred);
836 
837 	/*
838 	 * Handle deferred decrement of ref counts.
839 	 */
840 	if (textvp != NULL)
841 		vrele(textvp);
842 	if (error != 0)
843 		vrele(binvp);
844 #ifdef KTRACE
845 	if (tracevp != NULL)
846 		vrele(tracevp);
847 	if (tracecred != NULL)
848 		crfree(tracecred);
849 #endif
850 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
851 	pargs_drop(oldargs);
852 	pargs_drop(newargs);
853 	if (oldsigacts != NULL)
854 		sigacts_free(oldsigacts);
855 
856 exec_fail_dealloc:
857 
858 	/*
859 	 * free various allocated resources
860 	 */
861 	if (imgp->firstpage != NULL)
862 		exec_unmap_first_page(imgp);
863 
864 	if (imgp->vp != NULL) {
865 		if (args->fname)
866 			NDFREE(&nd, NDF_ONLY_PNBUF);
867 		if (imgp->opened)
868 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
869 		vput(imgp->vp);
870 	}
871 
872 	if (imgp->object != NULL)
873 		vm_object_deallocate(imgp->object);
874 
875 	free(imgp->freepath, M_TEMP);
876 
877 	if (error == 0) {
878 		PROC_LOCK(p);
879 		td->td_dbgflags |= TDB_EXEC;
880 		PROC_UNLOCK(p);
881 
882 		/*
883 		 * Stop the process here if its stop event mask has
884 		 * the S_EXEC bit set.
885 		 */
886 		STOPEVENT(p, S_EXEC, 0);
887 		goto done2;
888 	}
889 
890 exec_fail:
891 	/* we're done here, clear P_INEXEC */
892 	PROC_LOCK(p);
893 	p->p_flag &= ~P_INEXEC;
894 	PROC_UNLOCK(p);
895 
896 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
897 
898 done2:
899 #ifdef MAC
900 	mac_execve_exit(imgp);
901 	mac_execve_interpreter_exit(interpvplabel);
902 #endif
903 	exec_free_args(args);
904 
905 	if (error && imgp->vmspace_destroyed) {
906 		/* sorry, no more process anymore. exit gracefully */
907 		exit1(td, W_EXITCODE(0, SIGABRT));
908 		/* NOT REACHED */
909 	}
910 
911 #ifdef KTRACE
912 	if (error == 0)
913 		ktrprocctor(p);
914 #endif
915 
916 	return (error);
917 }
918 
919 int
920 exec_map_first_page(imgp)
921 	struct image_params *imgp;
922 {
923 	int rv, i;
924 	int initial_pagein;
925 	vm_page_t ma[VM_INITIAL_PAGEIN];
926 	vm_object_t object;
927 
928 	if (imgp->firstpage != NULL)
929 		exec_unmap_first_page(imgp);
930 
931 	object = imgp->vp->v_object;
932 	if (object == NULL)
933 		return (EACCES);
934 	VM_OBJECT_WLOCK(object);
935 #if VM_NRESERVLEVEL > 0
936 	vm_object_color(object, 0);
937 #endif
938 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
939 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
940 		initial_pagein = VM_INITIAL_PAGEIN;
941 		if (initial_pagein > object->size)
942 			initial_pagein = object->size;
943 		for (i = 1; i < initial_pagein; i++) {
944 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
945 				if (ma[i]->valid)
946 					break;
947 				if (vm_page_tryxbusy(ma[i]))
948 					break;
949 			} else {
950 				ma[i] = vm_page_alloc(object, i,
951 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
952 				if (ma[i] == NULL)
953 					break;
954 			}
955 		}
956 		initial_pagein = i;
957 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
958 		ma[0] = vm_page_lookup(object, 0);
959 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
960 			if (ma[0] != NULL) {
961 				vm_page_lock(ma[0]);
962 				vm_page_free(ma[0]);
963 				vm_page_unlock(ma[0]);
964 			}
965 			VM_OBJECT_WUNLOCK(object);
966 			return (EIO);
967 		}
968 	}
969 	vm_page_xunbusy(ma[0]);
970 	vm_page_lock(ma[0]);
971 	vm_page_hold(ma[0]);
972 	vm_page_activate(ma[0]);
973 	vm_page_unlock(ma[0]);
974 	VM_OBJECT_WUNLOCK(object);
975 
976 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
977 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
978 
979 	return (0);
980 }
981 
982 void
983 exec_unmap_first_page(imgp)
984 	struct image_params *imgp;
985 {
986 	vm_page_t m;
987 
988 	if (imgp->firstpage != NULL) {
989 		m = sf_buf_page(imgp->firstpage);
990 		sf_buf_free(imgp->firstpage);
991 		imgp->firstpage = NULL;
992 		vm_page_lock(m);
993 		vm_page_unhold(m);
994 		vm_page_unlock(m);
995 	}
996 }
997 
998 /*
999  * Destroy old address space, and allocate a new stack
1000  *	The new stack is only SGROWSIZ large because it is grown
1001  *	automatically in trap.c.
