xref: /freebsd/sys/kern/kern_exec.c (revision 06a31d6a6779b74405b41c8ad4579b5d81db4d4c)
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  * $FreeBSD$
27  */
28 
29 #include "opt_ktrace.h"
30 #include "opt_mac.h"
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/eventhandler.h>
35 #include <sys/lock.h>
36 #include <sys/mutex.h>
37 #include <sys/sysproto.h>
38 #include <sys/signalvar.h>
39 #include <sys/kernel.h>
40 #include <sys/mac.h>
41 #include <sys/mount.h>
42 #include <sys/filedesc.h>
43 #include <sys/fcntl.h>
44 #include <sys/acct.h>
45 #include <sys/exec.h>
46 #include <sys/imgact.h>
47 #include <sys/imgact_elf.h>
48 #include <sys/wait.h>
49 #include <sys/malloc.h>
50 #include <sys/proc.h>
51 #include <sys/pioctl.h>
52 #include <sys/namei.h>
53 #include <sys/sysent.h>
54 #include <sys/shm.h>
55 #include <sys/sysctl.h>
56 #include <sys/user.h>
57 #include <sys/vnode.h>
58 #ifdef KTRACE
59 #include <sys/ktrace.h>
60 #endif
61 
62 #include <vm/vm.h>
63 #include <vm/vm_param.h>
64 #include <vm/pmap.h>
65 #include <vm/vm_page.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_kern.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_pager.h>
71 
72 #include <machine/reg.h>
73 
74 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
75 
76 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
77 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
78 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
79 static int kern_execve(struct thread *td, char *fname, char **argv,
80 	char **envv, struct mac *mac_p);
81 
82 /* XXX This should be vm_size_t. */
83 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
84     NULL, 0, sysctl_kern_ps_strings, "LU", "");
85 
86 /* XXX This should be vm_size_t. */
87 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
88     NULL, 0, sysctl_kern_usrstack, "LU", "");
89 
90 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
91     NULL, 0, sysctl_kern_stackprot, "I", "");
92 
93 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
94 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
95     &ps_arg_cache_limit, 0, "");
96 
97 int ps_argsopen = 1;
98 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
99 
100 #ifdef __ia64__
101 /* XXX HACK */
102 static int regstkpages = 256;
103 SYSCTL_INT(_machdep, OID_AUTO, regstkpages, CTLFLAG_RW, &regstkpages, 0, "");
104 #endif
105 
106 static int
107 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
108 {
109 	struct proc *p;
110 
111 	p = curproc;
112 	return (SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
113 	   sizeof(p->p_sysent->sv_psstrings)));
114 }
115 
116 static int
117 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
118 {
119 	struct proc *p;
120 
121 	p = curproc;
122 	return (SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
123 	    sizeof(p->p_sysent->sv_usrstack)));
124 }
125 
126 static int
127 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
128 {
129 	struct proc *p;
130 
131 	p = curproc;
132 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
133 	    sizeof(p->p_sysent->sv_stackprot)));
134 }
135 
136 /*
137  * Each of the items is a pointer to a `const struct execsw', hence the
138  * double pointer here.
139  */
140 static const struct execsw **execsw;
141 
142 /*
143  * In-kernel implementation of execve().  All arguments are assumed to be
144  * userspace pointers from the passed thread.
145  *
146  * MPSAFE
147  */
148 static int
149 kern_execve(td, fname, argv, envv, mac_p)
150 	struct thread *td;
151 	char *fname;
152 	char **argv;
153 	char **envv;
154 	struct mac *mac_p;
155 {
156 	struct proc *p = td->td_proc;
157 	struct nameidata nd, *ndp;
158 	struct ucred *newcred = NULL, *oldcred;
159 	struct uidinfo *euip;
160 	register_t *stack_base;
161 	int error, len, i;
162 	struct image_params image_params, *imgp;
163 	struct vattr attr;
164 	int (*img_first)(struct image_params *);
165 	struct pargs *oldargs = NULL, *newargs = NULL;
166 	struct sigacts *oldsigacts, *newsigacts;
167 #ifdef KTRACE
168 	struct vnode *tracevp = NULL;
169 	struct ucred *tracecred = NULL;
170 #endif
171 	struct vnode *textvp = NULL;
172 	int credential_changing;
173 	int textset;
174 #ifdef MAC
175 	struct label interplabel;	/* label of the interpreted vnode */
176 	struct label execlabel;		/* optional label argument */
177 	int will_transition, interplabelvalid = 0;
178 #endif
179 
180 	imgp = &image_params;
181 
182 	/*
183 	 * Lock the process and set the P_INEXEC flag to indicate that
184 	 * it should be left alone until we're done here.  This is
185 	 * necessary to avoid race conditions - e.g. in ptrace() -
186 	 * that might allow a local user to illicitly obtain elevated
187 	 * privileges.
