xref: /freebsd/sys/kern/kern_exec.c (revision c128b2d129a8e305b673ef5e3da76af1fb91ae60)
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 			return (ERESTART);	/* Try again later. */
247 		}
248 		PROC_UNLOCK(p);
249 	}
250 
251 	error = do_execve(td, args, mac_p);
252 
253 	if (p->p_flag & P_HADTHREADS) {
254 		PROC_LOCK(p);
255 		/*
256 		 * If success, we upgrade to SINGLE_EXIT state to
257 		 * force other threads to suicide.
258 		 */
259 		if (error == 0)
260 			thread_single(SINGLE_EXIT);
261 		else
262 			thread_single_end();
263 		PROC_UNLOCK(p);
264 	}
265 
266 	return (error);
267 }
268 
269 /*
270  * In-kernel implementation of execve().  All arguments are assumed to be
271  * userspace pointers from the passed thread.
272  *
273  * MPSAFE
274  */
275 static int
276 do_execve(td, args, mac_p)
277 	struct thread *td;
278 	struct image_args *args;
279 	struct mac *mac_p;
280 {
281 	struct proc *p = td->td_proc;
282 	struct nameidata nd, *ndp;
283 	struct ucred *newcred = NULL, *oldcred;
284 	struct uidinfo *euip;
285 	register_t *stack_base;
286 	int error, len, i;
287 	struct image_params image_params, *imgp;
288 	struct vattr attr;
289 	int (*img_first)(struct image_params *);
290 	struct pargs *oldargs = NULL, *newargs = NULL;
291 	struct sigacts *oldsigacts, *newsigacts;
292 #ifdef KTRACE
293 	struct vnode *tracevp = NULL;
294 	struct ucred *tracecred = NULL;
295 #endif
296 	struct vnode *textvp = NULL;
297 	int credential_changing;
298 	int vfslocked;
299 	int textset;
300 #ifdef MAC
301 	struct label *interplabel = NULL;
302 	int will_transition;
303 #endif
304 #ifdef HWPMC_HOOKS
305 	struct pmckern_procexec pe;
306 #endif
307 
308 	vfslocked = 0;
309 	imgp = &image_params;
310 
311 	/*
312 	 * Lock the process and set the P_INEXEC flag to indicate that
313 	 * it should be left alone until we're done here.  This is
314 	 * necessary to avoid race conditions - e.g. in ptrace() -
315 	 * that might allow a local user to illicitly obtain elevated
316 	 * privileges.
317 	 */
318 	PROC_LOCK(p);
319 	KASSERT((p->p_flag & P_INEXEC) == 0,
320 	    ("%s(): process already has P_INEXEC flag", __func__));
321 	p->p_flag |= P_INEXEC;
322 	PROC_UNLOCK(p);
323 
324 	/*
325 	 * Initialize part of the common data
326 	 */
327 	imgp->proc = p;
328 	imgp->execlabel = NULL;
329 	imgp->attr = &attr;
330 	imgp->entry_addr = 0;
331 	imgp->vmspace_destroyed = 0;
332 	imgp->interpreted = 0;
333 	imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
334 	imgp->auxargs = NULL;
335 	imgp->vp = NULL;
336 	imgp->object = NULL;
337 	imgp->firstpage = NULL;
338 	imgp->ps_strings = 0;
339 	imgp->auxarg_size = 0;
340 	imgp->args = args;
341 
342 #ifdef MAC
343 	error = mac_execve_enter(imgp, mac_p);
344 	if (error)
345 		goto exec_fail;
346 #endif
347 
348 	imgp->image_header = NULL;
349 
350 	/*
351 	 * Translate the file name. namei() returns a vnode pointer
352 	 *	in ni_vp amoung other things.
