xref: /freebsd/sys/kern/kern_exec.c (revision 2357939bc239bd5334a169b62313806178dd8f30)
1 /*
2  * Copyright (c) 1993, David Greenman
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_ktrace.h"
31 #include "opt_mac.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/eventhandler.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/sysproto.h>
39 #include <sys/signalvar.h>
40 #include <sys/kernel.h>
41 #include <sys/mac.h>
42 #include <sys/mount.h>
43 #include <sys/filedesc.h>
44 #include <sys/fcntl.h>
45 #include <sys/acct.h>
46 #include <sys/exec.h>
47 #include <sys/imgact.h>
48 #include <sys/imgact_elf.h>
49 #include <sys/wait.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 #include <sys/pioctl.h>
53 #include <sys/namei.h>
54 #include <sys/sf_buf.h>
55 #include <sys/sysent.h>
56 #include <sys/shm.h>
57 #include <sys/sysctl.h>
58 #include <sys/user.h>
59 #include <sys/vnode.h>
60 #ifdef KTRACE
61 #include <sys/ktrace.h>
62 #endif
63 
64 #include <vm/vm.h>
65 #include <vm/vm_param.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_map.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_pager.h>
73 
74 #include <machine/reg.h>
75 
76 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
77 
78 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
79 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
80 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
81 static int kern_execve(struct thread *td, char *fname, char **argv,
82 	char **envv, struct mac *mac_p);
83 
84 /* XXX This should be vm_size_t. */
85 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
86     NULL, 0, sysctl_kern_ps_strings, "LU", "");
87 
88 /* XXX This should be vm_size_t. */
89 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
90     NULL, 0, sysctl_kern_usrstack, "LU", "");
91 
92 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
93     NULL, 0, sysctl_kern_stackprot, "I", "");
94 
95 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
96 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
97     &ps_arg_cache_limit, 0, "");
98 
99 static int
100 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
101 {
102 	struct proc *p;
103 	int error;
104 
105 	p = curproc;
106 #if defined(__amd64__) || defined(__ia64__)
107 	if (req->oldlen == sizeof(unsigned int)) {
108 		unsigned int val;
109 		val = (unsigned int)p->p_sysent->sv_psstrings;
110 		error = SYSCTL_OUT(req, &val, sizeof(val));
111 	} else
112 #endif
113 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
114 		   sizeof(p->p_sysent->sv_psstrings));
115 	return error;
116 }
117 
118 static int
119 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
120 {
121 	struct proc *p;
122 	int error;
123 
124 	p = curproc;
125 #if defined(__amd64__) || defined(__ia64__)
126 	if (req->oldlen == sizeof(unsigned int)) {
127 		unsigned int val;
128 		val = (unsigned int)p->p_sysent->sv_usrstack;
129 		error = SYSCTL_OUT(req, &val, sizeof(val));
130 	} else
131 #endif
132 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
133 		    sizeof(p->p_sysent->sv_usrstack));
134 	return error;
135 }
136 
137 static int
138 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
139 {
140 	struct proc *p;
141 
142 	p = curproc;
143 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
144 	    sizeof(p->p_sysent->sv_stackprot)));
145 }
146 
147 /*
148  * Each of the items is a pointer to a `const struct execsw', hence the
149  * double pointer here.
150  */
151 static const struct execsw **execsw;
152 
153 #ifndef _SYS_SYSPROTO_H_
154 struct execve_args {
155         char    *fname;
156         char    **argv;
157         char    **envv;
158 };
159 #endif
160 
161 /*
162  * MPSAFE
163  */
164 int
165 execve(td, uap)
166 	struct thread *td;
167 	struct execve_args /* {
168 		char *fname;
169 		char **argv;
170 		char **envv;
171 	} */ *uap;
172 {
173 
174 	return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
175 }
176 
177 #ifndef _SYS_SYSPROTO_H_
178 struct __mac_execve_args {
179 	char	*fname;
180 	char	**argv;
181 	char	**envv;
182 	struct mac	*mac_p;
183 };
184 #endif
185 
186 /*
187  * MPSAFE
188  */
189 int
190 __mac_execve(td, uap)
191 	struct thread *td;
192 	struct __mac_execve_args /* {
193 		char *fname;
194 		char **argv;
195 		char **envv;
196 		struct mac *mac_p;
197 	} */ *uap;
198 {
199 
200 #ifdef MAC
201 	return (kern_execve(td, uap->fname, uap->argv, uap->envv,
202 	    uap->mac_p));
203 #else
204 	return (ENOSYS);
205 #endif
206 }
207 
208 /*
209  * In-kernel implementation of execve().  All arguments are assumed to be
210  * userspace pointers from the passed thread.
