xref: /freebsd/sys/kern/kern_exec.c (revision e8e8c939350bdf3c228a411caa9660c607c27a11)
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_ktrace.h"
33 #include "opt_vm.h"
34 
35 #include <sys/param.h>
36 #include <sys/capsicum.h>
37 #include <sys/systm.h>
38 #include <sys/eventhandler.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/signalvar.h>
43 #include <sys/kernel.h>
44 #include <sys/mount.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
47 #include <sys/acct.h>
48 #include <sys/exec.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
51 #include <sys/wait.h>
52 #include <sys/malloc.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/pioctl.h>
56 #include <sys/namei.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/sf_buf.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysent.h>
64 #include <sys/shm.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/stat.h>
68 #ifdef KTRACE
69 #include <sys/ktrace.h>
70 #endif
71 
72 #include <vm/vm.h>
73 #include <vm/vm_param.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_pager.h>
81 
82 #ifdef	HWPMC_HOOKS
83 #include <sys/pmckern.h>
84 #endif
85 
86 #include <machine/reg.h>
87 
88 #include <security/audit/audit.h>
89 #include <security/mac/mac_framework.h>
90 
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t	dtrace_fasttrap_exec;
94 #endif
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
98 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
99 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
100 
101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
102 
103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106 static int do_execve(struct thread *td, struct image_args *args,
107     struct mac *mac_p);
108 
109 /* XXX This should be vm_size_t. */
110 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
111     NULL, 0, sysctl_kern_ps_strings, "LU", "");
112 
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
115     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
116 
117 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
118     NULL, 0, sysctl_kern_stackprot, "I", "");
119 
120 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
121 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
122     &ps_arg_cache_limit, 0, "");
123 
124 static int disallow_high_osrel;
125 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
126     &disallow_high_osrel, 0,
127     "Disallow execution of binaries built for higher version of the world");
128 
129 static int map_at_zero = 0;
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &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(struct thread *td, struct execve_args *uap)
197 {
198 	struct image_args args;
199 	struct vmspace *oldvmspace;
200 	int error;
201 
202 	error = pre_execve(td, &oldvmspace);
203 	if (error != 0)
204 		return (error);
205 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
206 	    uap->argv, uap->envv);
207 	if (error == 0)
208 		error = kern_execve(td, &args, NULL);
209 	post_execve(td, error, oldvmspace);
210 	return (error);
211 }
212 
213 #ifndef _SYS_SYSPROTO_H_
214 struct fexecve_args {
215 	int	fd;
216 	char	**argv;
217 	char	**envv;
218 }
219 #endif
220 int
221 sys_fexecve(struct thread *td, struct fexecve_args *uap)
222 {
223 	struct image_args args;
224 	struct vmspace *oldvmspace;
225 	int error;
226 
227 	error = pre_execve(td, &oldvmspace);
228 	if (error != 0)
229 		return (error);
230 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
231 	    uap->argv, uap->envv);
232 	if (error == 0) {
233 		args.fd = uap->fd;
234 		error = kern_execve(td, &args, NULL);
235 	}
236 	post_execve(td, error, oldvmspace);
237 	return (error);
238 }
239 
240 #ifndef _SYS_SYSPROTO_H_
241 struct __mac_execve_args {
242 	char	*fname;
243 	char	**argv;
244 	char	**envv;
245 	struct mac	*mac_p;
246 };
247 #endif
248 
249 int
250 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
251 {
252 #ifdef MAC
253 	struct image_args args;
254 	struct vmspace *oldvmspace;
255 	int error;
256 
257 	error = pre_execve(td, &oldvmspace);
258 	if (error != 0)
259 		return (error);
260 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
261 	    uap->argv, uap->envv);
262 	if (error == 0)
263 		error = kern_execve(td, &args, uap->mac_p);
264 	post_execve(td, error, oldvmspace);
265 	return (error);
266 #else
267 	return (ENOSYS);
268 #endif
269 }
270 
271 int
272 pre_execve(struct thread *td, struct vmspace **oldvmspace)
273 {
274 	struct proc *p;
275 	int error;
276 
277 	KASSERT(td == curthread, ("non-current thread %p", td));
278 	error = 0;
279 	p = td->td_proc;
280 	if ((p->p_flag & P_HADTHREADS) != 0) {
281 		PROC_LOCK(p);
282 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
283 			error = ERESTART;
284 		PROC_UNLOCK(p);
285 	}
286 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
287 	    ("nested execve"));
288 	*oldvmspace = p->p_vmspace;
289 	return (error);
290 }
291 
292 void
293 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
294 {
295 	struct proc *p;
296 
297 	KASSERT(td == curthread, ("non-current thread %p", td));
298 	p = td->td_proc;
299 	if ((p->p_flag & P_HADTHREADS) != 0) {
300 		PROC_LOCK(p);
301 		/*
302 		 * If success, we upgrade to SINGLE_EXIT state to
303 		 * force other threads to suicide.
304 		 */
305 		if (error == 0)
306 			thread_single(p, SINGLE_EXIT);
307 		else
308 			thread_single_end(p, SINGLE_BOUNDARY);
309 		PROC_UNLOCK(p);
310 	}
311 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
312 		KASSERT(p->p_vmspace != oldvmspace,
313 		    ("oldvmspace still used"));
314 		vmspace_free(oldvmspace);
315 		td->td_pflags &= ~TDP_EXECVMSPC;
316 	}
317 }
318 
319 /*
320  * XXX: kern_execve has the astonishing property of not always returning to
321  * the caller.  If sufficiently bad things happen during the call to
322  * do_execve(), it can end up calling exit1(); as a result, callers must
323  * avoid doing anything which they might need to undo (e.g., allocating
324  * memory).
