xref: /freebsd/sys/kern/kern_exec.c (revision 63f9a4cb2684a303e3eb2ffed39c03a2e2b28ae0)
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_ktrace.h"
31 #include "opt_mac.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/eventhandler.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/sysproto.h>
39 #include <sys/signalvar.h>
40 #include <sys/kernel.h>
41 #include <sys/mac.h>
42 #include <sys/mount.h>
43 #include <sys/filedesc.h>
44 #include <sys/fcntl.h>
45 #include <sys/acct.h>
46 #include <sys/exec.h>
47 #include <sys/imgact.h>
48 #include <sys/imgact_elf.h>
49 #include <sys/wait.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 #include <sys/pioctl.h>
53 #include <sys/namei.h>
54 #include <sys/resourcevar.h>
55 #include <sys/sf_buf.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysent.h>
58 #include <sys/shm.h>
59 #include <sys/sysctl.h>
60 #include <sys/vnode.h>
61 #ifdef KTRACE
62 #include <sys/ktrace.h>
63 #endif
64 
65 #include <vm/vm.h>
66 #include <vm/vm_param.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_kern.h>
71 #include <vm/vm_extern.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_pager.h>
74 
75 #include <machine/reg.h>
76 
77 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
78 
79 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
80 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
81 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
82 static int do_execve(struct thread *td, char *fname, char **argv,
83 	char **envv, struct mac *mac_p);
84 
85 /* XXX This should be vm_size_t. */
86 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
87     NULL, 0, sysctl_kern_ps_strings, "LU", "");
88 
89 /* XXX This should be vm_size_t. */
90 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
91     NULL, 0, sysctl_kern_usrstack, "LU", "");
92 
93 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
94     NULL, 0, sysctl_kern_stackprot, "I", "");
95 
96 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
97 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
98     &ps_arg_cache_limit, 0, "");
99 
100 static int
101 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
102 {
103 	struct proc *p;
104 	int error;
105 
106 	p = curproc;
107 #ifdef SCTL_MASK32
108 	if (req->flags & SCTL_MASK32) {
109 		unsigned int val;
110 		val = (unsigned int)p->p_sysent->sv_psstrings;
111 		error = SYSCTL_OUT(req, &val, sizeof(val));
112 	} else
113 #endif
114 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
115 		   sizeof(p->p_sysent->sv_psstrings));
116 	return error;
117 }
118 
119 static int
120 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
121 {
122 	struct proc *p;
123 	int error;
124 
125 	p = curproc;
126 #ifdef SCTL_MASK32
127 	if (req->flags & SCTL_MASK32) {
128 		unsigned int val;
129 		val = (unsigned int)p->p_sysent->sv_usrstack;
130 		error = SYSCTL_OUT(req, &val, sizeof(val));
131 	} else
132 #endif
133 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
134 		    sizeof(p->p_sysent->sv_usrstack));
135 	return error;
136 }
137 
138 static int
139 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
140 {
141 	struct proc *p;
142 
143 	p = curproc;
144 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
145 	    sizeof(p->p_sysent->sv_stackprot)));
146 }
147 
148 /*
149  * Each of the items is a pointer to a `const struct execsw', hence the
150  * double pointer here.
151  */
152 static const struct execsw **execsw;
153 
154 #ifndef _SYS_SYSPROTO_H_
155 struct execve_args {
156 	char    *fname;
157 	char    **argv;
158 	char    **envv;
159 };
160 #endif
161 
162 /*
163  * MPSAFE
164  */
165 int
166 execve(td, uap)
167 	struct thread *td;
168 	struct execve_args /* {
169 		char *fname;
170 		char **argv;
171 		char **envv;
172 	} */ *uap;
173 {
174 
175 	return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
176 }
177 
178 #ifndef _SYS_SYSPROTO_H_
179 struct __mac_execve_args {
180 	char	*fname;
181 	char	**argv;
182 	char	**envv;
183 	struct mac	*mac_p;
184 };
185 #endif
186 
187 /*
188  * MPSAFE
189  */
190 int
191 __mac_execve(td, uap)
192 	struct thread *td;
193 	struct __mac_execve_args /* {
194 		char *fname;
195 		char **argv;
196 		char **envv;
197 		struct mac *mac_p;
198 	} */ *uap;
199 {
200 
201 #ifdef MAC
202 	return (kern_execve(td, uap->fname, uap->argv, uap->envv,
203 	    uap->mac_p));
204 #else
205 	return (ENOSYS);
206 #endif
207 }
208 
209 int
210 kern_execve(td, fname, argv, envv, mac_p)
211 	struct thread *td;
212 	char *fname;
213 	char **argv;
214 	char **envv;
215 	struct mac *mac_p;
216 {
217 	struct proc *p = td->td_proc;
218 	int error;
219 
220 	if (p->p_flag & P_HADTHREADS) {
221 		PROC_LOCK(p);
222 		if (thread_single(SINGLE_BOUNDARY)) {
223 			PROC_UNLOCK(p);
224 			return (ERESTART);	/* Try again later. */
225 		}
226 		PROC_UNLOCK(p);
227 	}
228 
229 	error = do_execve(td, fname, argv, envv, mac_p);
230 
231 	if (p->p_flag & P_HADTHREADS) {
232 		PROC_LOCK(p);
233 		/*
234 		 * If success, we upgrade to SINGLE_EXIT state to
235 		 * force other threads to suicide.
