xref: /freebsd/sys/kern/kern_exec.c (revision 87569f75a91f298c52a71823c04d41cf53c88889)
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_hwpmc_hooks.h"
31 #include "opt_ktrace.h"
32 #include "opt_mac.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/eventhandler.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/sysproto.h>
40 #include <sys/signalvar.h>
41 #include <sys/kernel.h>
42 #include <sys/mac.h>
43 #include <sys/mount.h>
44 #include <sys/filedesc.h>
45 #include <sys/fcntl.h>
46 #include <sys/acct.h>
47 #include <sys/exec.h>
48 #include <sys/imgact.h>
49 #include <sys/imgact_elf.h>
50 #include <sys/wait.h>
51 #include <sys/malloc.h>
52 #include <sys/proc.h>
53 #include <sys/pioctl.h>
54 #include <sys/namei.h>
55 #include <sys/resourcevar.h>
56 #include <sys/sf_buf.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
59 #include <sys/shm.h>
60 #include <sys/sysctl.h>
61 #include <sys/vnode.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65 
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_pager.h>
75 
76 #ifdef	HWPMC_HOOKS
77 #include <sys/pmckern.h>
78 #endif
79 
80 #include <machine/reg.h>
81 
82 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
83 
84 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
85 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
86 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
87 static int do_execve(struct thread *td, struct image_args *args,
88     struct mac *mac_p);
89 static void exec_free_args(struct image_args *);
90 
91 /* XXX This should be vm_size_t. */
92 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
93     NULL, 0, sysctl_kern_ps_strings, "LU", "");
94 
95 /* XXX This should be vm_size_t. */
96 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
97     NULL, 0, sysctl_kern_usrstack, "LU", "");
98 
99 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
100     NULL, 0, sysctl_kern_stackprot, "I", "");
101 
102 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
103 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
104     &ps_arg_cache_limit, 0, "");
105 
106 static int
107 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
108 {
109 	struct proc *p;
110 	int error;
111 
112 	p = curproc;
113 #ifdef SCTL_MASK32
114 	if (req->flags & SCTL_MASK32) {
115 		unsigned int val;
116 		val = (unsigned int)p->p_sysent->sv_psstrings;
117 		error = SYSCTL_OUT(req, &val, sizeof(val));
118 	} else
119 #endif
120 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
121 		   sizeof(p->p_sysent->sv_psstrings));
122 	return error;
123 }
124 
125 static int
126 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
127 {
128 	struct proc *p;
129 	int error;
130 
131 	p = curproc;
132 #ifdef SCTL_MASK32
133 	if (req->flags & SCTL_MASK32) {
134 		unsigned int val;
135 		val = (unsigned int)p->p_sysent->sv_usrstack;
136 		error = SYSCTL_OUT(req, &val, sizeof(val));
137 	} else
138 #endif
139 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
140 		    sizeof(p->p_sysent->sv_usrstack));
141 	return error;
142 }
143 
144 static int
145 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
146 {
147 	struct proc *p;
148 
149 	p = curproc;
150 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
151 	    sizeof(p->p_sysent->sv_stackprot)));
152 }
153 
154 /*
155  * Each of the items is a pointer to a `const struct execsw', hence the
156  * double pointer here.
157  */
158 static const struct execsw **execsw;
159 
160 #ifndef _SYS_SYSPROTO_H_
161 struct execve_args {
162 	char    *fname;
163 	char    **argv;
164 	char    **envv;
165 };
166 #endif
167 
168 /*
169  * MPSAFE
170  */
171 int
172 execve(td, uap)
173 	struct thread *td;
174 	struct execve_args /* {
175 		char *fname;
176 		char **argv;
177 		char **envv;
178 	} */ *uap;
179 {
180 	int error;
181 	struct image_args args;
182 
183 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
184 	    uap->argv, uap->envv);
185 	if (error == 0)
186 		error = kern_execve(td, &args, NULL);
187 	return (error);
188 }
189 
190 #ifndef _SYS_SYSPROTO_H_
191 struct __mac_execve_args {
192 	char	*fname;
193 	char	**argv;
194 	char	**envv;
195 	struct mac	*mac_p;
196 };
197 #endif
198 
199 /*
200  * MPSAFE
201  */
202 int
203 __mac_execve(td, uap)
204 	struct thread *td;
205 	struct __mac_execve_args /* {
206 		char *fname;
207 		char **argv;
208 		char **envv;
209 		struct mac *mac_p;
210 	} */ *uap;
211 {
212 #ifdef MAC
213 	int error;
214 	struct image_args args;
215 
216 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
217 	    uap->argv, uap->envv);
218 	if (error == 0)
219 		error = kern_execve(td, &args, uap->mac_p);
220 	return (error);
221 #else
222 	return (ENOSYS);
223 #endif
224 }
225 
226 /*
227  * XXX: kern_execve has the astonishing property of not always
228  * returning to the caller.  If sufficiently bad things happen during
229  * the call to do_execve(), it can end up calling exit1(); as a result,
230  * callers must avoid doing anything which they might need to undo
231  * (e.g., allocating memory).