325  */
326 int
327 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
328 {
329 
330 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
331 	    args->begin_envv - args->begin_argv);
332 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
333 	    args->endp - args->begin_envv);
334 	return (do_execve(td, args, mac_p));
335 }
336 
337 /*
338  * In-kernel implementation of execve().  All arguments are assumed to be
339  * userspace pointers from the passed thread.
340  */
341 static int
342 do_execve(td, args, mac_p)
343 	struct thread *td;
344 	struct image_args *args;
345 	struct mac *mac_p;
346 {
347 	struct proc *p = td->td_proc;
348 	struct nameidata nd;
349 	struct ucred *newcred = NULL, *oldcred;
350 	struct uidinfo *euip = NULL;
351 	register_t *stack_base;
352 	int error, i;
353 	struct image_params image_params, *imgp;
354 	struct vattr attr;
355 	int (*img_first)(struct image_params *);
356 	struct pargs *oldargs = NULL, *newargs = NULL;
357 	struct sigacts *oldsigacts, *newsigacts;
358 #ifdef KTRACE
359 	struct vnode *tracevp = NULL;
360 	struct ucred *tracecred = NULL;
361 #endif
362 	struct vnode *textvp = NULL, *binvp;
363 	cap_rights_t rights;
364 	int credential_changing;
365 	int textset;
366 #ifdef MAC
367 	struct label *interpvplabel = NULL;
368 	int will_transition;
369 #endif
370 #ifdef HWPMC_HOOKS
371 	struct pmckern_procexec pe;
372 #endif
373 	static const char fexecv_proc_title[] = "(fexecv)";
374 
375 	imgp = &image_params;
376 
377 	/*
378 	 * Lock the process and set the P_INEXEC flag to indicate that
379 	 * it should be left alone until we're done here.  This is
380 	 * necessary to avoid race conditions - e.g. in ptrace() -
381 	 * that might allow a local user to illicitly obtain elevated
382 	 * privileges.
383 	 */
384 	PROC_LOCK(p);
385 	KASSERT((p->p_flag & P_INEXEC) == 0,
386 	    ("%s(): process already has P_INEXEC flag", __func__));
387 	p->p_flag |= P_INEXEC;
388 	PROC_UNLOCK(p);
389 
390 	/*
391 	 * Initialize part of the common data
392 	 */
393 	bzero(imgp, sizeof(*imgp));
394 	imgp->proc = p;
395 	imgp->attr = &attr;
396 	imgp->args = args;
397 
398 #ifdef MAC
399 	error = mac_execve_enter(imgp, mac_p);
400 	if (error)
401 		goto exec_fail;
402 #endif
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 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
442 		 */
443 		error = fgetvp_exec(td, args->fd,
444 		    cap_rights_init(&rights, CAP_FEXECVE), &binvp);
445 		if (error)
446 			goto exec_fail;
447 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
448 		AUDIT_ARG_VNODE1(binvp);
449 		imgp->vp = binvp;
450 	}
451 
452 	/*
453 	 * Check file permissions (also 'opens' file)
454 	 */
455 	error = exec_check_permissions(imgp);
456 	if (error)
457 		goto exec_fail_dealloc;
458 
459 	imgp->object = imgp->vp->v_object;
460 	if (imgp->object != NULL)
461 		vm_object_reference(imgp->object);
462 
463 	/*
464 	 * Set VV_TEXT now so no one can write to the executable while we're
465 	 * activating it.
466 	 *
467 	 * Remember if this was set before and unset it in case this is not
468 	 * actually an executable image.
469 	 */
470 	textset = VOP_IS_TEXT(imgp->vp);
471 	VOP_SET_TEXT(imgp->vp);
472 
473 	error = exec_map_first_page(imgp);
474 	if (error)
475 		goto exec_fail_dealloc;
476 
477 	imgp->proc->p_osrel = 0;
478 	/*
479 	 *	If the current process has a special image activator it
480 	 *	wants to try first, call it.   For example, emulating shell
481 	 *	scripts differently.
482 	 */
483 	error = -1;
484 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
485 		error = img_first(imgp);
486 
487 	/*
488 	 *	Loop through the list of image activators, calling each one.
489 	 *	An activator returns -1 if there is no match, 0 on success,
490 	 *	and an error otherwise.
491 	 */
492 	for (i = 0; error == -1 && execsw[i]; ++i) {
493 		if (execsw[i]->ex_imgact == NULL ||
494 		    execsw[i]->ex_imgact == img_first) {
495 			continue;
496 		}
497 		error = (*execsw[i]->ex_imgact)(imgp);
498 	}
499 
500 	if (error) {
501 		if (error == -1) {
502 			if (textset == 0)
503 				VOP_UNSET_TEXT(imgp->vp);
504 			error = ENOEXEC;
505 		}
506 		goto exec_fail_dealloc;
507 	}
508 
509 	/*
510 	 * Special interpreter operation, cleanup and loop up to try to
511 	 * activate the interpreter.