236 		 */
237 		if (error == 0)
238 			thread_single(SINGLE_EXIT);
239 		else
240 			thread_single_end();
241 		PROC_UNLOCK(p);
242 	}
243 
244 	return (error);
245 }
246 
247 /*
248  * In-kernel implementation of execve().  All arguments are assumed to be
249  * userspace pointers from the passed thread.
250  *
251  * MPSAFE
252  */
253 static int
254 do_execve(td, fname, argv, envv, mac_p)
255 	struct thread *td;
256 	char *fname;
257 	char **argv;
258 	char **envv;
259 	struct mac *mac_p;
260 {
261 	struct proc *p = td->td_proc;
262 	struct nameidata nd, *ndp;
263 	struct ucred *newcred = NULL, *oldcred;
264 	struct uidinfo *euip;
265 	register_t *stack_base;
266 	int error, len, i;
267 	struct image_params image_params, *imgp;
268 	struct vattr attr;
269 	int (*img_first)(struct image_params *);
270 	struct pargs *oldargs = NULL, *newargs = NULL;
271 	struct sigacts *oldsigacts, *newsigacts;
272 #ifdef KTRACE
273 	struct vnode *tracevp = NULL;
274 	struct ucred *tracecred = NULL;
275 #endif
276 	struct vnode *textvp = NULL;
277 	int credential_changing;
278 	int textset;
279 #ifdef MAC
280 	struct label *interplabel = NULL;
281 	int will_transition;
282 #endif
283 
284 	imgp = &image_params;
285 
286 	/*
287 	 * Lock the process and set the P_INEXEC flag to indicate that
288 	 * it should be left alone until we're done here.  This is
289 	 * necessary to avoid race conditions - e.g. in ptrace() -
290 	 * that might allow a local user to illicitly obtain elevated
291 	 * privileges.
292 	 */
293 	PROC_LOCK(p);
294 	KASSERT((p->p_flag & P_INEXEC) == 0,
295 	    ("%s(): process already has P_INEXEC flag", __func__));
296 	p->p_flag |= P_INEXEC;
297 	PROC_UNLOCK(p);
298 
299 	/*
300 	 * Initialize part of the common data
301 	 */
302 	imgp->proc = p;
303 	imgp->userspace_argv = argv;
304 	imgp->userspace_envv = envv;
305 	imgp->execlabel = NULL;
306 	imgp->attr = &attr;
307 	imgp->argc = imgp->envc = 0;
308 	imgp->argv0 = NULL;
309 	imgp->entry_addr = 0;
310 	imgp->vmspace_destroyed = 0;
311 	imgp->interpreted = 0;
312 	imgp->interpreter_name[0] = '\0';
313 	imgp->auxargs = NULL;
314 	imgp->vp = NULL;
315 	imgp->object = NULL;
316 	imgp->firstpage = NULL;
317 	imgp->ps_strings = 0;
318 	imgp->auxarg_size = 0;
319 
320 #ifdef MAC
321 	error = mac_execve_enter(imgp, mac_p);
322 	if (error) {
323 		mtx_lock(&Giant);
324 		goto exec_fail;
325 	}
326 #endif
327 
328 	/*
329 	 * Allocate temporary demand zeroed space for argument and
330 	 *	environment strings
331 	 */
332 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX);
333 	if (imgp->stringbase == NULL) {
334 		error = ENOMEM;
335 		mtx_lock(&Giant);
336 		goto exec_fail;
337 	}
338 	imgp->stringp = imgp->stringbase;
339 	imgp->stringspace = ARG_MAX;
340 	imgp->image_header = NULL;
341 
342 	/*
343 	 * Translate the file name. namei() returns a vnode pointer
344 	 *	in ni_vp amoung other things.
