xref: /freebsd/sys/kern/kern_exec.c (revision 6780ab54325a71e7e70112b11657973edde8655e)
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  * $FreeBSD$
27  */
28 
29 #include "opt_ktrace.h"
30 #include "opt_mac.h"
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/lock.h>
35 #include <sys/mutex.h>
36 #include <sys/sysproto.h>
37 #include <sys/signalvar.h>
38 #include <sys/kernel.h>
39 #include <sys/mac.h>
40 #include <sys/mount.h>
41 #include <sys/filedesc.h>
42 #include <sys/fcntl.h>
43 #include <sys/acct.h>
44 #include <sys/exec.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/wait.h>
48 #include <sys/malloc.h>
49 #include <sys/proc.h>
50 #include <sys/pioctl.h>
51 #include <sys/namei.h>
52 #include <sys/sysent.h>
53 #include <sys/shm.h>
54 #include <sys/sysctl.h>
55 #include <sys/user.h>
56 #include <sys/vnode.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60 
61 #include <vm/vm.h>
62 #include <vm/vm_param.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_page.h>
65 #include <vm/vm_map.h>
66 #include <vm/vm_kern.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_pager.h>
70 
71 #include <machine/reg.h>
72 
73 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
74 
75 static MALLOC_DEFINE(M_ATEXEC, "atexec", "atexec callback");
76 
77 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
78 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
79 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
80 static int kern_execve(struct thread *td, char *fname, char **argv,
81 	char **envv, struct mac *mac_p);
82 
83 /*
84  * callout list for things to do at exec time
85  */
86 struct execlist {
87 	execlist_fn function;
88 	TAILQ_ENTRY(execlist) next;
89 };
90 
91 TAILQ_HEAD(exec_list_head, execlist);
92 static struct exec_list_head exec_list = TAILQ_HEAD_INITIALIZER(exec_list);
93 
94 /* XXX This should be vm_size_t. */
95 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
96     NULL, 0, sysctl_kern_ps_strings, "LU", "");
97 
98 /* XXX This should be vm_size_t. */
99 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
100     NULL, 0, sysctl_kern_usrstack, "LU", "");
101 
102 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
103     NULL, 0, sysctl_kern_stackprot, "I", "");
104 
105 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
106 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
107     &ps_arg_cache_limit, 0, "");
108 
109 int ps_argsopen = 1;
110 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
111 
112 #ifdef __ia64__
113 /* XXX HACK */
114 static int regstkpages = 256;
115 SYSCTL_INT(_machdep, OID_AUTO, regstkpages, CTLFLAG_RW, &regstkpages, 0, "");
116 #endif
117 
118 static int
119 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
120 {
121 	struct proc *p;
122 
123 	p = curproc;
124 	return (SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
125 	   sizeof(p->p_sysent->sv_psstrings)));
126 }
127 
128 static int
129 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
130 {
131 	struct proc *p;
132 
133 	p = curproc;
134 	return (SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
135 	    sizeof(p->p_sysent->sv_usrstack)));
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 /*
155  * In-kernel implementation of execve().  All arguments are assumed to be
156  * userspace pointers from the passed thread.
157  *
158  * MPSAFE
159  */
160 static int
161 kern_execve(td, fname, argv, envv, mac_p)
162 	struct thread *td;
163 	char *fname;
164 	char **argv;
165 	char **envv;
166 	struct mac *mac_p;
167 {
168 	struct proc *p = td->td_proc;
169 	struct nameidata nd, *ndp;
170 	struct ucred *newcred = NULL, *oldcred;
171 	struct uidinfo *euip;
172 	register_t *stack_base;
173 	int error, len, i;
174 	struct image_params image_params, *imgp;
175 	struct vattr attr;
176 	int (*img_first)(struct image_params *);
177 	struct pargs *oldargs = NULL, *newargs = NULL;
178 	struct procsig *oldprocsig, *newprocsig;
179 #ifdef KTRACE
180 	struct vnode *tracevp = NULL;
181 #endif
182 	struct vnode *textvp = NULL;
183 	int credential_changing;
184 	int textset;
185 #ifdef MAC
186 	struct label interplabel;	/* label of the interpreted vnode */
187 	struct label execlabel;		/* optional label argument */
188 	int will_transition, interplabelvalid = 0;
189 #endif
190 
191 	imgp = &image_params;
192 
193 	/*
194 	 * Lock the process and set the P_INEXEC flag to indicate that
195 	 * it should be left alone until we're done here.  This is
196 	 * necessary to avoid race conditions - e.g. in ptrace() -
197 	 * that might allow a local user to illicitly obtain elevated
198 	 * privileges.