512 	 */
513 	if (imgp->interpreted) {
514 		exec_unmap_first_page(imgp);
515 		/*
516 		 * VV_TEXT needs to be unset for scripts.  There is a short
517 		 * period before we determine that something is a script where
518 		 * VV_TEXT will be set. The vnode lock is held over this
519 		 * entire period so nothing should illegitimately be blocked.
520 		 */
521 		VOP_UNSET_TEXT(imgp->vp);
522 		/* free name buffer and old vnode */
523 		if (args->fname != NULL)
524 			NDFREE(&nd, NDF_ONLY_PNBUF);
525 #ifdef MAC
526 		mac_execve_interpreter_enter(binvp, &interpvplabel);
527 #endif
528 		if (imgp->opened) {
529 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
530 			imgp->opened = 0;
531 		}
532 		vput(binvp);
533 		vm_object_deallocate(imgp->object);
534 		imgp->object = NULL;
535 		/* set new name to that of the interpreter */
536 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
537 		    UIO_SYSSPACE, imgp->interpreter_name, td);
538 		args->fname = imgp->interpreter_name;
539 		goto interpret;
540 	}
541 
542 	/*
543 	 * NB: We unlock the vnode here because it is believed that none
544 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
545 	 */
546 	VOP_UNLOCK(imgp->vp, 0);
547 
548 	/*
549 	 * Do the best to calculate the full path to the image file.
550 	 */
551 	if (imgp->auxargs != NULL &&
552 	    ((args->fname != NULL && args->fname[0] == '/') ||
553 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
554 		imgp->execpath = args->fname;
555 
556 	if (disallow_high_osrel &&
557 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
558 		error = ENOEXEC;
559 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
560 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
561 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
562 		goto exec_fail_dealloc;
563 	}
564 
565 	/*
566 	 * Copy out strings (args and env) and initialize stack base
567 	 */
568 	if (p->p_sysent->sv_copyout_strings)
569 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
570 	else
571 		stack_base = exec_copyout_strings(imgp);
572 
573 	/*
574 	 * If custom stack fixup routine present for this process
575 	 * let it do the stack setup.
576 	 * Else stuff argument count as first item on stack
577 	 */
578 	if (p->p_sysent->sv_fixup != NULL)
579 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
580 	else
581 		suword(--stack_base, imgp->args->argc);
582 
583 	/*
584 	 * For security and other reasons, the file descriptor table cannot
585 	 * be shared after an exec.
586 	 */
587 	fdunshare(td);
588 	/* close files on exec */
589 	fdcloseexec(td);
590 
591 	/*
592 	 * Malloc things before we need locks.
593 	 */
594 	i = imgp->args->begin_envv - imgp->args->begin_argv;
595 	/* Cache arguments if they fit inside our allowance */
596 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
597 		newargs = pargs_alloc(i);
598 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
599 	}
600 
601 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
602 
603 	/* Get a reference to the vnode prior to locking the proc */
604 	VREF(binvp);
605 
606 	/*
607 	 * For security and other reasons, signal handlers cannot
608 	 * be shared after an exec. The new process gets a copy of the old
609 	 * handlers. In execsigs(), the new process will have its signals
610 	 * reset.
611 	 */
612 	if (sigacts_shared(p->p_sigacts)) {
613 		oldsigacts = p->p_sigacts;
614 		newsigacts = sigacts_alloc();
615 		sigacts_copy(newsigacts, oldsigacts);
616 	} else {
617 		oldsigacts = NULL;
618 		newsigacts = NULL; /* satisfy gcc */
619 	}
620 
621 	PROC_LOCK(p);
622 	if (oldsigacts)
623 		p->p_sigacts = newsigacts;
624 	oldcred = p->p_ucred;
625 	/* Stop profiling */
626 	stopprofclock(p);
627 
628 	/* reset caught signals */
629 	execsigs(p);
630 
631 	/* name this process - nameiexec(p, ndp) */
632 	bzero(p->p_comm, sizeof(p->p_comm));
633 	if (args->fname)
634 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
635 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
636 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
637 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
638 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
639 #ifdef KTR
640 	sched_clear_tdname(td);
641 #endif
642 
643 	/*
644 	 * mark as execed, wakeup the process that vforked (if any) and tell
645 	 * it that it now has its own resources back
646 	 */
647 	p->p_flag |= P_EXEC;
648 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
649 		p->p_flag2 &= ~P2_NOTRACE;
650 	if (p->p_flag & P_PPWAIT) {
651 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
652 		cv_broadcast(&p->p_pwait);
653 	}
654 
655 	/*
656 	 * Implement image setuid/setgid.
657 	 *
658 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
659 	 * the process is being traced.
660 	 *
661 	 * We disable setuid/setgid/etc in compatibility mode on the basis
662 	 * that most setugid applications are not written with that
663 	 * environment in mind, and will therefore almost certainly operate
664 	 * incorrectly. In principle there's no reason that setugid
665 	 * applications might not be useful in capability mode, so we may want
666 	 * to reconsider this conservative design choice in the future.
667 	 *
668 	 * XXXMAC: For the time being, use NOSUID to also prohibit
669 	 * transitions on the file system.