211  *
212  * MPSAFE
213  */
214 static int
215 kern_execve(td, fname, argv, envv, mac_p)
216 	struct thread *td;
217 	char *fname;
218 	char **argv;
219 	char **envv;
220 	struct mac *mac_p;
221 {
222 	struct proc *p = td->td_proc;
223 	struct nameidata nd, *ndp;
224 	struct ucred *newcred = NULL, *oldcred;
225 	struct uidinfo *euip;
226 	register_t *stack_base;
227 	int error, len, i;
228 	struct image_params image_params, *imgp;
229 	struct vattr attr;
230 	int (*img_first)(struct image_params *);
231 	struct pargs *oldargs = NULL, *newargs = NULL;
232 	struct sigacts *oldsigacts, *newsigacts;
233 #ifdef KTRACE
234 	struct vnode *tracevp = NULL;
235 	struct ucred *tracecred = NULL;
236 #endif
237 	struct vnode *textvp = NULL;
238 	int credential_changing;
239 	int textset;
240 #ifdef MAC
241 	struct label *interplabel = NULL;
242 	int will_transition;
243 #endif
244 
245 	imgp = &image_params;
246 
247 	/*
248 	 * Lock the process and set the P_INEXEC flag to indicate that
249 	 * it should be left alone until we're done here.  This is
250 	 * necessary to avoid race conditions - e.g. in ptrace() -
251 	 * that might allow a local user to illicitly obtain elevated
252 	 * privileges.
253 	 */
254 	PROC_LOCK(p);
255 	KASSERT((p->p_flag & P_INEXEC) == 0,
256 	    ("%s(): process already has P_INEXEC flag", __func__));
257 	if (p->p_flag & P_SA || p->p_numthreads > 1) {
258 		if (thread_single(SINGLE_EXIT)) {
259 			PROC_UNLOCK(p);
260 			mtx_unlock(&Giant);
261 			return (ERESTART);	/* Try again later. */
262 		}
263 		/*
264 		 * If we get here all other threads are dead,
265 		 * so unset the associated flags and lose KSE mode.
266 		 */
267 		p->p_flag &= ~P_SA;
268 		td->td_mailbox = NULL;
269 		thread_single_end();
270 	}
271 	p->p_flag |= P_INEXEC;
272 	PROC_UNLOCK(p);
273 
274 	/*
275 	 * Initialize part of the common data
276 	 */
277 	imgp->proc = p;
278 	imgp->userspace_argv = argv;
279 	imgp->userspace_envv = envv;
280 	imgp->execlabel = NULL;
281 	imgp->attr = &attr;
282 	imgp->argc = imgp->envc = 0;
283 	imgp->argv0 = NULL;
284 	imgp->entry_addr = 0;
285 	imgp->vmspace_destroyed = 0;
286 	imgp->interpreted = 0;
287 	imgp->interpreter_name[0] = '\0';
288 	imgp->auxargs = NULL;
289 	imgp->vp = NULL;
290 	imgp->object = NULL;
291 	imgp->firstpage = NULL;
292 	imgp->ps_strings = 0;
293 	imgp->auxarg_size = 0;
294 
295 #ifdef MAC
296 	error = mac_execve_enter(imgp, mac_p);
297 	if (error) {
298 		mtx_lock(&Giant);
299 		goto exec_fail;
300 	}
301 #endif
302 
303 	/*
304 	 * Allocate temporary demand zeroed space for argument and
305 	 *	environment strings
306 	 */
307 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX);
308 	if (imgp->stringbase == NULL) {
309 		error = ENOMEM;
310 		mtx_lock(&Giant);
311 		goto exec_fail;
312 	}
313 	imgp->stringp = imgp->stringbase;
314 	imgp->stringspace = ARG_MAX;
315 	imgp->image_header = NULL;
316 
317 	/*
318 	 * Translate the file name. namei() returns a vnode pointer
319 	 *	in ni_vp amoung other things.
320 	 */
321 	ndp = &nd;
322 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
323 	    UIO_USERSPACE, fname, td);
324 
325 	mtx_lock(&Giant);
326 interpret:
327 
328 	error = namei(ndp);
329 	if (error) {
330 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
331 		    ARG_MAX);
332 		goto exec_fail;
333 	}
334 
335 	imgp->vp = ndp->ni_vp;
336 	imgp->fname = fname;
337 
338 	/*
339 	 * Check file permissions (also 'opens' file)
340 	 */
341 	error = exec_check_permissions(imgp);
342 	if (error)
343 		goto exec_fail_dealloc;
344 
345 	if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0)
346 		vm_object_reference(imgp->object);
347 
348 	/*
349 	 * Set VV_TEXT now so no one can write to the executable while we're
350 	 * activating it.