353 	 */
354 	ndp = &nd;
355 	NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
356 	    UIO_SYSSPACE, args->fname, td);
357 
358 interpret:
359 	error = namei(ndp);
360 	if (error)
361 		goto exec_fail;
362 
363 	vfslocked = NDHASGIANT(ndp);
364 	imgp->vp = ndp->ni_vp;
365 
366 	/*
367 	 * Check file permissions (also 'opens' file)
368 	 */
369 	error = exec_check_permissions(imgp);
370 	if (error)
371 		goto exec_fail_dealloc;
372 
373 	imgp->object = imgp->vp->v_object;
374 	if (imgp->object != NULL)
375 		vm_object_reference(imgp->object);
376 
377 	/*
378 	 * Set VV_TEXT now so no one can write to the executable while we're
379 	 * activating it.
380 	 *
381 	 * Remember if this was set before and unset it in case this is not
382 	 * actually an executable image.
383 	 */
384 	textset = imgp->vp->v_vflag & VV_TEXT;
385 	imgp->vp->v_vflag |= VV_TEXT;
386 
387 	error = exec_map_first_page(imgp);
388 	if (error)
389 		goto exec_fail_dealloc;
390 
391 	/*
392 	 *	If the current process has a special image activator it
393 	 *	wants to try first, call it.   For example, emulating shell
394 	 *	scripts differently.
395 	 */
396 	error = -1;
397 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
398 		error = img_first(imgp);
399 
400 	/*
401 	 *	Loop through the list of image activators, calling each one.
402 	 *	An activator returns -1 if there is no match, 0 on success,
403 	 *	and an error otherwise.
404 	 */
405 	for (i = 0; error == -1 && execsw[i]; ++i) {
406 		if (execsw[i]->ex_imgact == NULL ||
407 		    execsw[i]->ex_imgact == img_first) {
408 			continue;
409 		}
410 		error = (*execsw[i]->ex_imgact)(imgp);
411 	}
412 
413 	if (error) {
414 		if (error == -1) {
415 			if (textset == 0)
416 				imgp->vp->v_vflag &= ~VV_TEXT;
417 			error = ENOEXEC;
418 		}
419 		goto exec_fail_dealloc;
420 	}
421 
422 	/*
423 	 * Special interpreter operation, cleanup and loop up to try to
424 	 * activate the interpreter.
425 	 */
426 	if (imgp->interpreted) {
427 		exec_unmap_first_page(imgp);
428 		/*
429 		 * VV_TEXT needs to be unset for scripts.  There is a short
430 		 * period before we determine that something is a script where
431 		 * VV_TEXT will be set. The vnode lock is held over this
432 		 * entire period so nothing should illegitimately be blocked.
433 		 */
434 		imgp->vp->v_vflag &= ~VV_TEXT;
435 		/* free name buffer and old vnode */
436 		NDFREE(ndp, NDF_ONLY_PNBUF);
437 #ifdef MAC
438 		interplabel = mac_vnode_label_alloc();
439 		mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
440 #endif
441 		vput(ndp->ni_vp);
442 		vm_object_deallocate(imgp->object);
443 		imgp->object = NULL;
444 		VFS_UNLOCK_GIANT(vfslocked);
445 		vfslocked = 0;
446 		/* set new name to that of the interpreter */
447 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
448 		    UIO_SYSSPACE, imgp->interpreter_name, td);
449 		goto interpret;
450 	}
451 
452 	/*
453 	 * Copy out strings (args and env) and initialize stack base
454 	 */
455 	if (p->p_sysent->sv_copyout_strings)
456 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
457 	else
458 		stack_base = exec_copyout_strings(imgp);
459 
460 	/*
461 	 * If custom stack fixup routine present for this process
462 	 * let it do the stack setup.
463 	 * Else stuff argument count as first item on stack
464 	 */
465 	if (p->p_sysent->sv_fixup != NULL)
466 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
467 	else
468 		suword(--stack_base, imgp->args->argc);
469 
470 	/*
471 	 * For security and other reasons, the file descriptor table cannot
472 	 * be shared after an exec.
