xref: /freebsd/sys/kern/kern_exec.c (revision 3642298923e528d795e3a30ec165d2b469e28b40)
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_hwpmc_hooks.h"
31 #include "opt_ktrace.h"
32 #include "opt_mac.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/eventhandler.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/sysproto.h>
40 #include <sys/signalvar.h>
41 #include <sys/kernel.h>
42 #include <sys/mac.h>
43 #include <sys/mount.h>
44 #include <sys/filedesc.h>
45 #include <sys/fcntl.h>
46 #include <sys/acct.h>
47 #include <sys/exec.h>
48 #include <sys/imgact.h>
49 #include <sys/imgact_elf.h>
50 #include <sys/wait.h>
51 #include <sys/malloc.h>
52 #include <sys/proc.h>
53 #include <sys/pioctl.h>
54 #include <sys/namei.h>
55 #include <sys/resourcevar.h>
56 #include <sys/sf_buf.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
59 #include <sys/shm.h>
60 #include <sys/sysctl.h>
61 #include <sys/vnode.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65 
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_pager.h>
75 
76 #ifdef	HWPMC_HOOKS
77 #include <sys/pmckern.h>
78 #endif
79 
80 #include <machine/reg.h>
81 
82 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
83 
84 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
85 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
86 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
87 static int do_execve(struct thread *td, struct image_args *args,
88     struct mac *mac_p);
89 
90 /* XXX This should be vm_size_t. */
91 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
92     NULL, 0, sysctl_kern_ps_strings, "LU", "");
93 
94 /* XXX This should be vm_size_t. */
95 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
96     NULL, 0, sysctl_kern_usrstack, "LU", "");
97 
98 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
99     NULL, 0, sysctl_kern_stackprot, "I", "");
100 
101 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
102 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
103     &ps_arg_cache_limit, 0, "");
104 
105 static int
106 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
107 {
108 	struct proc *p;
109 	int error;
110 
111 	p = curproc;
112 #ifdef SCTL_MASK32
113 	if (req->flags & SCTL_MASK32) {
114 		unsigned int val;
115 		val = (unsigned int)p->p_sysent->sv_psstrings;
116 		error = SYSCTL_OUT(req, &val, sizeof(val));
117 	} else
118 #endif
119 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
120 		   sizeof(p->p_sysent->sv_psstrings));
121 	return error;
122 }
123 
124 static int
125 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
126 {
127 	struct proc *p;
128 	int error;
129 
130 	p = curproc;
131 #ifdef SCTL_MASK32
132 	if (req->flags & SCTL_MASK32) {
133 		unsigned int val;
134 		val = (unsigned int)p->p_sysent->sv_usrstack;
135 		error = SYSCTL_OUT(req, &val, sizeof(val));
136 	} else
137 #endif
138 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
139 		    sizeof(p->p_sysent->sv_usrstack));
140 	return error;
141 }
142 
143 static int
144 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
145 {
146 	struct proc *p;
147 
148 	p = curproc;
149 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
150 	    sizeof(p->p_sysent->sv_stackprot)));
151 }
152 
153 /*
154  * Each of the items is a pointer to a `const struct execsw', hence the
155  * double pointer here.
156  */
157 static const struct execsw **execsw;
158 
159 #ifndef _SYS_SYSPROTO_H_
160 struct execve_args {
161 	char    *fname;
162 	char    **argv;
163 	char    **envv;
164 };
165 #endif
166 
167 /*
168  * MPSAFE
169  */
170 int
171 execve(td, uap)
172 	struct thread *td;
173 	struct execve_args /* {
174 		char *fname;
175 		char **argv;
176 		char **envv;
177 	} */ *uap;
178 {
179 	int error;
180 	struct image_args args;
181 
182 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
183 	    uap->argv, uap->envv);
184 
185 	if (error == 0)
186 		error = kern_execve(td, &args, NULL);
187 
188 	exec_free_args(&args);
189 
190 	return (error);
191 }
192 
193 #ifndef _SYS_SYSPROTO_H_
194 struct __mac_execve_args {
195 	char	*fname;
196 	char	**argv;
197 	char	**envv;
198 	struct mac	*mac_p;
199 };
200 #endif
201 
202 /*
203  * MPSAFE
204  */
205 int
206 __mac_execve(td, uap)
207 	struct thread *td;
208 	struct __mac_execve_args /* {
209 		char *fname;
210 		char **argv;
211 		char **envv;
212 		struct mac *mac_p;
213 	} */ *uap;
214 {
215 #ifdef MAC
216 	int error;
217 	struct image_args args;
218 
219 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
220 	    uap->argv, uap->envv);
221 
222 	if (error == 0)
223 		error = kern_execve(td, &args, uap->mac_p);
224 
225 	exec_free_args(&args);
226 
227 	return (error);
228 #else
229 	return (ENOSYS);
230 #endif
231 }
232 
233 int
234 kern_execve(td, args, mac_p)
235 	struct thread *td;
236 	struct image_args *args;
237 	struct mac *mac_p;
238 {
239 	struct proc *p = td->td_proc;
240 	int error;
241 
242 	if (p->p_flag & P_HADTHREADS) {
243 		PROC_LOCK(p);
244 		if (thread_single(SINGLE_BOUNDARY)) {
245 			PROC_UNLOCK(p);
246 			return (ERESTART);	/* Try again later. */
247 		}
248 		PROC_UNLOCK(p);
249 	}
250 
251 	error = do_execve(td, args, mac_p);
252 
253 	if (p->p_flag & P_HADTHREADS) {
254 		PROC_LOCK(p);
255 		/*
256 		 * If success, we upgrade to SINGLE_EXIT state to
257 		 * force other threads to suicide.