199 	 */
200 	PROC_LOCK(p);
201 	KASSERT((p->p_flag & P_INEXEC) == 0,
202 	    ("%s(): process already has P_INEXEC flag", __func__));
203 	if (p->p_flag & P_KSES) {
204 		if (thread_single(SINGLE_EXIT)) {
205 			PROC_UNLOCK(p);
206 			return (ERESTART);	/* Try again later. */
207 		}
208 		/*
209 		 * If we get here all other threads are dead,
210 		 * so unset the associated flags and lose KSE mode.
211 		 */
212 		p->p_flag &= ~P_KSES;
213 		td->td_flags &= ~TDF_UNBOUND;
214 		td->td_mailbox = NULL;
215 		td->td_kse->ke_mailbox = NULL;
216 		td->td_kse->ke_flags &= ~KEF_DOUPCALL;
217 		thread_single_end();
218 	}
219 	p->p_flag |= P_INEXEC;
220 	PROC_UNLOCK(p);
221 
222 	/*
223 	 * Initialize part of the common data
224 	 */
225 	imgp->proc = p;
226 	imgp->userspace_argv = argv;
227 	imgp->userspace_envv = envv;
228 	imgp->execlabel = NULL;
229 	imgp->attr = &attr;
230 	imgp->argc = imgp->envc = 0;
231 	imgp->argv0 = NULL;
232 	imgp->entry_addr = 0;
233 	imgp->vmspace_destroyed = 0;
234 	imgp->interpreted = 0;
235 	imgp->interpreter_name[0] = '\0';
236 	imgp->auxargs = NULL;
237 	imgp->vp = NULL;
238 	imgp->object = NULL;
239 	imgp->firstpage = NULL;
240 	imgp->ps_strings = 0;
241 	imgp->auxarg_size = 0;
242 
243 #ifdef MAC
244 	error = mac_execve_enter(imgp, mac_p, &execlabel);
245 	if (error) {
246 		mtx_lock(&Giant);
247 		goto exec_fail;
248 	}
249 #endif
250 
251 	/*
252 	 * Allocate temporary demand zeroed space for argument and
253 	 *	environment strings
254 	 */
255 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX +
256 	    PAGE_SIZE);
257 	if (imgp->stringbase == NULL) {
258 		error = ENOMEM;
259 		mtx_lock(&Giant);
260 		goto exec_fail;
261 	}
262 	imgp->stringp = imgp->stringbase;
263 	imgp->stringspace = ARG_MAX;
264 	imgp->image_header = imgp->stringbase + ARG_MAX;
265 
266 	/*
267 	 * Translate the file name. namei() returns a vnode pointer
268 	 *	in ni_vp amoung other things.
269 	 */
270 	ndp = &nd;
271 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
272 	    UIO_USERSPACE, fname, td);
273 
274 	mtx_lock(&Giant);
275 interpret:
276 
277 	error = namei(ndp);
278 	if (error) {
279 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
280 		    ARG_MAX + PAGE_SIZE);
281 		goto exec_fail;
282 	}
283 
284 	imgp->vp = ndp->ni_vp;
285 	imgp->fname = fname;
286 
287 	/*
288 	 * Check file permissions (also 'opens' file)
289 	 */
290 	error = exec_check_permissions(imgp);
291 	if (error)
292 		goto exec_fail_dealloc;
293 
294 	if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0)
295 		vm_object_reference(imgp->object);
296 
297 	/*
298 	 * Set VV_TEXT now so no one can write to the executable while we're
299 	 * activating it.
300 	 *
301 	 * Remember if this was set before and unset it in case this is not
302 	 * actually an executable image.
303 	 */
304 	textset = imgp->vp->v_vflag & VV_TEXT;
305 	imgp->vp->v_vflag |= VV_TEXT;
306 
307 	error = exec_map_first_page(imgp);
308 	if (error)
309 		goto exec_fail_dealloc;
310 
311 	/*
312 	 *	If the current process has a special image activator it
313 	 *	wants to try first, call it.   For example, emulating shell
314 	 *	scripts differently.
315 	 */
316 	error = -1;
317 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
318 		error = img_first(imgp);
319 
320 	/*
321 	 *	Loop through the list of image activators, calling each one.
322 	 *	An activator returns -1 if there is no match, 0 on success,
323 	 *	and an error otherwise.
324 	 */
325 	for (i = 0; error == -1 && execsw[i]; ++i) {
326 		if (execsw[i]->ex_imgact == NULL ||
327 		    execsw[i]->ex_imgact == img_first) {
328 			continue;
329 		}
330 		error = (*execsw[i]->ex_imgact)(imgp);
331 	}
332 
333 	if (error) {
334 		if (error == -1) {
335 			if (textset == 0)
336 				imgp->vp->v_vflag &= ~VV_TEXT;
337 			error = ENOEXEC;
338 		}
339 		goto exec_fail_dealloc;
340 	}
341 
342 	/*
343 	 * Special interpreter operation, cleanup and loop up to try to
344 	 * activate the interpreter.