473 	 */
474 	fdunshare(p, td);
475 
476 	/*
477 	 * Malloc things before we need locks.
478 	 */
479 	newcred = crget();
480 	euip = uifind(attr.va_uid);
481 	i = imgp->args->begin_envv - imgp->args->begin_argv;
482 	/* Cache arguments if they fit inside our allowance */
483 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
484 		newargs = pargs_alloc(i);
485 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
486 	}
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(p, td, SIGTRAP, NULL);
663 
664 	/* clear "fork but no exec" flag, as we _are_ execing */
665 	p->p_acflag &= ~AFORK;
666 
667 	/*
668 	 * Free any previous argument cache and replace it with
669 	 * the new argument cache, if any.
670 	 */
671 	oldargs = p->p_args;
672 	p->p_args = newargs;
673 	newargs = NULL;
674 
675 #ifdef	HWPMC_HOOKS
676 	/*
677 	 * Check if system-wide sampling is in effect or if the
678 	 * current process is using PMCs.  If so, do exec() time
679 	 * processing.  This processing needs to happen AFTER the
680 	 * P_INEXEC flag is cleared.
681 	 *
682 	 * The proc lock needs to be released before taking the PMC
683 	 * SX.
684 	 */
685 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
686 		PROC_UNLOCK(p);
687 		pe.pm_credentialschanged = credential_changing;
688 		pe.pm_entryaddr = imgp->entry_addr;
689 
690 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
691 	} else
692 		PROC_UNLOCK(p);
693 #else  /* !HWPMC_HOOKS */
694 	PROC_UNLOCK(p);
695 #endif
696 
697 	/* Set values passed into the program in registers. */
698 	if (p->p_sysent->sv_setregs)
699 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
700 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
701 	else
702 		exec_setregs(td, imgp->entry_addr,
703 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
704 
705 	vfs_mark_atime(imgp->vp, td);
706 
707 done1:
708 	/*
709 	 * Free any resources malloc'd earlier that we didn't use.
710 	 */
711 	uifree(euip);
712 	if (newcred == NULL)
713 		crfree(oldcred);
714 	else
715 		crfree(newcred);
716 	/*
717 	 * Handle deferred decrement of ref counts.
718 	 */
719 	if (textvp != NULL) {
720 		int tvfslocked;
721 
722 		tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
723 		vrele(textvp);
724 		VFS_UNLOCK_GIANT(tvfslocked);
725 	}
726 	if (ndp->ni_vp && error != 0)
727 		vrele(ndp->ni_vp);
728 #ifdef KTRACE
729 	if (tracevp != NULL)
730 		vrele(tracevp);
731 	if (tracecred != NULL)
732 		crfree(tracecred);
733 #endif
734 	if (oldargs != NULL)
735 		pargs_drop(oldargs);
736 	if (newargs != NULL)
737 		pargs_drop(newargs);
738 	if (oldsigacts != NULL)
739 		sigacts_free(oldsigacts);
740 
741 exec_fail_dealloc:
742 
743 	/*
744 	 * free various allocated resources
745 	 */
746 	if (imgp->firstpage != NULL)
747 		exec_unmap_first_page(imgp);
748 
749 	if (imgp->vp != NULL) {
750 		NDFREE(ndp, NDF_ONLY_PNBUF);
751 		vput(imgp->vp);
752 	}
753 
754 	if (imgp->object != NULL)
755 		vm_object_deallocate(imgp->object);
756 
757 	if (error == 0) {
758 		/*
759 		 * Stop the process here if its stop event mask has
760 		 * the S_EXEC bit set.