345 	 */
346 	ndp = &nd;
347 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
348 	    UIO_USERSPACE, fname, td);
349 
350 	mtx_lock(&Giant);
351 interpret:
352 
353 	error = namei(ndp);
354 	if (error) {
355 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
356 		    ARG_MAX);
357 		goto exec_fail;
358 	}
359 
360 	imgp->vp = ndp->ni_vp;
361 	imgp->fname = fname;
362 
363 	/*
364 	 * Check file permissions (also 'opens' file)
365 	 */
366 	error = exec_check_permissions(imgp);
367 	if (error)
368 		goto exec_fail_dealloc;
369 
370 	if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0)
371 		vm_object_reference(imgp->object);
372 
373 	/*
374 	 * Set VV_TEXT now so no one can write to the executable while we're
375 	 * activating it.
376 	 *
377 	 * Remember if this was set before and unset it in case this is not
378 	 * actually an executable image.
379 	 */
380 	textset = imgp->vp->v_vflag & VV_TEXT;
381 	imgp->vp->v_vflag |= VV_TEXT;
382 
383 	error = exec_map_first_page(imgp);
384 	if (error)
385 		goto exec_fail_dealloc;
386 
387 	/*
388 	 *	If the current process has a special image activator it
389 	 *	wants to try first, call it.   For example, emulating shell
390 	 *	scripts differently.
391 	 */
392 	error = -1;
393 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
394 		error = img_first(imgp);
395 
396 	/*
397 	 *	Loop through the list of image activators, calling each one.
398 	 *	An activator returns -1 if there is no match, 0 on success,
399 	 *	and an error otherwise.
400 	 */
401 	for (i = 0; error == -1 && execsw[i]; ++i) {
402 		if (execsw[i]->ex_imgact == NULL ||
403 		    execsw[i]->ex_imgact == img_first) {
404 			continue;
405 		}
406 		error = (*execsw[i]->ex_imgact)(imgp);
407 	}
408 
409 	if (error) {
410 		if (error == -1) {
411 			if (textset == 0)
412 				imgp->vp->v_vflag &= ~VV_TEXT;
413 			error = ENOEXEC;
414 		}
415 		goto exec_fail_dealloc;
416 	}
417 
418 	/*
419 	 * Special interpreter operation, cleanup and loop up to try to
420 	 * activate the interpreter.
421 	 */
422 	if (imgp->interpreted) {
423 		exec_unmap_first_page(imgp);
424 		/*
425 		 * VV_TEXT needs to be unset for scripts.  There is a short
426 		 * period before we determine that something is a script where
427 		 * VV_TEXT will be set. The vnode lock is held over this
428 		 * entire period so nothing should illegitimately be blocked.
429 		 */
430 		imgp->vp->v_vflag &= ~VV_TEXT;
431 		/* free name buffer and old vnode */
432 		NDFREE(ndp, NDF_ONLY_PNBUF);
433 #ifdef MAC
434 		interplabel = mac_vnode_label_alloc();
435 		mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
436 #endif
437 		vput(ndp->ni_vp);
438 		vm_object_deallocate(imgp->object);
439 		imgp->object = NULL;
440 		/* set new name to that of the interpreter */
441 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
442 		    UIO_SYSSPACE, imgp->interpreter_name, td);
443 		goto interpret;
444 	}
445 
446 	/*
447 	 * Copy out strings (args and env) and initialize stack base
448 	 */
449 	if (p->p_sysent->sv_copyout_strings)
450 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
451 	else
452 		stack_base = exec_copyout_strings(imgp);
453 
454 	/*
455 	 * If custom stack fixup routine present for this process
456 	 * let it do the stack setup.
457 	 * Else stuff argument count as first item on stack
458 	 */
459 	if (p->p_sysent->sv_fixup != NULL)
460 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
461 	else
462 		suword(--stack_base, imgp->argc);
463 
464 	/*
465 	 * For security and other reasons, the file descriptor table cannot
466 	 * be shared after an exec.
467 	 */
468 	FILEDESC_LOCK_FAST(p->p_fd);
469 	if (p->p_fd->fd_refcnt > 1) {
470 		struct filedesc *tmp;
471 
472 		FILEDESC_UNLOCK_FAST(p->p_fd);
473 		tmp = fdcopy(p->p_fd);
474 		fdfree(td);
475 		p->p_fd = tmp;
476 	} else
477 		FILEDESC_UNLOCK_FAST(p->p_fd);
478 
479 	/*
480 	 * Malloc things before we need locks.
481 	 */
482 	newcred = crget();
483 	euip = uifind(attr.va_uid);
484 	i = imgp->endargs - imgp->stringbase;
485 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
486 		newargs = pargs_alloc(i);
487 
488 	/* close files on exec */
489 	fdcloseexec(td);
490 
491 	/* Get a reference to the vnode prior to locking the proc */
492 	VREF(ndp->ni_vp);
493 
494 	/*
495 	 * For security and other reasons, signal handlers cannot
496 	 * be shared after an exec. The new process gets a copy of the old
497 	 * handlers. In execsigs(), the new process will have its signals
498 	 * reset.
