xref: /freebsd/sys/kern/kern_exec.c (revision 8ec81e392654c03ebba7792e47bf3744d623cf54)
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_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
34 #include "opt_vm.h"
35 
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
49 #include <sys/acct.h>
50 #include <sys/exec.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
53 #include <sys/wait.h>
54 #include <sys/malloc.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
65 #include <sys/shm.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
68 #include <sys/stat.h>
69 #ifdef KTRACE
70 #include <sys/ktrace.h>
71 #endif
72 
73 #include <vm/vm.h>
74 #include <vm/vm_param.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_pager.h>
82 
83 #ifdef	HWPMC_HOOKS
84 #include <sys/pmckern.h>
85 #endif
86 
87 #include <machine/reg.h>
88 
89 #include <security/audit/audit.h>
90 #include <security/mac/mac_framework.h>
91 
92 #ifdef KDTRACE_HOOKS
93 #include <sys/dtrace_bsd.h>
94 dtrace_execexit_func_t	dtrace_fasttrap_exec;
95 #endif
96 
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
100 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
101 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
102 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
103 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
104 
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
106 
107 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
108 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
110 static int do_execve(struct thread *td, struct image_args *args,
111     struct mac *mac_p);
112 
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
115     NULL, 0, sysctl_kern_ps_strings, "LU", "");
116 
117 /* XXX This should be vm_size_t. */
118 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
119     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
120 
121 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
122     NULL, 0, sysctl_kern_stackprot, "I", "");
123 
124 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
125 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
126     &ps_arg_cache_limit, 0, "");
127 
128 static int map_at_zero = 0;
129 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
131     "Permit processes to map an object at virtual address 0.");
132 
133 static int
134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
135 {
136 	struct proc *p;
137 	int error;
138 
139 	p = curproc;
140 #ifdef SCTL_MASK32
141 	if (req->flags & SCTL_MASK32) {
142 		unsigned int val;
143 		val = (unsigned int)p->p_sysent->sv_psstrings;
144 		error = SYSCTL_OUT(req, &val, sizeof(val));
145 	} else
146 #endif
147 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148 		   sizeof(p->p_sysent->sv_psstrings));
149 	return error;
150 }
151 
152 static int
153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
154 {
155 	struct proc *p;
156 	int error;
157 
158 	p = curproc;
159 #ifdef SCTL_MASK32
160 	if (req->flags & SCTL_MASK32) {
161 		unsigned int val;
162 		val = (unsigned int)p->p_sysent->sv_usrstack;
163 		error = SYSCTL_OUT(req, &val, sizeof(val));
164 	} else
165 #endif
166 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167 		    sizeof(p->p_sysent->sv_usrstack));
168 	return error;
169 }
170 
171 static int
172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
173 {
174 	struct proc *p;
175 
176 	p = curproc;
177 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
178 	    sizeof(p->p_sysent->sv_stackprot)));
179 }
180 
181 /*
182  * Each of the items is a pointer to a `const struct execsw', hence the
183  * double pointer here.
184  */
185 static const struct execsw **execsw;
186 
187 #ifndef _SYS_SYSPROTO_H_
188 struct execve_args {
189 	char    *fname;
190 	char    **argv;
191 	char    **envv;
192 };
193 #endif
194 
195 int
196 sys_execve(td, uap)
197 	struct thread *td;
198 	struct execve_args /* {
199 		char *fname;
200 		char **argv;
201 		char **envv;
202 	} */ *uap;
203 {
204 	int error;
205 	struct image_args args;
206 
207 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
208 	    uap->argv, uap->envv);
209 	if (error == 0)
210 		error = kern_execve(td, &args, NULL);
211 	return (error);
212 }
213 
214 #ifndef _SYS_SYSPROTO_H_
215 struct fexecve_args {
216 	int	fd;
217 	char	**argv;
218 	char	**envv;
219 }
220 #endif
221 int
222 sys_fexecve(struct thread *td, struct fexecve_args *uap)
223 {
224 	int error;
225 	struct image_args args;
226 
227 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
228 	    uap->argv, uap->envv);
229 	if (error == 0) {
230 		args.fd = uap->fd;
231 		error = kern_execve(td, &args, NULL);
232 	}
233 	return (error);
234 }
235 
236 #ifndef _SYS_SYSPROTO_H_
237 struct __mac_execve_args {
238 	char	*fname;
239 	char	**argv;
240 	char	**envv;
241 	struct mac	*mac_p;
242 };
243 #endif
244 
245 int
246 sys___mac_execve(td, uap)
247 	struct thread *td;
248 	struct __mac_execve_args /* {
249 		char *fname;
250 		char **argv;
251 		char **envv;
252 		struct mac *mac_p;
253 	} */ *uap;
254 {
255 #ifdef MAC
256 	int error;
257 	struct image_args args;
258 
259 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
260 	    uap->argv, uap->envv);
261 	if (error == 0)
262 		error = kern_execve(td, &args, uap->mac_p);
263 	return (error);
264 #else
265 	return (ENOSYS);
266 #endif
267 }
268 
269 /*
270  * XXX: kern_execve has the astonishing property of not always returning to
271  * the caller.  If sufficiently bad things happen during the call to
272  * do_execve(), it can end up calling exit1(); as a result, callers must
273  * avoid doing anything which they might need to undo (e.g., allocating
274  * memory).