761 		 */
762 		STOPEVENT(p, S_EXEC, 0);
763 		goto done2;
764 	}
765 
766 exec_fail:
767 	/* we're done here, clear P_INEXEC */
768 	PROC_LOCK(p);
769 	p->p_flag &= ~P_INEXEC;
770 	PROC_UNLOCK(p);
771 
772 done2:
773 #ifdef MAC
774 	mac_execve_exit(imgp);
775 	if (interplabel != NULL)
776 		mac_vnode_label_free(interplabel);
777 #endif
778 	VFS_UNLOCK_GIANT(vfslocked);
779 	exec_free_args(args);
780 
781 	if (error && imgp->vmspace_destroyed) {
782 		/* sorry, no more process anymore. exit gracefully */
783 		exit1(td, W_EXITCODE(0, SIGABRT));
784 		/* NOT REACHED */
785 	}
786 	return (error);
787 }
788 
789 int
790 exec_map_first_page(imgp)
791 	struct image_params *imgp;
792 {
793 	int rv, i;
794 	int initial_pagein;
795 	vm_page_t ma[VM_INITIAL_PAGEIN];
796 	vm_object_t object;
797 
798 	if (imgp->firstpage != NULL)
799 		exec_unmap_first_page(imgp);
800 
801 	object = imgp->vp->v_object;
802 	if (object == NULL)
803 		return (EACCES);
804 	VM_OBJECT_LOCK(object);
805 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
806 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
807 		initial_pagein = VM_INITIAL_PAGEIN;
808 		if (initial_pagein > object->size)
809 			initial_pagein = object->size;
810 		for (i = 1; i < initial_pagein; i++) {
811 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
812 				if (ma[i]->valid)
813 					break;
814 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
815 					break;
816 				vm_page_lock_queues();
817 				vm_page_busy(ma[i]);
818 				vm_page_unlock_queues();
819 			} else {
820 				ma[i] = vm_page_alloc(object, i,
821 				    VM_ALLOC_NORMAL);
822 				if (ma[i] == NULL)
823 					break;
824 			}
825 		}
826 		initial_pagein = i;
827 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
828 		ma[0] = vm_page_lookup(object, 0);
829 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
830 		    (ma[0]->valid == 0)) {
831 			if (ma[0]) {
832 				vm_page_lock_queues();
833 				vm_page_free(ma[0]);
834 				vm_page_unlock_queues();
835 			}
836 			VM_OBJECT_UNLOCK(object);
837 			return (EIO);
838 		}
839 	}
840 	vm_page_lock_queues();
841 	vm_page_hold(ma[0]);
842 	vm_page_wakeup(ma[0]);
843 	vm_page_unlock_queues();
844 	VM_OBJECT_UNLOCK(object);
845 
846 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
847 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
848 
849 	return (0);
850 }
851 
852 void
853 exec_unmap_first_page(imgp)
854 	struct image_params *imgp;
855 {
856 	vm_page_t m;
857 
858 	if (imgp->firstpage != NULL) {
859 		m = sf_buf_page(imgp->firstpage);
860 		sf_buf_free(imgp->firstpage);
861 		imgp->firstpage = NULL;
862 		vm_page_lock_queues();
863 		vm_page_unhold(m);
864 		vm_page_unlock_queues();
865 	}
866 }
867 
868 /*
869  * Destroy old address space, and allocate a new stack
870  *	The new stack is only SGROWSIZ large because it is grown
871  *	automatically in trap.c.
872  */
873 int
874 exec_new_vmspace(imgp, sv)
875 	struct image_params *imgp;
876 	struct sysentvec *sv;
877 {
878 	int error;
879 	struct proc *p = imgp->proc;
880 	struct vmspace *vmspace = p->p_vmspace;
881 	vm_offset_t stack_addr;
882 	vm_map_t map;
883 
884 	imgp->vmspace_destroyed = 1;
885 
886 	/* Called with Giant held, do not depend on it! */
887 	EVENTHANDLER_INVOKE(process_exec, p);
888 
889 	/*
890 	 * Here is as good a place as any to do any resource limit cleanups.
891 	 * This is needed if a 64 bit binary exec's a 32 bit binary - the
892 	 * data size limit may need to be changed to a value that makes
893 	 * sense for the 32 bit binary.
