xref: /freebsd/sys/kern/kern_exec.c (revision 65dcb5bcb1993631f47d8acc207904411e135758)
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 int coredump_pack_fileinfo = 1;
104 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
105     &coredump_pack_fileinfo, 0,
106     "Enable file path packing in 'procstat -f' coredump notes");
107 
108 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
110 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
111 static int do_execve(struct thread *td, struct image_args *args,
112     struct mac *mac_p);
113 
114 /* XXX This should be vm_size_t. */
115 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
116     NULL, 0, sysctl_kern_ps_strings, "LU", "");
117 
118 /* XXX This should be vm_size_t. */
119 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
120     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
121 
122 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
123     NULL, 0, sysctl_kern_stackprot, "I", "");
124 
125 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
126 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
127     &ps_arg_cache_limit, 0, "");
128 
129 static int disallow_high_osrel;
130 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
131     &disallow_high_osrel, 0,
132     "Disallow execution of binaries built for higher version of the world");
133 
134 static int map_at_zero = 0;
135 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
136     "Permit processes to map an object at virtual address 0.");
137 
138 static int
139 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
140 {
141 	struct proc *p;
142 	int error;
143 
144 	p = curproc;
145 #ifdef SCTL_MASK32
146 	if (req->flags & SCTL_MASK32) {
147 		unsigned int val;
148 		val = (unsigned int)p->p_sysent->sv_psstrings;
149 		error = SYSCTL_OUT(req, &val, sizeof(val));
150 	} else
151 #endif
152 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
153 		   sizeof(p->p_sysent->sv_psstrings));
154 	return error;
155 }
156 
157 static int
158 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
159 {
160 	struct proc *p;
161 	int error;
162 
163 	p = curproc;
164 #ifdef SCTL_MASK32
165 	if (req->flags & SCTL_MASK32) {
166 		unsigned int val;
167 		val = (unsigned int)p->p_sysent->sv_usrstack;
168 		error = SYSCTL_OUT(req, &val, sizeof(val));
169 	} else
170 #endif
171 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
172 		    sizeof(p->p_sysent->sv_usrstack));
173 	return error;
174 }
175 
176 static int
177 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
178 {
179 	struct proc *p;
180 
181 	p = curproc;
182 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
183 	    sizeof(p->p_sysent->sv_stackprot)));
184 }
185 
186 /*
187  * Each of the items is a pointer to a `const struct execsw', hence the
188  * double pointer here.
189  */
190 static const struct execsw **execsw;
191 
192 #ifndef _SYS_SYSPROTO_H_
193 struct execve_args {
194 	char    *fname;
195 	char    **argv;
196 	char    **envv;
197 };
198 #endif
199 
200 int
201 sys_execve(struct thread *td, struct execve_args *uap)
202 {
203 	struct image_args args;
204 	struct vmspace *oldvmspace;
205 	int error;
206 
207 	error = pre_execve(td, &oldvmspace);
208 	if (error != 0)
209 		return (error);
210 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
211 	    uap->argv, uap->envv);
212 	if (error == 0)
213 		error = kern_execve(td, &args, NULL);
214 	post_execve(td, error, oldvmspace);
215 	return (error);
216 }
217 
218 #ifndef _SYS_SYSPROTO_H_
219 struct fexecve_args {
220 	int	fd;
221 	char	**argv;
222 	char	**envv;
223 }
224 #endif
225 int
226 sys_fexecve(struct thread *td, struct fexecve_args *uap)
227 {
228 	struct image_args args;
229 	struct vmspace *oldvmspace;
230 	int error;
231 
232 	error = pre_execve(td, &oldvmspace);
233 	if (error != 0)
234 		return (error);
235 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
236 	    uap->argv, uap->envv);
237 	if (error == 0) {
238 		args.fd = uap->fd;
239 		error = kern_execve(td, &args, NULL);
240 	}
241 	post_execve(td, error, oldvmspace);
242 	return (error);
243 }
244 
245 #ifndef _SYS_SYSPROTO_H_
246 struct __mac_execve_args {
247 	char	*fname;
248 	char	**argv;
249 	char	**envv;
250 	struct mac	*mac_p;
251 };
252 #endif
253 
254 int
255 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
256 {
257 #ifdef MAC
258 	struct image_args args;
259 	struct vmspace *oldvmspace;
260 	int error;
261 
262 	error = pre_execve(td, &oldvmspace);
263 	if (error != 0)
264 		return (error);
265 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
266 	    uap->argv, uap->envv);
267 	if (error == 0)
268 		error = kern_execve(td, &args, uap->mac_p);
269 	post_execve(td, error, oldvmspace);
270 	return (error);
271 #else
272 	return (ENOSYS);
273 #endif
274 }
275 
276 int
277 pre_execve(struct thread *td, struct vmspace **oldvmspace)
278 {
279 	struct proc *p;
280 	int error;
281 
282 	KASSERT(td == curthread, ("non-current thread %p", td));
283 	error = 0;
284 	p = td->td_proc;
285 	if ((p->p_flag & P_HADTHREADS) != 0) {
286 		PROC_LOCK(p);
287 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
288 			error = ERESTART;
289 		PROC_UNLOCK(p);
290 	}
291 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
292 	    ("nested execve"));
293 	*oldvmspace = p->p_vmspace;
294 	return (error);
295 }
296 
297 void
298 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
299 {
300 	struct proc *p;
301 
302 	KASSERT(td == curthread, ("non-current thread %p", td));
303 	p = td->td_proc;
304 	if ((p->p_flag & P_HADTHREADS) != 0) {
305 		PROC_LOCK(p);
306 		/*
307 		 * If success, we upgrade to SINGLE_EXIT state to
308 		 * force other threads to suicide.