232  */
233 int
234 kern_execve(td, args, mac_p)
235 	struct thread *td;
236 	struct image_args *args;
237 	struct mac *mac_p;
238 {
239 	struct proc *p = td->td_proc;
240 	int error;
241 
242 	if (p->p_flag & P_HADTHREADS) {
243 		PROC_LOCK(p);
244 		if (thread_single(SINGLE_BOUNDARY)) {
245 			PROC_UNLOCK(p);
246 	       		exec_free_args(args);
247 			return (ERESTART);	/* Try again later. */
248 		}
249 		PROC_UNLOCK(p);
250 	}
251 
252 	error = do_execve(td, args, mac_p);
253 
254 	if (p->p_flag & P_HADTHREADS) {
255 		PROC_LOCK(p);
256 		/*
257 		 * If success, we upgrade to SINGLE_EXIT state to
258 		 * force other threads to suicide.
259 		 */
260 		if (error == 0)
261 			thread_single(SINGLE_EXIT);
262 		else
263 			thread_single_end();
264 		PROC_UNLOCK(p);
265 	}
266 
267 	return (error);
268 }
269 
270 /*
271  * In-kernel implementation of execve().  All arguments are assumed to be
272  * userspace pointers from the passed thread.
273  *
274  * MPSAFE
275  */
276 static int
277 do_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 nameidata nd, *ndp;
284 	struct ucred *newcred = NULL, *oldcred;
285 	struct uidinfo *euip;
286 	register_t *stack_base;
287 	int error, len, i;
288 	struct image_params image_params, *imgp;
289 	struct vattr attr;
290 	int (*img_first)(struct image_params *);
291 	struct pargs *oldargs = NULL, *newargs = NULL;
292 	struct sigacts *oldsigacts, *newsigacts;
293 #ifdef KTRACE
294 	struct vnode *tracevp = NULL;
295 	struct ucred *tracecred = NULL;
296 #endif
297 	struct vnode *textvp = NULL;
298 	int credential_changing;
299 	int vfslocked;
300 	int textset;
301 #ifdef MAC
302 	struct label *interplabel = NULL;
303 	int will_transition;
304 #endif
305 #ifdef HWPMC_HOOKS
306 	struct pmckern_procexec pe;
307 #endif
308 
309 	vfslocked = 0;
310 	imgp = &image_params;
311 
312 	/*
313 	 * Lock the process and set the P_INEXEC flag to indicate that
314 	 * it should be left alone until we're done here.  This is
315 	 * necessary to avoid race conditions - e.g. in ptrace() -
316 	 * that might allow a local user to illicitly obtain elevated
317 	 * privileges.
318 	 */
319 	PROC_LOCK(p);
320 	KASSERT((p->p_flag & P_INEXEC) == 0,
321 	    ("%s(): process already has P_INEXEC flag", __func__));
322 	p->p_flag |= P_INEXEC;
323 	PROC_UNLOCK(p);
324 
325 	/*
326 	 * Initialize part of the common data
327 	 */
328 	imgp->proc = p;
329 	imgp->execlabel = NULL;
330 	imgp->attr = &attr;
331 	imgp->entry_addr = 0;
332 	imgp->vmspace_destroyed = 0;
333 	imgp->interpreted = 0;
334 	imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
335 	imgp->auxargs = NULL;
336 	imgp->vp = NULL;
337 	imgp->object = NULL;
338 	imgp->firstpage = NULL;
339 	imgp->ps_strings = 0;
340 	imgp->auxarg_size = 0;
341 	imgp->args = args;
342 
343 #ifdef MAC
344 	error = mac_execve_enter(imgp, mac_p);
345 	if (error)
346 		goto exec_fail;
347 #endif
348 
349 	imgp->image_header = NULL;
350 
351 	/*
352 	 * Translate the file name. namei() returns a vnode pointer
353 	 *	in ni_vp amoung other things.
354 	 */
355 	ndp = &nd;
356 	NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
357 	    UIO_SYSSPACE, args->fname, td);
358 
359 interpret:
360 	error = namei(ndp);
361 	if (error)
362 		goto exec_fail;
363 
364 	vfslocked = NDHASGIANT(ndp);
365 	imgp->vp = ndp->ni_vp;
366 
367 	/*
368 	 * Check file permissions (also 'opens' file)
369 	 */
370 	error = exec_check_permissions(imgp);
371 	if (error)
372 		goto exec_fail_dealloc;
373 
374 	imgp->object = imgp->vp->v_object;
375 	if (imgp->object != NULL)
376 		vm_object_reference(imgp->object);
377 
378 	/*
379 	 * Set VV_TEXT now so no one can write to the executable while we're
380 	 * activating it.