670 	 */
671 	credential_changing = 0;
672 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
673 	    attr.va_uid;
674 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
675 	    attr.va_gid;
676 #ifdef MAC
677 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
678 	    interpvplabel, imgp);
679 	credential_changing |= will_transition;
680 #endif
681 
682 	if (credential_changing &&
683 #ifdef CAPABILITY_MODE
684 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
685 #endif
686 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
687 	    (p->p_flag & P_TRACED) == 0) {
688 		/*
689 		 * Turn off syscall tracing for set-id programs, except for
690 		 * root.  Record any set-id flags first to make sure that
691 		 * we do not regain any tracing during a possible block.
692 		 */
693 		setsugid(p);
694 
695 #ifdef KTRACE
696 		if (p->p_tracecred != NULL &&
697 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
698 			ktrprocexec(p, &tracecred, &tracevp);
699 #endif
700 		/*
701 		 * Close any file descriptors 0..2 that reference procfs,
702 		 * then make sure file descriptors 0..2 are in use.
703 		 *
704 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
705 		 * taking sleepable locks, so temporarily drop our locks.
706 		 */
707 		PROC_UNLOCK(p);
708 		VOP_UNLOCK(imgp->vp, 0);
709 		fdsetugidsafety(td);
710 		error = fdcheckstd(td);
711 		if (error != 0)
712 			goto done1;
713 		newcred = crdup(oldcred);
714 		euip = uifind(attr.va_uid);
715 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
716 		PROC_LOCK(p);
717 		/*
718 		 * Set the new credentials.
719 		 */
720 		if (attr.va_mode & S_ISUID)
721 			change_euid(newcred, euip);
722 		if (attr.va_mode & S_ISGID)
723 			change_egid(newcred, attr.va_gid);
724 #ifdef MAC
725 		if (will_transition) {
726 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
727 			    interpvplabel, imgp);
728 		}
729 #endif
730 		/*
731 		 * Implement correct POSIX saved-id behavior.
732 		 *
733 		 * XXXMAC: Note that the current logic will save the
734 		 * uid and gid if a MAC domain transition occurs, even
735 		 * though maybe it shouldn't.
736 		 */
737 		change_svuid(newcred, newcred->cr_uid);
738 		change_svgid(newcred, newcred->cr_gid);
739 		proc_set_cred(p, newcred);
740 	} else {
741 		if (oldcred->cr_uid == oldcred->cr_ruid &&
742 		    oldcred->cr_gid == oldcred->cr_rgid)
743 			p->p_flag &= ~P_SUGID;
744 		/*
745 		 * Implement correct POSIX saved-id behavior.
746 		 *
747 		 * XXX: It's not clear that the existing behavior is
748 		 * POSIX-compliant.  A number of sources indicate that the
749 		 * saved uid/gid should only be updated if the new ruid is
750 		 * not equal to the old ruid, or the new euid is not equal
751 		 * to the old euid and the new euid is not equal to the old
752 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
753 		 * Also, this code uses the new (replaced) euid and egid as
754 		 * the source, which may or may not be the right ones to use.
755 		 */
756 		if (oldcred->cr_svuid != oldcred->cr_uid ||
757 		    oldcred->cr_svgid != oldcred->cr_gid) {
758 			PROC_UNLOCK(p);
759 			VOP_UNLOCK(imgp->vp, 0);
760 			newcred = crdup(oldcred);
761 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
762 			PROC_LOCK(p);
763 			change_svuid(newcred, newcred->cr_uid);
764 			change_svgid(newcred, newcred->cr_gid);
765 			proc_set_cred(p, newcred);
766 		}
767 	}
768 
769 	/*
770 	 * Store the vp for use in procfs.  This vnode was referenced prior
771 	 * to locking the proc lock.
772 	 */
773 	textvp = p->p_textvp;
774 	p->p_textvp = binvp;
775 
776 #ifdef KDTRACE_HOOKS
777 	/*
778 	 * Tell the DTrace fasttrap provider about the exec if it
779 	 * has declared an interest.
780 	 */
781 	if (dtrace_fasttrap_exec)
782 		dtrace_fasttrap_exec(p);
783 #endif
784 
785 	/*
786 	 * Notify others that we exec'd, and clear the P_INEXEC flag
787 	 * as we're now a bona fide freshly-execed process.
788 	 */
789 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
790 	p->p_flag &= ~P_INEXEC;
791 
792 	/* clear "fork but no exec" flag, as we _are_ execing */
793 	p->p_acflag &= ~AFORK;
794 
795 	/*
796 	 * Free any previous argument cache and replace it with
797 	 * the new argument cache, if any.
798 	 */
799 	oldargs = p->p_args;
800 	p->p_args = newargs;
801 	newargs = NULL;
802 
803 #ifdef	HWPMC_HOOKS
804 	/*
805 	 * Check if system-wide sampling is in effect or if the
806 	 * current process is using PMCs.  If so, do exec() time
807 	 * processing.  This processing needs to happen AFTER the
808 	 * P_INEXEC flag is cleared.
809 	 *
810 	 * The proc lock needs to be released before taking the PMC
811 	 * SX.
812 	 */
813 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
814 		PROC_UNLOCK(p);
815 		VOP_UNLOCK(imgp->vp, 0);
816 		pe.pm_credentialschanged = credential_changing;
817 		pe.pm_entryaddr = imgp->entry_addr;
818 
819 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
820 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
821 	} else
822 		PROC_UNLOCK(p);
823 #else  /* !HWPMC_HOOKS */
824 	PROC_UNLOCK(p);
825 #endif
826 
827 	/* Set values passed into the program in registers. */
828 	if (p->p_sysent->sv_setregs)
829 		(*p->p_sysent->sv_setregs)(td, imgp,
830 		    (u_long)(uintptr_t)stack_base);
831 	else
832 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
833 
834 	vfs_mark_atime(imgp->vp, td->td_ucred);
835 
836 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
837 
838 	VOP_UNLOCK(imgp->vp, 0);
839 done1:
840 	/*
841 	 * Free any resources malloc'd earlier that we didn't use.