309 		 */
310 		if (error == 0)
311 			thread_single(p, SINGLE_EXIT);
312 		else
313 			thread_single_end(p, SINGLE_BOUNDARY);
314 		PROC_UNLOCK(p);
315 	}
316 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
317 		KASSERT(p->p_vmspace != oldvmspace,
318 		    ("oldvmspace still used"));
319 		vmspace_free(oldvmspace);
320 		td->td_pflags &= ~TDP_EXECVMSPC;
321 	}
322 }
323 
324 /*
325  * XXX: kern_execve has the astonishing property of not always returning to
326  * the caller.  If sufficiently bad things happen during the call to
327  * do_execve(), it can end up calling exit1(); as a result, callers must
328  * avoid doing anything which they might need to undo (e.g., allocating
329  * memory).
330  */
331 int
332 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
333 {
334 
335 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
336 	    args->begin_envv - args->begin_argv);
337 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
338 	    args->endp - args->begin_envv);
339 	return (do_execve(td, args, mac_p));
340 }
341 
342 /*
343  * In-kernel implementation of execve().  All arguments are assumed to be
344  * userspace pointers from the passed thread.
345  */
346 static int
347 do_execve(td, args, mac_p)
348 	struct thread *td;
349 	struct image_args *args;
350 	struct mac *mac_p;
351 {
352 	struct proc *p = td->td_proc;
353 	struct nameidata nd;
354 	struct ucred *newcred = NULL, *oldcred;
355 	struct uidinfo *euip = NULL;
356 	register_t *stack_base;
357 	int error, i;
358 	struct image_params image_params, *imgp;
359 	struct vattr attr;
360 	int (*img_first)(struct image_params *);
361 	struct pargs *oldargs = NULL, *newargs = NULL;
362 	struct sigacts *oldsigacts, *newsigacts;
363 #ifdef KTRACE
364 	struct vnode *tracevp = NULL;
365 	struct ucred *tracecred = NULL;
366 #endif
367 	struct vnode *oldtextvp = NULL, *newtextvp;
368 	cap_rights_t rights;
369 	int credential_changing;
370 	int textset;
371 #ifdef MAC
372 	struct label *interpvplabel = NULL;
373 	int will_transition;
374 #endif
375 #ifdef HWPMC_HOOKS
376 	struct pmckern_procexec pe;
377 #endif
378 	static const char fexecv_proc_title[] = "(fexecv)";
379 
380 	imgp = &image_params;
381 
382 	/*
383 	 * Lock the process and set the P_INEXEC flag to indicate that
384 	 * it should be left alone until we're done here.  This is
385 	 * necessary to avoid race conditions - e.g. in ptrace() -
386 	 * that might allow a local user to illicitly obtain elevated
387 	 * privileges.
388 	 */
389 	PROC_LOCK(p);
390 	KASSERT((p->p_flag & P_INEXEC) == 0,
391 	    ("%s(): process already has P_INEXEC flag", __func__));
392 	p->p_flag |= P_INEXEC;
393 	PROC_UNLOCK(p);
394 
395 	/*
396 	 * Initialize part of the common data
397 	 */
398 	bzero(imgp, sizeof(*imgp));
399 	imgp->proc = p;
400 	imgp->attr = &attr;
401 	imgp->args = args;
402 
403 #ifdef MAC
404 	error = mac_execve_enter(imgp, mac_p);
405 	if (error)
406 		goto exec_fail;
407 #endif
408 
409 	/*
410 	 * Translate the file name. namei() returns a vnode pointer
411 	 *	in ni_vp amoung other things.
412 	 *
413 	 * XXXAUDIT: It would be desirable to also audit the name of the
414 	 * interpreter if this is an interpreted binary.
415 	 */
416 	if (args->fname != NULL) {
417 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
418 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
419 	}
420 
421 	SDT_PROBE1(proc, kernel, , exec, args->fname);
422 
423 interpret:
424 	if (args->fname != NULL) {
425 #ifdef CAPABILITY_MODE
426 		/*
427 		 * While capability mode can't reach this point via direct
428 		 * path arguments to execve(), we also don't allow
429 		 * interpreters to be used in capability mode (for now).
430 		 * Catch indirect lookups and return a permissions error.
431 		 */
432 		if (IN_CAPABILITY_MODE(td)) {
433 			error = ECAPMODE;
434 			goto exec_fail;
435 		}
436 #endif
437 		error = namei(&nd);
438 		if (error)
439 			goto exec_fail;
440 
441 		newtextvp = nd.ni_vp;
442 		imgp->vp = newtextvp;
443 	} else {
444 		AUDIT_ARG_FD(args->fd);
445 		/*
446 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
447 		 */
448 		error = fgetvp_exec(td, args->fd,
449 		    cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
450 		if (error)
451 			goto exec_fail;
452 		vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
453 		AUDIT_ARG_VNODE1(newtextvp);
454 		imgp->vp = newtextvp;
455 	}
456 
457 	/*
458 	 * Check file permissions (also 'opens' file)
459 	 */
460 	error = exec_check_permissions(imgp);
461 	if (error)
462 		goto exec_fail_dealloc;
463 
464 	imgp->object = imgp->vp->v_object;
465 	if (imgp->object != NULL)
466 		vm_object_reference(imgp->object);
467 
468 	/*
469 	 * Set VV_TEXT now so no one can write to the executable while we're
470 	 * activating it.
471 	 *
472 	 * Remember if this was set before and unset it in case this is not
473 	 * actually an executable image.
474 	 */
475 	textset = VOP_IS_TEXT(imgp->vp);
476 	VOP_SET_TEXT(imgp->vp);
477 
478 	error = exec_map_first_page(imgp);
479 	if (error)
480 		goto exec_fail_dealloc;
481 
482 	imgp->proc->p_osrel = 0;
483 	/*
484 	 *	If the current process has a special image activator it
485 	 *	wants to try first, call it.   For example, emulating shell
486 	 *	scripts differently.
487 	 */
488 	error = -1;
489 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
490 		error = img_first(imgp);
491 
492 	/*
493 	 *	Loop through the list of image activators, calling each one.