345 	 */
346 	if (imgp->interpreted) {
347 		exec_unmap_first_page(imgp);
348 		/*
349 		 * VV_TEXT needs to be unset for scripts.  There is a short
350 		 * period before we determine that something is a script where
351 		 * VV_TEXT will be set. The vnode lock is held over this
352 		 * entire period so nothing should illegitimately be blocked.
353 		 */
354 		imgp->vp->v_vflag &= ~VV_TEXT;
355 		/* free name buffer and old vnode */
356 		NDFREE(ndp, NDF_ONLY_PNBUF);
357 #ifdef MAC
358 		mac_init_vnode_label(&interplabel);
359 		mac_copy_vnode_label(&ndp->ni_vp->v_label, &interplabel);
360 		interplabelvalid = 1;
361 #endif
362 		vput(ndp->ni_vp);
363 		vm_object_deallocate(imgp->object);
364 		imgp->object = NULL;
365 		/* set new name to that of the interpreter */
366 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
367 		    UIO_SYSSPACE, imgp->interpreter_name, td);
368 		goto interpret;
369 	}
370 
371 	/*
372 	 * Copy out strings (args and env) and initialize stack base
373 	 */
374 	if (p->p_sysent->sv_copyout_strings)
375 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
376 	else
377 		stack_base = exec_copyout_strings(imgp);
378 
379 	/*
380 	 * If custom stack fixup routine present for this process
381 	 * let it do the stack setup.
382 	 * Else stuff argument count as first item on stack
383 	 */
384 	if (p->p_sysent->sv_fixup)
385 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
386 	else
387 		suword(--stack_base, imgp->argc);
388 
389 	/*
390 	 * For security and other reasons, the file descriptor table cannot
391 	 * be shared after an exec.
392 	 */
393 	FILEDESC_LOCK(p->p_fd);
394 	if (p->p_fd->fd_refcnt > 1) {
395 		struct filedesc *tmp;
396 
397 		tmp = fdcopy(td->td_proc->p_fd);
398 		FILEDESC_UNLOCK(p->p_fd);
399 		fdfree(td);
400 		p->p_fd = tmp;
401 	} else
402 		FILEDESC_UNLOCK(p->p_fd);
403 
404 	/*
405 	 * Malloc things before we need locks.
406 	 */
407 	newcred = crget();
408 	euip = uifind(attr.va_uid);
409 	i = imgp->endargs - imgp->stringbase;
410 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
411 		newargs = pargs_alloc(i);
412 
413 	/* close files on exec */
414 	fdcloseexec(td);
415 
416 	/* Get a reference to the vnode prior to locking the proc */
417 	VREF(ndp->ni_vp);
418 
419 	/*
420 	 * For security and other reasons, signal handlers cannot
421 	 * be shared after an exec. The new process gets a copy of the old
422 	 * handlers. In execsigs(), the new process will have its signals
423 	 * reset.
424 	 */
425 	PROC_LOCK(p);
426 	mp_fixme("procsig needs a lock");
427 	if (p->p_procsig->ps_refcnt > 1) {
428 		oldprocsig = p->p_procsig;
429 		PROC_UNLOCK(p);
430 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
431 		    M_SUBPROC, 0);
432 		bcopy(oldprocsig, newprocsig, sizeof(*newprocsig));
433 		newprocsig->ps_refcnt = 1;
434 		oldprocsig->ps_refcnt--;
435 		PROC_LOCK(p);
436 		p->p_procsig = newprocsig;
437 		if (p->p_sigacts == &p->p_uarea->u_sigacts)
438 			panic("shared procsig but private sigacts?");
439 
440 		p->p_uarea->u_sigacts = *p->p_sigacts;
441 		p->p_sigacts = &p->p_uarea->u_sigacts;
442 	}
443 	/* Stop profiling */
444 	stopprofclock(p);
445 
446 	/* reset caught signals */
447 	execsigs(p);
448 
449 	/* name this process - nameiexec(p, ndp) */
450 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
451 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
452 	p->p_comm[len] = 0;
453 
454 	/*
455 	 * mark as execed, wakeup the process that vforked (if any) and tell
456 	 * it that it now has its own resources back
457 	 */
458 	p->p_flag |= P_EXEC;
459 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
460 		p->p_flag &= ~P_PPWAIT;
461 		wakeup(p->p_pptr);
462 	}
463 
464 	/*
465 	 * Implement image setuid/setgid.
466 	 *
467 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
468 	 * the process is being traced.
469 	 *
470 	 * XXXMAC: For the time being, use NOSUID to also prohibit
471 	 * transitions on the file system.
472 	 */
473 	oldcred = p->p_ucred;
474 	credential_changing = 0;
475 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
476 	    attr.va_uid;
477 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
478 	    attr.va_gid;
479 #ifdef MAC
480 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
481 	    interplabelvalid ? &interplabel : NULL, imgp);
482 	credential_changing |= will_transition;
483 #endif
484 
485 	if (credential_changing &&
486 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
487 	    (p->p_flag & P_TRACED) == 0) {
488 		/*
489 		 * Turn off syscall tracing for set-id programs, except for
490 		 * root.  Record any set-id flags first to make sure that
491 		 * we do not regain any tracing during a possible block.