381 	 *
382 	 * Remember if this was set before and unset it in case this is not
383 	 * actually an executable image.
384 	 */
385 	textset = imgp->vp->v_vflag & VV_TEXT;
386 	imgp->vp->v_vflag |= VV_TEXT;
387 
388 	error = exec_map_first_page(imgp);
389 	if (error)
390 		goto exec_fail_dealloc;
391 
392 	/*
393 	 *	If the current process has a special image activator it
394 	 *	wants to try first, call it.   For example, emulating shell
395 	 *	scripts differently.
396 	 */
397 	error = -1;
398 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
399 		error = img_first(imgp);
400 
401 	/*
402 	 *	Loop through the list of image activators, calling each one.
403 	 *	An activator returns -1 if there is no match, 0 on success,
404 	 *	and an error otherwise.
405 	 */
406 	for (i = 0; error == -1 && execsw[i]; ++i) {
407 		if (execsw[i]->ex_imgact == NULL ||
408 		    execsw[i]->ex_imgact == img_first) {
409 			continue;
410 		}
411 		error = (*execsw[i]->ex_imgact)(imgp);
412 	}
413 
414 	if (error) {
415 		if (error == -1) {
416 			if (textset == 0)
417 				imgp->vp->v_vflag &= ~VV_TEXT;
418 			error = ENOEXEC;
419 		}
420 		goto exec_fail_dealloc;
421 	}
422 
423 	/*
424 	 * Special interpreter operation, cleanup and loop up to try to
425 	 * activate the interpreter.
426 	 */
427 	if (imgp->interpreted) {
428 		exec_unmap_first_page(imgp);
429 		/*
430 		 * VV_TEXT needs to be unset for scripts.  There is a short
431 		 * period before we determine that something is a script where
432 		 * VV_TEXT will be set. The vnode lock is held over this
433 		 * entire period so nothing should illegitimately be blocked.
434 		 */
435 		imgp->vp->v_vflag &= ~VV_TEXT;
436 		/* free name buffer and old vnode */
437 		NDFREE(ndp, NDF_ONLY_PNBUF);
438 #ifdef MAC
439 		interplabel = mac_vnode_label_alloc();
440 		mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
441 #endif
442 		vput(ndp->ni_vp);
443 		vm_object_deallocate(imgp->object);
444 		imgp->object = NULL;
445 		VFS_UNLOCK_GIANT(vfslocked);
446 		vfslocked = 0;
447 		/* set new name to that of the interpreter */
448 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
449 		    UIO_SYSSPACE, imgp->interpreter_name, td);
450 		goto interpret;
451 	}
452 
453 	/*
454 	 * Copy out strings (args and env) and initialize stack base
455 	 */
456 	if (p->p_sysent->sv_copyout_strings)
457 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
458 	else
459 		stack_base = exec_copyout_strings(imgp);
460 
461 	/*
462 	 * If custom stack fixup routine present for this process
463 	 * let it do the stack setup.
464 	 * Else stuff argument count as first item on stack
465 	 */
466 	if (p->p_sysent->sv_fixup != NULL)
467 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
468 	else
469 		suword(--stack_base, imgp->args->argc);
470 
471 	/*
472 	 * For security and other reasons, the file descriptor table cannot
473 	 * be shared after an exec.
474 	 */
475 	fdunshare(p, td);
476 
477 	/*
478 	 * Malloc things before we need locks.
479 	 */
480 	newcred = crget();
481 	euip = uifind(attr.va_uid);
482 	i = imgp->args->begin_envv - imgp->args->begin_argv;
483 	/* Cache arguments if they fit inside our allowance */
484 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
485 		newargs = pargs_alloc(i);
486 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
487 	}
488 
489 	/* close files on exec */
490 	fdcloseexec(td);
491 
492 	/* Get a reference to the vnode prior to locking the proc */
493 	VREF(ndp->ni_vp);
494 
495 	/*
496 	 * For security and other reasons, signal handlers cannot
497 	 * be shared after an exec. The new process gets a copy of the old
498 	 * handlers. In execsigs(), the new process will have its signals
499 	 * reset.
500 	 */
501 	PROC_LOCK(p);
502 	if (sigacts_shared(p->p_sigacts)) {
503 		oldsigacts = p->p_sigacts;
504 		PROC_UNLOCK(p);
505 		newsigacts = sigacts_alloc();
506 		sigacts_copy(newsigacts, oldsigacts);
507 		PROC_LOCK(p);
508 		p->p_sigacts = newsigacts;
509 	} else
510 		oldsigacts = NULL;
511 
512 	/* Stop profiling */
513 	stopprofclock(p);
514 
515 	/* reset caught signals */
516 	execsigs(p);
517 
518 	/* name this process - nameiexec(p, ndp) */
519 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
520 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
521 	p->p_comm[len] = 0;
522 
523 	/*
524 	 * mark as execed, wakeup the process that vforked (if any) and tell
525 	 * it that it now has its own resources back
526 	 */
527 	p->p_flag |= P_EXEC;
528 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
529 		p->p_flag &= ~P_PPWAIT;
530 		wakeup(p->p_pptr);
531 	}
532 
533 	/*
534 	 * Implement image setuid/setgid.