494 	 *	An activator returns -1 if there is no match, 0 on success,
495 	 *	and an error otherwise.
496 	 */
497 	for (i = 0; error == -1 && execsw[i]; ++i) {
498 		if (execsw[i]->ex_imgact == NULL ||
499 		    execsw[i]->ex_imgact == img_first) {
500 			continue;
501 		}
502 		error = (*execsw[i]->ex_imgact)(imgp);
503 	}
504 
505 	if (error) {
506 		if (error == -1) {
507 			if (textset == 0)
508 				VOP_UNSET_TEXT(imgp->vp);
509 			error = ENOEXEC;
510 		}
511 		goto exec_fail_dealloc;
512 	}
513 
514 	/*
515 	 * Special interpreter operation, cleanup and loop up to try to
516 	 * activate the interpreter.
517 	 */
518 	if (imgp->interpreted) {
519 		exec_unmap_first_page(imgp);
520 		/*
521 		 * VV_TEXT needs to be unset for scripts.  There is a short
522 		 * period before we determine that something is a script where
523 		 * VV_TEXT will be set. The vnode lock is held over this
524 		 * entire period so nothing should illegitimately be blocked.
525 		 */
526 		VOP_UNSET_TEXT(imgp->vp);
527 		/* free name buffer and old vnode */
528 		if (args->fname != NULL)
529 			NDFREE(&nd, NDF_ONLY_PNBUF);
530 #ifdef MAC
531 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
532 #endif
533 		if (imgp->opened) {
534 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
535 			imgp->opened = 0;
536 		}
537 		vput(newtextvp);
538 		vm_object_deallocate(imgp->object);
539 		imgp->object = NULL;
540 		/* set new name to that of the interpreter */
541 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
542 		    UIO_SYSSPACE, imgp->interpreter_name, td);
543 		args->fname = imgp->interpreter_name;
544 		goto interpret;
545 	}
546 
547 	/*
548 	 * NB: We unlock the vnode here because it is believed that none
549 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
550 	 */
551 	VOP_UNLOCK(imgp->vp, 0);
552 
553 	/*
554 	 * Do the best to calculate the full path to the image file.
555 	 */
556 	if (imgp->auxargs != NULL &&
557 	    ((args->fname != NULL && args->fname[0] == '/') ||
558 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
559 		imgp->execpath = args->fname;
560 
561 	if (disallow_high_osrel &&
562 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
563 		error = ENOEXEC;
564 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
565 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
566 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
567 		goto exec_fail_dealloc;
568 	}
569 
570 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
571 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
572 		sys_cap_enter(td, NULL);
573 
574 	/*
575 	 * Copy out strings (args and env) and initialize stack base
576 	 */
577 	if (p->p_sysent->sv_copyout_strings)
578 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
579 	else
580 		stack_base = exec_copyout_strings(imgp);
581 
582 	/*
583 	 * If custom stack fixup routine present for this process
584 	 * let it do the stack setup.
585 	 * Else stuff argument count as first item on stack
586 	 */
587 	if (p->p_sysent->sv_fixup != NULL)
588 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
589 	else
590 		suword(--stack_base, imgp->args->argc);
591 
592 	if (args->fdp != NULL) {
593 		/* Install a brand new file descriptor table. */
594 		fdinstall_remapped(td, args->fdp);
595 		args->fdp = NULL;
596 	} else {
597 		/*
598 		 * Keep on using the existing file descriptor table. For
599 		 * security and other reasons, the file descriptor table
600 		 * cannot be shared after an exec.
601 		 */
602 		fdunshare(td);
603 		/* close files on exec */
604 		fdcloseexec(td);
605 	}
606 
607 	/*
608 	 * Malloc things before we need locks.
609 	 */
610 	i = imgp->args->begin_envv - imgp->args->begin_argv;
611 	/* Cache arguments if they fit inside our allowance */
612 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
613 		newargs = pargs_alloc(i);
614 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
615 	}
616 
617 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
618 
619 	/*
620 	 * For security and other reasons, signal handlers cannot
621 	 * be shared after an exec. The new process gets a copy of the old
622 	 * handlers. In execsigs(), the new process will have its signals
623 	 * reset.
624 	 */
625 	if (sigacts_shared(p->p_sigacts)) {
626 		oldsigacts = p->p_sigacts;
627 		newsigacts = sigacts_alloc();
628 		sigacts_copy(newsigacts, oldsigacts);
629 	} else {
630 		oldsigacts = NULL;
631 		newsigacts = NULL; /* satisfy gcc */
632 	}
633 
634 	PROC_LOCK(p);
635 	if (oldsigacts)
636 		p->p_sigacts = newsigacts;
637 	oldcred = p->p_ucred;
638 	/* Stop profiling */
639 	stopprofclock(p);
640 
641 	/* reset caught signals */
642 	execsigs(p);
643 
644 	/* name this process - nameiexec(p, ndp) */
645 	bzero(p->p_comm, sizeof(p->p_comm));
646 	if (args->fname)
647 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
648 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
649 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
650 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
651 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
652 #ifdef KTR
653 	sched_clear_tdname(td);
654 #endif
655 
656 	/*
657 	 * mark as execed, wakeup the process that vforked (if any) and tell
658 	 * it that it now has its own resources back
659 	 */
660 	p->p_flag |= P_EXEC;
661 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
662 		p->p_flag2 &= ~P2_NOTRACE;
663 	if (p->p_flag & P_PPWAIT) {
664 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
665 		cv_broadcast(&p->p_pwait);
666 	}
667 
668 	/*
669 	 * Implement image setuid/setgid.
670 	 *
671 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
672 	 * the process is being traced.
673 	 *
674 	 * We disable setuid/setgid/etc in compatibility mode on the basis
675 	 * that most setugid applications are not written with that
676 	 * environment in mind, and will therefore almost certainly operate
677 	 * incorrectly. In principle there's no reason that setugid
678 	 * applications might not be useful in capability mode, so we may want
679 	 * to reconsider this conservative design choice in the future.