1002  */
1003 int
1004 exec_new_vmspace(imgp, sv)
1005 	struct image_params *imgp;
1006 	struct sysentvec *sv;
1007 {
1008 	int error;
1009 	struct proc *p = imgp->proc;
1010 	struct vmspace *vmspace = p->p_vmspace;
1011 	vm_object_t obj;
1012 	struct rlimit rlim_stack;
1013 	vm_offset_t sv_minuser, stack_addr;
1014 	vm_map_t map;
1015 	u_long ssiz;
1016 
1017 	imgp->vmspace_destroyed = 1;
1018 	imgp->sysent = sv;
1019 
1020 	/* May be called with Giant held */
1021 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1022 
1023 	/*
1024 	 * Blow away entire process VM, if address space not shared,
1025 	 * otherwise, create a new VM space so that other threads are
1026 	 * not disrupted
1027 	 */
1028 	map = &vmspace->vm_map;
1029 	if (map_at_zero)
1030 		sv_minuser = sv->sv_minuser;
1031 	else
1032 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1033 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1034 	    vm_map_max(map) == sv->sv_maxuser) {
1035 		shmexit(vmspace);
1036 		pmap_remove_pages(vmspace_pmap(vmspace));
1037 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1038 	} else {
1039 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1040 		if (error)
1041 			return (error);
1042 		vmspace = p->p_vmspace;
1043 		map = &vmspace->vm_map;
1044 	}
1045 
1046 	/* Map a shared page */
1047 	obj = sv->sv_shared_page_obj;
1048 	if (obj != NULL) {
1049 		vm_object_reference(obj);
1050 		error = vm_map_fixed(map, obj, 0,
1051 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1052 		    VM_PROT_READ | VM_PROT_EXECUTE,
1053 		    VM_PROT_READ | VM_PROT_EXECUTE,
1054 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1055 		if (error) {
1056 			vm_object_deallocate(obj);
1057 			return (error);
1058 		}
1059 	}
1060 
1061 	/* Allocate a new stack */
1062 	if (imgp->stack_sz != 0) {
1063 		ssiz = imgp->stack_sz;
1064 		PROC_LOCK(p);
1065 		lim_rlimit(p, RLIMIT_STACK, &rlim_stack);
1066 		PROC_UNLOCK(p);
1067 		if (ssiz > rlim_stack.rlim_max)
1068 			ssiz = rlim_stack.rlim_max;
1069 		if (ssiz > rlim_stack.rlim_cur) {
1070 			rlim_stack.rlim_cur = ssiz;
1071 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1072 		}
1073 	} else if (sv->sv_maxssiz != NULL) {
1074 		ssiz = *sv->sv_maxssiz;
1075 	} else {
1076 		ssiz = maxssiz;
1077 	}
1078 	stack_addr = sv->sv_usrstack - ssiz;
1079 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1080 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1081 		sv->sv_stackprot,
1082 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1083 	if (error)
1084 		return (error);
1085 
1086 	/*
1087 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1088 	 * are still used to enforce the stack rlimit on the process stack.
1089 	 */
1090 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1091 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1092 
1093 	return (0);
1094 }
1095 
1096 /*
1097  * Copy out argument and environment strings from the old process address
1098  * space into the temporary string buffer.