188 	 */
189 	PROC_LOCK(p);
190 	KASSERT((p->p_flag & P_INEXEC) == 0,
191 	    ("%s(): process already has P_INEXEC flag", __func__));
192 	if (p->p_flag & P_THREADED || p->p_numthreads > 1) {
193 		if (thread_single(SINGLE_EXIT)) {
194 			PROC_UNLOCK(p);
195 			return (ERESTART);	/* Try again later. */
196 		}
197 		/*
198 		 * If we get here all other threads are dead,
199 		 * so unset the associated flags and lose KSE mode.
200 		 */
201 		p->p_flag &= ~P_THREADED;
202 		td->td_mailbox = NULL;
203 		thread_single_end();
204 	}
205 	p->p_flag |= P_INEXEC;
206 	PROC_UNLOCK(p);
207 
208 	/*
209 	 * Initialize part of the common data
210 	 */
211 	imgp->proc = p;
212 	imgp->userspace_argv = argv;
213 	imgp->userspace_envv = envv;
214 	imgp->execlabel = NULL;
215 	imgp->attr = &attr;
216 	imgp->argc = imgp->envc = 0;
217 	imgp->argv0 = NULL;
218 	imgp->entry_addr = 0;
219 	imgp->vmspace_destroyed = 0;
220 	imgp->interpreted = 0;
221 	imgp->interpreter_name[0] = '\0';
222 	imgp->auxargs = NULL;
223 	imgp->vp = NULL;
224 	imgp->object = NULL;
225 	imgp->firstpage = NULL;
226 	imgp->ps_strings = 0;
227 	imgp->auxarg_size = 0;
228 
229 #ifdef MAC
230 	error = mac_execve_enter(imgp, mac_p, &execlabel);
231 	if (error) {
232 		mtx_lock(&Giant);
233 		goto exec_fail;
234 	}
235 #endif
236 
237 	/*
238 	 * Allocate temporary demand zeroed space for argument and
239 	 *	environment strings
240 	 */
241 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX +
242 	    PAGE_SIZE);
243 	if (imgp->stringbase == NULL) {
244 		error = ENOMEM;
245 		mtx_lock(&Giant);
246 		goto exec_fail;
247 	}
248 	imgp->stringp = imgp->stringbase;
249 	imgp->stringspace = ARG_MAX;
250 	imgp->image_header = imgp->stringbase + ARG_MAX;
251 
252 	/*
253 	 * Translate the file name. namei() returns a vnode pointer
254 	 *	in ni_vp amoung other things.
255 	 */
256 	ndp = &nd;
257 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
258 	    UIO_USERSPACE, fname, td);
259 
260 	mtx_lock(&Giant);
261 interpret:
262 
263 	error = namei(ndp);
264 	if (error) {
265 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
266 		    ARG_MAX + PAGE_SIZE);
267 		goto exec_fail;
268 	}
269 
270 	imgp->vp = ndp->ni_vp;
271 	imgp->fname = fname;
272 
273 	/*
274 	 * Check file permissions (also 'opens' file)
275 	 */
276 	error = exec_check_permissions(imgp);
277 	if (error)
278 		goto exec_fail_dealloc;
279 
280 	if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0)
281 		vm_object_reference(imgp->object);
282 
283 	/*
284 	 * Set VV_TEXT now so no one can write to the executable while we're
285 	 * activating it.
286 	 *
287 	 * Remember if this was set before and unset it in case this is not
288 	 * actually an executable image.
289 	 */
290 	textset = imgp->vp->v_vflag & VV_TEXT;
291 	imgp->vp->v_vflag |= VV_TEXT;
292 
293 	error = exec_map_first_page(imgp);
294 	if (error)
295 		goto exec_fail_dealloc;
296 
297 	/*
298 	 *	If the current process has a special image activator it
299 	 *	wants to try first, call it.   For example, emulating shell
300 	 *	scripts differently.
301 	 */
302 	error = -1;
303 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
304 		error = img_first(imgp);
305 
306 	/*
307 	 *	Loop through the list of image activators, calling each one.
308 	 *	An activator returns -1 if there is no match, 0 on success,
309 	 *	and an error otherwise.