535 	 *
536 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
537 	 * the process is being traced.
538 	 *
539 	 * XXXMAC: For the time being, use NOSUID to also prohibit
540 	 * transitions on the file system.
541 	 */
542 	oldcred = p->p_ucred;
543 	credential_changing = 0;
544 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
545 	    attr.va_uid;
546 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
547 	    attr.va_gid;
548 #ifdef MAC
549 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
550 	    interplabel, imgp);
551 	credential_changing |= will_transition;
552 #endif
553 
554 	if (credential_changing &&
555 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
556 	    (p->p_flag & P_TRACED) == 0) {
557 		/*
558 		 * Turn off syscall tracing for set-id programs, except for
559 		 * root.  Record any set-id flags first to make sure that
560 		 * we do not regain any tracing during a possible block.
561 		 */
562 		setsugid(p);
563 #ifdef KTRACE
564 		if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) {
565 			mtx_lock(&ktrace_mtx);
566 			p->p_traceflag = 0;
567 			tracevp = p->p_tracevp;
568 			p->p_tracevp = NULL;
569 			tracecred = p->p_tracecred;
570 			p->p_tracecred = NULL;
571 			mtx_unlock(&ktrace_mtx);
572 		}
573 #endif
574 		/*
575 		 * Close any file descriptors 0..2 that reference procfs,
576 		 * then make sure file descriptors 0..2 are in use.
577 		 *
578 		 * setugidsafety() may call closef() and then pfind()
579 		 * which may grab the process lock.
580 		 * fdcheckstd() may call falloc() which may block to
581 		 * allocate memory, so temporarily drop the process lock.
582 		 */
583 		PROC_UNLOCK(p);
584 		setugidsafety(td);
585 		error = fdcheckstd(td);
586 		if (error != 0)
587 			goto done1;
588 		PROC_LOCK(p);
589 		/*
590 		 * Set the new credentials.
591 		 */
592 		crcopy(newcred, oldcred);
593 		if (attr.va_mode & VSUID)
594 			change_euid(newcred, euip);
595 		if (attr.va_mode & VSGID)
596 			change_egid(newcred, attr.va_gid);
597 #ifdef MAC
598 		if (will_transition) {
599 			mac_execve_transition(oldcred, newcred, imgp->vp,
600 			    interplabel, imgp);
601 		}
602 #endif
603 		/*
604 		 * Implement correct POSIX saved-id behavior.
605 		 *
606 		 * XXXMAC: Note that the current logic will save the
607 		 * uid and gid if a MAC domain transition occurs, even
608 		 * though maybe it shouldn't.
609 		 */
610 		change_svuid(newcred, newcred->cr_uid);
611 		change_svgid(newcred, newcred->cr_gid);
612 		p->p_ucred = newcred;
613 		newcred = NULL;
614 	} else {
615 		if (oldcred->cr_uid == oldcred->cr_ruid &&
616 		    oldcred->cr_gid == oldcred->cr_rgid)
617 			p->p_flag &= ~P_SUGID;
618 		/*
619 		 * Implement correct POSIX saved-id behavior.
620 		 *
621 		 * XXX: It's not clear that the existing behavior is
622 		 * POSIX-compliant.  A number of sources indicate that the
623 		 * saved uid/gid should only be updated if the new ruid is
624 		 * not equal to the old ruid, or the new euid is not equal
625 		 * to the old euid and the new euid is not equal to the old
626 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
627 		 * Also, this code uses the new (replaced) euid and egid as
628 		 * the source, which may or may not be the right ones to use.
629 		 */
630 		if (oldcred->cr_svuid != oldcred->cr_uid ||
631 		    oldcred->cr_svgid != oldcred->cr_gid) {
632 			crcopy(newcred, oldcred);
633 			change_svuid(newcred, newcred->cr_uid);
634 			change_svgid(newcred, newcred->cr_gid);
635 			p->p_ucred = newcred;
636 			newcred = NULL;
637 		}
638 	}
639 
640 	/*
641 	 * Store the vp for use in procfs.  This vnode was referenced prior
642 	 * to locking the proc lock.
643 	 */
644 	textvp = p->p_textvp;
645 	p->p_textvp = ndp->ni_vp;
646 
647 	/*
648 	 * Notify others that we exec'd, and clear the P_INEXEC flag
649 	 * as we're now a bona fide freshly-execed process.
650 	 */
651 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
652 	p->p_flag &= ~P_INEXEC;
653 
654 	/*
655 	 * If tracing the process, trap to debugger so breakpoints
656 	 * can be set before the program executes.