894 	 */
895 	if (sv->sv_fixlimits != NULL)
896 		sv->sv_fixlimits(p);
897 
898 	/*
899 	 * Blow away entire process VM, if address space not shared,
900 	 * otherwise, create a new VM space so that other threads are
901 	 * not disrupted
902 	 */
903 	map = &vmspace->vm_map;
904 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
905 	    vm_map_max(map) == sv->sv_maxuser) {
906 		shmexit(vmspace);
907 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
908 		    vm_map_max(map));
909 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
910 	} else {
911 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
912 		vmspace = p->p_vmspace;
913 		map = &vmspace->vm_map;
914 	}
915 
916 	/* Allocate a new stack */
917 	stack_addr = sv->sv_usrstack - maxssiz;
918 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
919 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
920 	if (error)
921 		return (error);
922 
923 #ifdef __ia64__
924 	/* Allocate a new register stack */
925 	stack_addr = IA64_BACKINGSTORE;
926 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
927 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
928 	if (error)
929 		return (error);
930 #endif
931 
932 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
933 	 * VM_STACK case, but they are still used to monitor the size of the
934 	 * process stack so we can check the stack rlimit.
935 	 */
936 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
937 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
938 
939 	return (0);
940 }
941 
942 /*
943  * Copy out argument and environment strings from the old process
944  *	address space into the temporary string buffer.
945  */
946 int
947 exec_copyin_args(struct image_args *args, char *fname,
948     enum uio_seg segflg, char **argv, char **envv)
949 {
950 	char *argp, *envp;
951 	int error;
952 	size_t length;
953 
954 	error = 0;
955 
956 	bzero(args, sizeof(*args));
957 	if (argv == NULL)
958 		return (EFAULT);
959 	/*
960 	 * Allocate temporary demand zeroed space for argument and
961 	 *	environment strings:
962 	 *
963 	 * o ARG_MAX for argument and environment;
964 	 * o MAXSHELLCMDLEN for the name of interpreters.
965 	 */
966 	args->buf = (char *) kmem_alloc_wait(exec_map,
967 	    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
968 	if (args->buf == NULL)
969 		return (ENOMEM);
970 	args->begin_argv = args->buf;
971 	args->endp = args->begin_argv;
972 	args->stringspace = ARG_MAX;
973 
974 	args->fname = args->buf + ARG_MAX;
975 
976 	/*
977 	 * Copy the file name.
978 	 */
979 	error = (segflg == UIO_SYSSPACE) ?
980 	    copystr(fname, args->fname, PATH_MAX, &length) :
981 	    copyinstr(fname, args->fname, PATH_MAX, &length);
982 	if (error != 0)
983 		return (error);
984 
985 	/*
986 	 * extract arguments first
987 	 */
988 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
989 		if (argp == (caddr_t) -1)
990 			return (EFAULT);
991 		if ((error = copyinstr(argp, args->endp,
992 		    args->stringspace, &length))) {
993 			if (error == ENAMETOOLONG)
994 				return (E2BIG);
995 			return (error);
996 		}
997 		args->stringspace -= length;
998 		args->endp += length;
999 		args->argc++;
1000 	}
1001 
1002 	args->begin_envv = args->endp;
1003 
1004 	/*
1005 	 * extract environment strings
1006 	 */
1007 	if (envv) {
1008 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1009 			if (envp == (caddr_t)-1)
1010 				return (EFAULT);
1011 			if ((error = copyinstr(envp, args->endp,
1012 			    args->stringspace, &length))) {
1013 				if (error == ENAMETOOLONG)
1014 					return (E2BIG);
1015 				return (error);
1016 			}
1017 			args->stringspace -= length;
1018 			args->endp += length;
1019 			args->envc++;
1020 		}
1021 	}
1022 
1023 	return (0);
1024 }
1025 
1026 static void
1027 exec_free_args(struct image_args *args)
1028 {
1029 
1030 	if (args->buf) {
1031 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1032 		    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1033 		args->buf = NULL;
1034 	}
1035 }
1036 
1037 /*
1038  * Copy strings out to the new process address space, constructing
1039  *	new arg and env vector tables. Return a pointer to the base
1040  *	so that it can be used as the initial stack pointer.