680 	 *
681 	 * XXXMAC: For the time being, use NOSUID to also prohibit
682 	 * transitions on the file system.
683 	 */
684 	credential_changing = 0;
685 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
686 	    attr.va_uid;
687 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
688 	    attr.va_gid;
689 #ifdef MAC
690 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
691 	    interpvplabel, imgp);
692 	credential_changing |= will_transition;
693 #endif
694 
695 	if (credential_changing &&
696 #ifdef CAPABILITY_MODE
697 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
698 #endif
699 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
700 	    (p->p_flag & P_TRACED) == 0) {
701 		/*
702 		 * Turn off syscall tracing for set-id programs, except for
703 		 * root.  Record any set-id flags first to make sure that
704 		 * we do not regain any tracing during a possible block.
705 		 */
706 		setsugid(p);
707 
708 #ifdef KTRACE
709 		if (p->p_tracecred != NULL &&
710 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
711 			ktrprocexec(p, &tracecred, &tracevp);
712 #endif
713 		/*
714 		 * Close any file descriptors 0..2 that reference procfs,
715 		 * then make sure file descriptors 0..2 are in use.
716 		 *
717 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
718 		 * taking sleepable locks, so temporarily drop our locks.
719 		 */
720 		PROC_UNLOCK(p);
721 		VOP_UNLOCK(imgp->vp, 0);
722 		fdsetugidsafety(td);
723 		error = fdcheckstd(td);
724 		if (error != 0)
725 			goto done1;
726 		newcred = crdup(oldcred);
727 		euip = uifind(attr.va_uid);
728 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
729 		PROC_LOCK(p);
730 		/*
731 		 * Set the new credentials.
732 		 */
733 		if (attr.va_mode & S_ISUID)
734 			change_euid(newcred, euip);
735 		if (attr.va_mode & S_ISGID)
736 			change_egid(newcred, attr.va_gid);
737 #ifdef MAC
738 		if (will_transition) {
739 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
740 			    interpvplabel, imgp);
741 		}
742 #endif
743 		/*
744 		 * Implement correct POSIX saved-id behavior.
745 		 *
746 		 * XXXMAC: Note that the current logic will save the
747 		 * uid and gid if a MAC domain transition occurs, even
748 		 * though maybe it shouldn't.
749 		 */
750 		change_svuid(newcred, newcred->cr_uid);
751 		change_svgid(newcred, newcred->cr_gid);
752 		proc_set_cred(p, newcred);
753 	} else {
754 		if (oldcred->cr_uid == oldcred->cr_ruid &&
755 		    oldcred->cr_gid == oldcred->cr_rgid)
756 			p->p_flag &= ~P_SUGID;
757 		/*
758 		 * Implement correct POSIX saved-id behavior.
759 		 *
760 		 * XXX: It's not clear that the existing behavior is
761 		 * POSIX-compliant.  A number of sources indicate that the
762 		 * saved uid/gid should only be updated if the new ruid is
763 		 * not equal to the old ruid, or the new euid is not equal
764 		 * to the old euid and the new euid is not equal to the old
765 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
766 		 * Also, this code uses the new (replaced) euid and egid as
767 		 * the source, which may or may not be the right ones to use.
768 		 */
769 		if (oldcred->cr_svuid != oldcred->cr_uid ||
770 		    oldcred->cr_svgid != oldcred->cr_gid) {
771 			PROC_UNLOCK(p);
772 			VOP_UNLOCK(imgp->vp, 0);
773 			newcred = crdup(oldcred);
774 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
775 			PROC_LOCK(p);
776 			change_svuid(newcred, newcred->cr_uid);
777 			change_svgid(newcred, newcred->cr_gid);
778 			proc_set_cred(p, newcred);
779 		}
780 	}
781 
782 	/*
783 	 * Store the vp for use in procfs.  This vnode was referenced by namei
784 	 * or fgetvp_exec.
785 	 */
786 	oldtextvp = p->p_textvp;
787 	p->p_textvp = newtextvp;
788 
789 #ifdef KDTRACE_HOOKS
790 	/*
791 	 * Tell the DTrace fasttrap provider about the exec if it
792 	 * has declared an interest.
793 	 */
794 	if (dtrace_fasttrap_exec)
795 		dtrace_fasttrap_exec(p);
796 #endif
797 
798 	/*
799 	 * Notify others that we exec'd, and clear the P_INEXEC flag
800 	 * as we're now a bona fide freshly-execed process.
801 	 */
802 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
803 	p->p_flag &= ~P_INEXEC;
804 
805 	/* clear "fork but no exec" flag, as we _are_ execing */
806 	p->p_acflag &= ~AFORK;
807 
808 	/*
809 	 * Free any previous argument cache and replace it with
810 	 * the new argument cache, if any.
811 	 */
812 	oldargs = p->p_args;
813 	p->p_args = newargs;
814 	newargs = NULL;
815 
816 #ifdef	HWPMC_HOOKS
817 	/*
818 	 * Check if system-wide sampling is in effect or if the
819 	 * current process is using PMCs.  If so, do exec() time
820 	 * processing.  This processing needs to happen AFTER the
821 	 * P_INEXEC flag is cleared.
822 	 *
823 	 * The proc lock needs to be released before taking the PMC
824 	 * SX.
825 	 */
826 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
827 		PROC_UNLOCK(p);
828 		VOP_UNLOCK(imgp->vp, 0);
829 		pe.pm_credentialschanged = credential_changing;
830 		pe.pm_entryaddr = imgp->entry_addr;
831 
832 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
833 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
834 	} else
835 		PROC_UNLOCK(p);
836 #else  /* !HWPMC_HOOKS */
837 	PROC_UNLOCK(p);
838 #endif
839 
840 	/* Set values passed into the program in registers. */
841 	if (p->p_sysent->sv_setregs)
842 		(*p->p_sysent->sv_setregs)(td, imgp,
843 		    (u_long)(uintptr_t)stack_base);
844 	else
845 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
846 
847 	vfs_mark_atime(imgp->vp, td->td_ucred);
848 
849 	SDT_PROBE1(proc, kernel, , exec__success, args->fname);
850 
851 	VOP_UNLOCK(imgp->vp, 0);
852 done1:
853 	/*
854 	 * Free any resources malloc'd earlier that we didn't use.