657 	 * Use tdsignal to deliver signal to current thread, use
658 	 * psignal may cause the signal to be delivered to wrong thread
659 	 * because that thread will exit, remember we are going to enter
660 	 * single thread mode.
661 	 */
662 	if (p->p_flag & P_TRACED)
663 		tdsignal(p, td, SIGTRAP, NULL);
664 
665 	/* clear "fork but no exec" flag, as we _are_ execing */
666 	p->p_acflag &= ~AFORK;
667 
668 	/*
669 	 * Free any previous argument cache and replace it with
670 	 * the new argument cache, if any.
671 	 */
672 	oldargs = p->p_args;
673 	p->p_args = newargs;
674 	newargs = NULL;
675 
676 #ifdef	HWPMC_HOOKS
677 	/*
678 	 * Check if system-wide sampling is in effect or if the
679 	 * current process is using PMCs.  If so, do exec() time
680 	 * processing.  This processing needs to happen AFTER the
681 	 * P_INEXEC flag is cleared.
682 	 *
683 	 * The proc lock needs to be released before taking the PMC
684 	 * SX.
685 	 */
686 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
687 		PROC_UNLOCK(p);
688 		pe.pm_credentialschanged = credential_changing;
689 		pe.pm_entryaddr = imgp->entry_addr;
690 
691 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
692 	} else
693 		PROC_UNLOCK(p);
694 #else  /* !HWPMC_HOOKS */
695 	PROC_UNLOCK(p);
696 #endif
697 
698 	/* Set values passed into the program in registers. */
699 	if (p->p_sysent->sv_setregs)
700 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
701 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
702 	else
703 		exec_setregs(td, imgp->entry_addr,
704 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
705 
706 	vfs_mark_atime(imgp->vp, td);
707 
708 done1:
709 	/*
710 	 * Free any resources malloc'd earlier that we didn't use.
711 	 */
712 	uifree(euip);
713 	if (newcred == NULL)
714 		crfree(oldcred);
715 	else
716 		crfree(newcred);
717 	/*
718 	 * Handle deferred decrement of ref counts.
719 	 */
720 	if (textvp != NULL) {
721 		int tvfslocked;
722 
723 		tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
724 		vrele(textvp);
725 		VFS_UNLOCK_GIANT(tvfslocked);
726 	}
727 	if (ndp->ni_vp && error != 0)
728 		vrele(ndp->ni_vp);
729 #ifdef KTRACE
730 	if (tracevp != NULL)
731 		vrele(tracevp);
732 	if (tracecred != NULL)
733 		crfree(tracecred);
734 #endif
735 	if (oldargs != NULL)
736 		pargs_drop(oldargs);
737 	if (newargs != NULL)
738 		pargs_drop(newargs);
739 	if (oldsigacts != NULL)
740 		sigacts_free(oldsigacts);
741 
742 exec_fail_dealloc:
743 
744 	/*
745 	 * free various allocated resources
746 	 */
747 	if (imgp->firstpage != NULL)
748 		exec_unmap_first_page(imgp);
749 
750 	if (imgp->vp != NULL) {
751 		NDFREE(ndp, NDF_ONLY_PNBUF);
752 		vput(imgp->vp);
753 	}
754 
755 	if (imgp->object != NULL)
756 		vm_object_deallocate(imgp->object);
757 
758 	if (error == 0) {
759 		/*
760 		 * Stop the process here if its stop event mask has
761 		 * the S_EXEC bit set.