258 		 */
259 		if (error == 0)
260 			thread_single(SINGLE_EXIT);
261 		else
262 			thread_single_end();
263 		PROC_UNLOCK(p);
264 	}
265 
266 	return (error);
267 }
268 
269 /*
270  * In-kernel implementation of execve().  All arguments are assumed to be
271  * userspace pointers from the passed thread.
272  *
273  * MPSAFE
274  */
275 static int
276 do_execve(td, args, mac_p)
277 	struct thread *td;
278 	struct image_args *args;
279 	struct mac *mac_p;
280 {
281 	struct proc *p = td->td_proc;
282 	struct nameidata nd, *ndp;
283 	struct ucred *newcred = NULL, *oldcred;
284 	struct uidinfo *euip;
285 	register_t *stack_base;
286 	int error, len, i;
287 	struct image_params image_params, *imgp;
288 	struct vattr atimeattr, attr;
289 	int (*img_first)(struct image_params *);
290 	struct pargs *oldargs = NULL, *newargs = NULL;
291 	struct sigacts *oldsigacts, *newsigacts;
292 #ifdef KTRACE
293 	struct vnode *tracevp = NULL;
294 	struct ucred *tracecred = NULL;
295 #endif
296 	struct vnode *textvp = NULL;
297 	int credential_changing;
298 	int vfslocked;
299 	int textset;
300 #ifdef MAC
301 	struct label *interplabel = NULL;
302 	int will_transition;
303 #endif
304 #ifdef HWPMC_HOOKS
305 	struct pmckern_procexec pe;
306 #endif
307 
308 	vfslocked = 0;
309 	imgp = &image_params;
310 
311 	/*
312 	 * Lock the process and set the P_INEXEC flag to indicate that
313 	 * it should be left alone until we're done here.  This is
314 	 * necessary to avoid race conditions - e.g. in ptrace() -
315 	 * that might allow a local user to illicitly obtain elevated
316 	 * privileges.
317 	 */
318 	PROC_LOCK(p);
319 	KASSERT((p->p_flag & P_INEXEC) == 0,
320 	    ("%s(): process already has P_INEXEC flag", __func__));
321 	p->p_flag |= P_INEXEC;
322 	PROC_UNLOCK(p);
323 
324 	/*
325 	 * Initialize part of the common data
326 	 */
327 	imgp->proc = p;
328 	imgp->execlabel = NULL;
329 	imgp->attr = &attr;
330 	imgp->entry_addr = 0;
331 	imgp->vmspace_destroyed = 0;
332 	imgp->interpreted = 0;
333 	imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
334 	imgp->auxargs = NULL;
335 	imgp->vp = NULL;
336 	imgp->object = NULL;
337 	imgp->firstpage = NULL;
338 	imgp->ps_strings = 0;
339 	imgp->auxarg_size = 0;
340 	imgp->args = args;
341 
342 #ifdef MAC
343 	error = mac_execve_enter(imgp, mac_p);
344 	if (error)
345 		goto exec_fail;
346 #endif
347 
348 	imgp->image_header = NULL;
349 
350 	/*
351 	 * Translate the file name. namei() returns a vnode pointer
352 	 *	in ni_vp amoung other things.
353 	 */
354 	ndp = &nd;
355 	NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
356 	    UIO_SYSSPACE, args->fname, td);
357 
358 interpret:
359 	error = namei(ndp);
360 	if (error)
361 		goto exec_fail;
362 
363 	vfslocked = NDHASGIANT(ndp);
364 	imgp->vp = ndp->ni_vp;
365 
366 	/*
367 	 * Check file permissions (also 'opens' file)
368 	 */
369 	error = exec_check_permissions(imgp);
370 	if (error)
371 		goto exec_fail_dealloc;
372 
373 	imgp->object = imgp->vp->v_object;
374 	if (imgp->object != NULL)
375 		vm_object_reference(imgp->object);
376 
377 	/*
378 	 * Set VV_TEXT now so no one can write to the executable while we're
379 	 * activating it.
380 	 *
381 	 * Remember if this was set before and unset it in case this is not
382 	 * actually an executable image.