1041  */
1042 register_t *
1043 exec_copyout_strings(imgp)
1044 	struct image_params *imgp;
1045 {
1046 	int argc, envc;
1047 	char **vectp;
1048 	char *stringp, *destp;
1049 	register_t *stack_base;
1050 	struct ps_strings *arginfo;
1051 	struct proc *p;
1052 	int szsigcode;
1053 
1054 	/*
1055 	 * Calculate string base and vector table pointers.
1056 	 * Also deal with signal trampoline code for this exec type.
1057 	 */
1058 	p = imgp->proc;
1059 	szsigcode = 0;
1060 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1061 	if (p->p_sysent->sv_szsigcode != NULL)
1062 		szsigcode = *(p->p_sysent->sv_szsigcode);
1063 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1064 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1065 
1066 	/*
1067 	 * install sigcode
1068 	 */
1069 	if (szsigcode)
1070 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1071 		    szsigcode), szsigcode);
1072 
1073 	/*
1074 	 * If we have a valid auxargs ptr, prepare some room
1075 	 * on the stack.
1076 	 */
1077 	if (imgp->auxargs) {
1078 		/*
1079 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1080 		 * lower compatibility.
1081 		 */
1082 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1083 		    (AT_COUNT * 2);
1084 		/*
1085 		 * The '+ 2' is for the null pointers at the end of each of
1086 		 * the arg and env vector sets,and imgp->auxarg_size is room
1087 		 * for argument of Runtime loader.
1088 		 */
1089 		vectp = (char **)(destp - (imgp->args->argc +
1090 		    imgp->args->envc + 2 + imgp->auxarg_size) *
1091 		    sizeof(char *));
1092 
1093 	} else {
1094 		/*
1095 		 * The '+ 2' is for the null pointers at the end of each of
1096 		 * the arg and env vector sets
1097 		 */
1098 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1099 		    sizeof(char *));
1100 	}
1101 
1102 	/*
1103 	 * vectp also becomes our initial stack base
1104 	 */
1105 	stack_base = (register_t *)vectp;
1106 
1107 	stringp = imgp->args->begin_argv;
1108 	argc = imgp->args->argc;
1109 	envc = imgp->args->envc;
1110 
1111 	/*
1112 	 * Copy out strings - arguments and environment.
1113 	 */
1114 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1115 
1116 	/*
1117 	 * Fill in "ps_strings" struct for ps, w, etc.
1118 	 */
1119 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1120 	suword(&arginfo->ps_nargvstr, argc);
1121 
1122 	/*
1123 	 * Fill in argument portion of vector table.
1124 	 */
1125 	for (; argc > 0; --argc) {
1126 		suword(vectp++, (long)(intptr_t)destp);
1127 		while (*stringp++ != 0)
1128 			destp++;
1129 		destp++;
1130 	}
1131 
1132 	/* a null vector table pointer separates the argp's from the envp's */
1133 	suword(vectp++, 0);
1134 
1135 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1136 	suword(&arginfo->ps_nenvstr, envc);
1137 
1138 	/*
1139 	 * Fill in environment portion of vector table.
1140 	 */
1141 	for (; envc > 0; --envc) {
1142 		suword(vectp++, (long)(intptr_t)destp);
1143 		while (*stringp++ != 0)
1144 			destp++;
1145 		destp++;
1146 	}
1147 
1148 	/* end of vector table is a null pointer */
1149 	suword(vectp, 0);
1150 
1151 	return (stack_base);
1152 }
1153 
1154 /*
1155  * Check permissions of file to execute.