842 	 */
843 	if (euip != NULL)
844 		uifree(euip);
845 	if (newcred != NULL)
846 		crfree(oldcred);
847 
848 	/*
849 	 * Handle deferred decrement of ref counts.
850 	 */
851 	if (textvp != NULL)
852 		vrele(textvp);
853 	if (error != 0)
854 		vrele(binvp);
855 #ifdef KTRACE
856 	if (tracevp != NULL)
857 		vrele(tracevp);
858 	if (tracecred != NULL)
859 		crfree(tracecred);
860 #endif
861 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
862 	pargs_drop(oldargs);
863 	pargs_drop(newargs);
864 	if (oldsigacts != NULL)
865 		sigacts_free(oldsigacts);
866 
867 exec_fail_dealloc:
868 
869 	/*
870 	 * free various allocated resources
871 	 */
872 	if (imgp->firstpage != NULL)
873 		exec_unmap_first_page(imgp);
874 
875 	if (imgp->vp != NULL) {
876 		if (args->fname)
877 			NDFREE(&nd, NDF_ONLY_PNBUF);
878 		if (imgp->opened)
879 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
880 		vput(imgp->vp);
881 	}
882 
883 	if (imgp->object != NULL)
884 		vm_object_deallocate(imgp->object);
885 
886 	free(imgp->freepath, M_TEMP);
887 
888 	if (error == 0) {
889 		PROC_LOCK(p);
890 		td->td_dbgflags |= TDB_EXEC;
891 		PROC_UNLOCK(p);
892 
893 		/*
894 		 * Stop the process here if its stop event mask has
895 		 * the S_EXEC bit set.
896 		 */
897 		STOPEVENT(p, S_EXEC, 0);
898 		goto done2;
899 	}
900 
901 exec_fail:
902 	/* we're done here, clear P_INEXEC */
903 	PROC_LOCK(p);
904 	p->p_flag &= ~P_INEXEC;
905 	PROC_UNLOCK(p);
906 
907 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
908 
909 done2:
910 #ifdef MAC
911 	mac_execve_exit(imgp);
912 	mac_execve_interpreter_exit(interpvplabel);
913 #endif
914 	exec_free_args(args);
915 
916 	if (error && imgp->vmspace_destroyed) {
917 		/* sorry, no more process anymore. exit gracefully */
918 		exit1(td, W_EXITCODE(0, SIGABRT));
919 		/* NOT REACHED */
920 	}
921 
922 #ifdef KTRACE
923 	if (error == 0)
924 		ktrprocctor(p);
925 #endif
926 
927 	return (error);
928 }
929 
930 int
931 exec_map_first_page(imgp)
932 	struct image_params *imgp;
933 {
934 	int rv, i;
935 	int initial_pagein;
936 	vm_page_t ma[VM_INITIAL_PAGEIN];
937 	vm_object_t object;
938 
939 	if (imgp->firstpage != NULL)
940 		exec_unmap_first_page(imgp);
941 
942 	object = imgp->vp->v_object;
943 	if (object == NULL)
944 		return (EACCES);
945 	VM_OBJECT_WLOCK(object);
946 #if VM_NRESERVLEVEL > 0
947 	vm_object_color(object, 0);
948 #endif
949 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
950 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
951 		initial_pagein = VM_INITIAL_PAGEIN;
952 		if (initial_pagein > object->size)
953 			initial_pagein = object->size;
954 		for (i = 1; i < initial_pagein; i++) {
955 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
956 				if (ma[i]->valid)
957 					break;
958 				if (vm_page_tryxbusy(ma[i]))
959 					break;
960 			} else {
961 				ma[i] = vm_page_alloc(object, i,
962 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
963 				if (ma[i] == NULL)
964 					break;
965 			}
966 		}
967 		initial_pagein = i;
968 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
969 		if (rv != VM_PAGER_OK) {
970 			vm_page_lock(ma[0]);
971 			vm_page_free(ma[0]);
972 			vm_page_unlock(ma[0]);
973 			VM_OBJECT_WUNLOCK(object);
974 			return (EIO);
975 		}
976 	}
977 	vm_page_xunbusy(ma[0]);
978 	vm_page_lock(ma[0]);
979 	vm_page_hold(ma[0]);
980 	vm_page_activate(ma[0]);
981 	vm_page_unlock(ma[0]);
982 	VM_OBJECT_WUNLOCK(object);
983 
984 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
985 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
986 
987 	return (0);
988 }
989 
990 void
991 exec_unmap_first_page(imgp)
992 	struct image_params *imgp;
993 {
994 	vm_page_t m;
995 
996 	if (imgp->firstpage != NULL) {
997 		m = sf_buf_page(imgp->firstpage);
998 		sf_buf_free(imgp->firstpage);
999 		imgp->firstpage = NULL;
1000 		vm_page_lock(m);
1001 		vm_page_unhold(m);
1002 		vm_page_unlock(m);
1003 	}
1004 }
1005 
1006 /*
1007  * Destroy old address space, and allocate a new stack
1008  *	The new stack is only SGROWSIZ large because it is grown
1009  *	automatically in trap.c.