275  */
276 int
277 kern_execve(td, args, mac_p)
278 	struct thread *td;
279 	struct image_args *args;
280 	struct mac *mac_p;
281 {
282 	struct proc *p = td->td_proc;
283 	int error;
284 
285 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
286 	    args->begin_envv - args->begin_argv);
287 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
288 	    args->endp - args->begin_envv);
289 	if (p->p_flag & P_HADTHREADS) {
290 		PROC_LOCK(p);
291 		if (thread_single(SINGLE_BOUNDARY)) {
292 			PROC_UNLOCK(p);
293 	       		exec_free_args(args);
294 			return (ERESTART);	/* Try again later. */
295 		}
296 		PROC_UNLOCK(p);
297 	}
298 
299 	error = do_execve(td, args, mac_p);
300 
301 	if (p->p_flag & P_HADTHREADS) {
302 		PROC_LOCK(p);
303 		/*
304 		 * If success, we upgrade to SINGLE_EXIT state to
305 		 * force other threads to suicide.
306 		 */
307 		if (error == 0)
308 			thread_single(SINGLE_EXIT);
309 		else
310 			thread_single_end();
311 		PROC_UNLOCK(p);
312 	}
313 
314 	return (error);
315 }
316 
317 /*
318  * In-kernel implementation of execve().  All arguments are assumed to be
319  * userspace pointers from the passed thread.
320  */
321 static int
322 do_execve(td, args, mac_p)
323 	struct thread *td;
324 	struct image_args *args;
325 	struct mac *mac_p;
326 {
327 	struct proc *p = td->td_proc;
328 	struct nameidata nd;
329 	struct ucred *newcred = NULL, *oldcred;
330 	struct uidinfo *euip;
331 	register_t *stack_base;
332 	int error, i;
333 	struct image_params image_params, *imgp;
334 	struct vattr attr;
335 	int (*img_first)(struct image_params *);
336 	struct pargs *oldargs = NULL, *newargs = NULL;
337 	struct sigacts *oldsigacts, *newsigacts;
338 #ifdef KTRACE
339 	struct vnode *tracevp = NULL;
340 	struct ucred *tracecred = NULL;
341 #endif
342 	struct vnode *textvp = NULL, *binvp = NULL;
343 	int credential_changing;
344 	int textset;
345 #ifdef MAC
346 	struct label *interpvplabel = NULL;
347 	int will_transition;
348 #endif
349 #ifdef HWPMC_HOOKS
350 	struct pmckern_procexec pe;
351 #endif
352 	static const char fexecv_proc_title[] = "(fexecv)";
353 
354 	imgp = &image_params;
355 
356 	/*
357 	 * Lock the process and set the P_INEXEC flag to indicate that
358 	 * it should be left alone until we're done here.  This is
359 	 * necessary to avoid race conditions - e.g. in ptrace() -
360 	 * that might allow a local user to illicitly obtain elevated
361 	 * privileges.
362 	 */
363 	PROC_LOCK(p);
364 	KASSERT((p->p_flag & P_INEXEC) == 0,
365 	    ("%s(): process already has P_INEXEC flag", __func__));
366 	p->p_flag |= P_INEXEC;
367 	PROC_UNLOCK(p);
368 
369 	/*
370 	 * Initialize part of the common data
371 	 */
372 	imgp->proc = p;
373 	imgp->execlabel = NULL;
374 	imgp->attr = &attr;
375 	imgp->entry_addr = 0;
376 	imgp->reloc_base = 0;
377 	imgp->vmspace_destroyed = 0;
378 	imgp->interpreted = 0;
379 	imgp->opened = 0;
380 	imgp->interpreter_name = NULL;
381 	imgp->auxargs = NULL;
382 	imgp->vp = NULL;
383 	imgp->object = NULL;
384 	imgp->firstpage = NULL;
385 	imgp->ps_strings = 0;
386 	imgp->auxarg_size = 0;
387 	imgp->args = args;
388 	imgp->execpath = imgp->freepath = NULL;
389 	imgp->execpathp = 0;
390 	imgp->canary = 0;
391 	imgp->canarylen = 0;
392 	imgp->pagesizes = 0;
393 	imgp->pagesizeslen = 0;
394 	imgp->stack_prot = 0;
395 
396 #ifdef MAC
397 	error = mac_execve_enter(imgp, mac_p);
398 	if (error)
399 		goto exec_fail;
400 #endif
401 
402 	imgp->image_header = NULL;
403 
404 	/*
405 	 * Translate the file name. namei() returns a vnode pointer
406 	 *	in ni_vp amoung other things.
407 	 *
408 	 * XXXAUDIT: It would be desirable to also audit the name of the
409 	 * interpreter if this is an interpreted binary.
410 	 */
411 	if (args->fname != NULL) {
412 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
413 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
414 	}
415 
416 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
417 
418 interpret:
419 	if (args->fname != NULL) {
420 #ifdef CAPABILITY_MODE
421 		/*
422 		 * While capability mode can't reach this point via direct
423 		 * path arguments to execve(), we also don't allow
424 		 * interpreters to be used in capability mode (for now).
425 		 * Catch indirect lookups and return a permissions error.
426 		 */
427 		if (IN_CAPABILITY_MODE(td)) {
428 			error = ECAPMODE;
429 			goto exec_fail;
430 		}
431 #endif
432 		error = namei(&nd);
433 		if (error)
434 			goto exec_fail;
435 
436 		binvp  = nd.ni_vp;
437 		imgp->vp = binvp;
438 	} else {
439 		AUDIT_ARG_FD(args->fd);
440 		/*
441 		 * Some might argue that CAP_READ and/or CAP_MMAP should also
442 		 * be required here; such arguments will be entertained.