492 		 */
493 		setsugid(p);
494 #ifdef KTRACE
495 		if (p->p_tracep && suser_cred(oldcred, PRISON_ROOT)) {
496 			mtx_lock(&ktrace_mtx);
497 			p->p_traceflag = 0;
498 			tracevp = p->p_tracep;
499 			p->p_tracep = NULL;
500 			mtx_unlock(&ktrace_mtx);
501 		}
502 #endif
503 		/*
504 		 * Close any file descriptors 0..2 that reference procfs,
505 		 * then make sure file descriptors 0..2 are in use.
506 		 *
507 		 * setugidsafety() may call closef() and then pfind()
508 		 * which may grab the process lock.
509 		 * fdcheckstd() may call falloc() which may block to
510 		 * allocate memory, so temporarily drop the process lock.
511 		 */
512 		PROC_UNLOCK(p);
513 		setugidsafety(td);
514 		error = fdcheckstd(td);
515 		if (error != 0)
516 			goto done1;
517 		PROC_LOCK(p);
518 		/*
519 		 * Set the new credentials.
520 		 */
521 		crcopy(newcred, oldcred);
522 		if (attr.va_mode & VSUID)
523 			change_euid(newcred, euip);
524 		if (attr.va_mode & VSGID)
525 			change_egid(newcred, attr.va_gid);
526 #ifdef MAC
527 		if (will_transition) {
528 			mac_execve_transition(oldcred, newcred, imgp->vp,
529 			    interplabelvalid ? &interplabel : NULL, imgp);
530 		}
531 #endif
532 		/*
533 		 * Implement correct POSIX saved-id behavior.
534 		 *
535 		 * XXXMAC: Note that the current logic will save the
536 		 * uid and gid if a MAC domain transition occurs, even
537 		 * though maybe it shouldn't.
538 		 */
539 		change_svuid(newcred, newcred->cr_uid);
540 		change_svgid(newcred, newcred->cr_gid);
541 		p->p_ucred = newcred;
542 		newcred = NULL;
543 	} else {
544 		if (oldcred->cr_uid == oldcred->cr_ruid &&
545 		    oldcred->cr_gid == oldcred->cr_rgid)
546 			p->p_flag &= ~P_SUGID;
547 		/*
548 		 * Implement correct POSIX saved-id behavior.
549 		 *
550 		 * XXX: It's not clear that the existing behavior is
551 		 * POSIX-compliant.  A number of sources indicate that the
552 		 * saved uid/gid should only be updated if the new ruid is
553 		 * not equal to the old ruid, or the new euid is not equal
554 		 * to the old euid and the new euid is not equal to the old
555 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
556 		 * Also, this code uses the new (replaced) euid and egid as
557 		 * the source, which may or may not be the right ones to use.
558 		 */
559 		if (oldcred->cr_svuid != oldcred->cr_uid ||
560 		    oldcred->cr_svgid != oldcred->cr_gid) {
561 			crcopy(newcred, oldcred);
562 			change_svuid(newcred, newcred->cr_uid);
563 			change_svgid(newcred, newcred->cr_gid);
564 			p->p_ucred = newcred;
565 			newcred = NULL;
566 		}
567 	}
568 
569 	/*
570 	 * Store the vp for use in procfs.  This vnode was referenced prior
571 	 * to locking the proc lock.
572 	 */
573 	textvp = p->p_textvp;
574 	p->p_textvp = ndp->ni_vp;
575 
576 	/*
577 	 * Notify others that we exec'd, and clear the P_INEXEC flag
578 	 * as we're now a bona fide freshly-execed process.
579 	 */
580 	KNOTE(&p->p_klist, NOTE_EXEC);
581 	p->p_flag &= ~P_INEXEC;
582 
583 	/*
584 	 * If tracing the process, trap to debugger so breakpoints
585 	 * can be set before the program executes.
586 	 */
587 	if (p->p_flag & P_TRACED)
588 		psignal(p, SIGTRAP);
589 
590 	/* clear "fork but no exec" flag, as we _are_ execing */
591 	p->p_acflag &= ~AFORK;
592 
593 	/* Free any previous argument cache */
594 	oldargs = p->p_args;
595 	p->p_args = NULL;
596 
597 	/* Cache arguments if they fit inside our allowance */
598 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
599 		bcopy(imgp->stringbase, newargs->ar_args, i);
600 		p->p_args = newargs;
601 		newargs = NULL;
602 	}
603 	PROC_UNLOCK(p);
604 
605 	/* Set values passed into the program in registers. */
606 	if (p->p_sysent->sv_setregs)
607 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
608 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
609 	else
610 		exec_setregs(td, imgp->entry_addr,
611 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
612 
613 done1:
614 	/*
615 	 * Free any resources malloc'd earlier that we didn't use.