351 	 *
352 	 * Remember if this was set before and unset it in case this is not
353 	 * actually an executable image.
354 	 */
355 	textset = imgp->vp->v_vflag & VV_TEXT;
356 	imgp->vp->v_vflag |= VV_TEXT;
357 
358 	error = exec_map_first_page(imgp);
359 	if (error)
360 		goto exec_fail_dealloc;
361 
362 	/*
363 	 *	If the current process has a special image activator it
364 	 *	wants to try first, call it.   For example, emulating shell
365 	 *	scripts differently.
366 	 */
367 	error = -1;
368 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
369 		error = img_first(imgp);
370 
371 	/*
372 	 *	Loop through the list of image activators, calling each one.
373 	 *	An activator returns -1 if there is no match, 0 on success,
374 	 *	and an error otherwise.
375 	 */
376 	for (i = 0; error == -1 && execsw[i]; ++i) {
377 		if (execsw[i]->ex_imgact == NULL ||
378 		    execsw[i]->ex_imgact == img_first) {
379 			continue;
380 		}
381 		error = (*execsw[i]->ex_imgact)(imgp);
382 	}
383 
384 	if (error) {
385 		if (error == -1) {
386 			if (textset == 0)
387 				imgp->vp->v_vflag &= ~VV_TEXT;
388 			error = ENOEXEC;
389 		}
390 		goto exec_fail_dealloc;
391 	}
392 
393 	/*
394 	 * Special interpreter operation, cleanup and loop up to try to
395 	 * activate the interpreter.
396 	 */
397 	if (imgp->interpreted) {
398 		exec_unmap_first_page(imgp);
399 		/*
400 		 * VV_TEXT needs to be unset for scripts.  There is a short
401 		 * period before we determine that something is a script where
402 		 * VV_TEXT will be set. The vnode lock is held over this
403 		 * entire period so nothing should illegitimately be blocked.
404 		 */
405 		imgp->vp->v_vflag &= ~VV_TEXT;
406 		/* free name buffer and old vnode */
407 		NDFREE(ndp, NDF_ONLY_PNBUF);
408 #ifdef MAC
409 		interplabel = mac_vnode_label_alloc();
410 		mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
411 #endif
412 		vput(ndp->ni_vp);
413 		vm_object_deallocate(imgp->object);
414 		imgp->object = NULL;
415 		/* set new name to that of the interpreter */
416 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
417 		    UIO_SYSSPACE, imgp->interpreter_name, td);
418 		goto interpret;
419 	}
420 
421 	/*
422 	 * Copy out strings (args and env) and initialize stack base
423 	 */
424 	if (p->p_sysent->sv_copyout_strings)
425 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
426 	else
427 		stack_base = exec_copyout_strings(imgp);
428 
429 	/*
430 	 * If custom stack fixup routine present for this process
431 	 * let it do the stack setup.
432 	 * Else stuff argument count as first item on stack
433 	 */
434 	if (p->p_sysent->sv_fixup != NULL)
435 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
436 	else
437 		suword(--stack_base, imgp->argc);
438 
439 	/*
440 	 * For security and other reasons, the file descriptor table cannot
441 	 * be shared after an exec.
442 	 */
443 	FILEDESC_LOCK(p->p_fd);
444 	if (p->p_fd->fd_refcnt > 1) {
445 		struct filedesc *tmp;
446 
447 		tmp = fdcopy(td->td_proc->p_fd);
448 		FILEDESC_UNLOCK(p->p_fd);
449 		fdfree(td);
450 		p->p_fd = tmp;
451 	} else
452 		FILEDESC_UNLOCK(p->p_fd);
453 
454 	/*
455 	 * Malloc things before we need locks.
456 	 */
457 	newcred = crget();
458 	euip = uifind(attr.va_uid);
459 	i = imgp->endargs - imgp->stringbase;
460 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
461 		newargs = pargs_alloc(i);
462 
463 	/* close files on exec */
464 	fdcloseexec(td);
465 
466 	/* Get a reference to the vnode prior to locking the proc */
467 	VREF(ndp->ni_vp);
468 
469 	/*
470 	 * For security and other reasons, signal handlers cannot
471 	 * be shared after an exec. The new process gets a copy of the old
472 	 * handlers. In execsigs(), the new process will have its signals
473 	 * reset.
