xref: /freebsd/sys/kern/kern_exec.c (revision e653b48c80fb85b2a10372d664a4b55dbdc51dae)
1 /*
2  * Copyright (c) 1993, David Greenman
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_ktrace.h"
31 #include "opt_mac.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/eventhandler.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/sysproto.h>
39 #include <sys/signalvar.h>
40 #include <sys/kernel.h>
41 #include <sys/mac.h>
42 #include <sys/mount.h>
43 #include <sys/filedesc.h>
44 #include <sys/fcntl.h>
45 #include <sys/acct.h>
46 #include <sys/exec.h>
47 #include <sys/imgact.h>
48 #include <sys/imgact_elf.h>
49 #include <sys/wait.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 #include <sys/pioctl.h>
53 #include <sys/namei.h>
54 #include <sys/resourcevar.h>
55 #include <sys/sf_buf.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysent.h>
58 #include <sys/shm.h>
59 #include <sys/sysctl.h>
60 #include <sys/vnode.h>
61 #ifdef KTRACE
62 #include <sys/ktrace.h>
63 #endif
64 
65 #include <vm/vm.h>
66 #include <vm/vm_param.h>
67 #include <vm/pmap.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_kern.h>
71 #include <vm/vm_extern.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_pager.h>
74 
75 #include <machine/reg.h>
76 
77 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
78 
79 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
80 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
81 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
82 static int do_execve(struct thread *td, char *fname, char **argv,
83 	char **envv, struct mac *mac_p);
84 
85 /* XXX This should be vm_size_t. */
86 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
87     NULL, 0, sysctl_kern_ps_strings, "LU", "");
88 
89 /* XXX This should be vm_size_t. */
90 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
91     NULL, 0, sysctl_kern_usrstack, "LU", "");
92 
93 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
94     NULL, 0, sysctl_kern_stackprot, "I", "");
95 
96 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
97 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
98     &ps_arg_cache_limit, 0, "");
99 
100 static int
101 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
102 {
103 	struct proc *p;
104 	int error;
105 
106 	p = curproc;
107 #ifdef SCTL_MASK32
108 	if (req->flags & SCTL_MASK32) {
109 		unsigned int val;
110 		val = (unsigned int)p->p_sysent->sv_psstrings;
111 		error = SYSCTL_OUT(req, &val, sizeof(val));
112 	} else
113 #endif
114 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
115 		   sizeof(p->p_sysent->sv_psstrings));
116 	return error;
117 }
118 
119 static int
120 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
121 {
122 	struct proc *p;
123 	int error;
124 
125 	p = curproc;
126 #ifdef SCTL_MASK32
127 	if (req->flags & SCTL_MASK32) {
128 		unsigned int val;
129 		val = (unsigned int)p->p_sysent->sv_usrstack;
130 		error = SYSCTL_OUT(req, &val, sizeof(val));
131 	} else
132 #endif
133 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
134 		    sizeof(p->p_sysent->sv_usrstack));
135 	return error;
136 }
137 
138 static int
139 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
140 {
141 	struct proc *p;
142 
143 	p = curproc;
144 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
145 	    sizeof(p->p_sysent->sv_stackprot)));
146 }
147 
148 /*
149  * Each of the items is a pointer to a `const struct execsw', hence the
150  * double pointer here.
151  */
152 static const struct execsw **execsw;
153 
154 #ifndef _SYS_SYSPROTO_H_
155 struct execve_args {
156 	char    *fname;
157 	char    **argv;
158 	char    **envv;
159 };
160 #endif
161 
162 /*
163  * MPSAFE
164  */
165 int
166 execve(td, uap)
167 	struct thread *td;
168 	struct execve_args /* {
169 		char *fname;
170 		char **argv;
171 		char **envv;
172 	} */ *uap;
173 {
174 
175 	return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
176 }
177 
178 #ifndef _SYS_SYSPROTO_H_
179 struct __mac_execve_args {
180 	char	*fname;
181 	char	**argv;
182 	char	**envv;
183 	struct mac	*mac_p;
184 };
185 #endif
186 
187 /*
188  * MPSAFE
189  */
190 int
191 __mac_execve(td, uap)
192 	struct thread *td;
193 	struct __mac_execve_args /* {
194 		char *fname;
195 		char **argv;
196 		char **envv;
197 		struct mac *mac_p;
198 	} */ *uap;
199 {
200 
201 #ifdef MAC
202 	return (kern_execve(td, uap->fname, uap->argv, uap->envv,
203 	    uap->mac_p));
204 #else
205 	return (ENOSYS);
206 #endif
207 }
208 
209 int
210 kern_execve(td, fname, argv, envv, mac_p)
211 	struct thread *td;
212 	char *fname;
213 	char **argv;
214 	char **envv;
215 	struct mac *mac_p;
216 {
217 	struct proc *p = td->td_proc;
218 	int error;
219 
220 	if (p->p_flag & P_HADTHREADS) {
221 		PROC_LOCK(p);
222 		if (thread_single(SINGLE_BOUNDARY)) {
223 			PROC_UNLOCK(p);
224 			return (ERESTART);	/* Try again later. */
225 		}
226 		PROC_UNLOCK(p);
227 	}
228 
229 	error = do_execve(td, fname, argv, envv, mac_p);
230 
231 	if (p->p_flag & P_HADTHREADS) {
232 		PROC_LOCK(p);
233 		/*
234 		 * If success, we upgrade to SINGLE_EXIT state to
235 		 * force other threads to suicide.