762 		 */
763 		STOPEVENT(p, S_EXEC, 0);
764 		goto done2;
765 	}
766 
767 exec_fail:
768 	/* we're done here, clear P_INEXEC */
769 	PROC_LOCK(p);
770 	p->p_flag &= ~P_INEXEC;
771 	PROC_UNLOCK(p);
772 
773 done2:
774 #ifdef MAC
775 	mac_execve_exit(imgp);
776 	if (interplabel != NULL)
777 		mac_vnode_label_free(interplabel);
778 #endif
779 	VFS_UNLOCK_GIANT(vfslocked);
780 	exec_free_args(args);
781 
782 	if (error && imgp->vmspace_destroyed) {
783 		/* sorry, no more process anymore. exit gracefully */
784 		exit1(td, W_EXITCODE(0, SIGABRT));
785 		/* NOT REACHED */
786 	}
787 	return (error);
788 }
789 
790 int
791 exec_map_first_page(imgp)
792 	struct image_params *imgp;
793 {
794 	int rv, i;
795 	int initial_pagein;
796 	vm_page_t ma[VM_INITIAL_PAGEIN];
797 	vm_object_t object;
798 
799 	if (imgp->firstpage != NULL)
800 		exec_unmap_first_page(imgp);
801 
802 	object = imgp->vp->v_object;
803 	if (object == NULL)
804 		return (EACCES);
805 	VM_OBJECT_LOCK(object);
806 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
807 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
808 		initial_pagein = VM_INITIAL_PAGEIN;
809 		if (initial_pagein > object->size)
810 			initial_pagein = object->size;
811 		for (i = 1; i < initial_pagein; i++) {
812 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
813 				if (ma[i]->valid)
814 					break;
815 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
816 					break;
817 				vm_page_lock_queues();
818 				vm_page_busy(ma[i]);
819 				vm_page_unlock_queues();
820 			} else {
821 				ma[i] = vm_page_alloc(object, i,
822 				    VM_ALLOC_NORMAL);
823 				if (ma[i] == NULL)
824 					break;
825 			}
826 		}
827 		initial_pagein = i;
828 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
829 		ma[0] = vm_page_lookup(object, 0);
830 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
831 		    (ma[0]->valid == 0)) {
832 			if (ma[0]) {
833 				vm_page_lock_queues();
834 				vm_page_free(ma[0]);
835 				vm_page_unlock_queues();
836 			}
837 			VM_OBJECT_UNLOCK(object);
838 			return (EIO);
839 		}
840 	}
841 	vm_page_lock_queues();
842 	vm_page_hold(ma[0]);
843 	vm_page_wakeup(ma[0]);
844 	vm_page_unlock_queues();
845 	VM_OBJECT_UNLOCK(object);
846 
847 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
848 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
849 
850 	return (0);
851 }
852 
853 void
854 exec_unmap_first_page(imgp)
855 	struct image_params *imgp;
856 {
857 	vm_page_t m;
858 
859 	if (imgp->firstpage != NULL) {
860 		m = sf_buf_page(imgp->firstpage);
861 		sf_buf_free(imgp->firstpage);
862 		imgp->firstpage = NULL;
863 		vm_page_lock_queues();
864 		vm_page_unhold(m);
865 		vm_page_unlock_queues();
866 	}
867 }
868 
869 /*
870  * Destroy old address space, and allocate a new stack
871  *	The new stack is only SGROWSIZ large because it is grown
872  *	automatically in trap.c.
873  */
874 int
875 exec_new_vmspace(imgp, sv)
876 	struct image_params *imgp;
877 	struct sysentvec *sv;
878 {
879 	int error;
880 	struct proc *p = imgp->proc;
881 	struct vmspace *vmspace = p->p_vmspace;
882 	vm_offset_t stack_addr;
883 	vm_map_t map;
884 
885 	imgp->vmspace_destroyed = 1;
886 
887 	/* Called with Giant held, do not depend on it! */
888 	EVENTHANDLER_INVOKE(process_exec, p);
889 
890 	/*
891 	 * Here is as good a place as any to do any resource limit cleanups.
892 	 * This is needed if a 64 bit binary exec's a 32 bit binary - the
893 	 * data size limit may need to be changed to a value that makes
894 	 * sense for the 32 bit binary.
895 	 */
896 	if (sv->sv_fixlimits != NULL)
897 		sv->sv_fixlimits(p);
898 
899 	/*
900 	 * Blow away entire process VM, if address space not shared,
901 	 * otherwise, create a new VM space so that other threads are
902 	 * not disrupted
903 	 */
904 	map = &vmspace->vm_map;
905 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
906 	    vm_map_max(map) == sv->sv_maxuser) {
907 		shmexit(vmspace);
908 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
909 		    vm_map_max(map));
910 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
911 	} else {
912 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
913 		vmspace = p->p_vmspace;
914 		map = &vmspace->vm_map;
915 	}
916 
917 	/* Allocate a new stack */
918 	stack_addr = sv->sv_usrstack - maxssiz;
919 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
920 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
921 	if (error)
922 		return (error);
923 
924 #ifdef __ia64__
925 	/* Allocate a new register stack */
926 	stack_addr = IA64_BACKINGSTORE;
927 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
928 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
929 	if (error)
930 		return (error);
931 #endif
932 
933 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
934 	 * VM_STACK case, but they are still used to monitor the size of the
935 	 * process stack so we can check the stack rlimit.
936 	 */
937 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
938 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
939 
940 	return (0);
941 }
942 
943 /*
944  * Copy out argument and environment strings from the old process
945  *	address space into the temporary string buffer.
946  */
947 int
948 exec_copyin_args(struct image_args *args, char *fname,
949     enum uio_seg segflg, char **argv, char **envv)
950 {
951 	char *argp, *envp;
952 	int error;
953 	size_t length;
954 
955 	error = 0;
956 
957 	bzero(args, sizeof(*args));
958 	if (argv == NULL)
959 		return (EFAULT);
960 	/*
961 	 * Allocate temporary demand zeroed space for argument and
962 	 *	environment strings:
963 	 *
964 	 * o ARG_MAX for argument and environment;
965 	 * o MAXSHELLCMDLEN for the name of interpreters.