310 	 */
311 	for (i = 0; error == -1 && execsw[i]; ++i) {
312 		if (execsw[i]->ex_imgact == NULL ||
313 		    execsw[i]->ex_imgact == img_first) {
314 			continue;
315 		}
316 		error = (*execsw[i]->ex_imgact)(imgp);
317 	}
318 
319 	if (error) {
320 		if (error == -1) {
321 			if (textset == 0)
322 				imgp->vp->v_vflag &= ~VV_TEXT;
323 			error = ENOEXEC;
324 		}
325 		goto exec_fail_dealloc;
326 	}
327 
328 	/*
329 	 * Special interpreter operation, cleanup and loop up to try to
330 	 * activate the interpreter.
331 	 */
332 	if (imgp->interpreted) {
333 		exec_unmap_first_page(imgp);
334 		/*
335 		 * VV_TEXT needs to be unset for scripts.  There is a short
336 		 * period before we determine that something is a script where
337 		 * VV_TEXT will be set. The vnode lock is held over this
338 		 * entire period so nothing should illegitimately be blocked.
339 		 */
340 		imgp->vp->v_vflag &= ~VV_TEXT;
341 		/* free name buffer and old vnode */
342 		NDFREE(ndp, NDF_ONLY_PNBUF);
343 #ifdef MAC
344 		mac_init_vnode_label(&interplabel);
345 		mac_copy_vnode_label(&ndp->ni_vp->v_label, &interplabel);
346 		interplabelvalid = 1;
347 #endif
348 		vput(ndp->ni_vp);
349 		vm_object_deallocate(imgp->object);
350 		imgp->object = NULL;
351 		/* set new name to that of the interpreter */
352 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
353 		    UIO_SYSSPACE, imgp->interpreter_name, td);
354 		goto interpret;
355 	}
356 
357 	/*
358 	 * Copy out strings (args and env) and initialize stack base
359 	 */
360 	if (p->p_sysent->sv_copyout_strings)
361 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
362 	else
363 		stack_base = exec_copyout_strings(imgp);
364 
365 	/*
366 	 * If custom stack fixup routine present for this process
367 	 * let it do the stack setup.
368 	 * Else stuff argument count as first item on stack
369 	 */
370 	if (p->p_sysent->sv_fixup)
371 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
372 	else
373 		suword(--stack_base, imgp->argc);
374 
375 	/*
376 	 * For security and other reasons, the file descriptor table cannot
377 	 * be shared after an exec.
378 	 */
379 	FILEDESC_LOCK(p->p_fd);
380 	if (p->p_fd->fd_refcnt > 1) {
381 		struct filedesc *tmp;
382 
383 		tmp = fdcopy(td->td_proc->p_fd);
384 		FILEDESC_UNLOCK(p->p_fd);
385 		fdfree(td);
386 		p->p_fd = tmp;
387 	} else
388 		FILEDESC_UNLOCK(p->p_fd);
389 
390 	/*
391 	 * Malloc things before we need locks.
392 	 */
393 	newcred = crget();
394 	euip = uifind(attr.va_uid);
395 	i = imgp->endargs - imgp->stringbase;
396 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
397 		newargs = pargs_alloc(i);
398 
399 	/* close files on exec */
400 	fdcloseexec(td);
401 
402 	/* Get a reference to the vnode prior to locking the proc */
403 	VREF(ndp->ni_vp);
404 
405 	/*
406 	 * For security and other reasons, signal handlers cannot
407 	 * be shared after an exec. The new process gets a copy of the old
408 	 * handlers. In execsigs(), the new process will have its signals
409 	 * reset.
410 	 */
411 	PROC_LOCK(p);
412 	if (sigacts_shared(p->p_sigacts)) {
413 		oldsigacts = p->p_sigacts;
414 		PROC_UNLOCK(p);
415 		newsigacts = sigacts_alloc();
416 		sigacts_copy(newsigacts, oldsigacts);
417 		PROC_LOCK(p);
418 		p->p_sigacts = newsigacts;
419 	} else
420 		oldsigacts = NULL;
421 
422 	/* Stop profiling */
423 	stopprofclock(p);
424 
425 	/* reset caught signals */
426 	execsigs(p);
427 
428 	/* name this process - nameiexec(p, ndp) */
429 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
430 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
431 	p->p_comm[len] = 0;
432 
433 	/*
434 	 * mark as execed, wakeup the process that vforked (if any) and tell
435 	 * it that it now has its own resources back
436 	 */
437 	p->p_flag |= P_EXEC;
438 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
439 		p->p_flag &= ~P_PPWAIT;
440 		wakeup(p->p_pptr);
441 	}
442 
443 	/*
444 	 * Implement image setuid/setgid.
445 	 *
446 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
447 	 * the process is being traced.
448 	 *
449 	 * XXXMAC: For the time being, use NOSUID to also prohibit
450 	 * transitions on the file system.