855 	 */
856 	if (euip != NULL)
857 		uifree(euip);
858 	if (newcred != NULL)
859 		crfree(oldcred);
860 
861 	/*
862 	 * Handle deferred decrement of ref counts.
863 	 */
864 	if (oldtextvp != NULL)
865 		vrele(oldtextvp);
866 #ifdef KTRACE
867 	if (tracevp != NULL)
868 		vrele(tracevp);
869 	if (tracecred != NULL)
870 		crfree(tracecred);
871 #endif
872 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
873 	pargs_drop(oldargs);
874 	pargs_drop(newargs);
875 	if (oldsigacts != NULL)
876 		sigacts_free(oldsigacts);
877 
878 exec_fail_dealloc:
879 
880 	/*
881 	 * free various allocated resources
882 	 */
883 	if (imgp->firstpage != NULL)
884 		exec_unmap_first_page(imgp);
885 
886 	if (imgp->vp != NULL) {
887 		if (args->fname)
888 			NDFREE(&nd, NDF_ONLY_PNBUF);
889 		if (imgp->opened)
890 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
891 		if (error != 0)
892 			vput(imgp->vp);
893 		else
894 			VOP_UNLOCK(imgp->vp, 0);
895 	}
896 
897 	if (imgp->object != NULL)
898 		vm_object_deallocate(imgp->object);
899 
900 	free(imgp->freepath, M_TEMP);
901 
902 	if (error == 0) {
903 		PROC_LOCK(p);
904 		td->td_dbgflags |= TDB_EXEC;
905 		PROC_UNLOCK(p);
906 
907 		/*
908 		 * Stop the process here if its stop event mask has
909 		 * the S_EXEC bit set.
910 		 */
911 		STOPEVENT(p, S_EXEC, 0);
912 		goto done2;
913 	}
914 
915 exec_fail:
916 	/* we're done here, clear P_INEXEC */
917 	PROC_LOCK(p);
918 	p->p_flag &= ~P_INEXEC;
919 	PROC_UNLOCK(p);
920 
921 	SDT_PROBE1(proc, kernel, , exec__failure, error);
922 
923 done2:
924 #ifdef MAC
925 	mac_execve_exit(imgp);
926 	mac_execve_interpreter_exit(interpvplabel);
927 #endif
928 	exec_free_args(args);
929 
930 	if (error && imgp->vmspace_destroyed) {
931 		/* sorry, no more process anymore. exit gracefully */
932 		exit1(td, 0, SIGABRT);
933 		/* NOT REACHED */
934 	}
935 
936 #ifdef KTRACE
937 	if (error == 0)
938 		ktrprocctor(p);
939 #endif
940 
941 	return (error);
942 }
943 
944 int
945 exec_map_first_page(imgp)
946 	struct image_params *imgp;
947 {
948 	int rv, i;
949 	int initial_pagein;
950 	vm_page_t ma[VM_INITIAL_PAGEIN];
951 	vm_object_t object;
952 
953 	if (imgp->firstpage != NULL)
954 		exec_unmap_first_page(imgp);
955 
956 	object = imgp->vp->v_object;
957 	if (object == NULL)
958 		return (EACCES);
959 	VM_OBJECT_WLOCK(object);
960 #if VM_NRESERVLEVEL > 0
961 	vm_object_color(object, 0);
962 #endif
963 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
964 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
965 		initial_pagein = VM_INITIAL_PAGEIN;
966 		if (initial_pagein > object->size)
967 			initial_pagein = object->size;
968 		for (i = 1; i < initial_pagein; i++) {
969 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
970 				if (ma[i]->valid)
971 					break;
972 				if (vm_page_tryxbusy(ma[i]))
973 					break;
974 			} else {
975 				ma[i] = vm_page_alloc(object, i,
976 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
977 				if (ma[i] == NULL)
978 					break;
979 			}
980 		}
981 		initial_pagein = i;
982 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
983 		if (rv != VM_PAGER_OK) {
984 			vm_page_lock(ma[0]);
985 			vm_page_free(ma[0]);
986 			vm_page_unlock(ma[0]);
987 			VM_OBJECT_WUNLOCK(object);
988 			return (EIO);
989 		}
990 	}
991 	vm_page_xunbusy(ma[0]);
992 	vm_page_lock(ma[0]);
993 	vm_page_hold(ma[0]);
994 	vm_page_activate(ma[0]);
995 	vm_page_unlock(ma[0]);
996 	VM_OBJECT_WUNLOCK(object);
997 
998 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
999 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1000 
1001 	return (0);
1002 }
1003 
1004 void
1005 exec_unmap_first_page(imgp)
1006 	struct image_params *imgp;
1007 {
1008 	vm_page_t m;
1009 
1010 	if (imgp->firstpage != NULL) {
1011 		m = sf_buf_page(imgp->firstpage);
1012 		sf_buf_free(imgp->firstpage);
1013 		imgp->firstpage = NULL;
1014 		vm_page_lock(m);
1015 		vm_page_unhold(m);
1016 		vm_page_unlock(m);
1017 	}
1018 }
1019 
1020 /*
1021  * Destroy old address space, and allocate a new stack
1022  *	The new stack is only SGROWSIZ large because it is grown
1023  *	automatically in trap.c.