1099  */
1100 int
1101 exec_copyin_args(struct image_args *args, char *fname,
1102     enum uio_seg segflg, char **argv, char **envv)
1103 {
1104 	u_long argp, envp;
1105 	int error;
1106 	size_t length;
1107 
1108 	bzero(args, sizeof(*args));
1109 	if (argv == NULL)
1110 		return (EFAULT);
1111 
1112 	/*
1113 	 * Allocate demand-paged memory for the file name, argument, and
1114 	 * environment strings.
1115 	 */
1116 	error = exec_alloc_args(args);
1117 	if (error != 0)
1118 		return (error);
1119 
1120 	/*
1121 	 * Copy the file name.
1122 	 */
1123 	if (fname != NULL) {
1124 		args->fname = args->buf;
1125 		error = (segflg == UIO_SYSSPACE) ?
1126 		    copystr(fname, args->fname, PATH_MAX, &length) :
1127 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1128 		if (error != 0)
1129 			goto err_exit;
1130 	} else
1131 		length = 0;
1132 
1133 	args->begin_argv = args->buf + length;
1134 	args->endp = args->begin_argv;
1135 	args->stringspace = ARG_MAX;
1136 
1137 	/*
1138 	 * extract arguments first
1139 	 */
1140 	for (;;) {
1141 		error = fueword(argv++, &argp);
1142 		if (error == -1) {
1143 			error = EFAULT;
1144 			goto err_exit;
1145 		}
1146 		if (argp == 0)
1147 			break;
1148 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1149 		    args->stringspace, &length);
1150 		if (error != 0) {
1151 			if (error == ENAMETOOLONG)
1152 				error = E2BIG;
1153 			goto err_exit;
1154 		}
1155 		args->stringspace -= length;
1156 		args->endp += length;
1157 		args->argc++;
1158 	}
1159 
1160 	args->begin_envv = args->endp;
1161 
1162 	/*
1163 	 * extract environment strings
1164 	 */
1165 	if (envv) {
1166 		for (;;) {
1167 			error = fueword(envv++, &envp);
1168 			if (error == -1) {
1169 				error = EFAULT;
1170 				goto err_exit;
1171 			}
1172 			if (envp == 0)
1173 				break;
1174 			error = copyinstr((void *)(uintptr_t)envp,
1175 			    args->endp, args->stringspace, &length);
1176 			if (error != 0) {
1177 				if (error == ENAMETOOLONG)
1178 					error = E2BIG;
1179 				goto err_exit;
1180 			}
1181 			args->stringspace -= length;
1182 			args->endp += length;
1183 			args->envc++;
1184 		}
1185 	}
1186 
1187 	return (0);
1188 
1189 err_exit:
1190 	exec_free_args(args);
1191 	return (error);
1192 }
1193 
1194 /*
1195  * Allocate temporary demand-paged, zero-filled memory for the file name,
1196  * argument, and environment strings.  Returns zero if the allocation succeeds
1197  * and ENOMEM otherwise.
1198  */
1199 int
1200 exec_alloc_args(struct image_args *args)
1201 {
1202 
1203 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1204 	return (args->buf != NULL ? 0 : ENOMEM);
1205 }
1206 
1207 void
1208 exec_free_args(struct image_args *args)
1209 {
1210 
1211 	if (args->buf != NULL) {
1212 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1213 		    PATH_MAX + ARG_MAX);
1214 		args->buf = NULL;
1215 	}
1216 	if (args->fname_buf != NULL) {
1217 		free(args->fname_buf, M_TEMP);
1218 		args->fname_buf = NULL;
1219 	}
1220 }
1221 
1222 /*
1223  * Copy strings out to the new process address space, constructing new arg
1224  * and env vector tables. Return a pointer to the base so that it can be used
1225  * as the initial stack pointer.
1226  */
1227 register_t *
1228 exec_copyout_strings(imgp)
1229 	struct image_params *imgp;
1230 {
1231 	int argc, envc;
1232 	char **vectp;
1233 	char *stringp;
1234 	uintptr_t destp;
1235 	register_t *stack_base;
1236 	struct ps_strings *arginfo;
1237 	struct proc *p;
1238 	size_t execpath_len;
1239 	int szsigcode, szps;
1240 	char canary[sizeof(long) * 8];
1241 
1242 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1243 	/*
1244 	 * Calculate string base and vector table pointers.