616 	 */
617 	uifree(euip);
618 	if (newcred == NULL)
619 		crfree(oldcred);
620 	else
621 		crfree(newcred);
622 	/*
623 	 * Handle deferred decrement of ref counts.
624 	 */
625 	if (textvp != NULL)
626 		vrele(textvp);
627 	if (ndp->ni_vp && error != 0)
628 		vrele(ndp->ni_vp);
629 #ifdef KTRACE
630 	if (tracevp != NULL)
631 		vrele(tracevp);
632 #endif
633 	if (oldargs != NULL)
634 		pargs_drop(oldargs);
635 	if (newargs != NULL)
636 		pargs_drop(newargs);
637 
638 exec_fail_dealloc:
639 
640 	/*
641 	 * free various allocated resources
642 	 */
643 	if (imgp->firstpage)
644 		exec_unmap_first_page(imgp);
645 
646 	if (imgp->vp) {
647 		NDFREE(ndp, NDF_ONLY_PNBUF);
648 		vput(imgp->vp);
649 	}
650 
651 	if (imgp->stringbase != NULL)
652 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
653 		    ARG_MAX + PAGE_SIZE);
654 
655 	if (imgp->object)
656 		vm_object_deallocate(imgp->object);
657 
658 	if (error == 0) {
659 		/*
660 		 * Stop the process here if its stop event mask has
661 		 * the S_EXEC bit set.
662 		 */
663 		STOPEVENT(p, S_EXEC, 0);
664 		goto done2;
665 	}
666 
667 exec_fail:
668 	/* we're done here, clear P_INEXEC */
669 	PROC_LOCK(p);
670 	p->p_flag &= ~P_INEXEC;
671 	PROC_UNLOCK(p);
672 
673 	if (imgp->vmspace_destroyed) {
674 		/* sorry, no more process anymore. exit gracefully */
675 #ifdef MAC
676 		mac_execve_exit(imgp);
677 		if (interplabelvalid)
678 			mac_destroy_vnode_label(&interplabel);
679 #endif
680 		exit1(td, W_EXITCODE(0, SIGABRT));
681 		/* NOT REACHED */
682 		error = 0;
683 	}
684 done2:
685 #ifdef MAC
686 	mac_execve_exit(imgp);
687 	if (interplabelvalid)
688 		mac_destroy_vnode_label(&interplabel);
689 #endif
690 	mtx_unlock(&Giant);
691 	return (error);
692 }
693 
694 #ifndef _SYS_SYSPROTO_H_
695 struct execve_args {
696         char    *fname;
697         char    **argv;
698         char    **envv;
699 };
700 #endif
701 
702 /*
703  * MPSAFE
704  */
705 int
706 execve(td, uap)
707 	struct thread *td;
708 	struct execve_args /* {
709 		char *fname;
710 		char **argv;
711 		char **envv;
712 	} */ *uap;
713 {
714 
715 	return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
716 }
717 
718 #ifndef _SYS_SYSPROTO_H_
719 struct __mac_execve_args {
720 	char	*fname;
721 	char	**argv;
722 	char	**envv;
723 	struct mac	*mac_p;
724 };
725 #endif
726 
727 /*
728  * MPSAFE
729  */
730 int
731 __mac_execve(td, uap)
732 	struct thread *td;
733 	struct __mac_execve_args /* {
734 		char *fname;
735 		char **argv;
736 		char **envv;
737 		struct mac *mac_p;
738 	} */ *uap;
739 {
740 
741 #ifdef MAC
742 	return (kern_execve(td, uap->fname, uap->argv, uap->envv,
743 	    uap->mac_p));
744 #else
745 	return (ENOSYS);
746 #endif
747 }
748 
749 int
750 exec_map_first_page(imgp)
751 	struct image_params *imgp;
752 {
753 	int rv, i;
754 	int initial_pagein;
755 	vm_page_t ma[VM_INITIAL_PAGEIN];
756 	vm_object_t object;
757 
758 	GIANT_REQUIRED;
759 
760 	if (imgp->firstpage) {
761 		exec_unmap_first_page(imgp);
762 	}
763 
764 	VOP_GETVOBJECT(imgp->vp, &object);
765 
766 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
767 
768 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
769 		initial_pagein = VM_INITIAL_PAGEIN;
770 		if (initial_pagein > object->size)
771 			initial_pagein = object->size;
772 		for (i = 1; i < initial_pagein; i++) {
773 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
774 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
775 					break;
776 				if (ma[i]->valid)
777 					break;
778 				vm_page_lock_queues();
779 				vm_page_busy(ma[i]);
780 				vm_page_unlock_queues();
781 			} else {
782 				ma[i] = vm_page_alloc(object, i,
783 				    VM_ALLOC_NORMAL);
784 				if (ma[i] == NULL)
785 					break;
786 			}
787 		}
788 		initial_pagein = i;
789 
790 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
791 		ma[0] = vm_page_lookup(object, 0);
792 
793 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
794 		    (ma[0]->valid == 0)) {
795 			if (ma[0]) {
796 				vm_page_lock_queues();
797 				pmap_remove_all(ma[0]);
798 				vm_page_free(ma[0]);
799 				vm_page_unlock_queues();
800 			}
801 			return (EIO);
802 		}
803 	}
804 	vm_page_lock_queues();
805 	vm_page_wire(ma[0]);
806 	vm_page_wakeup(ma[0]);
807 	vm_page_unlock_queues();
808 
809 	pmap_qenter((vm_offset_t)imgp->image_header, ma, 1);
810 	imgp->firstpage = ma[0];
811 
812 	return (0);
813 }
814 
815 void
816 exec_unmap_first_page(imgp)
817 	struct image_params *imgp;
818 {
819 	GIANT_REQUIRED;
820 
821 	if (imgp->firstpage) {
822 		pmap_qremove((vm_offset_t)imgp->image_header, 1);
823 		vm_page_lock_queues();
824 		vm_page_unwire(imgp->firstpage, 1);
825 		vm_page_unlock_queues();
826 		imgp->firstpage = NULL;
827 	}
828 }
829 
830 /*
831  * Destroy old address space, and allocate a new stack
832  *	The new stack is only SGROWSIZ large because it is grown
833  *	automatically in trap.c.