236 		 */
237 		if (error == 0)
238 			thread_single(SINGLE_EXIT);
239 		else
240 			thread_single_end();
241 		PROC_UNLOCK(p);
242 	}
243 
244 	return (error);
245 }
246 
247 /*
248  * In-kernel implementation of execve().  All arguments are assumed to be
249  * userspace pointers from the passed thread.
250  *
251  * MPSAFE
252  */
253 static int
254 do_execve(td, fname, argv, envv, mac_p)
255 	struct thread *td;
256 	char *fname;
257 	char **argv;
258 	char **envv;
259 	struct mac *mac_p;
260 {
261 	struct proc *p = td->td_proc;
262 	struct nameidata nd, *ndp;
263 	struct ucred *newcred = NULL, *oldcred;
264 	struct uidinfo *euip;
265 	register_t *stack_base;
266 	int error, len, i;
267 	struct image_params image_params, *imgp;
268 	struct vattr attr;
269 	int (*img_first)(struct image_params *);
270 	struct pargs *oldargs = NULL, *newargs = NULL;
271 	struct sigacts *oldsigacts, *newsigacts;
272 #ifdef KTRACE
273 	struct vnode *tracevp = NULL;
274 	struct ucred *tracecred = NULL;
275 #endif
276 	struct vnode *textvp = NULL;
277 	int credential_changing;
278 	int textset;
279 #ifdef MAC
280 	struct label *interplabel = NULL;
281 	int will_transition;
282 #endif
283 
284 	imgp = &image_params;
285 
286 	/*
287 	 * Lock the process and set the P_INEXEC flag to indicate that
288 	 * it should be left alone until we're done here.  This is
289 	 * necessary to avoid race conditions - e.g. in ptrace() -
290 	 * that might allow a local user to illicitly obtain elevated
291 	 * privileges.
292 	 */
293 	PROC_LOCK(p);
294 	KASSERT((p->p_flag & P_INEXEC) == 0,
295 	    ("%s(): process already has P_INEXEC flag", __func__));
296 	p->p_flag |= P_INEXEC;
297 	PROC_UNLOCK(p);
298 
299 	/*
300 	 * Initialize part of the common data
301 	 */
302 	imgp->proc = p;
303 	imgp->userspace_argv = argv;
304 	imgp->userspace_envv = envv;
305 	imgp->execlabel = NULL;
306 	imgp->attr = &attr;
307 	imgp->argc = imgp->envc = 0;
308 	imgp->argv0 = NULL;
309 	imgp->entry_addr = 0;
310 	imgp->vmspace_destroyed = 0;
311 	imgp->interpreted = 0;
312 	imgp->interpreter_name[0] = '\0';
313 	imgp->auxargs = NULL;
314 	imgp->vp = NULL;
315 	imgp->object = NULL;
316 	imgp->firstpage = NULL;
317 	imgp->ps_strings = 0;
318 	imgp->auxarg_size = 0;
319 
320 #ifdef MAC
321 	error = mac_execve_enter(imgp, mac_p);
322 	if (error) {
323 		mtx_lock(&Giant);
324 		goto exec_fail;
325 	}
326 #endif
327 
328 	/*
329 	 * Allocate temporary demand zeroed space for argument and
330 	 *	environment strings
331 	 */
332 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX);
333 	if (imgp->stringbase == NULL) {
334 		error = ENOMEM;
335 		mtx_lock(&Giant);
336 		goto exec_fail;
337 	}
338 	imgp->stringp = imgp->stringbase;
339 	imgp->stringspace = ARG_MAX;
340 	imgp->image_header = NULL;
341 
342 	/*
343 	 * Translate the file name. namei() returns a vnode pointer
344 	 *	in ni_vp amoung other things.