451 	 */
452 	oldcred = p->p_ucred;
453 	credential_changing = 0;
454 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
455 	    attr.va_uid;
456 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
457 	    attr.va_gid;
458 #ifdef MAC
459 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
460 	    interplabelvalid ? &interplabel : NULL, imgp);
461 	credential_changing |= will_transition;
462 #endif
463 
464 	if (credential_changing &&
465 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
466 	    (p->p_flag & P_TRACED) == 0) {
467 		/*
468 		 * Turn off syscall tracing for set-id programs, except for
469 		 * root.  Record any set-id flags first to make sure that
470 		 * we do not regain any tracing during a possible block.
471 		 */
472 		setsugid(p);
473 #ifdef KTRACE
474 		if (p->p_tracevp != NULL && suser_cred(oldcred, PRISON_ROOT)) {
475 			mtx_lock(&ktrace_mtx);
476 			p->p_traceflag = 0;
477 			tracevp = p->p_tracevp;
478 			p->p_tracevp = NULL;
479 			tracecred = p->p_tracecred;
480 			p->p_tracecred = NULL;
481 			mtx_unlock(&ktrace_mtx);
482 		}
483 #endif
484 		/*
485 		 * Close any file descriptors 0..2 that reference procfs,
486 		 * then make sure file descriptors 0..2 are in use.
487 		 *
488 		 * setugidsafety() may call closef() and then pfind()
489 		 * which may grab the process lock.
490 		 * fdcheckstd() may call falloc() which may block to
491 		 * allocate memory, so temporarily drop the process lock.
492 		 */
493 		PROC_UNLOCK(p);
494 		setugidsafety(td);
495 		error = fdcheckstd(td);
496 		if (error != 0)
497 			goto done1;
498 		PROC_LOCK(p);
499 		/*
500 		 * Set the new credentials.
501 		 */
502 		crcopy(newcred, oldcred);
503 		if (attr.va_mode & VSUID)
504 			change_euid(newcred, euip);
505 		if (attr.va_mode & VSGID)
506 			change_egid(newcred, attr.va_gid);
507 #ifdef MAC
508 		if (will_transition) {
509 			mac_execve_transition(oldcred, newcred, imgp->vp,
510 			    interplabelvalid ? &interplabel : NULL, imgp);
511 		}
512 #endif
513 		/*
514 		 * Implement correct POSIX saved-id behavior.
515 		 *
516 		 * XXXMAC: Note that the current logic will save the
517 		 * uid and gid if a MAC domain transition occurs, even
518 		 * though maybe it shouldn't.
519 		 */
520 		change_svuid(newcred, newcred->cr_uid);
521 		change_svgid(newcred, newcred->cr_gid);
522 		p->p_ucred = newcred;
523 		newcred = NULL;
524 	} else {
525 		if (oldcred->cr_uid == oldcred->cr_ruid &&
526 		    oldcred->cr_gid == oldcred->cr_rgid)
527 			p->p_flag &= ~P_SUGID;
528 		/*
529 		 * Implement correct POSIX saved-id behavior.
530 		 *
531 		 * XXX: It's not clear that the existing behavior is
532 		 * POSIX-compliant.  A number of sources indicate that the
533 		 * saved uid/gid should only be updated if the new ruid is
534 		 * not equal to the old ruid, or the new euid is not equal
535 		 * to the old euid and the new euid is not equal to the old
536 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
537 		 * Also, this code uses the new (replaced) euid and egid as
538 		 * the source, which may or may not be the right ones to use.
539 		 */
540 		if (oldcred->cr_svuid != oldcred->cr_uid ||
541 		    oldcred->cr_svgid != oldcred->cr_gid) {
542 			crcopy(newcred, oldcred);
543 			change_svuid(newcred, newcred->cr_uid);
544 			change_svgid(newcred, newcred->cr_gid);
545 			p->p_ucred = newcred;
546 			newcred = NULL;
547 		}
548 	}
549 
550 	/*
551 	 * Store the vp for use in procfs.  This vnode was referenced prior
552 	 * to locking the proc lock.
553 	 */
554 	textvp = p->p_textvp;
555 	p->p_textvp = ndp->ni_vp;
556 
557 	/*
558 	 * Notify others that we exec'd, and clear the P_INEXEC flag
559 	 * as we're now a bona fide freshly-execed process.
560 	 */
561 	KNOTE(&p->p_klist, NOTE_EXEC);
562 	p->p_flag &= ~P_INEXEC;
563 
564 	/*
565 	 * If tracing the process, trap to debugger so breakpoints
566 	 * can be set before the program executes.