499 	 */
500 	PROC_LOCK(p);
501 	if (sigacts_shared(p->p_sigacts)) {
502 		oldsigacts = p->p_sigacts;
503 		PROC_UNLOCK(p);
504 		newsigacts = sigacts_alloc();
505 		sigacts_copy(newsigacts, oldsigacts);
506 		PROC_LOCK(p);
507 		p->p_sigacts = newsigacts;
508 	} else
509 		oldsigacts = NULL;
510 
511 	/* Stop profiling */
512 	stopprofclock(p);
513 
514 	/* reset caught signals */
515 	execsigs(p);
516 
517 	/* name this process - nameiexec(p, ndp) */
518 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
519 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
520 	p->p_comm[len] = 0;
521 
522 	/*
523 	 * mark as execed, wakeup the process that vforked (if any) and tell
524 	 * it that it now has its own resources back
525 	 */
526 	p->p_flag |= P_EXEC;
527 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
528 		p->p_flag &= ~P_PPWAIT;
529 		wakeup(p->p_pptr);
530 	}
531 
532 	/*
533 	 * Implement image setuid/setgid.
534 	 *
535 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
536 	 * the process is being traced.
537 	 *
538 	 * XXXMAC: For the time being, use NOSUID to also prohibit
539 	 * transitions on the file system.
540 	 */
541 	oldcred = p->p_ucred;
542 	credential_changing = 0;
543 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
544 	    attr.va_uid;
545 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
546 	    attr.va_gid;
547 #ifdef MAC
548 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
549 	    interplabel, imgp);
550 	credential_changing |= will_transition;
551 #endif
552 
553 	if (credential_changing &&
554 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
555 	    (p->p_flag & P_TRACED) == 0) {
556 		/*
557 		 * Turn off syscall tracing for set-id programs, except for
558 		 * root.  Record any set-id flags first to make sure that
559 		 * we do not regain any tracing during a possible block.
560 		 */
561 		setsugid(p);
562 #ifdef KTRACE
563 		if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) {
564 			mtx_lock(&ktrace_mtx);
565 			p->p_traceflag = 0;
566 			tracevp = p->p_tracevp;
567 			p->p_tracevp = NULL;
568 			tracecred = p->p_tracecred;
569 			p->p_tracecred = NULL;
570 			mtx_unlock(&ktrace_mtx);
571 		}
572 #endif
573 		/*
574 		 * Close any file descriptors 0..2 that reference procfs,
575 		 * then make sure file descriptors 0..2 are in use.
576 		 *
577 		 * setugidsafety() may call closef() and then pfind()
578 		 * which may grab the process lock.
579 		 * fdcheckstd() may call falloc() which may block to
580 		 * allocate memory, so temporarily drop the process lock.
581 		 */
582 		PROC_UNLOCK(p);
583 		setugidsafety(td);
584 		error = fdcheckstd(td);
585 		if (error != 0)
586 			goto done1;
587 		PROC_LOCK(p);
588 		/*
589 		 * Set the new credentials.
590 		 */
591 		crcopy(newcred, oldcred);
592 		if (attr.va_mode & VSUID)
593 			change_euid(newcred, euip);
594 		if (attr.va_mode & VSGID)
595 			change_egid(newcred, attr.va_gid);
596 #ifdef MAC
597 		if (will_transition) {
598 			mac_execve_transition(oldcred, newcred, imgp->vp,
599 			    interplabel, imgp);
600 		}
601 #endif
602 		/*
603 		 * Implement correct POSIX saved-id behavior.
604 		 *
605 		 * XXXMAC: Note that the current logic will save the
606 		 * uid and gid if a MAC domain transition occurs, even
607 		 * though maybe it shouldn't.
608 		 */
609 		change_svuid(newcred, newcred->cr_uid);
610 		change_svgid(newcred, newcred->cr_gid);
611 		p->p_ucred = newcred;
612 		newcred = NULL;
613 	} else {
614 		if (oldcred->cr_uid == oldcred->cr_ruid &&
615 		    oldcred->cr_gid == oldcred->cr_rgid)
616 			p->p_flag &= ~P_SUGID;
617 		/*
618 		 * Implement correct POSIX saved-id behavior.
619 		 *
620 		 * XXX: It's not clear that the existing behavior is
621 		 * POSIX-compliant.  A number of sources indicate that the
622 		 * saved uid/gid should only be updated if the new ruid is
623 		 * not equal to the old ruid, or the new euid is not equal
624 		 * to the old euid and the new euid is not equal to the old
625 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
626 		 * Also, this code uses the new (replaced) euid and egid as
627 		 * the source, which may or may not be the right ones to use.