345 	 */
346 	ndp = &nd;
347 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
348 	    UIO_USERSPACE, fname, td);
349 
350 	mtx_lock(&Giant);
351 interpret:
352 
353 	error = namei(ndp);
354 	if (error) {
355 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
356 		    ARG_MAX);
357 		goto exec_fail;
358 	}
359 
360 	imgp->vp = ndp->ni_vp;
361 	imgp->fname = fname;
362 
363 	/*
364 	 * Check file permissions (also 'opens' file)
365 	 */
366 	error = exec_check_permissions(imgp);
367 	if (error)
368 		goto exec_fail_dealloc;
369 
370 	if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0)
371 		vm_object_reference(imgp->object);
372 
373 	/*
374 	 * Set VV_TEXT now so no one can write to the executable while we're
375 	 * activating it.
376 	 *
377 	 * Remember if this was set before and unset it in case this is not
378 	 * actually an executable image.
379 	 */
380 	textset = imgp->vp->v_vflag & VV_TEXT;
381 	imgp->vp->v_vflag |= VV_TEXT;
382 
383 	error = exec_map_first_page(imgp);
384 	if (error)
385 		goto exec_fail_dealloc;
386 
387 	/*
388 	 *	If the current process has a special image activator it
389 	 *	wants to try first, call it.   For example, emulating shell
390 	 *	scripts differently.
391 	 */
392 	error = -1;
393 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
394 		error = img_first(imgp);
395 
396 	/*
397 	 *	Loop through the list of image activators, calling each one.
398 	 *	An activator returns -1 if there is no match, 0 on success,
399 	 *	and an error otherwise.
400 	 */
401 	for (i = 0; error == -1 && execsw[i]; ++i) {
402 		if (execsw[i]->ex_imgact == NULL ||
403 		    execsw[i]->ex_imgact == img_first) {
404 			continue;
405 		}
406 		error = (*execsw[i]->ex_imgact)(imgp);
407 	}
408 
409 	if (error) {
410 		if (error == -1) {
411 			if (textset == 0)
412 				imgp->vp->v_vflag &= ~VV_TEXT;
413 			error = ENOEXEC;
414 		}
415 		goto exec_fail_dealloc;
416 	}
417 
418 	/*
419 	 * Special interpreter operation, cleanup and loop up to try to
420 	 * activate the interpreter.
421 	 */
422 	if (imgp->interpreted) {
423 		exec_unmap_first_page(imgp);
424 		/*
425 		 * VV_TEXT needs to be unset for scripts.  There is a short
426 		 * period before we determine that something is a script where
427 		 * VV_TEXT will be set. The vnode lock is held over this
428 		 * entire period so nothing should illegitimately be blocked.
429 		 */
430 		imgp->vp->v_vflag &= ~VV_TEXT;
431 		/* free name buffer and old vnode */
432 		NDFREE(ndp, NDF_ONLY_PNBUF);
433 #ifdef MAC
434 		interplabel = mac_vnode_label_alloc();
435 		mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
436 #endif
437 		vput(ndp->ni_vp);
438 		vm_object_deallocate(imgp->object);
439 		imgp->object = NULL;
440 		/* set new name to that of the interpreter */
441 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
442 		    UIO_SYSSPACE, imgp->interpreter_name, td);
443 		goto interpret;
444 	}
445 
446 	/*
447 	 * Copy out strings (args and env) and initialize stack base
448 	 */
449 	if (p->p_sysent->sv_copyout_strings)
450 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
451 	else
452 		stack_base = exec_copyout_strings(imgp);
453 
454 	/*
455 	 * If custom stack fixup routine present for this process
456 	 * let it do the stack setup.
457 	 * Else stuff argument count as first item on stack
458 	 */
459 	if (p->p_sysent->sv_fixup != NULL)
460 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
461 	else
462 		suword(--stack_base, imgp->argc);
463 
464 	/*
465 	 * For security and other reasons, the file descriptor table cannot
466 	 * be shared after an exec.
467 	 */
468 	fdunshare(p, td);
469 
470 	/*
471 	 * Malloc things before we need locks.
472 	 */
473 	newcred = crget();
474 	euip = uifind(attr.va_uid);
475 	i = imgp->endargs - imgp->stringbase;
476 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
477 		newargs = pargs_alloc(i);
478 
479 	/* close files on exec */
480 	fdcloseexec(td);
481 
482 	/* Get a reference to the vnode prior to locking the proc */
483 	VREF(ndp->ni_vp);
484 
485 	/*
486 	 * For security and other reasons, signal handlers cannot
487 	 * be shared after an exec. The new process gets a copy of the old
488 	 * handlers. In execsigs(), the new process will have its signals
489 	 * reset.