1245 	 * Also deal with signal trampoline code for this exec type.
1246 	 */
1247 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1248 		execpath_len = strlen(imgp->execpath) + 1;
1249 	else
1250 		execpath_len = 0;
1251 	p = imgp->proc;
1252 	szsigcode = 0;
1253 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1254 	if (p->p_sysent->sv_sigcode_base == 0) {
1255 		if (p->p_sysent->sv_szsigcode != NULL)
1256 			szsigcode = *(p->p_sysent->sv_szsigcode);
1257 	}
1258 	destp =	(uintptr_t)arginfo;
1259 
1260 	/*
1261 	 * install sigcode
1262 	 */
1263 	if (szsigcode != 0) {
1264 		destp -= szsigcode;
1265 		destp = rounddown2(destp, sizeof(void *));
1266 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1267 	}
1268 
1269 	/*
1270 	 * Copy the image path for the rtld.
1271 	 */
1272 	if (execpath_len != 0) {
1273 		destp -= execpath_len;
1274 		imgp->execpathp = destp;
1275 		copyout(imgp->execpath, (void *)destp, execpath_len);
1276 	}
1277 
1278 	/*
1279 	 * Prepare the canary for SSP.
1280 	 */
1281 	arc4rand(canary, sizeof(canary), 0);
1282 	destp -= sizeof(canary);
1283 	imgp->canary = destp;
1284 	copyout(canary, (void *)destp, sizeof(canary));
1285 	imgp->canarylen = sizeof(canary);
1286 
1287 	/*
1288 	 * Prepare the pagesizes array.
1289 	 */
1290 	destp -= szps;
1291 	destp = rounddown2(destp, sizeof(void *));
1292 	imgp->pagesizes = destp;
1293 	copyout(pagesizes, (void *)destp, szps);
1294 	imgp->pagesizeslen = szps;
1295 
1296 	destp -= ARG_MAX - imgp->args->stringspace;
1297 	destp = rounddown2(destp, sizeof(void *));
1298 
1299 	/*
1300 	 * If we have a valid auxargs ptr, prepare some room
1301 	 * on the stack.
1302 	 */
1303 	if (imgp->auxargs) {
1304 		/*
1305 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1306 		 * lower compatibility.
1307 		 */
1308 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1309 		    (AT_COUNT * 2);
1310 		/*
1311 		 * The '+ 2' is for the null pointers at the end of each of
1312 		 * the arg and env vector sets,and imgp->auxarg_size is room
1313 		 * for argument of Runtime loader.
1314 		 */
1315 		vectp = (char **)(destp - (imgp->args->argc +
1316 		    imgp->args->envc + 2 + imgp->auxarg_size)
1317 		    * sizeof(char *));
1318 	} else {
1319 		/*
1320 		 * The '+ 2' is for the null pointers at the end of each of
1321 		 * the arg and env vector sets
1322 		 */
1323 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1324 		    + 2) * sizeof(char *));
1325 	}
1326 
1327 	/*
1328 	 * vectp also becomes our initial stack base
1329 	 */
1330 	stack_base = (register_t *)vectp;
1331 
1332 	stringp = imgp->args->begin_argv;
1333 	argc = imgp->args->argc;
1334 	envc = imgp->args->envc;
1335 
1336 	/*
1337 	 * Copy out strings - arguments and environment.
1338 	 */
1339 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1340 
1341 	/*
1342 	 * Fill in "ps_strings" struct for ps, w, etc.
1343 	 */
1344 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1345 	suword32(&arginfo->ps_nargvstr, argc);
1346 
1347 	/*
1348 	 * Fill in argument portion of vector table.
1349 	 */
1350 	for (; argc > 0; --argc) {
1351 		suword(vectp++, (long)(intptr_t)destp);
1352 		while (*stringp++ != 0)
1353 			destp++;
1354 		destp++;
1355 	}
1356 
1357 	/* a null vector table pointer separates the argp's from the envp's */
1358 	suword(vectp++, 0);
1359 
1360 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1361 	suword32(&arginfo->ps_nenvstr, envc);
1362 
1363 	/*
1364 	 * Fill in environment portion of vector table.