443 		 *
444 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
445 		 */
446 		error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
447 		if (error)
448 			goto exec_fail;
449 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
450 		AUDIT_ARG_VNODE1(binvp);
451 		imgp->vp = binvp;
452 	}
453 
454 	/*
455 	 * Check file permissions (also 'opens' file)
456 	 */
457 	error = exec_check_permissions(imgp);
458 	if (error)
459 		goto exec_fail_dealloc;
460 
461 	imgp->object = imgp->vp->v_object;
462 	if (imgp->object != NULL)
463 		vm_object_reference(imgp->object);
464 
465 	/*
466 	 * Set VV_TEXT now so no one can write to the executable while we're
467 	 * activating it.
468 	 *
469 	 * Remember if this was set before and unset it in case this is not
470 	 * actually an executable image.
471 	 */
472 	textset = VOP_IS_TEXT(imgp->vp);
473 	VOP_SET_TEXT(imgp->vp);
474 
475 	error = exec_map_first_page(imgp);
476 	if (error)
477 		goto exec_fail_dealloc;
478 
479 	imgp->proc->p_osrel = 0;
480 	/*
481 	 *	If the current process has a special image activator it
482 	 *	wants to try first, call it.   For example, emulating shell
483 	 *	scripts differently.
484 	 */
485 	error = -1;
486 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
487 		error = img_first(imgp);
488 
489 	/*
490 	 *	Loop through the list of image activators, calling each one.
491 	 *	An activator returns -1 if there is no match, 0 on success,
492 	 *	and an error otherwise.
493 	 */
494 	for (i = 0; error == -1 && execsw[i]; ++i) {
495 		if (execsw[i]->ex_imgact == NULL ||
496 		    execsw[i]->ex_imgact == img_first) {
497 			continue;
498 		}
499 		error = (*execsw[i]->ex_imgact)(imgp);
500 	}
501 
502 	if (error) {
503 		if (error == -1) {
504 			if (textset == 0)
505 				VOP_UNSET_TEXT(imgp->vp);
506 			error = ENOEXEC;
507 		}
508 		goto exec_fail_dealloc;
509 	}
510 
511 	/*
512 	 * Special interpreter operation, cleanup and loop up to try to
513 	 * activate the interpreter.
514 	 */
515 	if (imgp->interpreted) {
516 		exec_unmap_first_page(imgp);
517 		/*
518 		 * VV_TEXT needs to be unset for scripts.  There is a short
519 		 * period before we determine that something is a script where
520 		 * VV_TEXT will be set. The vnode lock is held over this
521 		 * entire period so nothing should illegitimately be blocked.
522 		 */
523 		VOP_UNSET_TEXT(imgp->vp);
524 		/* free name buffer and old vnode */
525 		if (args->fname != NULL)
526 			NDFREE(&nd, NDF_ONLY_PNBUF);
527 #ifdef MAC
528 		mac_execve_interpreter_enter(binvp, &interpvplabel);
529 #endif
530 		if (imgp->opened) {
531 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
532 			imgp->opened = 0;
533 		}
534 		vput(binvp);
535 		vm_object_deallocate(imgp->object);
536 		imgp->object = NULL;
537 		/* set new name to that of the interpreter */
538 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
539 		    UIO_SYSSPACE, imgp->interpreter_name, td);
540 		args->fname = imgp->interpreter_name;
541 		goto interpret;
542 	}
543 
544 	/*
545 	 * NB: We unlock the vnode here because it is believed that none
546 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
547 	 */
548 	VOP_UNLOCK(imgp->vp, 0);
549 
550 	/*
551 	 * Do the best to calculate the full path to the image file.
552 	 */
553 	if (imgp->auxargs != NULL &&
554 	    ((args->fname != NULL && args->fname[0] == '/') ||
555 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
556 		imgp->execpath = args->fname;
557 
558 	/*
559 	 * Copy out strings (args and env) and initialize stack base
560 	 */
561 	if (p->p_sysent->sv_copyout_strings)
562 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
563 	else
564 		stack_base = exec_copyout_strings(imgp);
565 
566 	/*
567 	 * If custom stack fixup routine present for this process
568 	 * let it do the stack setup.
569 	 * Else stuff argument count as first item on stack
570 	 */
571 	if (p->p_sysent->sv_fixup != NULL)
572 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
573 	else
574 		suword(--stack_base, imgp->args->argc);
575 
576 	/*
577 	 * For security and other reasons, the file descriptor table cannot
578 	 * be shared after an exec.
579 	 */
580 	fdunshare(p, td);
581 
582 	/*
583 	 * Malloc things before we need locks.
584 	 */
585 	newcred = crget();
586 	euip = uifind(attr.va_uid);
587 	i = imgp->args->begin_envv - imgp->args->begin_argv;
588 	/* Cache arguments if they fit inside our allowance */
589 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
590 		newargs = pargs_alloc(i);
591 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
592 	}
593 
594 	/* close files on exec */
595 	fdcloseexec(td);
596 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
597 
598 	/* Get a reference to the vnode prior to locking the proc */
599 	VREF(binvp);
600 
601 	/*
602 	 * For security and other reasons, signal handlers cannot
603 	 * be shared after an exec. The new process gets a copy of the old
604 	 * handlers. In execsigs(), the new process will have its signals
605 	 * reset.