490 	 */
491 	PROC_LOCK(p);
492 	if (sigacts_shared(p->p_sigacts)) {
493 		oldsigacts = p->p_sigacts;
494 		PROC_UNLOCK(p);
495 		newsigacts = sigacts_alloc();
496 		sigacts_copy(newsigacts, oldsigacts);
497 		PROC_LOCK(p);
498 		p->p_sigacts = newsigacts;
499 	} else
500 		oldsigacts = NULL;
501 
502 	/* Stop profiling */
503 	stopprofclock(p);
504 
505 	/* reset caught signals */
506 	execsigs(p);
507 
508 	/* name this process - nameiexec(p, ndp) */
509 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
510 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
511 	p->p_comm[len] = 0;
512 
513 	/*
514 	 * mark as execed, wakeup the process that vforked (if any) and tell
515 	 * it that it now has its own resources back
516 	 */
517 	p->p_flag |= P_EXEC;
518 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
519 		p->p_flag &= ~P_PPWAIT;
520 		wakeup(p->p_pptr);
521 	}
522 
523 	/*
524 	 * Implement image setuid/setgid.
525 	 *
526 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
527 	 * the process is being traced.
528 	 *
529 	 * XXXMAC: For the time being, use NOSUID to also prohibit
530 	 * transitions on the file system.
531 	 */
532 	oldcred = p->p_ucred;
533 	credential_changing = 0;
534 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
535 	    attr.va_uid;
536 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
537 	    attr.va_gid;
538 #ifdef MAC
539 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
540 	    interplabel, imgp);
541 	credential_changing |= will_transition;
542 #endif
543 
544 	if (credential_changing &&
545 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
546 	    (p->p_flag & P_TRACED) == 0) {
547 		/*
548 		 * Turn off syscall tracing for set-id programs, except for
549 		 * root.  Record any set-id flags first to make sure that
550 		 * we do not regain any tracing during a possible block.
551 		 */
552 		setsugid(p);
553 #ifdef KTRACE
554 		if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) {
555 			mtx_lock(&ktrace_mtx);
556 			p->p_traceflag = 0;
557 			tracevp = p->p_tracevp;
558 			p->p_tracevp = NULL;
559 			tracecred = p->p_tracecred;
560 			p->p_tracecred = NULL;
561 			mtx_unlock(&ktrace_mtx);
562 		}
563 #endif
564 		/*
565 		 * Close any file descriptors 0..2 that reference procfs,
566 		 * then make sure file descriptors 0..2 are in use.
567 		 *
568 		 * setugidsafety() may call closef() and then pfind()
569 		 * which may grab the process lock.
570 		 * fdcheckstd() may call falloc() which may block to
571 		 * allocate memory, so temporarily drop the process lock.
572 		 */
573 		PROC_UNLOCK(p);
574 		setugidsafety(td);
575 		error = fdcheckstd(td);
576 		if (error != 0)
577 			goto done1;
578 		PROC_LOCK(p);
579 		/*
580 		 * Set the new credentials.
581 		 */
582 		crcopy(newcred, oldcred);
583 		if (attr.va_mode & VSUID)
584 			change_euid(newcred, euip);
585 		if (attr.va_mode & VSGID)
586 			change_egid(newcred, attr.va_gid);
587 #ifdef MAC
588 		if (will_transition) {
589 			mac_execve_transition(oldcred, newcred, imgp->vp,
590 			    interplabel, imgp);
591 		}
592 #endif
593 		/*
594 		 * Implement correct POSIX saved-id behavior.
595 		 *
596 		 * XXXMAC: Note that the current logic will save the
597 		 * uid and gid if a MAC domain transition occurs, even
598 		 * though maybe it shouldn't.
599 		 */
600 		change_svuid(newcred, newcred->cr_uid);
601 		change_svgid(newcred, newcred->cr_gid);
602 		p->p_ucred = newcred;
603 		newcred = NULL;
604 	} else {
605 		if (oldcred->cr_uid == oldcred->cr_ruid &&
606 		    oldcred->cr_gid == oldcred->cr_rgid)
607 			p->p_flag &= ~P_SUGID;
608 		/*
609 		 * Implement correct POSIX saved-id behavior.
610 		 *
611 		 * XXX: It's not clear that the existing behavior is
612 		 * POSIX-compliant.  A number of sources indicate that the
613 		 * saved uid/gid should only be updated if the new ruid is
614 		 * not equal to the old ruid, or the new euid is not equal
615 		 * to the old euid and the new euid is not equal to the old
616 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
617 		 * Also, this code uses the new (replaced) euid and egid as
618 		 * the source, which may or may not be the right ones to use.