383 	 */
384 	textset = imgp->vp->v_vflag & VV_TEXT;
385 	imgp->vp->v_vflag |= VV_TEXT;
386 
387 	error = exec_map_first_page(imgp);
388 	if (error)
389 		goto exec_fail_dealloc;
390 
391 	/*
392 	 *	If the current process has a special image activator it
393 	 *	wants to try first, call it.   For example, emulating shell
394 	 *	scripts differently.
395 	 */
396 	error = -1;
397 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
398 		error = img_first(imgp);
399 
400 	/*
401 	 *	Loop through the list of image activators, calling each one.
402 	 *	An activator returns -1 if there is no match, 0 on success,
403 	 *	and an error otherwise.
404 	 */
405 	for (i = 0; error == -1 && execsw[i]; ++i) {
406 		if (execsw[i]->ex_imgact == NULL ||
407 		    execsw[i]->ex_imgact == img_first) {
408 			continue;
409 		}
410 		error = (*execsw[i]->ex_imgact)(imgp);
411 	}
412 
413 	if (error) {
414 		if (error == -1) {
415 			if (textset == 0)
416 				imgp->vp->v_vflag &= ~VV_TEXT;
417 			error = ENOEXEC;
418 		}
419 		goto exec_fail_dealloc;
420 	}
421 
422 	/*
423 	 * Special interpreter operation, cleanup and loop up to try to
424 	 * activate the interpreter.
425 	 */
426 	if (imgp->interpreted) {
427 		exec_unmap_first_page(imgp);
428 		/*
429 		 * VV_TEXT needs to be unset for scripts.  There is a short
430 		 * period before we determine that something is a script where
431 		 * VV_TEXT will be set. The vnode lock is held over this
432 		 * entire period so nothing should illegitimately be blocked.
433 		 */
434 		imgp->vp->v_vflag &= ~VV_TEXT;
435 		/* free name buffer and old vnode */
436 		NDFREE(ndp, NDF_ONLY_PNBUF);
437 #ifdef MAC
438 		interplabel = mac_vnode_label_alloc();
439 		mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
440 #endif
441 		vput(ndp->ni_vp);
442 		vm_object_deallocate(imgp->object);
443 		imgp->object = NULL;
444 		VFS_UNLOCK_GIANT(vfslocked);
445 		vfslocked = 0;
446 		/* set new name to that of the interpreter */
447 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
448 		    UIO_SYSSPACE, imgp->interpreter_name, td);
449 		goto interpret;
450 	}
451 
452 	/*
453 	 * Copy out strings (args and env) and initialize stack base
454 	 */
455 	if (p->p_sysent->sv_copyout_strings)
456 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
457 	else
458 		stack_base = exec_copyout_strings(imgp);
459 
460 	/*
461 	 * If custom stack fixup routine present for this process
462 	 * let it do the stack setup.
463 	 * Else stuff argument count as first item on stack
464 	 */
465 	if (p->p_sysent->sv_fixup != NULL)
466 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
467 	else
468 		suword(--stack_base, imgp->args->argc);
469 
470 	/*
471 	 * For security and other reasons, the file descriptor table cannot
472 	 * be shared after an exec.
473 	 */
474 	fdunshare(p, td);
475 
476 	/*
477 	 * Malloc things before we need locks.
478 	 */
479 	newcred = crget();
480 	euip = uifind(attr.va_uid);
481 	i = imgp->args->begin_envv - imgp->args->begin_argv;
482 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
483 		newargs = pargs_alloc(i);
484 
485 	/* close files on exec */
486 	fdcloseexec(td);
487 
488 	/* Get a reference to the vnode prior to locking the proc */
489 	VREF(ndp->ni_vp);
490 
491 	/*
492 	 * For security and other reasons, signal handlers cannot
493 	 * be shared after an exec. The new process gets a copy of the old
494 	 * handlers. In execsigs(), the new process will have its signals
495 	 * reset.
496 	 */
497 	PROC_LOCK(p);
498 	if (sigacts_shared(p->p_sigacts)) {
499 		oldsigacts = p->p_sigacts;
500 		PROC_UNLOCK(p);
501 		newsigacts = sigacts_alloc();
502 		sigacts_copy(newsigacts, oldsigacts);
503 		PROC_LOCK(p);
504 		p->p_sigacts = newsigacts;
505 	} else
506 		oldsigacts = NULL;
507 
508 	/* Stop profiling */
509 	stopprofclock(p);
510 
511 	/* reset caught signals */
512 	execsigs(p);
513 
514 	/* name this process - nameiexec(p, ndp) */
515 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
516 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
517 	p->p_comm[len] = 0;
518 
519 	/*
520 	 * mark as execed, wakeup the process that vforked (if any) and tell
521 	 * it that it now has its own resources back
522 	 */
523 	p->p_flag |= P_EXEC;
524 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
525 		p->p_flag &= ~P_PPWAIT;
526 		wakeup(p->p_pptr);
527 	}
528 
529 	/*
530 	 * Implement image setuid/setgid.