966 	 */
967 	args->buf = (char *) kmem_alloc_wait(exec_map,
968 	    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
969 	if (args->buf == NULL)
970 		return (ENOMEM);
971 	args->begin_argv = args->buf;
972 	args->endp = args->begin_argv;
973 	args->stringspace = ARG_MAX;
974 
975 	args->fname = args->buf + ARG_MAX;
976 
977 	/*
978 	 * Copy the file name.
979 	 */
980 	error = (segflg == UIO_SYSSPACE) ?
981 	    copystr(fname, args->fname, PATH_MAX, &length) :
982 	    copyinstr(fname, args->fname, PATH_MAX, &length);
983 	if (error != 0)
984 		goto err_exit;
985 
986 	/*
987 	 * extract arguments first
988 	 */
989 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
990 		if (argp == (caddr_t) -1) {
991 			error = EFAULT;
992 			goto err_exit;
993 		}
994 		if ((error = copyinstr(argp, args->endp,
995 		    args->stringspace, &length))) {
996 			if (error == ENAMETOOLONG)
997 				error = E2BIG;
998 			goto err_exit;
999 		}
1000 		args->stringspace -= length;
1001 		args->endp += length;
1002 		args->argc++;
1003 	}
1004 
1005 	args->begin_envv = args->endp;
1006 
1007 	/*
1008 	 * extract environment strings
1009 	 */
1010 	if (envv) {
1011 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1012 			if (envp == (caddr_t)-1) {
1013 				error = EFAULT;
1014 				goto err_exit;
1015 			}
1016 			if ((error = copyinstr(envp, args->endp,
1017 			    args->stringspace, &length))) {
1018 				if (error == ENAMETOOLONG)
1019 					error = E2BIG;
1020 				goto err_exit;
1021 			}
1022 			args->stringspace -= length;
1023 			args->endp += length;
1024 			args->envc++;
1025 		}
1026 	}
1027 
1028 	return (0);
1029 
1030 err_exit:
1031 	exec_free_args(args);
1032 	return (error);
1033 }
1034 
1035 static void
1036 exec_free_args(struct image_args *args)
1037 {
1038 
1039 	if (args->buf) {
1040 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1041 		    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1042 		args->buf = NULL;
1043 	}
1044 }
1045 
1046 /*
1047  * Copy strings out to the new process address space, constructing
1048  *	new arg and env vector tables. Return a pointer to the base
1049  *	so that it can be used as the initial stack pointer.
1050  */
1051 register_t *
1052 exec_copyout_strings(imgp)
1053 	struct image_params *imgp;
1054 {
1055 	int argc, envc;
1056 	char **vectp;
1057 	char *stringp, *destp;
1058 	register_t *stack_base;
1059 	struct ps_strings *arginfo;
1060 	struct proc *p;
1061 	int szsigcode;
1062 
1063 	/*
1064 	 * Calculate string base and vector table pointers.
1065 	 * Also deal with signal trampoline code for this exec type.
1066 	 */
1067 	p = imgp->proc;
1068 	szsigcode = 0;
1069 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1070 	if (p->p_sysent->sv_szsigcode != NULL)
1071 		szsigcode = *(p->p_sysent->sv_szsigcode);
1072 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1073 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1074 
1075 	/*
1076 	 * install sigcode
1077 	 */
1078 	if (szsigcode)
1079 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1080 		    szsigcode), szsigcode);
1081 
1082 	/*
1083 	 * If we have a valid auxargs ptr, prepare some room
1084 	 * on the stack.
1085 	 */
1086 	if (imgp->auxargs) {
1087 		/*
1088 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1089 		 * lower compatibility.
1090 		 */
1091 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1092 		    (AT_COUNT * 2);
1093 		/*
1094 		 * The '+ 2' is for the null pointers at the end of each of
1095 		 * the arg and env vector sets,and imgp->auxarg_size is room
1096 		 * for argument of Runtime loader.
1097 		 */
1098 		vectp = (char **)(destp - (imgp->args->argc +
1099 		    imgp->args->envc + 2 + imgp->auxarg_size) *
1100 		    sizeof(char *));
1101 
1102 	} else {
1103 		/*
1104 		 * The '+ 2' is for the null pointers at the end of each of
1105 		 * the arg and env vector sets
1106 		 */
1107 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1108 		    sizeof(char *));
1109 	}
1110 
1111 	/*
1112 	 * vectp also becomes our initial stack base
1113 	 */
1114 	stack_base = (register_t *)vectp;
1115 
1116 	stringp = imgp->args->begin_argv;
1117 	argc = imgp->args->argc;
1118 	envc = imgp->args->envc;
1119 
1120 	/*
1121 	 * Copy out strings - arguments and environment.
1122 	 */
1123 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1124 
1125 	/*
1126 	 * Fill in "ps_strings" struct for ps, w, etc.
1127 	 */
1128 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1129 	suword(&arginfo->ps_nargvstr, argc);
1130 
1131 	/*
1132 	 * Fill in argument portion of vector table.