619 		 */
620 		if (oldcred->cr_svuid != oldcred->cr_uid ||
621 		    oldcred->cr_svgid != oldcred->cr_gid) {
622 			crcopy(newcred, oldcred);
623 			change_svuid(newcred, newcred->cr_uid);
624 			change_svgid(newcred, newcred->cr_gid);
625 			p->p_ucred = newcred;
626 			newcred = NULL;
627 		}
628 	}
629 
630 	/*
631 	 * Store the vp for use in procfs.  This vnode was referenced prior
632 	 * to locking the proc lock.
633 	 */
634 	textvp = p->p_textvp;
635 	p->p_textvp = ndp->ni_vp;
636 
637 	/*
638 	 * Notify others that we exec'd, and clear the P_INEXEC flag
639 	 * as we're now a bona fide freshly-execed process.
640 	 */
641 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
642 	p->p_flag &= ~P_INEXEC;
643 
644 	/*
645 	 * If tracing the process, trap to debugger so breakpoints
646 	 * can be set before the program executes.
647 	 * Use tdsignal to deliver signal to current thread, use
648 	 * psignal may cause the signal to be delivered to wrong thread
649 	 * because that thread will exit, remember we are going to enter
650 	 * single thread mode.
651 	 */
652 	if (p->p_flag & P_TRACED)
653 		tdsignal(td, SIGTRAP, SIGTARGET_TD);
654 
655 	/* clear "fork but no exec" flag, as we _are_ execing */
656 	p->p_acflag &= ~AFORK;
657 
658 	/* Free any previous argument cache */
659 	oldargs = p->p_args;
660 	p->p_args = NULL;
661 
662 	/* Cache arguments if they fit inside our allowance */
663 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
664 		bcopy(imgp->stringbase, newargs->ar_args, i);
665 		p->p_args = newargs;
666 		newargs = NULL;
667 	}
668 	PROC_UNLOCK(p);
669 
670 	/* Set values passed into the program in registers. */
671 	if (p->p_sysent->sv_setregs)
672 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
673 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
674 	else
675 		exec_setregs(td, imgp->entry_addr,
676 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
677 
678 done1:
679 	/*
680 	 * Free any resources malloc'd earlier that we didn't use.
681 	 */
682 	uifree(euip);
683 	if (newcred == NULL)
684 		crfree(oldcred);
685 	else
686 		crfree(newcred);
687 	/*
688 	 * Handle deferred decrement of ref counts.
689 	 */
690 	if (textvp != NULL)
691 		vrele(textvp);
692 	if (ndp->ni_vp && error != 0)
693 		vrele(ndp->ni_vp);
694 #ifdef KTRACE
695 	if (tracevp != NULL)
696 		vrele(tracevp);
697 	if (tracecred != NULL)
698 		crfree(tracecred);
699 #endif
700 	if (oldargs != NULL)
701 		pargs_drop(oldargs);
702 	if (newargs != NULL)
703 		pargs_drop(newargs);
704 	if (oldsigacts != NULL)
705 		sigacts_free(oldsigacts);
706 
707 exec_fail_dealloc:
708 
709 	/*
710 	 * free various allocated resources
711 	 */
712 	if (imgp->firstpage != NULL)
713 		exec_unmap_first_page(imgp);
714 
715 	if (imgp->vp != NULL) {
716 		NDFREE(ndp, NDF_ONLY_PNBUF);
717 		vput(imgp->vp);
718 	}
719 
720 	if (imgp->stringbase != NULL)
721 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
722 		    ARG_MAX);
723 
724 	if (imgp->object != NULL)
725 		vm_object_deallocate(imgp->object);
726 
727 	if (error == 0) {
728 		/*
729 		 * Stop the process here if its stop event mask has
730 		 * the S_EXEC bit set.