531 	 *
532 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
533 	 * the process is being traced.
534 	 *
535 	 * XXXMAC: For the time being, use NOSUID to also prohibit
536 	 * transitions on the file system.
537 	 */
538 	oldcred = p->p_ucred;
539 	credential_changing = 0;
540 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
541 	    attr.va_uid;
542 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
543 	    attr.va_gid;
544 #ifdef MAC
545 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
546 	    interplabel, imgp);
547 	credential_changing |= will_transition;
548 #endif
549 
550 	if (credential_changing &&
551 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
552 	    (p->p_flag & P_TRACED) == 0) {
553 		/*
554 		 * Turn off syscall tracing for set-id programs, except for
555 		 * root.  Record any set-id flags first to make sure that
556 		 * we do not regain any tracing during a possible block.
557 		 */
558 		setsugid(p);
559 #ifdef KTRACE
560 		if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) {
561 			mtx_lock(&ktrace_mtx);
562 			p->p_traceflag = 0;
563 			tracevp = p->p_tracevp;
564 			p->p_tracevp = NULL;
565 			tracecred = p->p_tracecred;
566 			p->p_tracecred = NULL;
567 			mtx_unlock(&ktrace_mtx);
568 		}
569 #endif
570 		/*
571 		 * Close any file descriptors 0..2 that reference procfs,
572 		 * then make sure file descriptors 0..2 are in use.
573 		 *
574 		 * setugidsafety() may call closef() and then pfind()
575 		 * which may grab the process lock.
576 		 * fdcheckstd() may call falloc() which may block to
577 		 * allocate memory, so temporarily drop the process lock.
578 		 */
579 		PROC_UNLOCK(p);
580 		setugidsafety(td);
581 		error = fdcheckstd(td);
582 		if (error != 0)
583 			goto done1;
584 		PROC_LOCK(p);
585 		/*
586 		 * Set the new credentials.
587 		 */
588 		crcopy(newcred, oldcred);
589 		if (attr.va_mode & VSUID)
590 			change_euid(newcred, euip);
591 		if (attr.va_mode & VSGID)
592 			change_egid(newcred, attr.va_gid);
593 #ifdef MAC
594 		if (will_transition) {
595 			mac_execve_transition(oldcred, newcred, imgp->vp,
596 			    interplabel, imgp);
597 		}
598 #endif
599 		/*
600 		 * Implement correct POSIX saved-id behavior.
601 		 *
602 		 * XXXMAC: Note that the current logic will save the
603 		 * uid and gid if a MAC domain transition occurs, even
604 		 * though maybe it shouldn't.
605 		 */
606 		change_svuid(newcred, newcred->cr_uid);
607 		change_svgid(newcred, newcred->cr_gid);
608 		p->p_ucred = newcred;
609 		newcred = NULL;
610 	} else {
611 		if (oldcred->cr_uid == oldcred->cr_ruid &&
612 		    oldcred->cr_gid == oldcred->cr_rgid)
613 			p->p_flag &= ~P_SUGID;
614 		/*
615 		 * Implement correct POSIX saved-id behavior.
616 		 *
617 		 * XXX: It's not clear that the existing behavior is
618 		 * POSIX-compliant.  A number of sources indicate that the
619 		 * saved uid/gid should only be updated if the new ruid is
620 		 * not equal to the old ruid, or the new euid is not equal
621 		 * to the old euid and the new euid is not equal to the old
622 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
623 		 * Also, this code uses the new (replaced) euid and egid as
624 		 * the source, which may or may not be the right ones to use.
625 		 */
626 		if (oldcred->cr_svuid != oldcred->cr_uid ||
627 		    oldcred->cr_svgid != oldcred->cr_gid) {
628 			crcopy(newcred, oldcred);
629 			change_svuid(newcred, newcred->cr_uid);
630 			change_svgid(newcred, newcred->cr_gid);
631 			p->p_ucred = newcred;
632 			newcred = NULL;
633 		}
634 	}
635 
636 	/*
637 	 * Store the vp for use in procfs.  This vnode was referenced prior
638 	 * to locking the proc lock.
639 	 */
640 	textvp = p->p_textvp;
641 	p->p_textvp = ndp->ni_vp;
642 
643 	/*
644 	 * Notify others that we exec'd, and clear the P_INEXEC flag
645 	 * as we're now a bona fide freshly-execed process.
646 	 */
647 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
648 	p->p_flag &= ~P_INEXEC;
649 
650 	/*
651 	 * If tracing the process, trap to debugger so breakpoints
652 	 * can be set before the program executes.
653 	 * Use tdsignal to deliver signal to current thread, use
654 	 * psignal may cause the signal to be delivered to wrong thread
655 	 * because that thread will exit, remember we are going to enter
656 	 * single thread mode.