1010  */
1011 int
1012 exec_new_vmspace(imgp, sv)
1013 	struct image_params *imgp;
1014 	struct sysentvec *sv;
1015 {
1016 	int error;
1017 	struct proc *p = imgp->proc;
1018 	struct vmspace *vmspace = p->p_vmspace;
1019 	vm_object_t obj;
1020 	struct rlimit rlim_stack;
1021 	vm_offset_t sv_minuser, stack_addr;
1022 	vm_map_t map;
1023 	u_long ssiz;
1024 
1025 	imgp->vmspace_destroyed = 1;
1026 	imgp->sysent = sv;
1027 
1028 	/* May be called with Giant held */
1029 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1030 
1031 	/*
1032 	 * Blow away entire process VM, if address space not shared,
1033 	 * otherwise, create a new VM space so that other threads are
1034 	 * not disrupted
1035 	 */
1036 	map = &vmspace->vm_map;
1037 	if (map_at_zero)
1038 		sv_minuser = sv->sv_minuser;
1039 	else
1040 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1041 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1042 	    vm_map_max(map) == sv->sv_maxuser) {
1043 		shmexit(vmspace);
1044 		pmap_remove_pages(vmspace_pmap(vmspace));
1045 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1046 	} else {
1047 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1048 		if (error)
1049 			return (error);
1050 		vmspace = p->p_vmspace;
1051 		map = &vmspace->vm_map;
1052 	}
1053 
1054 	/* Map a shared page */
1055 	obj = sv->sv_shared_page_obj;
1056 	if (obj != NULL) {
1057 		vm_object_reference(obj);
1058 		error = vm_map_fixed(map, obj, 0,
1059 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1060 		    VM_PROT_READ | VM_PROT_EXECUTE,
1061 		    VM_PROT_READ | VM_PROT_EXECUTE,
1062 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1063 		if (error) {
1064 			vm_object_deallocate(obj);
1065 			return (error);
1066 		}
1067 	}
1068 
1069 	/* Allocate a new stack */
1070 	if (imgp->stack_sz != 0) {
1071 		ssiz = trunc_page(imgp->stack_sz);
1072 		PROC_LOCK(p);
1073 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1074 		PROC_UNLOCK(p);
1075 		if (ssiz > rlim_stack.rlim_max)
1076 			ssiz = rlim_stack.rlim_max;
1077 		if (ssiz > rlim_stack.rlim_cur) {
1078 			rlim_stack.rlim_cur = ssiz;
1079 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1080 		}
1081 	} else if (sv->sv_maxssiz != NULL) {
1082 		ssiz = *sv->sv_maxssiz;
1083 	} else {
1084 		ssiz = maxssiz;
1085 	}
1086 	stack_addr = sv->sv_usrstack - ssiz;
1087 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1088 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1089 		sv->sv_stackprot,
1090 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1091 	if (error)
1092 		return (error);
1093 
1094 	/*
1095 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1096 	 * are still used to enforce the stack rlimit on the process stack.
1097 	 */
1098 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1099 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1100 
1101 	return (0);
1102 }
1103 
1104 /*
1105  * Copy out argument and environment strings from the old process address
1106  * space into the temporary string buffer.
1107  */
1108 int
1109 exec_copyin_args(struct image_args *args, char *fname,
1110     enum uio_seg segflg, char **argv, char **envv)
1111 {
1112 	u_long argp, envp;
1113 	int error;
1114 	size_t length;
1115 
1116 	bzero(args, sizeof(*args));
1117 	if (argv == NULL)
1118 		return (EFAULT);
1119 
1120 	/*
1121 	 * Allocate demand-paged memory for the file name, argument, and
1122 	 * environment strings.
1123 	 */
1124 	error = exec_alloc_args(args);
1125 	if (error != 0)
1126 		return (error);
1127 
1128 	/*
1129 	 * Copy the file name.
1130 	 */
1131 	if (fname != NULL) {
1132 		args->fname = args->buf;
1133 		error = (segflg == UIO_SYSSPACE) ?
1134 		    copystr(fname, args->fname, PATH_MAX, &length) :
1135 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1136 		if (error != 0)
1137 			goto err_exit;
1138 	} else
1139 		length = 0;
1140 
1141 	args->begin_argv = args->buf + length;
1142 	args->endp = args->begin_argv;
1143 	args->stringspace = ARG_MAX;
1144 
1145 	/*
1146 	 * extract arguments first
1147 	 */
1148 	for (;;) {
1149 		error = fueword(argv++, &argp);
1150 		if (error == -1) {
1151 			error = EFAULT;
1152 			goto err_exit;
1153 		}
1154 		if (argp == 0)
1155 			break;
1156 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1157 		    args->stringspace, &length);
1158 		if (error != 0) {
1159 			if (error == ENAMETOOLONG)
1160 				error = E2BIG;
1161 			goto err_exit;
1162 		}
1163 		args->stringspace -= length;
1164 		args->endp += length;
1165 		args->argc++;
1166 	}
1167 
1168 	args->begin_envv = args->endp;
1169 
1170 	/*
1171 	 * extract environment strings
1172 	 */
1173 	if (envv) {
1174 		for (;;) {
1175 			error = fueword(envv++, &envp);
1176 			if (error == -1) {
1177 				error = EFAULT;
1178 				goto err_exit;
1179 			}
1180 			if (envp == 0)
1181 				break;
1182 			error = copyinstr((void *)(uintptr_t)envp,
1183 			    args->endp, args->stringspace, &length);
1184 			if (error != 0) {
1185 				if (error == ENAMETOOLONG)
1186 					error = E2BIG;
1187 				goto err_exit;
1188 			}
1189 			args->stringspace -= length;
1190 			args->endp += length;
1191 			args->envc++;
1192 		}
1193 	}
1194 
1195 	return (0);
1196 
1197 err_exit:
1198 	exec_free_args(args);
1199 	return (error);
1200 }
1201 
1202 /*
1203  * Allocate temporary demand-paged, zero-filled memory for the file name,
1204  * argument, and environment strings.  Returns zero if the allocation succeeds
1205  * and ENOMEM otherwise.