474 	 */
475 	PROC_LOCK(p);
476 	if (sigacts_shared(p->p_sigacts)) {
477 		oldsigacts = p->p_sigacts;
478 		PROC_UNLOCK(p);
479 		newsigacts = sigacts_alloc();
480 		sigacts_copy(newsigacts, oldsigacts);
481 		PROC_LOCK(p);
482 		p->p_sigacts = newsigacts;
483 	} else
484 		oldsigacts = NULL;
485 
486 	/* Stop profiling */
487 	stopprofclock(p);
488 
489 	/* reset caught signals */
490 	execsigs(p);
491 
492 	/* name this process - nameiexec(p, ndp) */
493 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
494 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
495 	p->p_comm[len] = 0;
496 
497 	/*
498 	 * mark as execed, wakeup the process that vforked (if any) and tell
499 	 * it that it now has its own resources back
500 	 */
501 	p->p_flag |= P_EXEC;
502 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
503 		p->p_flag &= ~P_PPWAIT;
504 		wakeup(p->p_pptr);
505 	}
506 
507 	/*
508 	 * Implement image setuid/setgid.
509 	 *
510 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
511 	 * the process is being traced.
512 	 *
513 	 * XXXMAC: For the time being, use NOSUID to also prohibit
514 	 * transitions on the file system.
515 	 */
516 	oldcred = p->p_ucred;
517 	credential_changing = 0;
518 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
519 	    attr.va_uid;
520 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
521 	    attr.va_gid;
522 #ifdef MAC
523 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
524 	    interplabel, imgp);
525 	credential_changing |= will_transition;
526 #endif
527 
528 	if (credential_changing &&
529 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
530 	    (p->p_flag & P_TRACED) == 0) {
531 		/*
532 		 * Turn off syscall tracing for set-id programs, except for
533 		 * root.  Record any set-id flags first to make sure that
534 		 * we do not regain any tracing during a possible block.
535 		 */
536 		setsugid(p);
537 #ifdef KTRACE
538 		if (p->p_tracevp != NULL && suser_cred(oldcred, PRISON_ROOT)) {
539 			mtx_lock(&ktrace_mtx);
540 			p->p_traceflag = 0;
541 			tracevp = p->p_tracevp;
542 			p->p_tracevp = NULL;
543 			tracecred = p->p_tracecred;
544 			p->p_tracecred = NULL;
545 			mtx_unlock(&ktrace_mtx);
546 		}
547 #endif
548 		/*
549 		 * Close any file descriptors 0..2 that reference procfs,
550 		 * then make sure file descriptors 0..2 are in use.
551 		 *
552 		 * setugidsafety() may call closef() and then pfind()
553 		 * which may grab the process lock.
554 		 * fdcheckstd() may call falloc() which may block to
555 		 * allocate memory, so temporarily drop the process lock.
556 		 */
557 		PROC_UNLOCK(p);
558 		setugidsafety(td);
559 		error = fdcheckstd(td);
560 		if (error != 0)
561 			goto done1;
562 		PROC_LOCK(p);
563 		/*
564 		 * Set the new credentials.
565 		 */
566 		crcopy(newcred, oldcred);
567 		if (attr.va_mode & VSUID)
568 			change_euid(newcred, euip);
569 		if (attr.va_mode & VSGID)
570 			change_egid(newcred, attr.va_gid);
571 #ifdef MAC
572 		if (will_transition) {
573 			mac_execve_transition(oldcred, newcred, imgp->vp,
574 			    interplabel, imgp);
575 		}
576 #endif
577 		/*
578 		 * Implement correct POSIX saved-id behavior.
579 		 *
580 		 * XXXMAC: Note that the current logic will save the
581 		 * uid and gid if a MAC domain transition occurs, even
582 		 * though maybe it shouldn't.
583 		 */
584 		change_svuid(newcred, newcred->cr_uid);
585 		change_svgid(newcred, newcred->cr_gid);
586 		p->p_ucred = newcred;
587 		newcred = NULL;
588 	} else {
589 		if (oldcred->cr_uid == oldcred->cr_ruid &&
590 		    oldcred->cr_gid == oldcred->cr_rgid)
591 			p->p_flag &= ~P_SUGID;
592 		/*
593 		 * Implement correct POSIX saved-id behavior.
594 		 *
595 		 * XXX: It's not clear that the existing behavior is
596 		 * POSIX-compliant.  A number of sources indicate that the
597 		 * saved uid/gid should only be updated if the new ruid is
598 		 * not equal to the old ruid, or the new euid is not equal
599 		 * to the old euid and the new euid is not equal to the old
600 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
601 		 * Also, this code uses the new (replaced) euid and egid as
602 		 * the source, which may or may not be the right ones to use.