606 	 */
607 	PROC_LOCK(p);
608 	oldcred = crcopysafe(p, newcred);
609 	if (sigacts_shared(p->p_sigacts)) {
610 		oldsigacts = p->p_sigacts;
611 		PROC_UNLOCK(p);
612 		newsigacts = sigacts_alloc();
613 		sigacts_copy(newsigacts, oldsigacts);
614 		PROC_LOCK(p);
615 		p->p_sigacts = newsigacts;
616 	} else
617 		oldsigacts = NULL;
618 
619 	/* Stop profiling */
620 	stopprofclock(p);
621 
622 	/* reset caught signals */
623 	execsigs(p);
624 
625 	/* name this process - nameiexec(p, ndp) */
626 	bzero(p->p_comm, sizeof(p->p_comm));
627 	if (args->fname)
628 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
629 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
630 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
631 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
632 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
633 #ifdef KTR
634 	sched_clear_tdname(td);
635 #endif
636 
637 	/*
638 	 * mark as execed, wakeup the process that vforked (if any) and tell
639 	 * it that it now has its own resources back
640 	 */
641 	p->p_flag |= P_EXEC;
642 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
643 		p->p_flag &= ~P_PPWAIT;
644 		cv_broadcast(&p->p_pwait);
645 	}
646 
647 	/*
648 	 * Implement image setuid/setgid.
649 	 *
650 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
651 	 * the process is being traced.
652 	 *
653 	 * We disable setuid/setgid/etc in compatibility mode on the basis
654 	 * that most setugid applications are not written with that
655 	 * environment in mind, and will therefore almost certainly operate
656 	 * incorrectly. In principle there's no reason that setugid
657 	 * applications might not be useful in capability mode, so we may want
658 	 * to reconsider this conservative design choice in the future.
659 	 *
660 	 * XXXMAC: For the time being, use NOSUID to also prohibit
661 	 * transitions on the file system.
662 	 */
663 	credential_changing = 0;
664 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
665 	    attr.va_uid;
666 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
667 	    attr.va_gid;
668 #ifdef MAC
669 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
670 	    interpvplabel, imgp);
671 	credential_changing |= will_transition;
672 #endif
673 
674 	if (credential_changing &&
675 #ifdef CAPABILITY_MODE
676 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
677 #endif
678 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
679 	    (p->p_flag & P_TRACED) == 0) {
680 		/*
681 		 * Turn off syscall tracing for set-id programs, except for
682 		 * root.  Record any set-id flags first to make sure that
683 		 * we do not regain any tracing during a possible block.
684 		 */
685 		setsugid(p);
686 
687 #ifdef KTRACE
688 		if (p->p_tracecred != NULL &&
689 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
690 			ktrprocexec(p, &tracecred, &tracevp);
691 #endif
692 		/*
693 		 * Close any file descriptors 0..2 that reference procfs,
694 		 * then make sure file descriptors 0..2 are in use.
695 		 *
696 		 * setugidsafety() may call closef() and then pfind()
697 		 * which may grab the process lock.
698 		 * fdcheckstd() may call falloc() which may block to
699 		 * allocate memory, so temporarily drop the process lock.
700 		 */
701 		PROC_UNLOCK(p);
702 		VOP_UNLOCK(imgp->vp, 0);
703 		setugidsafety(td);
704 		error = fdcheckstd(td);
705 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
706 		if (error != 0)
707 			goto done1;
708 		PROC_LOCK(p);
709 		/*
710 		 * Set the new credentials.
711 		 */
712 		if (attr.va_mode & S_ISUID)
713 			change_euid(newcred, euip);
714 		if (attr.va_mode & S_ISGID)
715 			change_egid(newcred, attr.va_gid);
716 #ifdef MAC
717 		if (will_transition) {
718 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
719 			    interpvplabel, imgp);
720 		}
721 #endif
722 		/*
723 		 * Implement correct POSIX saved-id behavior.
724 		 *
725 		 * XXXMAC: Note that the current logic will save the
726 		 * uid and gid if a MAC domain transition occurs, even
727 		 * though maybe it shouldn't.
728 		 */
729 		change_svuid(newcred, newcred->cr_uid);
730 		change_svgid(newcred, newcred->cr_gid);
731 		p->p_ucred = newcred;
732 		newcred = NULL;
733 	} else {
734 		if (oldcred->cr_uid == oldcred->cr_ruid &&
735 		    oldcred->cr_gid == oldcred->cr_rgid)
736 			p->p_flag &= ~P_SUGID;
737 		/*
738 		 * Implement correct POSIX saved-id behavior.
739 		 *
740 		 * XXX: It's not clear that the existing behavior is
741 		 * POSIX-compliant.  A number of sources indicate that the
742 		 * saved uid/gid should only be updated if the new ruid is
743 		 * not equal to the old ruid, or the new euid is not equal
744 		 * to the old euid and the new euid is not equal to the old
745 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
746 		 * Also, this code uses the new (replaced) euid and egid as
747 		 * the source, which may or may not be the right ones to use.
748 		 */
749 		if (oldcred->cr_svuid != oldcred->cr_uid ||
750 		    oldcred->cr_svgid != oldcred->cr_gid) {
751 			change_svuid(newcred, newcred->cr_uid);
752 			change_svgid(newcred, newcred->cr_gid);
753 			p->p_ucred = newcred;
754 			newcred = NULL;
755 		}
756 	}
757 
758 	/*
759 	 * Store the vp for use in procfs.  This vnode was referenced prior
760 	 * to locking the proc lock.
761 	 */
762 	textvp = p->p_textvp;
763 	p->p_textvp = binvp;
764 
765 #ifdef KDTRACE_HOOKS
766 	/*
767 	 * Tell the DTrace fasttrap provider about the exec if it
768 	 * has declared an interest.
769 	 */
770 	if (dtrace_fasttrap_exec)
771 		dtrace_fasttrap_exec(p);
772 #endif
773 
774 	/*
775 	 * Notify others that we exec'd, and clear the P_INEXEC flag
776 	 * as we're now a bona fide freshly-execed process.
