xref: /freebsd/sys/kern/kern_exec.c (revision 5773cccf19ef7b97e56c1101aa481c43149224da)
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/lock.h>
35 #include <sys/mutex.h>
36 #include <sys/sysproto.h>
37 #include <sys/signalvar.h>
38 #include <sys/kernel.h>
39 #include <sys/mac.h>
40 #include <sys/mount.h>
41 #include <sys/filedesc.h>
42 #include <sys/fcntl.h>
43 #include <sys/acct.h>
44 #include <sys/exec.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/wait.h>
48 #include <sys/malloc.h>
49 #include <sys/proc.h>
50 #include <sys/pioctl.h>
51 #include <sys/namei.h>
52 #include <sys/sysent.h>
53 #include <sys/shm.h>
54 #include <sys/sysctl.h>
55 #include <sys/user.h>
56 #include <sys/vnode.h>
57 #ifdef KTRACE
58 #include <sys/ktrace.h>
59 #endif
60 
61 #include <vm/vm.h>
62 #include <vm/vm_param.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_page.h>
65 #include <vm/vm_map.h>
66 #include <vm/vm_kern.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_pager.h>
70 
71 #include <machine/reg.h>
72 
73 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
74 
75 static MALLOC_DEFINE(M_ATEXEC, "atexec", "atexec callback");
76 
77 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
78 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
79 static int kern_execve(struct thread *td, char *fname, char **argv,
80 	char **envv, struct mac *mac_p);
81 
82 /*
83  * callout list for things to do at exec time
84  */
85 struct execlist {
86 	execlist_fn function;
87 	TAILQ_ENTRY(execlist) next;
88 };
89 
90 TAILQ_HEAD(exec_list_head, execlist);
91 static struct exec_list_head exec_list = TAILQ_HEAD_INITIALIZER(exec_list);
92 
93 /* XXX This should be vm_size_t. */
94 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
95     NULL, 0, sysctl_kern_ps_strings, "LU", "");
96 
97 /* XXX This should be vm_size_t. */
98 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
99     NULL, 0, sysctl_kern_usrstack, "LU", "");
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 int ps_argsopen = 1;
106 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
107 
108 #ifdef __ia64__
109 /* XXX HACK */
110 static int regstkpages = 256;
111 SYSCTL_INT(_machdep, OID_AUTO, regstkpages, CTLFLAG_RW, &regstkpages, 0, "");
112 #endif
113 
114 static int
115 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
116 {
117 	struct proc *p;
118 
119 	p = curproc;
120 	return (SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
121 	   sizeof(p->p_sysent->sv_psstrings)));
122 }
123 
124 static int
125 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
126 {
127 	struct proc *p;
128 
129 	p = curproc;
130 	return (SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
131 	    sizeof(p->p_sysent->sv_usrstack)));
132 }
133 
134 /*
135  * Each of the items is a pointer to a `const struct execsw', hence the
136  * double pointer here.
137  */
138 static const struct execsw **execsw;
139 
140 /*
141  * In-kernel implementation of execve().  All arguments are assumed to be
142  * userspace pointers from the passed thread.
143  *
144  * MPSAFE
145  */
146 static int
147 kern_execve(td, fname, argv, envv, mac_p)
148 	struct thread *td;
149 	char *fname;
150 	char **argv;
151 	char **envv;
152 	struct mac *mac_p;
153 {
154 	struct proc *p = td->td_proc;
155 	struct nameidata nd, *ndp;
156 	struct ucred *newcred = NULL, *oldcred;
157 	struct uidinfo *euip;
158 	register_t *stack_base;
159 	int error, len, i;
160 	struct image_params image_params, *imgp;
161 	struct vattr attr;
162 	int (*img_first)(struct image_params *);
163 	struct pargs *oldargs = NULL, *newargs = NULL;
164 	struct procsig *oldprocsig, *newprocsig;
165 #ifdef KTRACE
166 	struct vnode *tracevp = NULL;
167 #endif
168 	struct vnode *textvp = NULL;
169 	int credential_changing;
170 	int textset;
171 #ifdef MAC
172 	struct label interplabel;	/* label of the interpreted vnode */
173 	struct label execlabel;		/* optional label argument */
174 	int will_transition, interplabelvalid = 0;
175 #endif
176 
177 	imgp = &image_params;
178 
179 	/*
180 	 * Lock the process and set the P_INEXEC flag to indicate that
181 	 * it should be left alone until we're done here.  This is
182 	 * necessary to avoid race conditions - e.g. in ptrace() -
183 	 * that might allow a local user to illicitly obtain elevated
184 	 * privileges.
185 	 */
186 	PROC_LOCK(p);
187 	KASSERT((p->p_flag & P_INEXEC) == 0,
188 	    ("%s(): process already has P_INEXEC flag", __func__));
189 	if (p->p_flag & P_KSES) {
190 		if (thread_single(SINGLE_EXIT)) {
191 			PROC_UNLOCK(p);
192 			return (ERESTART);	/* Try again later. */
193 		}
194 		/*
195 		 * If we get here all other threads are dead,
196 		 * so unset the associated flags and lose KSE mode.