567 	 */
568 	if (p->p_flag & P_TRACED)
569 		psignal(p, SIGTRAP);
570 
571 	/* clear "fork but no exec" flag, as we _are_ execing */
572 	p->p_acflag &= ~AFORK;
573 
574 	/* Free any previous argument cache */
575 	oldargs = p->p_args;
576 	p->p_args = NULL;
577 
578 	/* Cache arguments if they fit inside our allowance */
579 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
580 		bcopy(imgp->stringbase, newargs->ar_args, i);
581 		p->p_args = newargs;
582 		newargs = NULL;
583 	}
584 	PROC_UNLOCK(p);
585 
586 	/* Set values passed into the program in registers. */
587 	if (p->p_sysent->sv_setregs)
588 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
589 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
590 	else
591 		exec_setregs(td, imgp->entry_addr,
592 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
593 
594 done1:
595 	/*
596 	 * Free any resources malloc'd earlier that we didn't use.
597 	 */
598 	uifree(euip);
599 	if (newcred == NULL)
600 		crfree(oldcred);
601 	else
602 		crfree(newcred);
603 	/*
604 	 * Handle deferred decrement of ref counts.
605 	 */
606 	if (textvp != NULL)
607 		vrele(textvp);
608 	if (ndp->ni_vp && error != 0)
609 		vrele(ndp->ni_vp);
610 #ifdef KTRACE
611 	if (tracevp != NULL)
612 		vrele(tracevp);
613 	if (tracecred != NULL)
614 		crfree(tracecred);
615 #endif
616 	if (oldargs != NULL)
617 		pargs_drop(oldargs);
618 	if (newargs != NULL)
619 		pargs_drop(newargs);
620 	if (oldsigacts != NULL)
621 		sigacts_free(oldsigacts);
622 
623 exec_fail_dealloc:
624 
625 	/*
626 	 * free various allocated resources
627 	 */
628 	if (imgp->firstpage)
629 		exec_unmap_first_page(imgp);
630 
631 	if (imgp->vp) {
632 		NDFREE(ndp, NDF_ONLY_PNBUF);
633 		vput(imgp->vp);
634 	}
635 
636 	if (imgp->stringbase != NULL)
637 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
638 		    ARG_MAX + PAGE_SIZE);
639 
640 	if (imgp->object)
641 		vm_object_deallocate(imgp->object);
642 
643 	if (error == 0) {
644 		/*
645 		 * Stop the process here if its stop event mask has
646 		 * the S_EXEC bit set.
647 		 */
648 		STOPEVENT(p, S_EXEC, 0);
649 		goto done2;
650 	}
651 
652 exec_fail:
653 	/* we're done here, clear P_INEXEC */
654 	PROC_LOCK(p);
655 	p->p_flag &= ~P_INEXEC;
656 	PROC_UNLOCK(p);
657 
658 	if (imgp->vmspace_destroyed) {
659 		/* sorry, no more process anymore. exit gracefully */
660 #ifdef MAC
661 		mac_execve_exit(imgp);
662 		if (interplabelvalid)
663 			mac_destroy_vnode_label(&interplabel);
664 #endif
665 		exit1(td, W_EXITCODE(0, SIGABRT));
666 		/* NOT REACHED */
667 		error = 0;
668 	}
669 done2:
670 #ifdef MAC
671 	mac_execve_exit(imgp);
672 	if (interplabelvalid)
673 		mac_destroy_vnode_label(&interplabel);
674 #endif
675 	mtx_unlock(&Giant);
676 	return (error);
677 }
678 
679 #ifndef _SYS_SYSPROTO_H_
680 struct execve_args {
681         char    *fname;
682         char    **argv;
683         char    **envv;
684 };
685 #endif
686 
687 /*
688  * MPSAFE
689  */
690 int
691 execve(td, uap)
692 	struct thread *td;
693 	struct execve_args /* {
694 		char *fname;
695 		char **argv;
696 		char **envv;
697 	} */ *uap;
698 {
699 
700 	return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
701 }
702 
703 #ifndef _SYS_SYSPROTO_H_
704 struct __mac_execve_args {
705 	char	*fname;
706 	char	**argv;
707 	char	**envv;
708 	struct mac	*mac_p;
709 };
710 #endif
711 
712 /*
713  * MPSAFE
714  */
715 int
716 __mac_execve(td, uap)
717 	struct thread *td;
718 	struct __mac_execve_args /* {
719 		char *fname;
720 		char **argv;
721 		char **envv;
722 		struct mac *mac_p;
723 	} */ *uap;
724 {
725 
726 #ifdef MAC
727 	return (kern_execve(td, uap->fname, uap->argv, uap->envv,
728 	    uap->mac_p));
729 #else
730 	return (ENOSYS);
731 #endif
732 }
733 
734 int
735 exec_map_first_page(imgp)
736 	struct image_params *imgp;
737 {
738 	int rv, i;
739 	int initial_pagein;
740 	vm_page_t ma[VM_INITIAL_PAGEIN];
741 	vm_object_t object;
742 