1024  */
1025 int
1026 exec_new_vmspace(imgp, sv)
1027 	struct image_params *imgp;
1028 	struct sysentvec *sv;
1029 {
1030 	int error;
1031 	struct proc *p = imgp->proc;
1032 	struct vmspace *vmspace = p->p_vmspace;
1033 	vm_object_t obj;
1034 	struct rlimit rlim_stack;
1035 	vm_offset_t sv_minuser, stack_addr;
1036 	vm_map_t map;
1037 	u_long ssiz;
1038 
1039 	imgp->vmspace_destroyed = 1;
1040 	imgp->sysent = sv;
1041 
1042 	/* May be called with Giant held */
1043 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1044 
1045 	/*
1046 	 * Blow away entire process VM, if address space not shared,
1047 	 * otherwise, create a new VM space so that other threads are
1048 	 * not disrupted
1049 	 */
1050 	map = &vmspace->vm_map;
1051 	if (map_at_zero)
1052 		sv_minuser = sv->sv_minuser;
1053 	else
1054 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1055 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1056 	    vm_map_max(map) == sv->sv_maxuser) {
1057 		shmexit(vmspace);
1058 		pmap_remove_pages(vmspace_pmap(vmspace));
1059 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1060 	} else {
1061 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1062 		if (error)
1063 			return (error);
1064 		vmspace = p->p_vmspace;
1065 		map = &vmspace->vm_map;
1066 	}
1067 
1068 	/* Map a shared page */
1069 	obj = sv->sv_shared_page_obj;
1070 	if (obj != NULL) {
1071 		vm_object_reference(obj);
1072 		error = vm_map_fixed(map, obj, 0,
1073 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1074 		    VM_PROT_READ | VM_PROT_EXECUTE,
1075 		    VM_PROT_READ | VM_PROT_EXECUTE,
1076 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1077 		if (error) {
1078 			vm_object_deallocate(obj);
1079 			return (error);
1080 		}
1081 	}
1082 
1083 	/* Allocate a new stack */
1084 	if (imgp->stack_sz != 0) {
1085 		ssiz = trunc_page(imgp->stack_sz);
1086 		PROC_LOCK(p);
1087 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1088 		PROC_UNLOCK(p);
1089 		if (ssiz > rlim_stack.rlim_max)
1090 			ssiz = rlim_stack.rlim_max;
1091 		if (ssiz > rlim_stack.rlim_cur) {
1092 			rlim_stack.rlim_cur = ssiz;
1093 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1094 		}
1095 	} else if (sv->sv_maxssiz != NULL) {
1096 		ssiz = *sv->sv_maxssiz;
1097 	} else {
1098 		ssiz = maxssiz;
1099 	}
1100 	stack_addr = sv->sv_usrstack - ssiz;
1101 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1102 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1103 		sv->sv_stackprot,
1104 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1105 	if (error)
1106 		return (error);
1107 
1108 	/*
1109 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1110 	 * are still used to enforce the stack rlimit on the process stack.
1111 	 */
1112 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1113 	vmspace->vm_maxsaddr = (char *)stack_addr;
1114 
1115 	return (0);
1116 }
1117 
1118 /*
1119  * Copy out argument and environment strings from the old process address
1120  * space into the temporary string buffer.
1121  */
1122 int
1123 exec_copyin_args(struct image_args *args, char *fname,
1124     enum uio_seg segflg, char **argv, char **envv)
1125 {
1126 	u_long argp, envp;
1127 	int error;
1128 	size_t length;
1129 
1130 	bzero(args, sizeof(*args));
1131 	if (argv == NULL)
1132 		return (EFAULT);
1133 
1134 	/*
1135 	 * Allocate demand-paged memory for the file name, argument, and
1136 	 * environment strings.
1137 	 */
1138 	error = exec_alloc_args(args);
1139 	if (error != 0)
1140 		return (error);
1141 
1142 	/*
1143 	 * Copy the file name.
1144 	 */
1145 	if (fname != NULL) {
1146 		args->fname = args->buf;
1147 		error = (segflg == UIO_SYSSPACE) ?
1148 		    copystr(fname, args->fname, PATH_MAX, &length) :
1149 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1150 		if (error != 0)
1151 			goto err_exit;
1152 	} else
1153 		length = 0;
1154 
1155 	args->begin_argv = args->buf + length;
1156 	args->endp = args->begin_argv;
1157 	args->stringspace = ARG_MAX;
1158 
1159 	/*
1160 	 * extract arguments first
1161 	 */
1162 	for (;;) {
1163 		error = fueword(argv++, &argp);
1164 		if (error == -1) {
1165 			error = EFAULT;
1166 			goto err_exit;
1167 		}
1168 		if (argp == 0)
1169 			break;
1170 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1171 		    args->stringspace, &length);
1172 		if (error != 0) {
1173 			if (error == ENAMETOOLONG)
1174 				error = E2BIG;
1175 			goto err_exit;
1176 		}
1177 		args->stringspace -= length;
1178 		args->endp += length;
1179 		args->argc++;
1180 	}
1181 
1182 	args->begin_envv = args->endp;
1183 
1184 	/*
1185 	 * extract environment strings
1186 	 */
1187 	if (envv) {
1188 		for (;;) {
1189 			error = fueword(envv++, &envp);
1190 			if (error == -1) {
1191 				error = EFAULT;
1192 				goto err_exit;
1193 			}
1194 			if (envp == 0)
1195 				break;
1196 			error = copyinstr((void *)(uintptr_t)envp,
1197 			    args->endp, args->stringspace, &length);
1198 			if (error != 0) {
1199 				if (error == ENAMETOOLONG)
1200 					error = E2BIG;
1201 				goto err_exit;
1202 			}
1203 			args->stringspace -= length;
1204 			args->endp += length;
1205 			args->envc++;
1206 		}
1207 	}
1208 
1209 	return (0);
1210 
1211 err_exit:
1212 	exec_free_args(args);
1213 	return (error);
1214 }
1215 
1216 int
1217 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1218     const void *data, size_t datalen, const int *fds, size_t fdslen)
1219 {
1220 	struct filedesc *ofdp;
1221 	const char *p;
1222 	int *kfds;
1223 	int error;
1224 
1225 	memset(args, '\0', sizeof(*args));
1226 	ofdp = td->td_proc->p_fd;
1227 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1228 		return (E2BIG);
1229 	error = exec_alloc_args(args);
1230 	if (error != 0)
1231 		return (error);
1232 
1233 	args->begin_argv = args->buf;
1234 	args->stringspace = ARG_MAX;
1235 
1236 	if (datalen > 0) {
1237 		/*
1238 		 * Argument buffer has been provided. Copy it into the
1239 		 * kernel as a single string and add a terminating null
1240 		 * byte.