1133 	 */
1134 	for (; argc > 0; --argc) {
1135 		suword(vectp++, (long)(intptr_t)destp);
1136 		while (*stringp++ != 0)
1137 			destp++;
1138 		destp++;
1139 	}
1140 
1141 	/* a null vector table pointer separates the argp's from the envp's */
1142 	suword(vectp++, 0);
1143 
1144 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1145 	suword(&arginfo->ps_nenvstr, envc);
1146 
1147 	/*
1148 	 * Fill in environment portion of vector table.
1149 	 */
1150 	for (; envc > 0; --envc) {
1151 		suword(vectp++, (long)(intptr_t)destp);
1152 		while (*stringp++ != 0)
1153 			destp++;
1154 		destp++;
1155 	}
1156 
1157 	/* end of vector table is a null pointer */
1158 	suword(vectp, 0);
1159 
1160 	return (stack_base);
1161 }
1162 
1163 /*
1164  * Check permissions of file to execute.
1165  *	Called with imgp->vp locked.
1166  *	Return 0 for success or error code on failure.
1167  */
1168 int
1169 exec_check_permissions(imgp)
1170 	struct image_params *imgp;
1171 {
1172 	struct vnode *vp = imgp->vp;
1173 	struct vattr *attr = imgp->attr;
1174 	struct thread *td;
1175 	int error;
1176 
1177 	td = curthread;			/* XXXKSE */
1178 
1179 	/* Get file attributes */
1180 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1181 	if (error)
1182 		return (error);
1183 
1184 #ifdef MAC
1185 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1186 	if (error)
1187 		return (error);
1188 #endif
1189 
1190 	/*
1191 	 * 1) Check if file execution is disabled for the filesystem that this
1192 	 *	file resides on.
1193 	 * 2) Insure that at least one execute bit is on - otherwise root
1194 	 *	will always succeed, and we don't want to happen unless the
1195 	 *	file really is executable.
1196 	 * 3) Insure that the file is a regular file.
1197 	 */
1198 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1199 	    ((attr->va_mode & 0111) == 0) ||
1200 	    (attr->va_type != VREG))
1201 		return (EACCES);
1202 
1203 	/*
1204 	 * Zero length files can't be exec'd
1205 	 */
1206 	if (attr->va_size == 0)
1207 		return (ENOEXEC);
1208 
1209 	/*
1210 	 *  Check for execute permission to file based on current credentials.
1211 	 */
1212 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1213 	if (error)
1214 		return (error);
1215 
1216 	/*
1217 	 * Check number of open-for-writes on the file and deny execution
1218 	 * if there are any.
1219 	 */
1220 	if (vp->v_writecount)
1221 		return (ETXTBSY);
1222 
1223 	/*
1224 	 * Call filesystem specific open routine (which does nothing in the
1225 	 * general case).
1226 	 */
1227 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1228 	return (error);
1229 }
1230 
1231 /*
1232  * Exec handler registration
1233  */
1234 int
1235 exec_register(execsw_arg)
1236 	const struct execsw *execsw_arg;
1237 {
1238 	const struct execsw **es, **xs, **newexecsw;
1239 	int count = 2;	/* New slot and trailing NULL */
1240 
1241 	if (execsw)
1242 		for (es = execsw; *es; es++)
1243 			count++;
1244 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1245 	if (newexecsw == NULL)
1246 		return (ENOMEM);
1247 	xs = newexecsw;
1248 	if (execsw)
1249 		for (es = execsw; *es; es++)
1250 			*xs++ = *es;
1251 	*xs++ = execsw_arg;
1252 	*xs = NULL;
1253 	if (execsw)
1254 		free(execsw, M_TEMP);
1255 	execsw = newexecsw;
1256 	return (0);
1257 }
1258 
1259 int
1260 exec_unregister(execsw_arg)
1261 	const struct execsw *execsw_arg;
1262 {
1263 	const struct execsw **es, **xs, **newexecsw;
1264 	int count = 1;
1265 
1266 	if (execsw == NULL)
1267 		panic("unregister with no handlers left?\n");
1268 
1269 	for (es = execsw; *es; es++) {
1270 		if (*es == execsw_arg)
1271 			break;
1272 	}
1273 	if (*es == NULL)
1274 		return (ENOENT);
1275 	for (es = execsw; *es; es++)
1276 		if (*es != execsw_arg)
1277 			count++;
1278 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1279 	if (newexecsw == NULL)
1280 		return (ENOMEM);
1281 	xs = newexecsw;
1282 	for (es = execsw; *es; es++)
1283 		if (*es != execsw_arg)
1284 			*xs++ = *es;
1285 	*xs = NULL;
1286 	if (execsw)
1287 		free(execsw, M_TEMP);
1288 	execsw = newexecsw;
1289 	return (0);
1290 }
1291