777 	 */
778 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
779 	p->p_flag &= ~P_INEXEC;
780 
781 	/* clear "fork but no exec" flag, as we _are_ execing */
782 	p->p_acflag &= ~AFORK;
783 
784 	/*
785 	 * Free any previous argument cache and replace it with
786 	 * the new argument cache, if any.
787 	 */
788 	oldargs = p->p_args;
789 	p->p_args = newargs;
790 	newargs = NULL;
791 
792 #ifdef	HWPMC_HOOKS
793 	/*
794 	 * Check if system-wide sampling is in effect or if the
795 	 * current process is using PMCs.  If so, do exec() time
796 	 * processing.  This processing needs to happen AFTER the
797 	 * P_INEXEC flag is cleared.
798 	 *
799 	 * The proc lock needs to be released before taking the PMC
800 	 * SX.
801 	 */
802 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
803 		PROC_UNLOCK(p);
804 		VOP_UNLOCK(imgp->vp, 0);
805 		pe.pm_credentialschanged = credential_changing;
806 		pe.pm_entryaddr = imgp->entry_addr;
807 
808 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
809 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
810 	} else
811 		PROC_UNLOCK(p);
812 #else  /* !HWPMC_HOOKS */
813 	PROC_UNLOCK(p);
814 #endif
815 
816 	/* Set values passed into the program in registers. */
817 	if (p->p_sysent->sv_setregs)
818 		(*p->p_sysent->sv_setregs)(td, imgp,
819 		    (u_long)(uintptr_t)stack_base);
820 	else
821 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
822 
823 	vfs_mark_atime(imgp->vp, td->td_ucred);
824 
825 	SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
826 
827 done1:
828 	/*
829 	 * Free any resources malloc'd earlier that we didn't use.
830 	 */
831 	uifree(euip);
832 	if (newcred == NULL)
833 		crfree(oldcred);
834 	else
835 		crfree(newcred);
836 	VOP_UNLOCK(imgp->vp, 0);
837 
838 	/*
839 	 * Handle deferred decrement of ref counts.
840 	 */
841 	if (textvp != NULL)
842 		vrele(textvp);
843 	if (binvp && error != 0)
844 		vrele(binvp);
845 #ifdef KTRACE
846 	if (tracevp != NULL)
847 		vrele(tracevp);
848 	if (tracecred != NULL)
849 		crfree(tracecred);
850 #endif
851 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
852 	pargs_drop(oldargs);
853 	pargs_drop(newargs);
854 	if (oldsigacts != NULL)
855 		sigacts_free(oldsigacts);
856 
857 exec_fail_dealloc:
858 
859 	/*
860 	 * free various allocated resources
861 	 */
862 	if (imgp->firstpage != NULL)
863 		exec_unmap_first_page(imgp);
864 
865 	if (imgp->vp != NULL) {
866 		if (args->fname)
867 			NDFREE(&nd, NDF_ONLY_PNBUF);
868 		if (imgp->opened)
869 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
870 		vput(imgp->vp);
871 	}
872 
873 	if (imgp->object != NULL)
874 		vm_object_deallocate(imgp->object);
875 
876 	free(imgp->freepath, M_TEMP);
877 
878 	if (error == 0) {
879 		PROC_LOCK(p);
880 		td->td_dbgflags |= TDB_EXEC;
881 		PROC_UNLOCK(p);
882 
883 		/*
884 		 * Stop the process here if its stop event mask has
885 		 * the S_EXEC bit set.
886 		 */
887 		STOPEVENT(p, S_EXEC, 0);
888 		goto done2;
889 	}
890 
891 exec_fail:
892 	/* we're done here, clear P_INEXEC */
893 	PROC_LOCK(p);
894 	p->p_flag &= ~P_INEXEC;
895 	PROC_UNLOCK(p);
896 
897 	SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
898 
899 done2:
900 #ifdef MAC
901 	mac_execve_exit(imgp);
902 	mac_execve_interpreter_exit(interpvplabel);
903 #endif
904 	exec_free_args(args);
905 
906 	if (error && imgp->vmspace_destroyed) {
907 		/* sorry, no more process anymore. exit gracefully */
908 		exit1(td, W_EXITCODE(0, SIGABRT));
909 		/* NOT REACHED */
910 	}
911 
912 #ifdef KTRACE
913 	if (error == 0)
914 		ktrprocctor(p);
915 #endif
916 
917 	return (error);
918 }
919 
920 int
921 exec_map_first_page(imgp)
922 	struct image_params *imgp;
923 {
924 	int rv, i;
925 	int initial_pagein;
926 	vm_page_t ma[VM_INITIAL_PAGEIN];
927 	vm_object_t object;
928 
929 	if (imgp->firstpage != NULL)
930 		exec_unmap_first_page(imgp);
931 
932 	object = imgp->vp->v_object;
933 	if (object == NULL)
934 		return (EACCES);
935 	VM_OBJECT_LOCK(object);
936 #if VM_NRESERVLEVEL > 0
937 	if ((object->flags & OBJ_COLORED) == 0) {
938 		object->flags |= OBJ_COLORED;
939 		object->pg_color = 0;
940 	}
941 #endif
942 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
943 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
944 		initial_pagein = VM_INITIAL_PAGEIN;
945 		if (initial_pagein > object->size)
946 			initial_pagein = object->size;
947 		for (i = 1; i < initial_pagein; i++) {
948 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
949 				if (ma[i]->valid)
950 					break;
951 				if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
952 					break;
953 				vm_page_busy(ma[i]);
954 			} else {
955 				ma[i] = vm_page_alloc(object, i,
956 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
957 				if (ma[i] == NULL)
958 					break;
959 			}
960 		}
961 		initial_pagein = i;
962 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
963 		ma[0] = vm_page_lookup(object, 0);
964 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
965 			if (ma[0] != NULL) {
966 				vm_page_lock(ma[0]);
967 				vm_page_free(ma[0]);
968 				vm_page_unlock(ma[0]);
969 			}
970 			VM_OBJECT_UNLOCK(object);
971 			return (EIO);
972 		}
973 	}
974 	vm_page_lock(ma[0]);
975 	vm_page_hold(ma[0]);
976 	vm_page_unlock(ma[0]);
977 	vm_page_wakeup(ma[0]);
978 	VM_OBJECT_UNLOCK(object);
979 
980 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
981 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
982 
983 	return (0);
984 }
985 
986 void
987 exec_unmap_first_page(imgp)
988 	struct image_params *imgp;
989 {
990 	vm_page_t m;
991 
992 	if (imgp->firstpage != NULL) {
993 		m = sf_buf_page(imgp->firstpage);
994 		sf_buf_free(imgp->firstpage);
995 		imgp->firstpage = NULL;
996 		vm_page_lock(m);
997 		vm_page_unhold(m);
998 		vm_page_unlock(m);
999 	}
1000 }
1001 
1002 /*
1003  * Destroy old address space, and allocate a new stack
1004  *	The new stack is only SGROWSIZ large because it is grown
1005  *	automatically in trap.c.