657 	 */
658 	if (p->p_flag & P_TRACED)
659 		tdsignal(td, SIGTRAP, SIGTARGET_TD);
660 
661 	/* clear "fork but no exec" flag, as we _are_ execing */
662 	p->p_acflag &= ~AFORK;
663 
664 	/* Free any previous argument cache */
665 	oldargs = p->p_args;
666 	p->p_args = NULL;
667 
668 	/* Cache arguments if they fit inside our allowance */
669 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
670 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
671 		p->p_args = newargs;
672 		newargs = NULL;
673 	}
674 
675 #ifdef	HWPMC_HOOKS
676 	/*
677 	 * Check if system-wide sampling is in effect or if the
678 	 * current process is using PMCs.  If so, do exec() time
679 	 * processing.  This processing needs to happen AFTER the
680 	 * P_INEXEC flag is cleared.
681 	 *
682 	 * The proc lock needs to be released before taking the PMC
683 	 * SX.
684 	 */
685 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
686 		PROC_UNLOCK(p);
687 		pe.pm_credentialschanged = credential_changing;
688 		pe.pm_entryaddr = imgp->entry_addr;
689 
690 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
691 	} else
692 		PROC_UNLOCK(p);
693 #else  /* !HWPMC_HOOKS */
694 	PROC_UNLOCK(p);
695 #endif
696 
697 	/* Set values passed into the program in registers. */
698 	if (p->p_sysent->sv_setregs)
699 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
700 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
701 	else
702 		exec_setregs(td, imgp->entry_addr,
703 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
704 
705 	/*
706 	 * Here we should update the access time of the file.  This must
707 	 * be implemented by the underlying filesystem in the same way as
708 	 * access timestamps for a VOP_READ() because we want to avoid
709 	 * blocking and/or I/O, and have not called vn_start_write().
710 	 */
711 	if ((imgp->vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
712 		VATTR_NULL(&atimeattr);
713 		atimeattr.va_vaflags |= VA_EXECVE_ATIME;
714 		(void)VOP_SETATTR(imgp->vp, &atimeattr, td->td_ucred, td);
715 	}
716 
717 done1:
718 	/*
719 	 * Free any resources malloc'd earlier that we didn't use.
720 	 */
721 	uifree(euip);
722 	if (newcred == NULL)
723 		crfree(oldcred);
724 	else
725 		crfree(newcred);
726 	/*
727 	 * Handle deferred decrement of ref counts.
728 	 */
729 	if (textvp != NULL)
730 		vrele(textvp);
731 	if (ndp->ni_vp && error != 0)
732 		vrele(ndp->ni_vp);
733 #ifdef KTRACE
734 	if (tracevp != NULL)
735 		vrele(tracevp);
736 	if (tracecred != NULL)
737 		crfree(tracecred);
738 #endif
739 	if (oldargs != NULL)
740 		pargs_drop(oldargs);
741 	if (newargs != NULL)
742 		pargs_drop(newargs);
743 	if (oldsigacts != NULL)
744 		sigacts_free(oldsigacts);
745 
746 exec_fail_dealloc:
747 
748 	/*
749 	 * free various allocated resources
750 	 */
751 	if (imgp->firstpage != NULL)
752 		exec_unmap_first_page(imgp);
753 
754 	if (imgp->vp != NULL) {
755 		NDFREE(ndp, NDF_ONLY_PNBUF);
756 		vput(imgp->vp);
757 	}
758 
759 	if (imgp->object != NULL)
760 		vm_object_deallocate(imgp->object);
761 
762 	if (error == 0) {
763 		/*
764 		 * Stop the process here if its stop event mask has
765 		 * the S_EXEC bit set.