603 		 */
604 		if (oldcred->cr_svuid != oldcred->cr_uid ||
605 		    oldcred->cr_svgid != oldcred->cr_gid) {
606 			crcopy(newcred, oldcred);
607 			change_svuid(newcred, newcred->cr_uid);
608 			change_svgid(newcred, newcred->cr_gid);
609 			p->p_ucred = newcred;
610 			newcred = NULL;
611 		}
612 	}
613 
614 	/*
615 	 * Store the vp for use in procfs.  This vnode was referenced prior
616 	 * to locking the proc lock.
617 	 */
618 	textvp = p->p_textvp;
619 	p->p_textvp = ndp->ni_vp;
620 
621 	/*
622 	 * Notify others that we exec'd, and clear the P_INEXEC flag
623 	 * as we're now a bona fide freshly-execed process.
624 	 */
625 	KNOTE(&p->p_klist, NOTE_EXEC);
626 	p->p_flag &= ~P_INEXEC;
627 
628 	/*
629 	 * If tracing the process, trap to debugger so breakpoints
630 	 * can be set before the program executes.
631 	 */
632 	if (p->p_flag & P_TRACED)
633 		psignal(p, SIGTRAP);
634 
635 	/* clear "fork but no exec" flag, as we _are_ execing */
636 	p->p_acflag &= ~AFORK;
637 
638 	/* Free any previous argument cache */
639 	oldargs = p->p_args;
640 	p->p_args = NULL;
641 
642 	/* Cache arguments if they fit inside our allowance */
643 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
644 		bcopy(imgp->stringbase, newargs->ar_args, i);
645 		p->p_args = newargs;
646 		newargs = NULL;
647 	}
648 	PROC_UNLOCK(p);
649 
650 	/* Set values passed into the program in registers. */
651 	if (p->p_sysent->sv_setregs)
652 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
653 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
654 	else
655 		exec_setregs(td, imgp->entry_addr,
656 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
657 
658 done1:
659 	/*
660 	 * Free any resources malloc'd earlier that we didn't use.
661 	 */
662 	uifree(euip);
663 	if (newcred == NULL)
664 		crfree(oldcred);
665 	else
666 		crfree(newcred);
667 	/*
668 	 * Handle deferred decrement of ref counts.
669 	 */
670 	if (textvp != NULL)
671 		vrele(textvp);
672 	if (ndp->ni_vp && error != 0)
673 		vrele(ndp->ni_vp);
674 #ifdef KTRACE
675 	if (tracevp != NULL)
676 		vrele(tracevp);
677 	if (tracecred != NULL)
678 		crfree(tracecred);
679 #endif
680 	if (oldargs != NULL)
681 		pargs_drop(oldargs);
682 	if (newargs != NULL)
683 		pargs_drop(newargs);
684 	if (oldsigacts != NULL)
685 		sigacts_free(oldsigacts);
686 
687 exec_fail_dealloc:
688 
689 	/*
690 	 * free various allocated resources
691 	 */
692 	if (imgp->firstpage != NULL)
693 		exec_unmap_first_page(imgp);
694 
695 	if (imgp->vp != NULL) {
696 		NDFREE(ndp, NDF_ONLY_PNBUF);
697 		vput(imgp->vp);
698 	}
699 
700 	if (imgp->stringbase != NULL)
701 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
702 		    ARG_MAX);
703 
704 	if (imgp->object != NULL)
705 		vm_object_deallocate(imgp->object);
706 
707 	if (error == 0) {
708 		/*
709 		 * Stop the process here if its stop event mask has
710 		 * the S_EXEC bit set.