1006  */
1007 int
1008 exec_new_vmspace(imgp, sv)
1009 	struct image_params *imgp;
1010 	struct sysentvec *sv;
1011 {
1012 	int error;
1013 	struct proc *p = imgp->proc;
1014 	struct vmspace *vmspace = p->p_vmspace;
1015 	vm_object_t obj;
1016 	vm_offset_t sv_minuser, stack_addr;
1017 	vm_map_t map;
1018 	u_long ssiz;
1019 
1020 	imgp->vmspace_destroyed = 1;
1021 	imgp->sysent = sv;
1022 
1023 	/* May be called with Giant held */
1024 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1025 
1026 	/*
1027 	 * Blow away entire process VM, if address space not shared,
1028 	 * otherwise, create a new VM space so that other threads are
1029 	 * not disrupted
1030 	 */
1031 	map = &vmspace->vm_map;
1032 	if (map_at_zero)
1033 		sv_minuser = sv->sv_minuser;
1034 	else
1035 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1036 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1037 	    vm_map_max(map) == sv->sv_maxuser) {
1038 		shmexit(vmspace);
1039 		pmap_remove_pages(vmspace_pmap(vmspace));
1040 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1041 	} else {
1042 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1043 		if (error)
1044 			return (error);
1045 		vmspace = p->p_vmspace;
1046 		map = &vmspace->vm_map;
1047 	}
1048 
1049 	/* Map a shared page */
1050 	obj = sv->sv_shared_page_obj;
1051 	if (obj != NULL) {
1052 		vm_object_reference(obj);
1053 		error = vm_map_fixed(map, obj, 0,
1054 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1055 		    VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL,
1056 		    MAP_COPY_ON_WRITE | MAP_ACC_NO_CHARGE);
1057 		if (error) {
1058 			vm_object_deallocate(obj);
1059 			return (error);
1060 		}
1061 	}
1062 
1063 	/* Allocate a new stack */
1064 	if (sv->sv_maxssiz != NULL)
1065 		ssiz = *sv->sv_maxssiz;
1066 	else
1067 		ssiz = maxssiz;
1068 	stack_addr = sv->sv_usrstack - ssiz;
1069 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1070 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1071 		sv->sv_stackprot,
1072 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1073 	if (error)
1074 		return (error);
1075 
1076 #ifdef __ia64__
1077 	/* Allocate a new register stack */
1078 	stack_addr = IA64_BACKINGSTORE;
1079 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1080 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1081 	if (error)
1082 		return (error);
1083 #endif
1084 
1085 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1086 	 * VM_STACK case, but they are still used to monitor the size of the
1087 	 * process stack so we can check the stack rlimit.
1088 	 */
1089 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1090 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1091 
1092 	return (0);
1093 }
1094 
1095 /*
1096  * Copy out argument and environment strings from the old process address
1097  * space into the temporary string buffer.
1098  */
1099 int
1100 exec_copyin_args(struct image_args *args, char *fname,
1101     enum uio_seg segflg, char **argv, char **envv)
1102 {
1103 	char *argp, *envp;
1104 	int error;
1105 	size_t length;
1106 
1107 	bzero(args, sizeof(*args));
1108 	if (argv == NULL)
1109 		return (EFAULT);
1110 
1111 	/*
1112 	 * Allocate demand-paged memory for the file name, argument, and
1113 	 * environment strings.
1114 	 */
1115 	error = exec_alloc_args(args);
1116 	if (error != 0)
1117 		return (error);
1118 
1119 	/*
1120 	 * Copy the file name.
1121 	 */
1122 	if (fname != NULL) {
1123 		args->fname = args->buf;
1124 		error = (segflg == UIO_SYSSPACE) ?