834  */
835 int
836 exec_new_vmspace(imgp, sv)
837 	struct image_params *imgp;
838 	struct sysentvec *sv;
839 {
840 	int error;
841 	struct execlist *ep;
842 	struct proc *p = imgp->proc;
843 	struct vmspace *vmspace = p->p_vmspace;
844 	vm_offset_t stack_addr;
845 	vm_map_t map;
846 
847 	GIANT_REQUIRED;
848 
849 	stack_addr = sv->sv_usrstack - maxssiz;
850 
851 	imgp->vmspace_destroyed = 1;
852 
853 	/*
854 	 * Perform functions registered with at_exec().
855 	 */
856 	TAILQ_FOREACH(ep, &exec_list, next)
857 		(*ep->function)(p);
858 
859 	/*
860 	 * Blow away entire process VM, if address space not shared,
861 	 * otherwise, create a new VM space so that other threads are
862 	 * not disrupted
863 	 */
864 	map = &vmspace->vm_map;
865 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
866 	    vm_map_max(map) == sv->sv_maxuser) {
867 		shmexit(vmspace);
868 		vm_page_lock_queues();
869 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
870 		    vm_map_max(map));
871 		vm_page_unlock_queues();
872 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
873 	} else {
874 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
875 		vmspace = p->p_vmspace;
876 		map = &vmspace->vm_map;
877 	}
878 
879 	/* Allocate a new stack */
880 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
881 	    sv->sv_stackprot, VM_PROT_ALL, 0);
882 	if (error)
883 		return (error);
884 
885 #ifdef __ia64__
886 	{
887 		/*
888 		 * Allocate backing store. We really need something
889 		 * similar to vm_map_stack which can allow the backing
890 		 * store to grow upwards. This will do for now.
891 		 */
892 		vm_offset_t bsaddr;
893 		bsaddr = p->p_sysent->sv_usrstack - 2 * maxssiz;
894 		error = vm_map_find(map, 0, 0, &bsaddr,
895 		    regstkpages * PAGE_SIZE, 0, VM_PROT_ALL, VM_PROT_ALL, 0);
896 		FIRST_THREAD_IN_PROC(p)->td_md.md_bspstore = bsaddr;
897 	}
898 #endif
899 
900 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
901 	 * VM_STACK case, but they are still used to monitor the size of the
902 	 * process stack so we can check the stack rlimit.
903 	 */
904 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
905 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
906 
907 	return (0);
908 }
909 
910 /*
911  * Copy out argument and environment strings from the old process
912  *	address space into the temporary string buffer.