743 	GIANT_REQUIRED;
744 
745 	if (imgp->firstpage) {
746 		exec_unmap_first_page(imgp);
747 	}
748 
749 	VOP_GETVOBJECT(imgp->vp, &object);
750 
751 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
752 
753 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
754 		initial_pagein = VM_INITIAL_PAGEIN;
755 		if (initial_pagein > object->size)
756 			initial_pagein = object->size;
757 		for (i = 1; i < initial_pagein; i++) {
758 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
759 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
760 					break;
761 				if (ma[i]->valid)
762 					break;
763 				vm_page_lock_queues();
764 				vm_page_busy(ma[i]);
765 				vm_page_unlock_queues();
766 			} else {
767 				ma[i] = vm_page_alloc(object, i,
768 				    VM_ALLOC_NORMAL);
769 				if (ma[i] == NULL)
770 					break;
771 			}
772 		}
773 		initial_pagein = i;
774 
775 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
776 		ma[0] = vm_page_lookup(object, 0);
777 
778 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
779 		    (ma[0]->valid == 0)) {
780 			if (ma[0]) {
781 				vm_page_lock_queues();
782 				pmap_remove_all(ma[0]);
783 				vm_page_free(ma[0]);
784 				vm_page_unlock_queues();
785 			}
786 			return (EIO);
787 		}
788 	}
789 	vm_page_lock_queues();
790 	vm_page_wire(ma[0]);
791 	vm_page_wakeup(ma[0]);
792 	vm_page_unlock_queues();
793 
794 	pmap_qenter((vm_offset_t)imgp->image_header, ma, 1);
795 	imgp->firstpage = ma[0];
796 
797 	return (0);
798 }
799 
800 void
801 exec_unmap_first_page(imgp)
802 	struct image_params *imgp;
803 {
804 	GIANT_REQUIRED;
805 
806 	if (imgp->firstpage) {
807 		pmap_qremove((vm_offset_t)imgp->image_header, 1);
808 		vm_page_lock_queues();
809 		vm_page_unwire(imgp->firstpage, 1);
810 		vm_page_unlock_queues();
811 		imgp->firstpage = NULL;
812 	}
813 }
814 
815 /*
816  * Destroy old address space, and allocate a new stack
817  *	The new stack is only SGROWSIZ large because it is grown
818  *	automatically in trap.c.
819  */
820 int
821 exec_new_vmspace(imgp, sv)
822 	struct image_params *imgp;
823 	struct sysentvec *sv;
824 {
825 	int error;
826 	struct proc *p = imgp->proc;
827 	struct vmspace *vmspace = p->p_vmspace;
828 	vm_offset_t stack_addr;
829 	vm_map_t map;
830 
831 	GIANT_REQUIRED;
832 
833 	stack_addr = sv->sv_usrstack - maxssiz;
834 
835 	imgp->vmspace_destroyed = 1;
836 
837 	EVENTHANDLER_INVOKE(process_exec, p);
838 
839 	/*
840 	 * Blow away entire process VM, if address space not shared,
841 	 * otherwise, create a new VM space so that other threads are
842 	 * not disrupted
843 	 */
844 	map = &vmspace->vm_map;
845 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
846 	    vm_map_max(map) == sv->sv_maxuser) {
847 		shmexit(vmspace);
848 		vm_page_lock_queues();
849 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
850 		    vm_map_max(map));
851 		vm_page_unlock_queues();
852 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
853 	} else {
854 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
855 		vmspace = p->p_vmspace;
856 		map = &vmspace->vm_map;
857 	}
858 
859 	/* Allocate a new stack */
860 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
861 	    sv->sv_stackprot, VM_PROT_ALL, 0);
862 	if (error)
863 		return (error);
864 
865 #ifdef __ia64__
866 	{
867 		/*
868 		 * Allocate backing store. We really need something
869 		 * similar to vm_map_stack which can allow the backing
870 		 * store to grow upwards. This will do for now.
871 		 */
872 		vm_offset_t bsaddr;
873 		bsaddr = p->p_sysent->sv_usrstack - 2 * maxssiz;
874 		error = vm_map_find(map, 0, 0, &bsaddr,
875 		    regstkpages * PAGE_SIZE, 0, VM_PROT_ALL, VM_PROT_ALL, 0);
876 		FIRST_THREAD_IN_PROC(p)->td_md.md_bspstore = bsaddr;
877 	}
878 #endif
879 
880 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
881 	 * VM_STACK case, but they are still used to monitor the size of the
882 	 * process stack so we can check the stack rlimit.