628 		 */
629 		if (oldcred->cr_svuid != oldcred->cr_uid ||
630 		    oldcred->cr_svgid != oldcred->cr_gid) {
631 			crcopy(newcred, oldcred);
632 			change_svuid(newcred, newcred->cr_uid);
633 			change_svgid(newcred, newcred->cr_gid);
634 			p->p_ucred = newcred;
635 			newcred = NULL;
636 		}
637 	}
638 
639 	/*
640 	 * Store the vp for use in procfs.  This vnode was referenced prior
641 	 * to locking the proc lock.
642 	 */
643 	textvp = p->p_textvp;
644 	p->p_textvp = ndp->ni_vp;
645 
646 	/*
647 	 * Notify others that we exec'd, and clear the P_INEXEC flag
648 	 * as we're now a bona fide freshly-execed process.
649 	 */
650 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
651 	p->p_flag &= ~P_INEXEC;
652 
653 	/*
654 	 * If tracing the process, trap to debugger so breakpoints
655 	 * can be set before the program executes.
656 	 * Use tdsignal to deliver signal to current thread, use
657 	 * psignal may cause the signal to be delivered to wrong thread
658 	 * because that thread will exit, remember we are going to enter
659 	 * single thread mode.
660 	 */
661 	if (p->p_flag & P_TRACED)
662 		tdsignal(td, SIGTRAP, SIGTARGET_TD);
663 
664 	/* clear "fork but no exec" flag, as we _are_ execing */
665 	p->p_acflag &= ~AFORK;
666 
667 	/* Free any previous argument cache */
668 	oldargs = p->p_args;
669 	p->p_args = NULL;
670 
671 	/* Cache arguments if they fit inside our allowance */
672 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
673 		bcopy(imgp->stringbase, newargs->ar_args, i);
674 		p->p_args = newargs;
675 		newargs = NULL;
676 	}
677 	PROC_UNLOCK(p);
678 
679 	/* Set values passed into the program in registers. */
680 	if (p->p_sysent->sv_setregs)
681 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
682 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
683 	else
684 		exec_setregs(td, imgp->entry_addr,
685 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
686 
687 done1:
688 	/*
689 	 * Free any resources malloc'd earlier that we didn't use.
690 	 */
691 	uifree(euip);
692 	if (newcred == NULL)
693 		crfree(oldcred);
694 	else
695 		crfree(newcred);
696 	/*
697 	 * Handle deferred decrement of ref counts.
698 	 */
699 	if (textvp != NULL)
700 		vrele(textvp);
701 	if (ndp->ni_vp && error != 0)
702 		vrele(ndp->ni_vp);
703 #ifdef KTRACE
704 	if (tracevp != NULL)
705 		vrele(tracevp);
706 	if (tracecred != NULL)
707 		crfree(tracecred);
708 #endif
709 	if (oldargs != NULL)
710 		pargs_drop(oldargs);
711 	if (newargs != NULL)
712 		pargs_drop(newargs);
713 	if (oldsigacts != NULL)
714 		sigacts_free(oldsigacts);
715 
716 exec_fail_dealloc:
717 
718 	/*
719 	 * free various allocated resources
720 	 */
721 	if (imgp->firstpage != NULL)
722 		exec_unmap_first_page(imgp);
723 
724 	if (imgp->vp != NULL) {
725 		NDFREE(ndp, NDF_ONLY_PNBUF);
726 		vput(imgp->vp);
727 	}
728 
729 	if (imgp->stringbase != NULL)
730 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
731 		    ARG_MAX);
732 
733 	if (imgp->object != NULL)
734 		vm_object_deallocate(imgp->object);
735 
736 	if (error == 0) {
737 		/*
738 		 * Stop the process here if its stop event mask has
739 		 * the S_EXEC bit set.