766 		 */
767 		STOPEVENT(p, S_EXEC, 0);
768 		goto done2;
769 	}
770 
771 exec_fail:
772 	/* we're done here, clear P_INEXEC */
773 	PROC_LOCK(p);
774 	p->p_flag &= ~P_INEXEC;
775 	PROC_UNLOCK(p);
776 
777 	if (imgp->vmspace_destroyed) {
778 		/* sorry, no more process anymore. exit gracefully */
779 #ifdef MAC
780 		mac_execve_exit(imgp);
781 		if (interplabel != NULL)
782 			mac_vnode_label_free(interplabel);
783 #endif
784 		VFS_UNLOCK_GIANT(vfslocked);
785 		exit1(td, W_EXITCODE(0, SIGABRT));
786 		/* NOT REACHED */
787 		error = 0;
788 	}
789 done2:
790 #ifdef MAC
791 	mac_execve_exit(imgp);
792 	if (interplabel != NULL)
793 		mac_vnode_label_free(interplabel);
794 #endif
795 	VFS_UNLOCK_GIANT(vfslocked);
796 	return (error);
797 }
798 
799 int
800 exec_map_first_page(imgp)
801 	struct image_params *imgp;
802 {
803 	int rv, i;
804 	int initial_pagein;
805 	vm_page_t ma[VM_INITIAL_PAGEIN];
806 	vm_object_t object;
807 
808 	if (imgp->firstpage != NULL)
809 		exec_unmap_first_page(imgp);
810 
811 	object = imgp->vp->v_object;
812 	if (object == NULL)
813 		return (EACCES);
814 	VM_OBJECT_LOCK(object);
815 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
816 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
817 		initial_pagein = VM_INITIAL_PAGEIN;
818 		if (initial_pagein > object->size)
819 			initial_pagein = object->size;
820 		for (i = 1; i < initial_pagein; i++) {
821 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
822 				if (ma[i]->valid)
823 					break;
824 				vm_page_lock_queues();
825 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
826 					vm_page_unlock_queues();
827 					break;
828 				}
829 				vm_page_busy(ma[i]);
830 				vm_page_unlock_queues();
831 			} else {
832 				ma[i] = vm_page_alloc(object, i,
833 				    VM_ALLOC_NORMAL);
834 				if (ma[i] == NULL)
835 					break;
836 			}
837 		}
838 		initial_pagein = i;
839 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
840 		ma[0] = vm_page_lookup(object, 0);
841 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
842 		    (ma[0]->valid == 0)) {
843 			if (ma[0]) {
844 				vm_page_lock_queues();
845 				pmap_remove_all(ma[0]);
846 				vm_page_free(ma[0]);
847 				vm_page_unlock_queues();
848 			}
849 			VM_OBJECT_UNLOCK(object);
850 			return (EIO);
851 		}
852 	}
853 	vm_page_lock_queues();
854 	vm_page_hold(ma[0]);
855 	vm_page_wakeup(ma[0]);
856 	vm_page_unlock_queues();
857 	VM_OBJECT_UNLOCK(object);
858 
859 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
860 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
861 
862 	return (0);
863 }
864 
865 void
866 exec_unmap_first_page(imgp)
867 	struct image_params *imgp;
868 {
869 	vm_page_t m;
870 
871 	if (imgp->firstpage != NULL) {
872 		m = sf_buf_page(imgp->firstpage);
873 		sf_buf_free(imgp->firstpage);
874 		imgp->firstpage = NULL;
875 		vm_page_lock_queues();
876 		vm_page_unhold(m);
877 		vm_page_unlock_queues();
878 	}
879 }
880 
881 /*
882  * Destroy old address space, and allocate a new stack
883  *	The new stack is only SGROWSIZ large because it is grown
884  *	automatically in trap.c.
885  */
886 int
887 exec_new_vmspace(imgp, sv)
888 	struct image_params *imgp;
889 	struct sysentvec *sv;
890 {
891 	int error;
892 	struct proc *p = imgp->proc;
893 	struct vmspace *vmspace = p->p_vmspace;
894 	vm_offset_t stack_addr;
895 	vm_map_t map;
896 
897 	imgp->vmspace_destroyed = 1;
898 
899 	/* Called with Giant held, do not depend on it! */
900 	EVENTHANDLER_INVOKE(process_exec, p);
901 
902 	/*
903 	 * Here is as good a place as any to do any resource limit cleanups.
904 	 * This is needed if a 64 bit binary exec's a 32 bit binary - the
905 	 * data size limit may need to be changed to a value that makes
906 	 * sense for the 32 bit binary.
907 	 */
908 	if (sv->sv_fixlimits != NULL)
909 		sv->sv_fixlimits(imgp);
910 
911 	/*
912 	 * Blow away entire process VM, if address space not shared,
913 	 * otherwise, create a new VM space so that other threads are
914 	 * not disrupted
915 	 */
916 	map = &vmspace->vm_map;
917 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
918 	    vm_map_max(map) == sv->sv_maxuser) {
919 		shmexit(vmspace);
920 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
921 		    vm_map_max(map));
922 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
923 	} else {
924 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
925 		vmspace = p->p_vmspace;
926 		map = &vmspace->vm_map;
927 	}
928 
929 	/* Allocate a new stack */
930 	stack_addr = sv->sv_usrstack - maxssiz;
931 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
932 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
933 	if (error)
934 		return (error);
935 
936 #ifdef __ia64__
937 	/* Allocate a new register stack */
938 	stack_addr = IA64_BACKINGSTORE;
939 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
940 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
941 	if (error)
942 		return (error);
943 #endif
944 
945 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
946 	 * VM_STACK case, but they are still used to monitor the size of the
947 	 * process stack so we can check the stack rlimit.
948 	 */
949 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
950 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
951 
952 	return (0);
953 }
954 
955 /*
956  * Copy out argument and environment strings from the old process
957  *	address space into the temporary string buffer.
958  */
959 int
960 exec_copyin_args(struct image_args *args, char *fname,
961     enum uio_seg segflg, char **argv, char **envv)
962 {
963 	char *argp, *envp;
964 	int error;
965 	size_t length;
966 
967 	error = 0;
968 
969 	bzero(args, sizeof(*args));
970 	if (argv == NULL)
971 		return (EFAULT);
972 	/*
973 	 * Allocate temporary demand zeroed space for argument and
974 	 *	environment strings:
975 	 *
976 	 * o ARG_MAX for argument and environment;
977 	 * o MAXSHELLCMDLEN for the name of interpreters.