1365 	 */
1366 	for (; envc > 0; --envc) {
1367 		suword(vectp++, (long)(intptr_t)destp);
1368 		while (*stringp++ != 0)
1369 			destp++;
1370 		destp++;
1371 	}
1372 
1373 	/* end of vector table is a null pointer */
1374 	suword(vectp, 0);
1375 
1376 	return (stack_base);
1377 }
1378 
1379 /*
1380  * Check permissions of file to execute.
1381  *	Called with imgp->vp locked.
1382  *	Return 0 for success or error code on failure.
1383  */
1384 int
1385 exec_check_permissions(imgp)
1386 	struct image_params *imgp;
1387 {
1388 	struct vnode *vp = imgp->vp;
1389 	struct vattr *attr = imgp->attr;
1390 	struct thread *td;
1391 	int error, writecount;
1392 
1393 	td = curthread;
1394 
1395 	/* Get file attributes */
1396 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1397 	if (error)
1398 		return (error);
1399 
1400 #ifdef MAC
1401 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1402 	if (error)
1403 		return (error);
1404 #endif
1405 
1406 	/*
1407 	 * 1) Check if file execution is disabled for the filesystem that
1408 	 *    this file resides on.
1409 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1410 	 *    privileged user will always succeed, and we don't want this
1411 	 *    to happen unless the file really is executable.
1412 	 * 3) Ensure that the file is a regular file.
1413 	 */
1414 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1415 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1416 	    (attr->va_type != VREG))
1417 		return (EACCES);
1418 
1419 	/*
1420 	 * Zero length files can't be exec'd
1421 	 */
1422 	if (attr->va_size == 0)
1423 		return (ENOEXEC);
1424 
1425 	/*
1426 	 *  Check for execute permission to file based on current credentials.
1427 	 */
1428 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1429 	if (error)
1430 		return (error);
1431 
1432 	/*
1433 	 * Check number of open-for-writes on the file and deny execution
1434 	 * if there are any.
1435 	 */
1436 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1437 	if (error != 0)
1438 		return (error);
1439 	if (writecount != 0)
1440 		return (ETXTBSY);
1441 
1442 	/*
1443 	 * Call filesystem specific open routine (which does nothing in the
1444 	 * general case).
1445 	 */
1446 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1447 	if (error == 0)
1448 		imgp->opened = 1;
1449 	return (error);
1450 }
1451 
1452 /*
1453  * Exec handler registration
1454  */
1455 int
1456 exec_register(execsw_arg)
1457 	const struct execsw *execsw_arg;
1458 {
1459 	const struct execsw **es, **xs, **newexecsw;
1460 	int count = 2;	/* New slot and trailing NULL */
1461 
1462 	if (execsw)
1463 		for (es = execsw; *es; es++)
1464 			count++;
1465 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1466 	if (newexecsw == NULL)
1467 		return (ENOMEM);
1468 	xs = newexecsw;
1469 	if (execsw)
1470 		for (es = execsw; *es; es++)
1471 			*xs++ = *es;
1472 	*xs++ = execsw_arg;
1473 	*xs = NULL;
1474 	if (execsw)
1475 		free(execsw, M_TEMP);
1476 	execsw = newexecsw;
1477 	return (0);
1478 }
1479 
1480 int
1481 exec_unregister(execsw_arg)
1482 	const struct execsw *execsw_arg;
1483 {
1484 	const struct execsw **es, **xs, **newexecsw;
1485 	int count = 1;
1486 
1487 	if (execsw == NULL)
1488 		panic("unregister with no handlers left?\n");
1489 
1490 	for (es = execsw; *es; es++) {
1491 		if (*es == execsw_arg)
1492 			break;
1493 	}
1494 	if (*es == NULL)
1495 		return (ENOENT);
1496 	for (es = execsw; *es; es++)
1497 		if (*es != execsw_arg)
1498 			count++;
1499 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1500 	if (newexecsw == NULL)
1501 		return (ENOMEM);
1502 	xs = newexecsw;
1503 	for (es = execsw; *es; es++)
1504 		if (*es != execsw_arg)
1505 			*xs++ = *es;
1506 	*xs = NULL;
1507 	if (execsw)
1508 		free(execsw, M_TEMP);
1509 	execsw = newexecsw;
1510 	return (0);
1511 }
1512