740 		 */
741 		STOPEVENT(p, S_EXEC, 0);
742 		goto done2;
743 	}
744 
745 exec_fail:
746 	/* we're done here, clear P_INEXEC */
747 	PROC_LOCK(p);
748 	p->p_flag &= ~P_INEXEC;
749 	PROC_UNLOCK(p);
750 
751 	if (imgp->vmspace_destroyed) {
752 		/* sorry, no more process anymore. exit gracefully */
753 #ifdef MAC
754 		mac_execve_exit(imgp);
755 		if (interplabel != NULL)
756 			mac_vnode_label_free(interplabel);
757 #endif
758 		mtx_unlock(&Giant);
759 		exit1(td, W_EXITCODE(0, SIGABRT));
760 		/* NOT REACHED */
761 		error = 0;
762 	}
763 done2:
764 #ifdef MAC
765 	mac_execve_exit(imgp);
766 	if (interplabel != NULL)
767 		mac_vnode_label_free(interplabel);
768 #endif
769 	mtx_unlock(&Giant);
770 	return (error);
771 }
772 
773 int
774 exec_map_first_page(imgp)
775 	struct image_params *imgp;
776 {
777 	int rv, i;
778 	int initial_pagein;
779 	vm_page_t ma[VM_INITIAL_PAGEIN];
780 	vm_object_t object;
781 
782 	GIANT_REQUIRED;
783 
784 	if (imgp->firstpage != NULL)
785 		exec_unmap_first_page(imgp);
786 
787 	VOP_GETVOBJECT(imgp->vp, &object);
788 	VM_OBJECT_LOCK(object);
789 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
790 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
791 		initial_pagein = VM_INITIAL_PAGEIN;
792 		if (initial_pagein > object->size)
793 			initial_pagein = object->size;
794 		for (i = 1; i < initial_pagein; i++) {
795 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
796 				if (ma[i]->valid)
797 					break;
798 				vm_page_lock_queues();
799 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
800 					vm_page_unlock_queues();
801 					break;
802 				}
803 				vm_page_busy(ma[i]);
804 				vm_page_unlock_queues();
805 			} else {
806 				ma[i] = vm_page_alloc(object, i,
807 				    VM_ALLOC_NORMAL);
808 				if (ma[i] == NULL)
809 					break;
810 			}
811 		}
812 		initial_pagein = i;
813 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
814 		ma[0] = vm_page_lookup(object, 0);
815 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
816 		    (ma[0]->valid == 0)) {
817 			if (ma[0]) {
818 				vm_page_lock_queues();
819 				pmap_remove_all(ma[0]);
820 				vm_page_free(ma[0]);
821 				vm_page_unlock_queues();
822 			}
823 			VM_OBJECT_UNLOCK(object);
824 			return (EIO);
825 		}
826 	}
827 	vm_page_lock_queues();
828 	vm_page_hold(ma[0]);
829 	vm_page_wakeup(ma[0]);
830 	vm_page_unlock_queues();
831 	VM_OBJECT_UNLOCK(object);
832 
833 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
834 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
835 
836 	return (0);
837 }
838 
839 void
840 exec_unmap_first_page(imgp)
841 	struct image_params *imgp;
842 {
843 	vm_page_t m;
844 
845 	if (imgp->firstpage != NULL) {
846 		m = sf_buf_page(imgp->firstpage);
847 		sf_buf_free(imgp->firstpage);
848 		imgp->firstpage = NULL;
849 		vm_page_lock_queues();
850 		vm_page_unhold(m);
851 		vm_page_unlock_queues();
852 	}
853 }
854 
855 /*
856  * Destroy old address space, and allocate a new stack
857  *	The new stack is only SGROWSIZ large because it is grown
858  *	automatically in trap.c.
859  */
860 int
861 exec_new_vmspace(imgp, sv)
862 	struct image_params *imgp;
863 	struct sysentvec *sv;
864 {
865 	int error;
866 	struct proc *p = imgp->proc;
867 	struct vmspace *vmspace = p->p_vmspace;
868 	vm_offset_t stack_addr;
869 	vm_map_t map;
870 
871 	GIANT_REQUIRED;
872 
873 	imgp->vmspace_destroyed = 1;
874 
875 	/* Called with Giant held, do not depend on it! */
876 	EVENTHANDLER_INVOKE(process_exec, p);
877 
878 	/*
879 	 * Here is as good a place as any to do any resource limit cleanups.
880 	 * This is needed if a 64 bit binary exec's a 32 bit binary - the
881 	 * data size limit may need to be changed to a value that makes
882 	 * sense for the 32 bit binary.
883 	 */
884 	if (sv->sv_fixlimits != NULL)
885 		sv->sv_fixlimits(imgp);
886 
887 	/*
888 	 * Blow away entire process VM, if address space not shared,
889 	 * otherwise, create a new VM space so that other threads are
890 	 * not disrupted
891 	 */
892 	map = &vmspace->vm_map;
893 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
894 	    vm_map_max(map) == sv->sv_maxuser) {
895 		shmexit(vmspace);
896 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
897 		    vm_map_max(map));
898 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
899 	} else {
900 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
901 		vmspace = p->p_vmspace;
902 		map = &vmspace->vm_map;
903 	}
904 
905 	/* Allocate a new stack */
906 	stack_addr = sv->sv_usrstack - maxssiz;
907 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
908 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
909 	if (error)
910 		return (error);
911 
912 #ifdef __ia64__
913 	/* Allocate a new register stack */
914 	stack_addr = IA64_BACKINGSTORE;
915 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
916 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
917 	if (error)
918 		return (error);
919 #endif
920 
921 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
922 	 * VM_STACK case, but they are still used to monitor the size of the
923 	 * process stack so we can check the stack rlimit.