978 	 */
979 	args->buf = (char *) kmem_alloc_wait(exec_map,
980 	    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
981 	if (args->buf == NULL)
982 		return (ENOMEM);
983 	args->begin_argv = args->buf;
984 	args->endp = args->begin_argv;
985 	args->stringspace = ARG_MAX;
986 
987 	args->fname = args->buf + ARG_MAX;
988 
989 	/*
990 	 * Copy the file name.
991 	 */
992 	error = (segflg == UIO_SYSSPACE) ?
993 	    copystr(fname, args->fname, PATH_MAX, &length) :
994 	    copyinstr(fname, args->fname, PATH_MAX, &length);
995 	if (error != 0)
996 		return (error);
997 
998 	/*
999 	 * extract arguments first
1000 	 */
1001 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1002 		if (argp == (caddr_t) -1)
1003 			return (EFAULT);
1004 		if ((error = copyinstr(argp, args->endp,
1005 		    args->stringspace, &length))) {
1006 			if (error == ENAMETOOLONG)
1007 				return (E2BIG);
1008 			return (error);
1009 		}
1010 		args->stringspace -= length;
1011 		args->endp += length;
1012 		args->argc++;
1013 	}
1014 
1015 	args->begin_envv = args->endp;
1016 
1017 	/*
1018 	 * extract environment strings
1019 	 */
1020 	if (envv) {
1021 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1022 			if (envp == (caddr_t)-1)
1023 				return (EFAULT);
1024 			if ((error = copyinstr(envp, args->endp,
1025 			    args->stringspace, &length))) {
1026 				if (error == ENAMETOOLONG)
1027 					return (E2BIG);
1028 				return (error);
1029 			}
1030 			args->stringspace -= length;
1031 			args->endp += length;
1032 			args->envc++;
1033 		}
1034 	}
1035 
1036 	return (0);
1037 }
1038 
1039 void
1040 exec_free_args(struct image_args *args)
1041 {
1042 
1043 	if (args->buf) {
1044 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1045 		    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1046 		args->buf = NULL;
1047 	}
1048 }
1049 
1050 /*
1051  * Copy strings out to the new process address space, constructing
1052  *	new arg and env vector tables. Return a pointer to the base
1053  *	so that it can be used as the initial stack pointer.
1054  */
1055 register_t *
1056 exec_copyout_strings(imgp)
1057 	struct image_params *imgp;
1058 {
1059 	int argc, envc;
1060 	char **vectp;
1061 	char *stringp, *destp;
1062 	register_t *stack_base;
1063 	struct ps_strings *arginfo;
1064 	struct proc *p;
1065 	int szsigcode;
1066 
1067 	/*
1068 	 * Calculate string base and vector table pointers.
1069 	 * Also deal with signal trampoline code for this exec type.
1070 	 */
1071 	p = imgp->proc;
1072 	szsigcode = 0;
1073 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1074 	if (p->p_sysent->sv_szsigcode != NULL)
1075 		szsigcode = *(p->p_sysent->sv_szsigcode);
1076 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1077 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1078 
1079 	/*
1080 	 * install sigcode
1081 	 */
1082 	if (szsigcode)
1083 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1084 		    szsigcode), szsigcode);
1085 
1086 	/*
1087 	 * If we have a valid auxargs ptr, prepare some room
1088 	 * on the stack.
1089 	 */
1090 	if (imgp->auxargs) {
1091 		/*
1092 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1093 		 * lower compatibility.
1094 		 */
1095 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1096 		    (AT_COUNT * 2);
1097 		/*
1098 		 * The '+ 2' is for the null pointers at the end of each of
1099 		 * the arg and env vector sets,and imgp->auxarg_size is room
1100 		 * for argument of Runtime loader.
1101 		 */
1102 		vectp = (char **)(destp - (imgp->args->argc +
1103 		    imgp->args->envc + 2 + imgp->auxarg_size) *
1104 		    sizeof(char *));
1105 
1106 	} else {
1107 		/*
1108 		 * The '+ 2' is for the null pointers at the end of each of
1109 		 * the arg and env vector sets
1110 		 */
1111 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1112 		    sizeof(char *));
1113 	}
1114 
1115 	/*
1116 	 * vectp also becomes our initial stack base
1117 	 */
1118 	stack_base = (register_t *)vectp;
1119 
1120 	stringp = imgp->args->begin_argv;
1121 	argc = imgp->args->argc;
1122 	envc = imgp->args->envc;
1123 
1124 	/*
1125 	 * Copy out strings - arguments and environment.
1126 	 */
1127 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1128 
1129 	/*
1130 	 * Fill in "ps_strings" struct for ps, w, etc.