1241 		 */
1242 		error = copyin(data, args->begin_argv, datalen);
1243 		if (error != 0)
1244 			goto err_exit;
1245 		args->begin_argv[datalen] = '\0';
1246 		args->endp = args->begin_argv + datalen + 1;
1247 		args->stringspace -= datalen + 1;
1248 
1249 		/*
1250 		 * Traditional argument counting. Count the number of
1251 		 * null bytes.
1252 		 */
1253 		for (p = args->begin_argv; p < args->endp; ++p)
1254 			if (*p == '\0')
1255 				++args->argc;
1256 	} else {
1257 		/* No argument buffer provided. */
1258 		args->endp = args->begin_argv;
1259 	}
1260 	/* There are no environment variables. */
1261 	args->begin_envv = args->endp;
1262 
1263 	/* Create new file descriptor table. */
1264 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1265 	error = copyin(fds, kfds, fdslen * sizeof(int));
1266 	if (error != 0) {
1267 		free(kfds, M_TEMP);
1268 		goto err_exit;
1269 	}
1270 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1271 	free(kfds, M_TEMP);
1272 	if (error != 0)
1273 		goto err_exit;
1274 
1275 	return (0);
1276 err_exit:
1277 	exec_free_args(args);
1278 	return (error);
1279 }
1280 
1281 /*
1282  * Allocate temporary demand-paged, zero-filled memory for the file name,
1283  * argument, and environment strings.  Returns zero if the allocation succeeds
1284  * and ENOMEM otherwise.
1285  */
1286 int
1287 exec_alloc_args(struct image_args *args)
1288 {
1289 
1290 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1291 	return (args->buf != NULL ? 0 : ENOMEM);
1292 }
1293 
1294 void
1295 exec_free_args(struct image_args *args)
1296 {
1297 
1298 	if (args->buf != NULL) {
1299 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1300 		    PATH_MAX + ARG_MAX);
1301 		args->buf = NULL;
1302 	}
1303 	if (args->fname_buf != NULL) {
1304 		free(args->fname_buf, M_TEMP);
1305 		args->fname_buf = NULL;
1306 	}
1307 	if (args->fdp != NULL)
1308 		fdescfree_remapped(args->fdp);
1309 }
1310 
1311 /*
1312  * Copy strings out to the new process address space, constructing new arg
1313  * and env vector tables. Return a pointer to the base so that it can be used
1314  * as the initial stack pointer.
1315  */
1316 register_t *
1317 exec_copyout_strings(imgp)
1318 	struct image_params *imgp;
1319 {
1320 	int argc, envc;
1321 	char **vectp;
1322 	char *stringp;
1323 	uintptr_t destp;
1324 	register_t *stack_base;
1325 	struct ps_strings *arginfo;
1326 	struct proc *p;
1327 	size_t execpath_len;
1328 	int szsigcode, szps;
1329 	char canary[sizeof(long) * 8];
1330 
1331 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1332 	/*
1333 	 * Calculate string base and vector table pointers.
1334 	 * Also deal with signal trampoline code for this exec type.
1335 	 */
1336 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1337 		execpath_len = strlen(imgp->execpath) + 1;
1338 	else
1339 		execpath_len = 0;
1340 	p = imgp->proc;
1341 	szsigcode = 0;
1342 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1343 	if (p->p_sysent->sv_sigcode_base == 0) {
1344 		if (p->p_sysent->sv_szsigcode != NULL)
1345 			szsigcode = *(p->p_sysent->sv_szsigcode);
1346 	}
1347 	destp =	(uintptr_t)arginfo;
1348 
1349 	/*
1350 	 * install sigcode
1351 	 */
1352 	if (szsigcode != 0) {
1353 		destp -= szsigcode;
1354 		destp = rounddown2(destp, sizeof(void *));
1355 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1356 	}
1357 
1358 	/*
1359 	 * Copy the image path for the rtld.
1360 	 */
1361 	if (execpath_len != 0) {
1362 		destp -= execpath_len;
1363 		imgp->execpathp = destp;
1364 		copyout(imgp->execpath, (void *)destp, execpath_len);
1365 	}
1366 
1367 	/*
1368 	 * Prepare the canary for SSP.
1369 	 */
1370 	arc4rand(canary, sizeof(canary), 0);
1371 	destp -= sizeof(canary);
1372 	imgp->canary = destp;
1373 	copyout(canary, (void *)destp, sizeof(canary));
1374 	imgp->canarylen = sizeof(canary);
1375 
1376 	/*
1377 	 * Prepare the pagesizes array.
1378 	 */
1379 	destp -= szps;
1380 	destp = rounddown2(destp, sizeof(void *));
1381 	imgp->pagesizes = destp;
1382 	copyout(pagesizes, (void *)destp, szps);
1383 	imgp->pagesizeslen = szps;
1384 
1385 	destp -= ARG_MAX - imgp->args->stringspace;
1386 	destp = rounddown2(destp, sizeof(void *));
1387 
1388 	/*
1389 	 * If we have a valid auxargs ptr, prepare some room
1390 	 * on the stack.
1391 	 */
1392 	if (imgp->auxargs) {
1393 		/*
1394 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1395 		 * lower compatibility.
1396 		 */
1397 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1398 		    (AT_COUNT * 2);
1399 		/*
1400 		 * The '+ 2' is for the null pointers at the end of each of
1401 		 * the arg and env vector sets,and imgp->auxarg_size is room
1402 		 * for argument of Runtime loader.