1156  *	Called with imgp->vp locked.
1157  *	Return 0 for success or error code on failure.
1158  */
1159 int
1160 exec_check_permissions(imgp)
1161 	struct image_params *imgp;
1162 {
1163 	struct vnode *vp = imgp->vp;
1164 	struct vattr *attr = imgp->attr;
1165 	struct thread *td;
1166 	int error;
1167 
1168 	td = curthread;			/* XXXKSE */
1169 
1170 	/* Get file attributes */
1171 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1172 	if (error)
1173 		return (error);
1174 
1175 #ifdef MAC
1176 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1177 	if (error)
1178 		return (error);
1179 #endif
1180 
1181 	/*
1182 	 * 1) Check if file execution is disabled for the filesystem that this
1183 	 *	file resides on.
1184 	 * 2) Insure that at least one execute bit is on - otherwise root
1185 	 *	will always succeed, and we don't want to happen unless the
1186 	 *	file really is executable.
1187 	 * 3) Insure that the file is a regular file.
1188 	 */
1189 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1190 	    ((attr->va_mode & 0111) == 0) ||
1191 	    (attr->va_type != VREG))
1192 		return (EACCES);
1193 
1194 	/*
1195 	 * Zero length files can't be exec'd
1196 	 */
1197 	if (attr->va_size == 0)
1198 		return (ENOEXEC);
1199 
1200 	/*
1201 	 *  Check for execute permission to file based on current credentials.
1202 	 */
1203 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1204 	if (error)
1205 		return (error);
1206 
1207 	/*
1208 	 * Check number of open-for-writes on the file and deny execution
1209 	 * if there are any.
1210 	 */
1211 	if (vp->v_writecount)
1212 		return (ETXTBSY);
1213 
1214 	/*
1215 	 * Call filesystem specific open routine (which does nothing in the
1216 	 * general case).
1217 	 */
1218 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1219 	return (error);
1220 }
1221 
1222 /*
1223  * Exec handler registration
1224  */
1225 int
1226 exec_register(execsw_arg)
1227 	const struct execsw *execsw_arg;
1228 {
1229 	const struct execsw **es, **xs, **newexecsw;
1230 	int count = 2;	/* New slot and trailing NULL */
1231 
1232 	if (execsw)
1233 		for (es = execsw; *es; es++)
1234 			count++;
1235 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1236 	if (newexecsw == NULL)
1237 		return (ENOMEM);
1238 	xs = newexecsw;
1239 	if (execsw)
1240 		for (es = execsw; *es; es++)
1241 			*xs++ = *es;
1242 	*xs++ = execsw_arg;
1243 	*xs = NULL;
1244 	if (execsw)
1245 		free(execsw, M_TEMP);
1246 	execsw = newexecsw;
1247 	return (0);
1248 }
1249 
1250 int
1251 exec_unregister(execsw_arg)
1252 	const struct execsw *execsw_arg;
1253 {
1254 	const struct execsw **es, **xs, **newexecsw;
1255 	int count = 1;
1256 
1257 	if (execsw == NULL)
1258 		panic("unregister with no handlers left?\n");
1259 
1260 	for (es = execsw; *es; es++) {
1261 		if (*es == execsw_arg)
1262 			break;
1263 	}
1264 	if (*es == NULL)
1265 		return (ENOENT);
1266 	for (es = execsw; *es; es++)
1267 		if (*es != execsw_arg)
1268 			count++;
1269 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1270 	if (newexecsw == NULL)
1271 		return (ENOMEM);
1272 	xs = newexecsw;
1273 	for (es = execsw; *es; es++)
1274 		if (*es != execsw_arg)
1275 			*xs++ = *es;
1276 	*xs = NULL;
1277 	if (execsw)
1278 		free(execsw, M_TEMP);
1279 	execsw = newexecsw;
1280 	return (0);
1281 }
1282