711 		 */
712 		STOPEVENT(p, S_EXEC, 0);
713 		goto done2;
714 	}
715 
716 exec_fail:
717 	/* we're done here, clear P_INEXEC */
718 	PROC_LOCK(p);
719 	p->p_flag &= ~P_INEXEC;
720 	PROC_UNLOCK(p);
721 
722 	if (imgp->vmspace_destroyed) {
723 		/* sorry, no more process anymore. exit gracefully */
724 #ifdef MAC
725 		mac_execve_exit(imgp);
726 		if (interplabel != NULL)
727 			mac_vnode_label_free(interplabel);
728 #endif
729 		exit1(td, W_EXITCODE(0, SIGABRT));
730 		/* NOT REACHED */
731 		error = 0;
732 	}
733 done2:
734 #ifdef MAC
735 	mac_execve_exit(imgp);
736 	if (interplabel != NULL)
737 		mac_vnode_label_free(interplabel);
738 #endif
739 	mtx_unlock(&Giant);
740 	return (error);
741 }
742 
743 int
744 exec_map_first_page(imgp)
745 	struct image_params *imgp;
746 {
747 	int rv, i;
748 	int initial_pagein;
749 	vm_page_t ma[VM_INITIAL_PAGEIN];
750 	vm_object_t object;
751 
752 	GIANT_REQUIRED;
753 
754 	if (imgp->firstpage != NULL)
755 		exec_unmap_first_page(imgp);
756 
757 	VOP_GETVOBJECT(imgp->vp, &object);
758 	VM_OBJECT_LOCK(object);
759 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
760 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
761 		initial_pagein = VM_INITIAL_PAGEIN;
762 		if (initial_pagein > object->size)
763 			initial_pagein = object->size;
764 		for (i = 1; i < initial_pagein; i++) {
765 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
766 				if (ma[i]->valid)
767 					break;
768 				vm_page_lock_queues();
769 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
770 					vm_page_unlock_queues();
771 					break;
772 				}
773 				vm_page_busy(ma[i]);
774 				vm_page_unlock_queues();
775 			} else {
776 				ma[i] = vm_page_alloc(object, i,
777 				    VM_ALLOC_NORMAL);
778 				if (ma[i] == NULL)
779 					break;
780 			}
781 		}
782 		initial_pagein = i;
783 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
784 		ma[0] = vm_page_lookup(object, 0);
785 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
786 		    (ma[0]->valid == 0)) {
787 			if (ma[0]) {
788 				vm_page_lock_queues();
789 				pmap_remove_all(ma[0]);
790 				vm_page_free(ma[0]);
791 				vm_page_unlock_queues();
792 			}
793 			VM_OBJECT_UNLOCK(object);
794 			return (EIO);
795 		}
796 	}
797 	vm_page_lock_queues();
798 	vm_page_hold(ma[0]);
799 	vm_page_wakeup(ma[0]);
800 	vm_page_unlock_queues();
801 	VM_OBJECT_UNLOCK(object);
802 
803 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
804 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
805 
806 	return (0);
807 }
808 
809 void
810 exec_unmap_first_page(imgp)
811 	struct image_params *imgp;
812 {
813 	vm_page_t m;
814 
815 	if (imgp->firstpage != NULL) {
816 		m = sf_buf_page(imgp->firstpage);
817 		sf_buf_free(imgp->firstpage);
818 		imgp->firstpage = NULL;
819 		vm_page_lock_queues();
820 		vm_page_unhold(m);
821 		vm_page_unlock_queues();
822 	}
823 }
824 
825 /*
826  * Destroy old address space, and allocate a new stack
827  *	The new stack is only SGROWSIZ large because it is grown
828  *	automatically in trap.c.
829  */
830 int
831 exec_new_vmspace(imgp, sv)
832 	struct image_params *imgp;
833 	struct sysentvec *sv;
834 {
835 	int error;
836 	struct proc *p = imgp->proc;
837 	struct vmspace *vmspace = p->p_vmspace;
838 	vm_offset_t stack_addr;
839 	vm_map_t map;
840 
841 	GIANT_REQUIRED;
842 
843 	imgp->vmspace_destroyed = 1;
844 
845 	/* Called with Giant held, do not depend on it! */
846 	EVENTHANDLER_INVOKE(process_exec, p);
847 
848 	/*
849 	 * Here is as good a place as any to do any resource limit cleanups.
850 	 * This is needed if a 64 bit binary exec's a 32 bit binary - the
851 	 * data size limit may need to be changed to a value that makes
852 	 * sense for the 32 bit binary.
853 	 */
854 	if (sv->sv_fixlimits != NULL)
855 		sv->sv_fixlimits(imgp);
856 
857 	/*
858 	 * Blow away entire process VM, if address space not shared,
859 	 * otherwise, create a new VM space so that other threads are
860 	 * not disrupted
861 	 */
862 	map = &vmspace->vm_map;
863 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
864 	    vm_map_max(map) == sv->sv_maxuser) {
865 		shmexit(vmspace);
866 		vm_page_lock_queues();
867 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
868 		    vm_map_max(map));
869 		vm_page_unlock_queues();
870 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
871 	} else {
872 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
873 		vmspace = p->p_vmspace;
874 		map = &vmspace->vm_map;
875 	}
876 
877 	/* Allocate a new stack */
878 	stack_addr = sv->sv_usrstack - maxssiz;
879 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
880 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
881 	if (error)
882 		return (error);
883 
884 #ifdef __ia64__
885 	/* Allocate a new register stack */
886 	stack_addr = IA64_BACKINGSTORE;
887 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
888 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
889 	if (error)
890 		return (error);
891 #endif
892 
893 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
894 	 * VM_STACK case, but they are still used to monitor the size of the
895 	 * process stack so we can check the stack rlimit.
896 	 */
897 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
898 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
899 
900 	return (0);
901 }
902 
903 /*
904  * Copy out argument and environment strings from the old process
905  *	address space into the temporary string buffer.