1125 		    copystr(fname, args->fname, PATH_MAX, &length) :
1126 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1127 		if (error != 0)
1128 			goto err_exit;
1129 	} else
1130 		length = 0;
1131 
1132 	args->begin_argv = args->buf + length;
1133 	args->endp = args->begin_argv;
1134 	args->stringspace = ARG_MAX;
1135 
1136 	/*
1137 	 * extract arguments first
1138 	 */
1139 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1140 		if (argp == (caddr_t) -1) {
1141 			error = EFAULT;
1142 			goto err_exit;
1143 		}
1144 		if ((error = copyinstr(argp, args->endp,
1145 		    args->stringspace, &length))) {
1146 			if (error == ENAMETOOLONG)
1147 				error = E2BIG;
1148 			goto err_exit;
1149 		}
1150 		args->stringspace -= length;
1151 		args->endp += length;
1152 		args->argc++;
1153 	}
1154 
1155 	args->begin_envv = args->endp;
1156 
1157 	/*
1158 	 * extract environment strings
1159 	 */
1160 	if (envv) {
1161 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1162 			if (envp == (caddr_t)-1) {
1163 				error = EFAULT;
1164 				goto err_exit;
1165 			}
1166 			if ((error = copyinstr(envp, args->endp,
1167 			    args->stringspace, &length))) {
1168 				if (error == ENAMETOOLONG)
1169 					error = E2BIG;
1170 				goto err_exit;
1171 			}
1172 			args->stringspace -= length;
1173 			args->endp += length;
1174 			args->envc++;
1175 		}
1176 	}
1177 
1178 	return (0);
1179 
1180 err_exit:
1181 	exec_free_args(args);
1182 	return (error);
1183 }
1184 
1185 /*
1186  * Allocate temporary demand-paged, zero-filled memory for the file name,
1187  * argument, and environment strings.  Returns zero if the allocation succeeds
1188  * and ENOMEM otherwise.
1189  */
1190 int
1191 exec_alloc_args(struct image_args *args)
1192 {
1193 
1194 	args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1195 	return (args->buf != NULL ? 0 : ENOMEM);
1196 }
1197 
1198 void
1199 exec_free_args(struct image_args *args)
1200 {
1201 
1202 	if (args->buf != NULL) {
1203 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1204 		    PATH_MAX + ARG_MAX);
1205 		args->buf = NULL;
1206 	}
1207 	if (args->fname_buf != NULL) {
1208 		free(args->fname_buf, M_TEMP);
1209 		args->fname_buf = NULL;
1210 	}
1211 }
1212 
1213 /*
1214  * Copy strings out to the new process address space, constructing new arg
1215  * and env vector tables. Return a pointer to the base so that it can be used
1216  * as the initial stack pointer.
1217  */
1218 register_t *
1219 exec_copyout_strings(imgp)
1220 	struct image_params *imgp;
1221 {
1222 	int argc, envc;
1223 	char **vectp;
1224 	char *stringp, *destp;
1225 	register_t *stack_base;
1226 	struct ps_strings *arginfo;
1227 	struct proc *p;
1228 	size_t execpath_len;
1229 	int szsigcode, szps;
1230 	char canary[sizeof(long) * 8];
1231 
1232 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1233 	/*
1234 	 * Calculate string base and vector table pointers.
1235 	 * Also deal with signal trampoline code for this exec type.
1236 	 */
1237 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1238 		execpath_len = strlen(imgp->execpath) + 1;
1239 	else
1240 		execpath_len = 0;
1241 	p = imgp->proc;
1242 	szsigcode = 0;
1243 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1244 	if (p->p_sysent->sv_sigcode_base == 0) {
1245 		if (p->p_sysent->sv_szsigcode != NULL)
1246 			szsigcode = *(p->p_sysent->sv_szsigcode);
1247 	}
1248 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1249 	    roundup(execpath_len, sizeof(char *)) -
1250 	    roundup(sizeof(canary), sizeof(char *)) -
1251 	    roundup(szps, sizeof(char *)) -
1252 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1253 
1254 	/*
1255 	 * install sigcode
1256 	 */
1257 	if (szsigcode != 0)
1258 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1259 		    szsigcode), szsigcode);
1260 
1261 	/*
1262 	 * Copy the image path for the rtld.
1263 	 */
1264 	if (execpath_len != 0) {
1265 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1266 		copyout(imgp->execpath, (void *)imgp->execpathp,
1267 		    execpath_len);
1268 	}
1269 
1270 	/*
1271 	 * Prepare the canary for SSP.
1272 	 */
1273 	arc4rand(canary, sizeof(canary), 0);
1274 	imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1275 	    sizeof(canary);
1276 	copyout(canary, (void *)imgp->canary, sizeof(canary));
1277 	imgp->canarylen = sizeof(canary);
1278 
1279 	/*
1280 	 * Prepare the pagesizes array.
1281 	 */
1282 	imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1283 	    roundup(sizeof(canary), sizeof(char *)) - szps;
1284 	copyout(pagesizes, (void *)imgp->pagesizes, szps);
1285 	imgp->pagesizeslen = szps;
1286 
1287 	/*
1288 	 * If we have a valid auxargs ptr, prepare some room
1289 	 * on the stack.
1290 	 */
1291 	if (imgp->auxargs) {
1292 		/*
1293 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1294 		 * lower compatibility.
1295 		 */
1296 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1297 		    (AT_COUNT * 2);
1298 		/*
1299 		 * The '+ 2' is for the null pointers at the end of each of
1300 		 * the arg and env vector sets,and imgp->auxarg_size is room
1301 		 * for argument of Runtime loader.