1403 		 */
1404 		vectp = (char **)(destp - (imgp->args->argc +
1405 		    imgp->args->envc + 2 + imgp->auxarg_size)
1406 		    * sizeof(char *));
1407 	} else {
1408 		/*
1409 		 * The '+ 2' is for the null pointers at the end of each of
1410 		 * the arg and env vector sets
1411 		 */
1412 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1413 		    + 2) * sizeof(char *));
1414 	}
1415 
1416 	/*
1417 	 * vectp also becomes our initial stack base
1418 	 */
1419 	stack_base = (register_t *)vectp;
1420 
1421 	stringp = imgp->args->begin_argv;
1422 	argc = imgp->args->argc;
1423 	envc = imgp->args->envc;
1424 
1425 	/*
1426 	 * Copy out strings - arguments and environment.
1427 	 */
1428 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1429 
1430 	/*
1431 	 * Fill in "ps_strings" struct for ps, w, etc.
1432 	 */
1433 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1434 	suword32(&arginfo->ps_nargvstr, argc);
1435 
1436 	/*
1437 	 * Fill in argument portion of vector table.
1438 	 */
1439 	for (; argc > 0; --argc) {
1440 		suword(vectp++, (long)(intptr_t)destp);
1441 		while (*stringp++ != 0)
1442 			destp++;
1443 		destp++;
1444 	}
1445 
1446 	/* a null vector table pointer separates the argp's from the envp's */
1447 	suword(vectp++, 0);
1448 
1449 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1450 	suword32(&arginfo->ps_nenvstr, envc);
1451 
1452 	/*
1453 	 * Fill in environment portion of vector table.
1454 	 */
1455 	for (; envc > 0; --envc) {
1456 		suword(vectp++, (long)(intptr_t)destp);
1457 		while (*stringp++ != 0)
1458 			destp++;
1459 		destp++;
1460 	}
1461 
1462 	/* end of vector table is a null pointer */
1463 	suword(vectp, 0);
1464 
1465 	return (stack_base);
1466 }
1467 
1468 /*
1469  * Check permissions of file to execute.
1470  *	Called with imgp->vp locked.
1471  *	Return 0 for success or error code on failure.
1472  */
1473 int
1474 exec_check_permissions(imgp)
1475 	struct image_params *imgp;
1476 {
1477 	struct vnode *vp = imgp->vp;
1478 	struct vattr *attr = imgp->attr;
1479 	struct thread *td;
1480 	int error, writecount;
1481 
1482 	td = curthread;
1483 
1484 	/* Get file attributes */
1485 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1486 	if (error)
1487 		return (error);
1488 
1489 #ifdef MAC
1490 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1491 	if (error)
1492 		return (error);
1493 #endif
1494 
1495 	/*
1496 	 * 1) Check if file execution is disabled for the filesystem that
1497 	 *    this file resides on.
1498 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1499 	 *    privileged user will always succeed, and we don't want this
1500 	 *    to happen unless the file really is executable.
1501 	 * 3) Ensure that the file is a regular file.
1502 	 */
1503 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1504 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1505 	    (attr->va_type != VREG))
1506 		return (EACCES);
1507 
1508 	/*
1509 	 * Zero length files can't be exec'd
1510 	 */
1511 	if (attr->va_size == 0)
1512 		return (ENOEXEC);
1513 
1514 	/*
1515 	 *  Check for execute permission to file based on current credentials.
1516 	 */
1517 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1518 	if (error)
1519 		return (error);
1520 
1521 	/*
1522 	 * Check number of open-for-writes on the file and deny execution
1523 	 * if there are any.
1524 	 */
1525 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1526 	if (error != 0)
1527 		return (error);
1528 	if (writecount != 0)
1529 		return (ETXTBSY);
1530 
1531 	/*
1532 	 * Call filesystem specific open routine (which does nothing in the
1533 	 * general case).
1534 	 */
1535 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1536 	if (error == 0)
1537 		imgp->opened = 1;
1538 	return (error);
1539 }
1540 
1541 /*
1542  * Exec handler registration
1543  */
1544 int
1545 exec_register(execsw_arg)
1546 	const struct execsw *execsw_arg;
1547 {
1548 	const struct execsw **es, **xs, **newexecsw;
1549 	int count = 2;	/* New slot and trailing NULL */
1550 
1551 	if (execsw)
1552 		for (es = execsw; *es; es++)
1553 			count++;
1554 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1555 	if (newexecsw == NULL)
1556 		return (ENOMEM);
1557 	xs = newexecsw;
1558 	if (execsw)
1559 		for (es = execsw; *es; es++)
1560 			*xs++ = *es;
1561 	*xs++ = execsw_arg;
1562 	*xs = NULL;
1563 	if (execsw)
1564 		free(execsw, M_TEMP);
1565 	execsw = newexecsw;
1566 	return (0);
1567 }
1568 
1569 int
1570 exec_unregister(execsw_arg)
1571 	const struct execsw *execsw_arg;
1572 {
1573 	const struct execsw **es, **xs, **newexecsw;
1574 	int count = 1;
1575 
1576 	if (execsw == NULL)
1577 		panic("unregister with no handlers left?\n");
1578 
1579 	for (es = execsw; *es; es++) {
1580 		if (*es == execsw_arg)
1581 			break;
1582 	}
1583 	if (*es == NULL)
1584 		return (ENOENT);
1585 	for (es = execsw; *es; es++)
1586 		if (*es != execsw_arg)
1587 			count++;
1588 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1589 	if (newexecsw == NULL)
1590 		return (ENOMEM);
1591 	xs = newexecsw;
1592 	for (es = execsw; *es; es++)
1593 		if (*es != execsw_arg)
1594 			*xs++ = *es;
1595 	*xs = NULL;
1596 	if (execsw)
1597 		free(execsw, M_TEMP);
1598 	execsw = newexecsw;
1599 	return (0);
1600 }
1601