906  */
907 int
908 exec_extract_strings(imgp)
909 	struct image_params *imgp;
910 {
911 	char	**argv, **envv;
912 	char	*argp, *envp;
913 	int	error;
914 	size_t	length;
915 
916 	/*
917 	 * extract arguments first
918 	 */
919 
920 	argv = imgp->userspace_argv;
921 
922 	if (argv) {
923 		argp = (caddr_t)(intptr_t)fuword(argv);
924 		if (argp == (caddr_t)-1)
925 			return (EFAULT);
926 		if (argp)
927 			argv++;
928 		if (imgp->argv0)
929 			argp = imgp->argv0;
930 		if (argp) {
931 			do {
932 				if (argp == (caddr_t)-1)
933 					return (EFAULT);
934 				if ((error = copyinstr(argp, imgp->stringp,
935 				    imgp->stringspace, &length))) {
936 					if (error == ENAMETOOLONG)
937 						return (E2BIG);
938 					return (error);
939 				}
940 				imgp->stringspace -= length;
941 				imgp->stringp += length;
942 				imgp->argc++;
943 			} while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
944 		}
945 	} else
946 		return (EFAULT);
947 
948 	imgp->endargs = imgp->stringp;
949 
950 	/*
951 	 * extract environment strings
952 	 */
953 
954 	envv = imgp->userspace_envv;
955 
956 	if (envv) {
957 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
958 			if (envp == (caddr_t)-1)
959 				return (EFAULT);
960 			if ((error = copyinstr(envp, imgp->stringp,
961 			    imgp->stringspace, &length))) {
962 				if (error == ENAMETOOLONG)
963 					return (E2BIG);
964 				return (error);
965 			}
966 			imgp->stringspace -= length;
967 			imgp->stringp += length;
968 			imgp->envc++;
969 		}
970 	}
971 
972 	return (0);
973 }
974 
975 /*
976  * Copy strings out to the new process address space, constructing
977  *	new arg and env vector tables. Return a pointer to the base
978  *	so that it can be used as the initial stack pointer.
979  */
980 register_t *
981 exec_copyout_strings(imgp)
982 	struct image_params *imgp;
983 {
984 	int argc, envc;
985 	char **vectp;
986 	char *stringp, *destp;
987 	register_t *stack_base;
988 	struct ps_strings *arginfo;
989 	struct proc *p;
990 	int szsigcode;
991 
992 	/*
993 	 * Calculate string base and vector table pointers.
994 	 * Also deal with signal trampoline code for this exec type.
995 	 */
996 	p = imgp->proc;
997 	szsigcode = 0;
998 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
999 	if (p->p_sysent->sv_szsigcode != NULL)
1000 		szsigcode = *(p->p_sysent->sv_szsigcode);
1001 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1002 	    roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
1003 
1004 	/*
1005 	 * install sigcode
1006 	 */
1007 	if (szsigcode)
1008 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1009 		    szsigcode), szsigcode);
1010 
1011 	/*
1012 	 * If we have a valid auxargs ptr, prepare some room
1013 	 * on the stack.
1014 	 */
1015 	if (imgp->auxargs) {
1016 		/*
1017 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1018 		 * lower compatibility.
1019 		 */
1020 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1021 		    (AT_COUNT * 2);
1022 		/*
1023 		 * The '+ 2' is for the null pointers at the end of each of
1024 		 * the arg and env vector sets,and imgp->auxarg_size is room
1025 		 * for argument of Runtime loader.
1026 		 */
1027 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1028 		    imgp->auxarg_size) * sizeof(char *));
1029 
1030 	} else
1031 		/*
1032 		 * The '+ 2' is for the null pointers at the end of each of
1033 		 * the arg and env vector sets
1034 		 */
1035 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1036 		    sizeof(char *));
1037 
1038 	/*
1039 	 * vectp also becomes our initial stack base
1040 	 */
1041 	stack_base = (register_t *)vectp;
1042 
1043 	stringp = imgp->stringbase;
1044 	argc = imgp->argc;
1045 	envc = imgp->envc;
1046 
1047 	/*
1048 	 * Copy out strings - arguments and environment.
1049 	 */
1050 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1051 
1052 	/*
1053 	 * Fill in "ps_strings" struct for ps, w, etc.
1054 	 */
1055 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1056 	suword(&arginfo->ps_nargvstr, argc);
1057 
1058 	/*
1059 	 * Fill in argument portion of vector table.