1302 		 */
1303 		vectp = (char **)(destp - (imgp->args->argc +
1304 		    imgp->args->envc + 2 + imgp->auxarg_size)
1305 		    * sizeof(char *));
1306 	} else {
1307 		/*
1308 		 * The '+ 2' is for the null pointers at the end of each of
1309 		 * the arg and env vector sets
1310 		 */
1311 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1312 		    sizeof(char *));
1313 	}
1314 
1315 	/*
1316 	 * vectp also becomes our initial stack base
1317 	 */
1318 	stack_base = (register_t *)vectp;
1319 
1320 	stringp = imgp->args->begin_argv;
1321 	argc = imgp->args->argc;
1322 	envc = imgp->args->envc;
1323 
1324 	/*
1325 	 * Copy out strings - arguments and environment.
1326 	 */
1327 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1328 
1329 	/*
1330 	 * Fill in "ps_strings" struct for ps, w, etc.
1331 	 */
1332 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1333 	suword32(&arginfo->ps_nargvstr, argc);
1334 
1335 	/*
1336 	 * Fill in argument portion of vector table.
1337 	 */
1338 	for (; argc > 0; --argc) {
1339 		suword(vectp++, (long)(intptr_t)destp);
1340 		while (*stringp++ != 0)
1341 			destp++;
1342 		destp++;
1343 	}
1344 
1345 	/* a null vector table pointer separates the argp's from the envp's */
1346 	suword(vectp++, 0);
1347 
1348 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1349 	suword32(&arginfo->ps_nenvstr, envc);
1350 
1351 	/*
1352 	 * Fill in environment portion of vector table.
1353 	 */
1354 	for (; envc > 0; --envc) {
1355 		suword(vectp++, (long)(intptr_t)destp);
1356 		while (*stringp++ != 0)
1357 			destp++;
1358 		destp++;
1359 	}
1360 
1361 	/* end of vector table is a null pointer */
1362 	suword(vectp, 0);
1363 
1364 	return (stack_base);
1365 }
1366 
1367 /*
1368  * Check permissions of file to execute.
1369  *	Called with imgp->vp locked.
1370  *	Return 0 for success or error code on failure.
1371  */
1372 int
1373 exec_check_permissions(imgp)
1374 	struct image_params *imgp;
1375 {
1376 	struct vnode *vp = imgp->vp;
1377 	struct vattr *attr = imgp->attr;
1378 	struct thread *td;
1379 	int error, writecount;
1380 
1381 	td = curthread;
1382 
1383 	/* Get file attributes */
1384 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1385 	if (error)
1386 		return (error);
1387 
1388 #ifdef MAC
1389 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1390 	if (error)
1391 		return (error);
1392 #endif
1393 
1394 	/*
1395 	 * 1) Check if file execution is disabled for the filesystem that
1396 	 *    this file resides on.
1397 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1398 	 *    privileged user will always succeed, and we don't want this
1399 	 *    to happen unless the file really is executable.
1400 	 * 3) Ensure that the file is a regular file.
1401 	 */
1402 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1403 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1404 	    (attr->va_type != VREG))
1405 		return (EACCES);
1406 
1407 	/*
1408 	 * Zero length files can't be exec'd
1409 	 */
1410 	if (attr->va_size == 0)
1411 		return (ENOEXEC);
1412 
1413 	/*
1414 	 *  Check for execute permission to file based on current credentials.
1415 	 */
1416 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1417 	if (error)
1418 		return (error);
1419 
1420 	/*
1421 	 * Check number of open-for-writes on the file and deny execution
1422 	 * if there are any.
1423 	 */
1424 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1425 	if (error != 0)
1426 		return (error);
1427 	if (writecount != 0)
1428 		return (ETXTBSY);
1429 
1430 	/*
1431 	 * Call filesystem specific open routine (which does nothing in the
1432 	 * general case).
1433 	 */
1434 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1435 	if (error == 0)
1436 		imgp->opened = 1;
1437 	return (error);
1438 }
1439 
1440 /*
1441  * Exec handler registration
1442  */
1443 int
1444 exec_register(execsw_arg)
1445 	const struct execsw *execsw_arg;
1446 {
1447 	const struct execsw **es, **xs, **newexecsw;
1448 	int count = 2;	/* New slot and trailing NULL */
1449 
1450 	if (execsw)
1451 		for (es = execsw; *es; es++)
1452 			count++;
1453 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1454 	if (newexecsw == NULL)
1455 		return (ENOMEM);
1456 	xs = newexecsw;
1457 	if (execsw)
1458 		for (es = execsw; *es; es++)
1459 			*xs++ = *es;
1460 	*xs++ = execsw_arg;
1461 	*xs = NULL;
1462 	if (execsw)
1463 		free(execsw, M_TEMP);
1464 	execsw = newexecsw;
1465 	return (0);
1466 }
1467 
1468 int
1469 exec_unregister(execsw_arg)
1470 	const struct execsw *execsw_arg;
1471 {
1472 	const struct execsw **es, **xs, **newexecsw;
1473 	int count = 1;
1474 
1475 	if (execsw == NULL)
1476 		panic("unregister with no handlers left?\n");
1477 
1478 	for (es = execsw; *es; es++) {
1479 		if (*es == execsw_arg)
1480 			break;
1481 	}
1482 	if (*es == NULL)
1483 		return (ENOENT);
1484 	for (es = execsw; *es; es++)
1485 		if (*es != execsw_arg)
1486 			count++;
1487 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1488 	if (newexecsw == NULL)
1489 		return (ENOMEM);
1490 	xs = newexecsw;
1491 	for (es = execsw; *es; es++)
1492 		if (*es != execsw_arg)
1493 			*xs++ = *es;
1494 	*xs = NULL;
1495 	if (execsw)
1496 		free(execsw, M_TEMP);
1497 	execsw = newexecsw;
1498 	return (0);
1499 }
1500