731 		 */
732 		STOPEVENT(p, S_EXEC, 0);
733 		goto done2;
734 	}
735 
736 exec_fail:
737 	/* we're done here, clear P_INEXEC */
738 	PROC_LOCK(p);
739 	p->p_flag &= ~P_INEXEC;
740 	PROC_UNLOCK(p);
741 
742 	if (imgp->vmspace_destroyed) {
743 		/* sorry, no more process anymore. exit gracefully */
744 #ifdef MAC
745 		mac_execve_exit(imgp);
746 		if (interplabel != NULL)
747 			mac_vnode_label_free(interplabel);
748 #endif
749 		mtx_unlock(&Giant);
750 		exit1(td, W_EXITCODE(0, SIGABRT));
751 		/* NOT REACHED */
752 		error = 0;
753 	}
754 done2:
755 #ifdef MAC
756 	mac_execve_exit(imgp);
757 	if (interplabel != NULL)
758 		mac_vnode_label_free(interplabel);
759 #endif
760 	mtx_unlock(&Giant);
761 	return (error);
762 }
763 
764 int
765 exec_map_first_page(imgp)
766 	struct image_params *imgp;
767 {
768 	int rv, i;
769 	int initial_pagein;
770 	vm_page_t ma[VM_INITIAL_PAGEIN];
771 	vm_object_t object;
772 
773 	GIANT_REQUIRED;
774 
775 	if (imgp->firstpage != NULL)
776 		exec_unmap_first_page(imgp);
777 
778 	VOP_GETVOBJECT(imgp->vp, &object);
779 	VM_OBJECT_LOCK(object);
780 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
781 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
782 		initial_pagein = VM_INITIAL_PAGEIN;
783 		if (initial_pagein > object->size)
784 			initial_pagein = object->size;
785 		for (i = 1; i < initial_pagein; i++) {
786 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
787 				if (ma[i]->valid)
788 					break;
789 				vm_page_lock_queues();
790 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
791 					vm_page_unlock_queues();
792 					break;
793 				}
794 				vm_page_busy(ma[i]);
795 				vm_page_unlock_queues();
796 			} else {
797 				ma[i] = vm_page_alloc(object, i,
798 				    VM_ALLOC_NORMAL);
799 				if (ma[i] == NULL)
800 					break;
801 			}
802 		}
803 		initial_pagein = i;
804 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
805 		ma[0] = vm_page_lookup(object, 0);
806 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
807 		    (ma[0]->valid == 0)) {
808 			if (ma[0]) {
809 				vm_page_lock_queues();
810 				pmap_remove_all(ma[0]);
811 				vm_page_free(ma[0]);
812 				vm_page_unlock_queues();
813 			}
814 			VM_OBJECT_UNLOCK(object);
815 			return (EIO);
816 		}
817 	}
818 	vm_page_lock_queues();
819 	vm_page_hold(ma[0]);
820 	vm_page_wakeup(ma[0]);
821 	vm_page_unlock_queues();
822 	VM_OBJECT_UNLOCK(object);
823 
824 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
825 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
826 
827 	return (0);
828 }
829 
830 void
831 exec_unmap_first_page(imgp)
832 	struct image_params *imgp;
833 {
834 	vm_page_t m;
835 
836 	if (imgp->firstpage != NULL) {
837 		m = sf_buf_page(imgp->firstpage);
838 		sf_buf_free(imgp->firstpage);
839 		imgp->firstpage = NULL;
840 		vm_page_lock_queues();
841 		vm_page_unhold(m);
842 		vm_page_unlock_queues();
843 	}
844 }
845 
846 /*
847  * Destroy old address space, and allocate a new stack
848  *	The new stack is only SGROWSIZ large because it is grown
849  *	automatically in trap.c.
850  */
851 int
852 exec_new_vmspace(imgp, sv)
853 	struct image_params *imgp;
854 	struct sysentvec *sv;
855 {
856 	int error;
857 	struct proc *p = imgp->proc;
858 	struct vmspace *vmspace = p->p_vmspace;
859 	vm_offset_t stack_addr;
860 	vm_map_t map;
861 
862 	GIANT_REQUIRED;
863 
864 	imgp->vmspace_destroyed = 1;
865 
866 	/* Called with Giant held, do not depend on it! */
867 	EVENTHANDLER_INVOKE(process_exec, p);
868 
869 	/*
870 	 * Here is as good a place as any to do any resource limit cleanups.
871 	 * This is needed if a 64 bit binary exec's a 32 bit binary - the
872 	 * data size limit may need to be changed to a value that makes
873 	 * sense for the 32 bit binary.
874 	 */
875 	if (sv->sv_fixlimits != NULL)
876 		sv->sv_fixlimits(imgp);
877 
878 	/*
879 	 * Blow away entire process VM, if address space not shared,
880 	 * otherwise, create a new VM space so that other threads are
881 	 * not disrupted
882 	 */
883 	map = &vmspace->vm_map;
884 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
885 	    vm_map_max(map) == sv->sv_maxuser) {
886 		shmexit(vmspace);
887 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
888 		    vm_map_max(map));
889 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
890 	} else {
891 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
892 		vmspace = p->p_vmspace;
893 		map = &vmspace->vm_map;
894 	}
895 
896 	/* Allocate a new stack */
897 	stack_addr = sv->sv_usrstack - maxssiz;
898 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
899 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
900 	if (error)
901 		return (error);
902 
903 #ifdef __ia64__
904 	/* Allocate a new register stack */
905 	stack_addr = IA64_BACKINGSTORE;
906 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
907 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
908 	if (error)
909 		return (error);
910 #endif
911 
912 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
913 	 * VM_STACK case, but they are still used to monitor the size of the
914 	 * process stack so we can check the stack rlimit.