1060 	 */
1061 	for (; argc > 0; --argc) {
1062 		suword(vectp++, (long)(intptr_t)destp);
1063 		while (*stringp++ != 0)
1064 			destp++;
1065 		destp++;
1066 	}
1067 
1068 	/* a null vector table pointer separates the argp's from the envp's */
1069 	suword(vectp++, 0);
1070 
1071 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1072 	suword(&arginfo->ps_nenvstr, envc);
1073 
1074 	/*
1075 	 * Fill in environment portion of vector table.
1076 	 */
1077 	for (; envc > 0; --envc) {
1078 		suword(vectp++, (long)(intptr_t)destp);
1079 		while (*stringp++ != 0)
1080 			destp++;
1081 		destp++;
1082 	}
1083 
1084 	/* end of vector table is a null pointer */
1085 	suword(vectp, 0);
1086 
1087 	return (stack_base);
1088 }
1089 
1090 /*
1091  * Check permissions of file to execute.
1092  *	Called with imgp->vp locked.
1093  *	Return 0 for success or error code on failure.
1094  */
1095 int
1096 exec_check_permissions(imgp)
1097 	struct image_params *imgp;
1098 {
1099 	struct vnode *vp = imgp->vp;
1100 	struct vattr *attr = imgp->attr;
1101 	struct thread *td;
1102 	int error;
1103 
1104 	td = curthread;			/* XXXKSE */
1105 
1106 	/* Get file attributes */
1107 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1108 	if (error)
1109 		return (error);
1110 
1111 #ifdef MAC
1112 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1113 	if (error)
1114 		return (error);
1115 #endif
1116 
1117 	/*
1118 	 * 1) Check if file execution is disabled for the filesystem that this
1119 	 *	file resides on.
1120 	 * 2) Insure that at least one execute bit is on - otherwise root
1121 	 *	will always succeed, and we don't want to happen unless the
1122 	 *	file really is executable.
1123 	 * 3) Insure that the file is a regular file.
1124 	 */
1125 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1126 	    ((attr->va_mode & 0111) == 0) ||
1127 	    (attr->va_type != VREG))
1128 		return (EACCES);
1129 
1130 	/*
1131 	 * Zero length files can't be exec'd
1132 	 */
1133 	if (attr->va_size == 0)
1134 		return (ENOEXEC);
1135 
1136 	/*
1137 	 *  Check for execute permission to file based on current credentials.
1138 	 */
1139 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1140 	if (error)
1141 		return (error);
1142 
1143 	/*
1144 	 * Check number of open-for-writes on the file and deny execution
1145 	 * if there are any.
1146 	 */
1147 	if (vp->v_writecount)
1148 		return (ETXTBSY);
1149 
1150 	/*
1151 	 * Call filesystem specific open routine (which does nothing in the
1152 	 * general case).
1153 	 */
1154 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1155 	return (error);
1156 }
1157 
1158 /*
1159  * Exec handler registration
1160  */
1161 int
1162 exec_register(execsw_arg)
1163 	const struct execsw *execsw_arg;
1164 {
1165 	const struct execsw **es, **xs, **newexecsw;
1166 	int count = 2;	/* New slot and trailing NULL */
1167 
1168 	if (execsw)
1169 		for (es = execsw; *es; es++)
1170 			count++;
1171 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1172 	if (newexecsw == NULL)
1173 		return (ENOMEM);
1174 	xs = newexecsw;
1175 	if (execsw)
1176 		for (es = execsw; *es; es++)
1177 			*xs++ = *es;
1178 	*xs++ = execsw_arg;
1179 	*xs = NULL;
1180 	if (execsw)
1181 		free(execsw, M_TEMP);
1182 	execsw = newexecsw;
1183 	return (0);
1184 }
1185 
1186 int
1187 exec_unregister(execsw_arg)
1188 	const struct execsw *execsw_arg;
1189 {
1190 	const struct execsw **es, **xs, **newexecsw;
1191 	int count = 1;
1192 
1193 	if (execsw == NULL)
1194 		panic("unregister with no handlers left?\n");
1195 
1196 	for (es = execsw; *es; es++) {
1197 		if (*es == execsw_arg)
1198 			break;
1199 	}
1200 	if (*es == NULL)
1201 		return (ENOENT);
1202 	for (es = execsw; *es; es++)
1203 		if (*es != execsw_arg)
1204 			count++;
1205 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1206 	if (newexecsw == NULL)
1207 		return (ENOMEM);
1208 	xs = newexecsw;
1209 	for (es = execsw; *es; es++)
1210 		if (*es != execsw_arg)
1211 			*xs++ = *es;
1212 	*xs = NULL;
1213 	if (execsw)
1214 		free(execsw, M_TEMP);
1215 	execsw = newexecsw;
1216 	return (0);
1217 }
1218