1206  */
1207 int
1208 exec_alloc_args(struct image_args *args)
1209 {
1210 
1211 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1212 	return (args->buf != NULL ? 0 : ENOMEM);
1213 }
1214 
1215 void
1216 exec_free_args(struct image_args *args)
1217 {
1218 
1219 	if (args->buf != NULL) {
1220 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1221 		    PATH_MAX + ARG_MAX);
1222 		args->buf = NULL;
1223 	}
1224 	if (args->fname_buf != NULL) {
1225 		free(args->fname_buf, M_TEMP);
1226 		args->fname_buf = NULL;
1227 	}
1228 }
1229 
1230 /*
1231  * Copy strings out to the new process address space, constructing new arg
1232  * and env vector tables. Return a pointer to the base so that it can be used
1233  * as the initial stack pointer.
1234  */
1235 register_t *
1236 exec_copyout_strings(imgp)
1237 	struct image_params *imgp;
1238 {
1239 	int argc, envc;
1240 	char **vectp;
1241 	char *stringp;
1242 	uintptr_t destp;
1243 	register_t *stack_base;
1244 	struct ps_strings *arginfo;
1245 	struct proc *p;
1246 	size_t execpath_len;
1247 	int szsigcode, szps;
1248 	char canary[sizeof(long) * 8];
1249 
1250 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1251 	/*
1252 	 * Calculate string base and vector table pointers.
1253 	 * Also deal with signal trampoline code for this exec type.
1254 	 */
1255 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1256 		execpath_len = strlen(imgp->execpath) + 1;
1257 	else
1258 		execpath_len = 0;
1259 	p = imgp->proc;
1260 	szsigcode = 0;
1261 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1262 	if (p->p_sysent->sv_sigcode_base == 0) {
1263 		if (p->p_sysent->sv_szsigcode != NULL)
1264 			szsigcode = *(p->p_sysent->sv_szsigcode);
1265 	}
1266 	destp =	(uintptr_t)arginfo;
1267 
1268 	/*
1269 	 * install sigcode
1270 	 */
1271 	if (szsigcode != 0) {
1272 		destp -= szsigcode;
1273 		destp = rounddown2(destp, sizeof(void *));
1274 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1275 	}
1276 
1277 	/*
1278 	 * Copy the image path for the rtld.
1279 	 */
1280 	if (execpath_len != 0) {
1281 		destp -= execpath_len;
1282 		imgp->execpathp = destp;
1283 		copyout(imgp->execpath, (void *)destp, execpath_len);
1284 	}
1285 
1286 	/*
1287 	 * Prepare the canary for SSP.
1288 	 */
1289 	arc4rand(canary, sizeof(canary), 0);
1290 	destp -= sizeof(canary);
1291 	imgp->canary = destp;
1292 	copyout(canary, (void *)destp, sizeof(canary));
1293 	imgp->canarylen = sizeof(canary);
1294 
1295 	/*
1296 	 * Prepare the pagesizes array.
1297 	 */
1298 	destp -= szps;
1299 	destp = rounddown2(destp, sizeof(void *));
1300 	imgp->pagesizes = destp;
1301 	copyout(pagesizes, (void *)destp, szps);
1302 	imgp->pagesizeslen = szps;
1303 
1304 	destp -= ARG_MAX - imgp->args->stringspace;
1305 	destp = rounddown2(destp, sizeof(void *));
1306 
1307 	/*
1308 	 * If we have a valid auxargs ptr, prepare some room
1309 	 * on the stack.
1310 	 */
1311 	if (imgp->auxargs) {
1312 		/*
1313 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1314 		 * lower compatibility.
1315 		 */
1316 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1317 		    (AT_COUNT * 2);
1318 		/*
1319 		 * The '+ 2' is for the null pointers at the end of each of
1320 		 * the arg and env vector sets,and imgp->auxarg_size is room
1321 		 * for argument of Runtime loader.
1322 		 */
1323 		vectp = (char **)(destp - (imgp->args->argc +
1324 		    imgp->args->envc + 2 + imgp->auxarg_size)
1325 		    * sizeof(char *));
1326 	} else {
1327 		/*
1328 		 * The '+ 2' is for the null pointers at the end of each of
1329 		 * the arg and env vector sets
1330 		 */
1331 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1332 		    + 2) * sizeof(char *));
1333 	}
1334 
1335 	/*
1336 	 * vectp also becomes our initial stack base
1337 	 */
1338 	stack_base = (register_t *)vectp;
1339 
1340 	stringp = imgp->args->begin_argv;
1341 	argc = imgp->args->argc;
1342 	envc = imgp->args->envc;
1343 
1344 	/*
1345 	 * Copy out strings - arguments and environment.