913  */
914 int
915 exec_extract_strings(imgp)
916 	struct image_params *imgp;
917 {
918 	char	**argv, **envv;
919 	char	*argp, *envp;
920 	int	error;
921 	size_t	length;
922 
923 	/*
924 	 * extract arguments first
925 	 */
926 
927 	argv = imgp->userspace_argv;
928 
929 	if (argv) {
930 		argp = (caddr_t)(intptr_t)fuword(argv);
931 		if (argp == (caddr_t)-1)
932 			return (EFAULT);
933 		if (argp)
934 			argv++;
935 		if (imgp->argv0)
936 			argp = imgp->argv0;
937 		if (argp) {
938 			do {
939 				if (argp == (caddr_t)-1)
940 					return (EFAULT);
941 				if ((error = copyinstr(argp, imgp->stringp,
942 				    imgp->stringspace, &length))) {
943 					if (error == ENAMETOOLONG)
944 						return (E2BIG);
945 					return (error);
946 				}
947 				imgp->stringspace -= length;
948 				imgp->stringp += length;
949 				imgp->argc++;
950 			} while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
951 		}
952 	}
953 
954 	imgp->endargs = imgp->stringp;
955 
956 	/*
957 	 * extract environment strings
958 	 */
959 
960 	envv = imgp->userspace_envv;
961 
962 	if (envv) {
963 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
964 			if (envp == (caddr_t)-1)
965 				return (EFAULT);
966 			if ((error = copyinstr(envp, imgp->stringp,
967 			    imgp->stringspace, &length))) {
968 				if (error == ENAMETOOLONG)
969 					return (E2BIG);
970 				return (error);
971 			}
972 			imgp->stringspace -= length;
973 			imgp->stringp += length;
974 			imgp->envc++;
975 		}
976 	}
977 
978 	return (0);
979 }
980 
981 /*
982  * Copy strings out to the new process address space, constructing
983  *	new arg and env vector tables. Return a pointer to the base
984  *	so that it can be used as the initial stack pointer.
985  */
986 register_t *
987 exec_copyout_strings(imgp)
988 	struct image_params *imgp;
989 {
990 	int argc, envc;
991 	char **vectp;
992 	char *stringp, *destp;
993 	register_t *stack_base;
994 	struct ps_strings *arginfo;
995 	struct proc *p;
996 	int szsigcode;
997 
998 	/*
999 	 * Calculate string base and vector table pointers.
1000 	 * Also deal with signal trampoline code for this exec type.
1001 	 */
1002 	p = imgp->proc;
1003 	szsigcode = 0;
1004 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1005 	if (p->p_sysent->sv_szsigcode != NULL)
1006 		szsigcode = *(p->p_sysent->sv_szsigcode);
1007 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1008 	    roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
1009 
1010 	/*
1011 	 * install sigcode
1012 	 */
1013 	if (szsigcode)
1014 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1015 		    szsigcode), szsigcode);
1016 
1017 	/*
1018 	 * If we have a valid auxargs ptr, prepare some room
1019 	 * on the stack.
1020 	 */
1021 	if (imgp->auxargs) {
1022 		/*
1023 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1024 		 * lower compatibility.
1025 		 */
1026 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1027 		    (AT_COUNT * 2);
1028 		/*
1029 		 * The '+ 2' is for the null pointers at the end of each of
1030 		 * the arg and env vector sets,and imgp->auxarg_size is room
1031 		 * for argument of Runtime loader.
1032 		 */
1033 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1034 		    imgp->auxarg_size) * sizeof(char *));
1035 
1036 	} else
1037 		/*
1038 		 * The '+ 2' is for the null pointers at the end of each of
1039 		 * the arg and env vector sets
1040 		 */
1041 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1042 		    sizeof(char *));
1043 
1044 	/*
1045 	 * vectp also becomes our initial stack base
1046 	 */
1047 	stack_base = (register_t *)vectp;
1048 
1049 	stringp = imgp->stringbase;
1050 	argc = imgp->argc;
1051 	envc = imgp->envc;
1052 
1053 	/*
1054 	 * Copy out strings - arguments and environment.
1055 	 */
1056 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1057 
1058 	/*
1059 	 * Fill in "ps_strings" struct for ps, w, etc.
1060 	 */
1061 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1062 	suword(&arginfo->ps_nargvstr, argc);
1063 
1064 	/*
1065 	 * Fill in argument portion of vector table.
1066 	 */
1067 	for (; argc > 0; --argc) {
1068 		suword(vectp++, (long)(intptr_t)destp);
1069 		while (*stringp++ != 0)
1070 			destp++;
1071 		destp++;
1072 	}
1073 
1074 	/* a null vector table pointer separates the argp's from the envp's */
1075 	suword(vectp++, 0);
1076 
1077 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1078 	suword(&arginfo->ps_nenvstr, envc);
1079 
1080 	/*
1081 	 * Fill in environment portion of vector table.
1082 	 */
1083 	for (; envc > 0; --envc) {
1084 		suword(vectp++, (long)(intptr_t)destp);
1085 		while (*stringp++ != 0)
1086 			destp++;
1087 		destp++;
1088 	}
1089 
1090 	/* end of vector table is a null pointer */
1091 	suword(vectp, 0);
1092 
1093 	return (stack_base);
1094 }
1095 
1096 /*
1097  * Check permissions of file to execute.
1098  *	Called with imgp->vp locked.
1099  *	Return 0 for success or error code on failure.