883 	 */
884 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
885 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
886 
887 	return (0);
888 }
889 
890 /*
891  * Copy out argument and environment strings from the old process
892  *	address space into the temporary string buffer.
893  */
894 int
895 exec_extract_strings(imgp)
896 	struct image_params *imgp;
897 {
898 	char	**argv, **envv;
899 	char	*argp, *envp;
900 	int	error;
901 	size_t	length;
902 
903 	/*
904 	 * extract arguments first
905 	 */
906 
907 	argv = imgp->userspace_argv;
908 
909 	if (argv) {
910 		argp = (caddr_t)(intptr_t)fuword(argv);
911 		if (argp == (caddr_t)-1)
912 			return (EFAULT);
913 		if (argp)
914 			argv++;
915 		if (imgp->argv0)
916 			argp = imgp->argv0;
917 		if (argp) {
918 			do {
919 				if (argp == (caddr_t)-1)
920 					return (EFAULT);
921 				if ((error = copyinstr(argp, imgp->stringp,
922 				    imgp->stringspace, &length))) {
923 					if (error == ENAMETOOLONG)
924 						return (E2BIG);
925 					return (error);
926 				}
927 				imgp->stringspace -= length;
928 				imgp->stringp += length;
929 				imgp->argc++;
930 			} while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
931 		}
932 	}
933 
934 	imgp->endargs = imgp->stringp;
935 
936 	/*
937 	 * extract environment strings
938 	 */
939 
940 	envv = imgp->userspace_envv;
941 
942 	if (envv) {
943 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
944 			if (envp == (caddr_t)-1)
945 				return (EFAULT);
946 			if ((error = copyinstr(envp, imgp->stringp,
947 			    imgp->stringspace, &length))) {
948 				if (error == ENAMETOOLONG)
949 					return (E2BIG);
950 				return (error);
951 			}
952 			imgp->stringspace -= length;
953 			imgp->stringp += length;
954 			imgp->envc++;
955 		}
956 	}
957 
958 	return (0);
959 }
960 
961 /*
962  * Copy strings out to the new process address space, constructing
963  *	new arg and env vector tables. Return a pointer to the base
964  *	so that it can be used as the initial stack pointer.
965  */
966 register_t *
967 exec_copyout_strings(imgp)
968 	struct image_params *imgp;
969 {
970 	int argc, envc;
971 	char **vectp;
972 	char *stringp, *destp;
973 	register_t *stack_base;
974 	struct ps_strings *arginfo;
975 	struct proc *p;
976 	int szsigcode;
977 
978 	/*
979 	 * Calculate string base and vector table pointers.
980 	 * Also deal with signal trampoline code for this exec type.
981 	 */
982 	p = imgp->proc;
983 	szsigcode = 0;
984 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
985 	if (p->p_sysent->sv_szsigcode != NULL)
986 		szsigcode = *(p->p_sysent->sv_szsigcode);
987 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
988 	    roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
989 
990 	/*
991 	 * install sigcode
992 	 */
993 	if (szsigcode)
994 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
995 		    szsigcode), szsigcode);
996 
997 	/*
998 	 * If we have a valid auxargs ptr, prepare some room
999 	 * on the stack.
1000 	 */
1001 	if (imgp->auxargs) {
1002 		/*
1003 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1004 		 * lower compatibility.
1005 		 */
1006 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1007 		    (AT_COUNT * 2);
1008 		/*
1009 		 * The '+ 2' is for the null pointers at the end of each of
1010 		 * the arg and env vector sets,and imgp->auxarg_size is room
1011 		 * for argument of Runtime loader.
1012 		 */
1013 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1014 		    imgp->auxarg_size) * sizeof(char *));
1015 
1016 	} else
1017 		/*
1018 		 * The '+ 2' is for the null pointers at the end of each of
1019 		 * the arg and env vector sets
1020 		 */
1021 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1022 		    sizeof(char *));
1023 
1024 	/*
1025 	 * vectp also becomes our initial stack base
1026 	 */
1027 	stack_base = (register_t *)vectp;
1028 
1029 	stringp = imgp->stringbase;
1030 	argc = imgp->argc;
1031 	envc = imgp->envc;
1032 
1033 	/*
1034 	 * Copy out strings - arguments and environment.
1035 	 */
1036 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1037 
1038 	/*
1039 	 * Fill in "ps_strings" struct for ps, w, etc.
1040 	 */
1041 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1042 	suword(&arginfo->ps_nargvstr, argc);
1043 
1044 	/*
1045 	 * Fill in argument portion of vector table.