1346 	 */
1347 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1348 
1349 	/*
1350 	 * Fill in "ps_strings" struct for ps, w, etc.
1351 	 */
1352 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1353 	suword32(&arginfo->ps_nargvstr, argc);
1354 
1355 	/*
1356 	 * Fill in argument portion of vector table.
1357 	 */
1358 	for (; argc > 0; --argc) {
1359 		suword(vectp++, (long)(intptr_t)destp);
1360 		while (*stringp++ != 0)
1361 			destp++;
1362 		destp++;
1363 	}
1364 
1365 	/* a null vector table pointer separates the argp's from the envp's */
1366 	suword(vectp++, 0);
1367 
1368 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1369 	suword32(&arginfo->ps_nenvstr, envc);
1370 
1371 	/*
1372 	 * Fill in environment portion of vector table.
1373 	 */
1374 	for (; envc > 0; --envc) {
1375 		suword(vectp++, (long)(intptr_t)destp);
1376 		while (*stringp++ != 0)
1377 			destp++;
1378 		destp++;
1379 	}
1380 
1381 	/* end of vector table is a null pointer */
1382 	suword(vectp, 0);
1383 
1384 	return (stack_base);
1385 }
1386 
1387 /*
1388  * Check permissions of file to execute.
1389  *	Called with imgp->vp locked.
1390  *	Return 0 for success or error code on failure.
1391  */
1392 int
1393 exec_check_permissions(imgp)
1394 	struct image_params *imgp;
1395 {
1396 	struct vnode *vp = imgp->vp;
1397 	struct vattr *attr = imgp->attr;
1398 	struct thread *td;
1399 	int error, writecount;
1400 
1401 	td = curthread;
1402 
1403 	/* Get file attributes */
1404 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1405 	if (error)
1406 		return (error);
1407 
1408 #ifdef MAC
1409 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1410 	if (error)
1411 		return (error);
1412 #endif
1413 
1414 	/*
1415 	 * 1) Check if file execution is disabled for the filesystem that
1416 	 *    this file resides on.
1417 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1418 	 *    privileged user will always succeed, and we don't want this
1419 	 *    to happen unless the file really is executable.
1420 	 * 3) Ensure that the file is a regular file.
1421 	 */
1422 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1423 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1424 	    (attr->va_type != VREG))
1425 		return (EACCES);
1426 
1427 	/*
1428 	 * Zero length files can't be exec'd
1429 	 */
1430 	if (attr->va_size == 0)
1431 		return (ENOEXEC);
1432 
1433 	/*
1434 	 *  Check for execute permission to file based on current credentials.
1435 	 */
1436 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1437 	if (error)
1438 		return (error);
1439 
1440 	/*
1441 	 * Check number of open-for-writes on the file and deny execution
1442 	 * if there are any.
1443 	 */
1444 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1445 	if (error != 0)
1446 		return (error);
1447 	if (writecount != 0)
1448 		return (ETXTBSY);
1449 
1450 	/*
1451 	 * Call filesystem specific open routine (which does nothing in the
1452 	 * general case).
1453 	 */
1454 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1455 	if (error == 0)
1456 		imgp->opened = 1;
1457 	return (error);
1458 }
1459 
1460 /*
1461  * Exec handler registration
1462  */
1463 int
1464 exec_register(execsw_arg)
1465 	const struct execsw *execsw_arg;
1466 {
1467 	const struct execsw **es, **xs, **newexecsw;
1468 	int count = 2;	/* New slot and trailing NULL */
1469 
1470 	if (execsw)
1471 		for (es = execsw; *es; es++)
1472 			count++;
1473 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1474 	if (newexecsw == NULL)
1475 		return (ENOMEM);
1476 	xs = newexecsw;
1477 	if (execsw)
1478 		for (es = execsw; *es; es++)
1479 			*xs++ = *es;
1480 	*xs++ = execsw_arg;
1481 	*xs = NULL;
1482 	if (execsw)
1483 		free(execsw, M_TEMP);
1484 	execsw = newexecsw;
1485 	return (0);
1486 }
1487 
1488 int
1489 exec_unregister(execsw_arg)
1490 	const struct execsw *execsw_arg;
1491 {
1492 	const struct execsw **es, **xs, **newexecsw;
1493 	int count = 1;
1494 
1495 	if (execsw == NULL)
1496 		panic("unregister with no handlers left?\n");
1497 
1498 	for (es = execsw; *es; es++) {
1499 		if (*es == execsw_arg)
1500 			break;
1501 	}
1502 	if (*es == NULL)
1503 		return (ENOENT);
1504 	for (es = execsw; *es; es++)
1505 		if (*es != execsw_arg)
1506 			count++;
1507 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1508 	if (newexecsw == NULL)
1509 		return (ENOMEM);
1510 	xs = newexecsw;
1511 	for (es = execsw; *es; es++)
1512 		if (*es != execsw_arg)
1513 			*xs++ = *es;
1514 	*xs = NULL;
1515 	if (execsw)
1516 		free(execsw, M_TEMP);
1517 	execsw = newexecsw;
1518 	return (0);
1519 }
1520