1100  */
1101 int
1102 exec_check_permissions(imgp)
1103 	struct image_params *imgp;
1104 {
1105 	struct vnode *vp = imgp->vp;
1106 	struct vattr *attr = imgp->attr;
1107 	struct thread *td;
1108 	int error;
1109 
1110 	td = curthread;			/* XXXKSE */
1111 
1112 	/* Get file attributes */
1113 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1114 	if (error)
1115 		return (error);
1116 
1117 #ifdef MAC
1118 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1119 	if (error)
1120 		return (error);
1121 #endif
1122 
1123 	/*
1124 	 * 1) Check if file execution is disabled for the filesystem that this
1125 	 *	file resides on.
1126 	 * 2) Insure that at least one execute bit is on - otherwise root
1127 	 *	will always succeed, and we don't want to happen unless the
1128 	 *	file really is executable.
1129 	 * 3) Insure that the file is a regular file.
1130 	 */
1131 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1132 	    ((attr->va_mode & 0111) == 0) ||
1133 	    (attr->va_type != VREG))
1134 		return (EACCES);
1135 
1136 	/*
1137 	 * Zero length files can't be exec'd
1138 	 */
1139 	if (attr->va_size == 0)
1140 		return (ENOEXEC);
1141 
1142 	/*
1143 	 *  Check for execute permission to file based on current credentials.
1144 	 */
1145 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1146 	if (error)
1147 		return (error);
1148 
1149 	/*
1150 	 * Check number of open-for-writes on the file and deny execution
1151 	 * if there are any.
1152 	 */
1153 	if (vp->v_writecount)
1154 		return (ETXTBSY);
1155 
1156 	/*
1157 	 * Call filesystem specific open routine (which does nothing in the
1158 	 * general case).
1159 	 */
1160 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td);
1161 	return (error);
1162 }
1163 
1164 /*
1165  * Exec handler registration
1166  */
1167 int
1168 exec_register(execsw_arg)
1169 	const struct execsw *execsw_arg;
1170 {
1171 	const struct execsw **es, **xs, **newexecsw;
1172 	int count = 2;	/* New slot and trailing NULL */
1173 
1174 	if (execsw)
1175 		for (es = execsw; *es; es++)
1176 			count++;
1177 	newexecsw = malloc(count * sizeof(*es), M_TEMP, 0);
1178 	if (newexecsw == NULL)
1179 		return (ENOMEM);
1180 	xs = newexecsw;
1181 	if (execsw)
1182 		for (es = execsw; *es; es++)
1183 			*xs++ = *es;
1184 	*xs++ = execsw_arg;
1185 	*xs = NULL;
1186 	if (execsw)
1187 		free(execsw, M_TEMP);
1188 	execsw = newexecsw;
1189 	return (0);
1190 }
1191 
1192 int
1193 exec_unregister(execsw_arg)
1194 	const struct execsw *execsw_arg;
1195 {
1196 	const struct execsw **es, **xs, **newexecsw;
1197 	int count = 1;
1198 
1199 	if (execsw == NULL)
1200 		panic("unregister with no handlers left?\n");
1201 
1202 	for (es = execsw; *es; es++) {
1203 		if (*es == execsw_arg)
1204 			break;
1205 	}
1206 	if (*es == NULL)
1207 		return (ENOENT);
1208 	for (es = execsw; *es; es++)
1209 		if (*es != execsw_arg)
1210 			count++;
1211 	newexecsw = malloc(count * sizeof(*es), M_TEMP, 0);
1212 	if (newexecsw == NULL)
1213 		return (ENOMEM);
1214 	xs = newexecsw;
1215 	for (es = execsw; *es; es++)
1216 		if (*es != execsw_arg)
1217 			*xs++ = *es;
1218 	*xs = NULL;
1219 	if (execsw)
1220 		free(execsw, M_TEMP);
1221 	execsw = newexecsw;
1222 	return (0);
1223 }
1224 
1225 int
1226 at_exec(function)
1227 	execlist_fn function;
1228 {
1229 	struct execlist *ep;
1230 
1231 #ifdef INVARIANTS
1232 	/* Be noisy if the programmer has lost track of things */
1233 	if (rm_at_exec(function))
1234 		printf("WARNING: exec callout entry (%p) already present\n",
1235 		    function);
1236 #endif
1237 	ep = malloc(sizeof(*ep), M_ATEXEC, M_NOWAIT);
1238 	if (ep == NULL)
1239 		return (ENOMEM);
1240 	ep->function = function;
1241 	TAILQ_INSERT_TAIL(&exec_list, ep, next);
1242 	return (0);
1243 }
1244 
1245 /*
1246  * Scan the exec callout list for the given item and remove it.
1247  * Returns the number of items removed (0 or 1)
1248  */
1249 int
1250 rm_at_exec(function)
1251 	execlist_fn function;
1252 {
1253 	struct execlist *ep;
1254 
1255 	TAILQ_FOREACH(ep, &exec_list, next) {
1256 		if (ep->function == function) {
1257 			TAILQ_REMOVE(&exec_list, ep, next);
1258 			free(ep, M_ATEXEC);
1259 			return (1);
1260 		}
1261 	}
1262 	return (0);
1263 }
1264