924 	 */
925 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
926 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
927 
928 	return (0);
929 }
930 
931 /*
932  * Copy out argument and environment strings from the old process
933  *	address space into the temporary string buffer.
934  */
935 int
936 exec_extract_strings(imgp)
937 	struct image_params *imgp;
938 {
939 	char	**argv, **envv;
940 	char	*argp, *envp;
941 	int	error;
942 	size_t	length;
943 
944 	/*
945 	 * extract arguments first
946 	 */
947 
948 	argv = imgp->userspace_argv;
949 
950 	if (argv) {
951 		argp = (caddr_t)(intptr_t)fuword(argv);
952 		if (argp == (caddr_t)-1)
953 			return (EFAULT);
954 		if (argp)
955 			argv++;
956 		if (imgp->argv0)
957 			argp = imgp->argv0;
958 		if (argp) {
959 			do {
960 				if (argp == (caddr_t)-1)
961 					return (EFAULT);
962 				if ((error = copyinstr(argp, imgp->stringp,
963 				    imgp->stringspace, &length))) {
964 					if (error == ENAMETOOLONG)
965 						return (E2BIG);
966 					return (error);
967 				}
968 				imgp->stringspace -= length;
969 				imgp->stringp += length;
970 				imgp->argc++;
971 			} while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
972 		}
973 	} else
974 		return (EFAULT);
975 
976 	imgp->endargs = imgp->stringp;
977 
978 	/*
979 	 * extract environment strings
980 	 */
981 
982 	envv = imgp->userspace_envv;
983 
984 	if (envv) {
985 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
986 			if (envp == (caddr_t)-1)
987 				return (EFAULT);
988 			if ((error = copyinstr(envp, imgp->stringp,
989 			    imgp->stringspace, &length))) {
990 				if (error == ENAMETOOLONG)
991 					return (E2BIG);
992 				return (error);
993 			}
994 			imgp->stringspace -= length;
995 			imgp->stringp += length;
996 			imgp->envc++;
997 		}
998 	}
999 
1000 	return (0);
1001 }
1002 
1003 /*
1004  * Copy strings out to the new process address space, constructing
1005  *	new arg and env vector tables. Return a pointer to the base
1006  *	so that it can be used as the initial stack pointer.
1007  */
1008 register_t *
1009 exec_copyout_strings(imgp)
1010 	struct image_params *imgp;
1011 {
1012 	int argc, envc;
1013 	char **vectp;
1014 	char *stringp, *destp;
1015 	register_t *stack_base;
1016 	struct ps_strings *arginfo;
1017 	struct proc *p;
1018 	int szsigcode;
1019 
1020 	/*
1021 	 * Calculate string base and vector table pointers.
1022 	 * Also deal with signal trampoline code for this exec type.
1023 	 */
1024 	p = imgp->proc;
1025 	szsigcode = 0;
1026 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1027 	if (p->p_sysent->sv_szsigcode != NULL)
1028 		szsigcode = *(p->p_sysent->sv_szsigcode);
1029 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1030 	    roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
1031 
1032 	/*
1033 	 * install sigcode
1034 	 */
1035 	if (szsigcode)
1036 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1037 		    szsigcode), szsigcode);
1038 
1039 	/*
1040 	 * If we have a valid auxargs ptr, prepare some room
1041 	 * on the stack.
1042 	 */
1043 	if (imgp->auxargs) {
1044 		/*
1045 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1046 		 * lower compatibility.
1047 		 */
1048 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1049 		    (AT_COUNT * 2);
1050 		/*
1051 		 * The '+ 2' is for the null pointers at the end of each of
1052 		 * the arg and env vector sets,and imgp->auxarg_size is room
1053 		 * for argument of Runtime loader.
1054 		 */
1055 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1056 		    imgp->auxarg_size) * sizeof(char *));
1057 
1058 	} else
1059 		/*
1060 		 * The '+ 2' is for the null pointers at the end of each of
1061 		 * the arg and env vector sets
1062 		 */
1063 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1064 		    sizeof(char *));
1065 
1066 	/*
1067 	 * vectp also becomes our initial stack base
1068 	 */
1069 	stack_base = (register_t *)vectp;
1070 
1071 	stringp = imgp->stringbase;
1072 	argc = imgp->argc;
1073 	envc = imgp->envc;
1074 
1075 	/*
1076 	 * Copy out strings - arguments and environment.
1077 	 */
1078 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1079 
1080 	/*
1081 	 * Fill in "ps_strings" struct for ps, w, etc.
1082 	 */
1083 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1084 	suword(&arginfo->ps_nargvstr, argc);
1085 
1086 	/*
1087 	 * Fill in argument portion of vector table.