915 	 */
916 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
917 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
918 
919 	return (0);
920 }
921 
922 /*
923  * Copy out argument and environment strings from the old process
924  *	address space into the temporary string buffer.
925  */
926 int
927 exec_extract_strings(imgp)
928 	struct image_params *imgp;
929 {
930 	char	**argv, **envv;
931 	char	*argp, *envp;
932 	int	error;
933 	size_t	length;
934 
935 	/*
936 	 * extract arguments first
937 	 */
938 
939 	argv = imgp->userspace_argv;
940 
941 	if (argv) {
942 		argp = (caddr_t)(intptr_t)fuword(argv);
943 		if (argp == (caddr_t)-1)
944 			return (EFAULT);
945 		if (argp)
946 			argv++;
947 		if (imgp->argv0)
948 			argp = imgp->argv0;
949 		if (argp) {
950 			do {
951 				if (argp == (caddr_t)-1)
952 					return (EFAULT);
953 				if ((error = copyinstr(argp, imgp->stringp,
954 				    imgp->stringspace, &length))) {
955 					if (error == ENAMETOOLONG)
956 						return (E2BIG);
957 					return (error);
958 				}
959 				imgp->stringspace -= length;
960 				imgp->stringp += length;
961 				imgp->argc++;
962 			} while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
963 		}
964 	} else
965 		return (EFAULT);
966 
967 	imgp->endargs = imgp->stringp;
968 
969 	/*
970 	 * extract environment strings
971 	 */
972 
973 	envv = imgp->userspace_envv;
974 
975 	if (envv) {
976 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
977 			if (envp == (caddr_t)-1)
978 				return (EFAULT);
979 			if ((error = copyinstr(envp, imgp->stringp,
980 			    imgp->stringspace, &length))) {
981 				if (error == ENAMETOOLONG)
982 					return (E2BIG);
983 				return (error);
984 			}
985 			imgp->stringspace -= length;
986 			imgp->stringp += length;
987 			imgp->envc++;
988 		}
989 	}
990 
991 	return (0);
992 }
993 
994 /*
995  * Copy strings out to the new process address space, constructing
996  *	new arg and env vector tables. Return a pointer to the base
997  *	so that it can be used as the initial stack pointer.
998  */
999 register_t *
1000 exec_copyout_strings(imgp)
1001 	struct image_params *imgp;
1002 {
1003 	int argc, envc;
1004 	char **vectp;
1005 	char *stringp, *destp;
1006 	register_t *stack_base;
1007 	struct ps_strings *arginfo;
1008 	struct proc *p;
1009 	int szsigcode;
1010 
1011 	/*
1012 	 * Calculate string base and vector table pointers.
1013 	 * Also deal with signal trampoline code for this exec type.
1014 	 */
1015 	p = imgp->proc;
1016 	szsigcode = 0;
1017 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1018 	if (p->p_sysent->sv_szsigcode != NULL)
1019 		szsigcode = *(p->p_sysent->sv_szsigcode);
1020 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1021 	    roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
1022 
1023 	/*
1024 	 * install sigcode
1025 	 */
1026 	if (szsigcode)
1027 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1028 		    szsigcode), szsigcode);
1029 
1030 	/*
1031 	 * If we have a valid auxargs ptr, prepare some room
1032 	 * on the stack.
1033 	 */
1034 	if (imgp->auxargs) {
1035 		/*
1036 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1037 		 * lower compatibility.
1038 		 */
1039 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1040 		    (AT_COUNT * 2);
1041 		/*
1042 		 * The '+ 2' is for the null pointers at the end of each of
1043 		 * the arg and env vector sets,and imgp->auxarg_size is room
1044 		 * for argument of Runtime loader.
1045 		 */
1046 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1047 		    imgp->auxarg_size) * sizeof(char *));
1048 
1049 	} else
1050 		/*
1051 		 * The '+ 2' is for the null pointers at the end of each of
1052 		 * the arg and env vector sets
1053 		 */
1054 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1055 		    sizeof(char *));
1056 
1057 	/*
1058 	 * vectp also becomes our initial stack base
1059 	 */
1060 	stack_base = (register_t *)vectp;
1061 
1062 	stringp = imgp->stringbase;
1063 	argc = imgp->argc;
1064 	envc = imgp->envc;
1065 
1066 	/*
1067 	 * Copy out strings - arguments and environment.
1068 	 */
1069 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1070 
1071 	/*
1072 	 * Fill in "ps_strings" struct for ps, w, etc.
1073 	 */
1074 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1075 	suword(&arginfo->ps_nargvstr, argc);
1076 
1077 	/*
1078 	 * Fill in argument portion of vector table.