1046 	 */
1047 	for (; argc > 0; --argc) {
1048 		suword(vectp++, (long)(intptr_t)destp);
1049 		while (*stringp++ != 0)
1050 			destp++;
1051 		destp++;
1052 	}
1053 
1054 	/* a null vector table pointer separates the argp's from the envp's */
1055 	suword(vectp++, 0);
1056 
1057 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1058 	suword(&arginfo->ps_nenvstr, envc);
1059 
1060 	/*
1061 	 * Fill in environment portion of vector table.
1062 	 */
1063 	for (; envc > 0; --envc) {
1064 		suword(vectp++, (long)(intptr_t)destp);
1065 		while (*stringp++ != 0)
1066 			destp++;
1067 		destp++;
1068 	}
1069 
1070 	/* end of vector table is a null pointer */
1071 	suword(vectp, 0);
1072 
1073 	return (stack_base);
1074 }
1075 
1076 /*
1077  * Check permissions of file to execute.
1078  *	Called with imgp->vp locked.
1079  *	Return 0 for success or error code on failure.
1080  */
1081 int
1082 exec_check_permissions(imgp)
1083 	struct image_params *imgp;
1084 {
1085 	struct vnode *vp = imgp->vp;
1086 	struct vattr *attr = imgp->attr;
1087 	struct thread *td;
1088 	int error;
1089 
1090 	td = curthread;			/* XXXKSE */
1091 
1092 	/* Get file attributes */
1093 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1094 	if (error)
1095 		return (error);
1096 
1097 #ifdef MAC
1098 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1099 	if (error)
1100 		return (error);
1101 #endif
1102 
1103 	/*
1104 	 * 1) Check if file execution is disabled for the filesystem that this
1105 	 *	file resides on.
1106 	 * 2) Insure that at least one execute bit is on - otherwise root
1107 	 *	will always succeed, and we don't want to happen unless the
1108 	 *	file really is executable.
1109 	 * 3) Insure that the file is a regular file.
1110 	 */
1111 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1112 	    ((attr->va_mode & 0111) == 0) ||
1113 	    (attr->va_type != VREG))
1114 		return (EACCES);
1115 
1116 	/*
1117 	 * Zero length files can't be exec'd
1118 	 */
1119 	if (attr->va_size == 0)
1120 		return (ENOEXEC);
1121 
1122 	/*
1123 	 *  Check for execute permission to file based on current credentials.
1124 	 */
1125 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1126 	if (error)
1127 		return (error);
1128 
1129 	/*
1130 	 * Check number of open-for-writes on the file and deny execution
1131 	 * if there are any.
1132 	 */
1133 	if (vp->v_writecount)
1134 		return (ETXTBSY);
1135 
1136 	/*
1137 	 * Call filesystem specific open routine (which does nothing in the
1138 	 * general case).
1139 	 */
1140 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td);
1141 	return (error);
1142 }
1143 
1144 /*
1145  * Exec handler registration
1146  */
1147 int
1148 exec_register(execsw_arg)
1149 	const struct execsw *execsw_arg;
1150 {
1151 	const struct execsw **es, **xs, **newexecsw;
1152 	int count = 2;	/* New slot and trailing NULL */
1153 
1154 	if (execsw)
1155 		for (es = execsw; *es; es++)
1156 			count++;
1157 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1158 	if (newexecsw == NULL)
1159 		return (ENOMEM);
1160 	xs = newexecsw;
1161 	if (execsw)
1162 		for (es = execsw; *es; es++)
1163 			*xs++ = *es;
1164 	*xs++ = execsw_arg;
1165 	*xs = NULL;
1166 	if (execsw)
1167 		free(execsw, M_TEMP);
1168 	execsw = newexecsw;
1169 	return (0);
1170 }
1171 
1172 int
1173 exec_unregister(execsw_arg)
1174 	const struct execsw *execsw_arg;
1175 {
1176 	const struct execsw **es, **xs, **newexecsw;
1177 	int count = 1;
1178 
1179 	if (execsw == NULL)
1180 		panic("unregister with no handlers left?\n");
1181 
1182 	for (es = execsw; *es; es++) {
1183 		if (*es == execsw_arg)
1184 			break;
1185 	}
1186 	if (*es == NULL)
1187 		return (ENOENT);
1188 	for (es = execsw; *es; es++)
1189 		if (*es != execsw_arg)
1190 			count++;
1191 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1192 	if (newexecsw == NULL)
1193 		return (ENOMEM);
1194 	xs = newexecsw;
1195 	for (es = execsw; *es; es++)
1196 		if (*es != execsw_arg)
1197 			*xs++ = *es;
1198 	*xs = NULL;
1199 	if (execsw)
1200 		free(execsw, M_TEMP);
1201 	execsw = newexecsw;
1202 	return (0);
1203 }
1204