1131 	 */
1132 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1133 	suword(&arginfo->ps_nargvstr, argc);
1134 
1135 	/*
1136 	 * Fill in argument portion of vector table.
1137 	 */
1138 	for (; argc > 0; --argc) {
1139 		suword(vectp++, (long)(intptr_t)destp);
1140 		while (*stringp++ != 0)
1141 			destp++;
1142 		destp++;
1143 	}
1144 
1145 	/* a null vector table pointer separates the argp's from the envp's */
1146 	suword(vectp++, 0);
1147 
1148 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1149 	suword(&arginfo->ps_nenvstr, envc);
1150 
1151 	/*
1152 	 * Fill in environment portion of vector table.
1153 	 */
1154 	for (; envc > 0; --envc) {
1155 		suword(vectp++, (long)(intptr_t)destp);
1156 		while (*stringp++ != 0)
1157 			destp++;
1158 		destp++;
1159 	}
1160 
1161 	/* end of vector table is a null pointer */
1162 	suword(vectp, 0);
1163 
1164 	return (stack_base);
1165 }
1166 
1167 /*
1168  * Check permissions of file to execute.
1169  *	Called with imgp->vp locked.
1170  *	Return 0 for success or error code on failure.
1171  */
1172 int
1173 exec_check_permissions(imgp)
1174 	struct image_params *imgp;
1175 {
1176 	struct vnode *vp = imgp->vp;
1177 	struct vattr *attr = imgp->attr;
1178 	struct thread *td;
1179 	int error;
1180 
1181 	td = curthread;			/* XXXKSE */
1182 
1183 	/* Get file attributes */
1184 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1185 	if (error)
1186 		return (error);
1187 
1188 #ifdef MAC
1189 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1190 	if (error)
1191 		return (error);
1192 #endif
1193 
1194 	/*
1195 	 * 1) Check if file execution is disabled for the filesystem that this
1196 	 *	file resides on.
1197 	 * 2) Insure that at least one execute bit is on - otherwise root
1198 	 *	will always succeed, and we don't want to happen unless the
1199 	 *	file really is executable.
1200 	 * 3) Insure that the file is a regular file.
1201 	 */
1202 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1203 	    ((attr->va_mode & 0111) == 0) ||
1204 	    (attr->va_type != VREG))
1205 		return (EACCES);
1206 
1207 	/*
1208 	 * Zero length files can't be exec'd
1209 	 */
1210 	if (attr->va_size == 0)
1211 		return (ENOEXEC);
1212 
1213 	/*
1214 	 *  Check for execute permission to file based on current credentials.
1215 	 */
1216 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1217 	if (error)
1218 		return (error);
1219 
1220 	/*
1221 	 * Check number of open-for-writes on the file and deny execution
1222 	 * if there are any.
1223 	 */
1224 	if (vp->v_writecount)
1225 		return (ETXTBSY);
1226 
1227 	/*
1228 	 * Call filesystem specific open routine (which does nothing in the
1229 	 * general case).
1230 	 */
1231 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1232 	return (error);
1233 }
1234 
1235 /*
1236  * Exec handler registration
1237  */
1238 int
1239 exec_register(execsw_arg)
1240 	const struct execsw *execsw_arg;
1241 {
1242 	const struct execsw **es, **xs, **newexecsw;
1243 	int count = 2;	/* New slot and trailing NULL */
1244 
1245 	if (execsw)
1246 		for (es = execsw; *es; es++)
1247 			count++;
1248 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1249 	if (newexecsw == NULL)
1250 		return (ENOMEM);
1251 	xs = newexecsw;
1252 	if (execsw)
1253 		for (es = execsw; *es; es++)
1254 			*xs++ = *es;
1255 	*xs++ = execsw_arg;
1256 	*xs = NULL;
1257 	if (execsw)
1258 		free(execsw, M_TEMP);
1259 	execsw = newexecsw;
1260 	return (0);
1261 }
1262 
1263 int
1264 exec_unregister(execsw_arg)
1265 	const struct execsw *execsw_arg;
1266 {
1267 	const struct execsw **es, **xs, **newexecsw;
1268 	int count = 1;
1269 
1270 	if (execsw == NULL)
1271 		panic("unregister with no handlers left?\n");
1272 
1273 	for (es = execsw; *es; es++) {
1274 		if (*es == execsw_arg)
1275 			break;
1276 	}
1277 	if (*es == NULL)
1278 		return (ENOENT);
1279 	for (es = execsw; *es; es++)
1280 		if (*es != execsw_arg)
1281 			count++;
1282 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1283 	if (newexecsw == NULL)
1284 		return (ENOMEM);
1285 	xs = newexecsw;
1286 	for (es = execsw; *es; es++)
1287 		if (*es != execsw_arg)
1288 			*xs++ = *es;
1289 	*xs = NULL;
1290 	if (execsw)
1291 		free(execsw, M_TEMP);
1292 	execsw = newexecsw;
1293 	return (0);
1294 }
1295