1079 	 */
1080 	for (; argc > 0; --argc) {
1081 		suword(vectp++, (long)(intptr_t)destp);
1082 		while (*stringp++ != 0)
1083 			destp++;
1084 		destp++;
1085 	}
1086 
1087 	/* a null vector table pointer separates the argp's from the envp's */
1088 	suword(vectp++, 0);
1089 
1090 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1091 	suword(&arginfo->ps_nenvstr, envc);
1092 
1093 	/*
1094 	 * Fill in environment portion of vector table.
1095 	 */
1096 	for (; envc > 0; --envc) {
1097 		suword(vectp++, (long)(intptr_t)destp);
1098 		while (*stringp++ != 0)
1099 			destp++;
1100 		destp++;
1101 	}
1102 
1103 	/* end of vector table is a null pointer */
1104 	suword(vectp, 0);
1105 
1106 	return (stack_base);
1107 }
1108 
1109 /*
1110  * Check permissions of file to execute.
1111  *	Called with imgp->vp locked.
1112  *	Return 0 for success or error code on failure.
1113  */
1114 int
1115 exec_check_permissions(imgp)
1116 	struct image_params *imgp;
1117 {
1118 	struct vnode *vp = imgp->vp;
1119 	struct vattr *attr = imgp->attr;
1120 	struct thread *td;
1121 	int error;
1122 
1123 	td = curthread;			/* XXXKSE */
1124 
1125 	/* Get file attributes */
1126 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1127 	if (error)
1128 		return (error);
1129 
1130 #ifdef MAC
1131 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1132 	if (error)
1133 		return (error);
1134 #endif
1135 
1136 	/*
1137 	 * 1) Check if file execution is disabled for the filesystem that this
1138 	 *	file resides on.
1139 	 * 2) Insure that at least one execute bit is on - otherwise root
1140 	 *	will always succeed, and we don't want to happen unless the
1141 	 *	file really is executable.
1142 	 * 3) Insure that the file is a regular file.
1143 	 */
1144 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1145 	    ((attr->va_mode & 0111) == 0) ||
1146 	    (attr->va_type != VREG))
1147 		return (EACCES);
1148 
1149 	/*
1150 	 * Zero length files can't be exec'd
1151 	 */
1152 	if (attr->va_size == 0)
1153 		return (ENOEXEC);
1154 
1155 	/*
1156 	 *  Check for execute permission to file based on current credentials.
1157 	 */
1158 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1159 	if (error)
1160 		return (error);
1161 
1162 	/*
1163 	 * Check number of open-for-writes on the file and deny execution
1164 	 * if there are any.
1165 	 */
1166 	if (vp->v_writecount)
1167 		return (ETXTBSY);
1168 
1169 	/*
1170 	 * Call filesystem specific open routine (which does nothing in the
1171 	 * general case).
1172 	 */
1173 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1174 	return (error);
1175 }
1176 
1177 /*
1178  * Exec handler registration
1179  */
1180 int
1181 exec_register(execsw_arg)
1182 	const struct execsw *execsw_arg;
1183 {
1184 	const struct execsw **es, **xs, **newexecsw;
1185 	int count = 2;	/* New slot and trailing NULL */
1186 
1187 	if (execsw)
1188 		for (es = execsw; *es; es++)
1189 			count++;
1190 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1191 	if (newexecsw == NULL)
1192 		return (ENOMEM);
1193 	xs = newexecsw;
1194 	if (execsw)
1195 		for (es = execsw; *es; es++)
1196 			*xs++ = *es;
1197 	*xs++ = execsw_arg;
1198 	*xs = NULL;
1199 	if (execsw)
1200 		free(execsw, M_TEMP);
1201 	execsw = newexecsw;
1202 	return (0);
1203 }
1204 
1205 int
1206 exec_unregister(execsw_arg)
1207 	const struct execsw *execsw_arg;
1208 {
1209 	const struct execsw **es, **xs, **newexecsw;
1210 	int count = 1;
1211 
1212 	if (execsw == NULL)
1213 		panic("unregister with no handlers left?\n");
1214 
1215 	for (es = execsw; *es; es++) {
1216 		if (*es == execsw_arg)
1217 			break;
1218 	}
1219 	if (*es == NULL)
1220 		return (ENOENT);
1221 	for (es = execsw; *es; es++)
1222 		if (*es != execsw_arg)
1223 			count++;
1224 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1225 	if (newexecsw == NULL)
1226 		return (ENOMEM);
1227 	xs = newexecsw;
1228 	for (es = execsw; *es; es++)
1229 		if (*es != execsw_arg)
1230 			*xs++ = *es;
1231 	*xs = NULL;
1232 	if (execsw)
1233 		free(execsw, M_TEMP);
1234 	execsw = newexecsw;
1235 	return (0);
1236 }
1237