1088 	 */
1089 	for (; argc > 0; --argc) {
1090 		suword(vectp++, (long)(intptr_t)destp);
1091 		while (*stringp++ != 0)
1092 			destp++;
1093 		destp++;
1094 	}
1095 
1096 	/* a null vector table pointer separates the argp's from the envp's */
1097 	suword(vectp++, 0);
1098 
1099 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1100 	suword(&arginfo->ps_nenvstr, envc);
1101 
1102 	/*
1103 	 * Fill in environment portion of vector table.
1104 	 */
1105 	for (; envc > 0; --envc) {
1106 		suword(vectp++, (long)(intptr_t)destp);
1107 		while (*stringp++ != 0)
1108 			destp++;
1109 		destp++;
1110 	}
1111 
1112 	/* end of vector table is a null pointer */
1113 	suword(vectp, 0);
1114 
1115 	return (stack_base);
1116 }
1117 
1118 /*
1119  * Check permissions of file to execute.
1120  *	Called with imgp->vp locked.
1121  *	Return 0 for success or error code on failure.
1122  */
1123 int
1124 exec_check_permissions(imgp)
1125 	struct image_params *imgp;
1126 {
1127 	struct vnode *vp = imgp->vp;
1128 	struct vattr *attr = imgp->attr;
1129 	struct thread *td;
1130 	int error;
1131 
1132 	td = curthread;			/* XXXKSE */
1133 
1134 	/* Get file attributes */
1135 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1136 	if (error)
1137 		return (error);
1138 
1139 #ifdef MAC
1140 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1141 	if (error)
1142 		return (error);
1143 #endif
1144 
1145 	/*
1146 	 * 1) Check if file execution is disabled for the filesystem that this
1147 	 *	file resides on.
1148 	 * 2) Insure that at least one execute bit is on - otherwise root
1149 	 *	will always succeed, and we don't want to happen unless the
1150 	 *	file really is executable.
1151 	 * 3) Insure that the file is a regular file.
1152 	 */
1153 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1154 	    ((attr->va_mode & 0111) == 0) ||
1155 	    (attr->va_type != VREG))
1156 		return (EACCES);
1157 
1158 	/*
1159 	 * Zero length files can't be exec'd
1160 	 */
1161 	if (attr->va_size == 0)
1162 		return (ENOEXEC);
1163 
1164 	/*
1165 	 *  Check for execute permission to file based on current credentials.
1166 	 */
1167 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1168 	if (error)
1169 		return (error);
1170 
1171 	/*
1172 	 * Check number of open-for-writes on the file and deny execution
1173 	 * if there are any.
1174 	 */
1175 	if (vp->v_writecount)
1176 		return (ETXTBSY);
1177 
1178 	/*
1179 	 * Call filesystem specific open routine (which does nothing in the
1180 	 * general case).
1181 	 */
1182 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1183 	return (error);
1184 }
1185 
1186 /*
1187  * Exec handler registration
1188  */
1189 int
1190 exec_register(execsw_arg)
1191 	const struct execsw *execsw_arg;
1192 {
1193 	const struct execsw **es, **xs, **newexecsw;
1194 	int count = 2;	/* New slot and trailing NULL */
1195 
1196 	if (execsw)
1197 		for (es = execsw; *es; es++)
1198 			count++;
1199 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1200 	if (newexecsw == NULL)
1201 		return (ENOMEM);
1202 	xs = newexecsw;
1203 	if (execsw)
1204 		for (es = execsw; *es; es++)
1205 			*xs++ = *es;
1206 	*xs++ = execsw_arg;
1207 	*xs = NULL;
1208 	if (execsw)
1209 		free(execsw, M_TEMP);
1210 	execsw = newexecsw;
1211 	return (0);
1212 }
1213 
1214 int
1215 exec_unregister(execsw_arg)
1216 	const struct execsw *execsw_arg;
1217 {
1218 	const struct execsw **es, **xs, **newexecsw;
1219 	int count = 1;
1220 
1221 	if (execsw == NULL)
1222 		panic("unregister with no handlers left?\n");
1223 
1224 	for (es = execsw; *es; es++) {
1225 		if (*es == execsw_arg)
1226 			break;
1227 	}
1228 	if (*es == NULL)
1229 		return (ENOENT);
1230 	for (es = execsw; *es; es++)
1231 		if (*es != execsw_arg)
1232 			count++;
1233 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1234 	if (newexecsw == NULL)
1235 		return (ENOMEM);
1236 	xs = newexecsw;
1237 	for (es = execsw; *es; es++)
1238 		if (*es != execsw_arg)
1239 			*xs++ = *es;
1240 	*xs = NULL;
1241 	if (execsw)
1242 		free(execsw, M_TEMP);
1243 	execsw = newexecsw;
1244 	return (0);
1245 }
1246