197 		 */
198 		p->p_flag &= ~P_KSES;
199 		td->td_flags &= ~TDF_UNBOUND;
200 		thread_single_end();
201 	}
202 	p->p_flag |= P_INEXEC;
203 	PROC_UNLOCK(p);
204 
205 	/*
206 	 * Initialize part of the common data
207 	 */
208 	imgp->proc = p;
209 	imgp->userspace_argv = argv;
210 	imgp->userspace_envv = envv;
211 	imgp->execlabel = NULL;
212 	imgp->attr = &attr;
213 	imgp->argc = imgp->envc = 0;
214 	imgp->argv0 = NULL;
215 	imgp->entry_addr = 0;
216 	imgp->vmspace_destroyed = 0;
217 	imgp->interpreted = 0;
218 	imgp->interpreter_name[0] = '\0';
219 	imgp->auxargs = NULL;
220 	imgp->vp = NULL;
221 	imgp->object = NULL;
222 	imgp->firstpage = NULL;
223 	imgp->ps_strings = 0;
224 	imgp->auxarg_size = 0;
225 
226 #ifdef MAC
227 	error = mac_execve_enter(imgp, mac_p, &execlabel);
228 	if (error) {
229 		mtx_lock(&Giant);
230 		goto exec_fail;
231 	}
232 #endif
233 
234 	/*
235 	 * Allocate temporary demand zeroed space for argument and
236 	 *	environment strings
237 	 */
238 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX +
239 	    PAGE_SIZE);
240 	if (imgp->stringbase == NULL) {
241 		error = ENOMEM;
242 		mtx_lock(&Giant);
243 		goto exec_fail;
244 	}
245 	imgp->stringp = imgp->stringbase;
246 	imgp->stringspace = ARG_MAX;
247 	imgp->image_header = imgp->stringbase + ARG_MAX;
248 
249 	/*
250 	 * Translate the file name. namei() returns a vnode pointer
251 	 *	in ni_vp amoung other things.
252 	 */
253 	ndp = &nd;
254 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
255 	    UIO_USERSPACE, fname, td);
256 
257 	mtx_lock(&Giant);
258 interpret:
259 
260 	error = namei(ndp);
261 	if (error) {
262 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
263 		    ARG_MAX + PAGE_SIZE);
264 		goto exec_fail;
265 	}
266 
267 	imgp->vp = ndp->ni_vp;
268 	imgp->fname = fname;
269 
270 	/*
271 	 * Check file permissions (also 'opens' file)
272 	 */
273 	error = exec_check_permissions(imgp);
274 	if (error)
275 		goto exec_fail_dealloc;
276 
277 	if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0)
278 		vm_object_reference(imgp->object);
279 
280 	/*
281 	 * Set VV_TEXT now so no one can write to the executable while we're
282 	 * activating it.
283 	 *
284 	 * Remember if this was set before and unset it in case this is not
285 	 * actually an executable image.
286 	 */
287 	textset = imgp->vp->v_vflag & VV_TEXT;
288 	imgp->vp->v_vflag |= VV_TEXT;
289 
290 	error = exec_map_first_page(imgp);
291 	if (error)
292 		goto exec_fail_dealloc;
293 
294 	/*
295 	 *	If the current process has a special image activator it
296 	 *	wants to try first, call it.   For example, emulating shell
297 	 *	scripts differently.
298 	 */
299 	error = -1;
300 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
301 		error = img_first(imgp);
302 
303 	/*
304 	 *	Loop through the list of image activators, calling each one.
305 	 *	An activator returns -1 if there is no match, 0 on success,
306 	 *	and an error otherwise.
307 	 */
308 	for (i = 0; error == -1 && execsw[i]; ++i) {
309 		if (execsw[i]->ex_imgact == NULL ||
310 		    execsw[i]->ex_imgact == img_first) {
311 			continue;
312 		}
313 		error = (*execsw[i]->ex_imgact)(imgp);
314 	}
315 
316 	if (error) {
317 		if (error == -1) {
318 			if (textset == 0)
319 				imgp->vp->v_vflag &= ~VV_TEXT;
320 			error = ENOEXEC;
321 		}
322 		goto exec_fail_dealloc;
323 	}
324 
325 	/*
326 	 * Special interpreter operation, cleanup and loop up to try to
327 	 * activate the interpreter.
328 	 */
329 	if (imgp->interpreted) {
330 		exec_unmap_first_page(imgp);
331 		/*
332 		 * VV_TEXT needs to be unset for scripts.  There is a short
333 		 * period before we determine that something is a script where
334 		 * VV_TEXT will be set. The vnode lock is held over this
335 		 * entire period so nothing should illegitimately be blocked.
336 		 */
337 		imgp->vp->v_vflag &= ~VV_TEXT;
338 		/* free name buffer and old vnode */
339 		NDFREE(ndp, NDF_ONLY_PNBUF);
340 #ifdef MAC
341 		mac_init_vnode_label(&interplabel);
342 		mac_copy_vnode_label(&ndp->ni_vp->v_label, &interplabel);
343 		interplabelvalid = 1;
344 #endif
345 		vput(ndp->ni_vp);
346 		vm_object_deallocate(imgp->object);
347 		imgp->object = NULL;
348 		/* set new name to that of the interpreter */
349 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
350 		    UIO_SYSSPACE, imgp->interpreter_name, td);
351 		goto interpret;
352 	}
353 
354 	/*
355 	 * Copy out strings (args and env) and initialize stack base
356 	 */
357 	if (p->p_sysent->sv_copyout_strings)
358 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
359 	else
360 		stack_base = exec_copyout_strings(imgp);
361 
362 	/*
363 	 * If custom stack fixup routine present for this process
364 	 * let it do the stack setup.
365 	 * Else stuff argument count as first item on stack
366 	 */
367 	if (p->p_sysent->sv_fixup)
368 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
369 	else
370 		suword(--stack_base, imgp->argc);
371 
372 	/*
373 	 * For security and other reasons, the file descriptor table cannot
374 	 * be shared after an exec.
375 	 */
376 	FILEDESC_LOCK(p->p_fd);
377 	if (p->p_fd->fd_refcnt > 1) {
378 		struct filedesc *tmp;
379 
380 		tmp = fdcopy(td);
381 		FILEDESC_UNLOCK(p->p_fd);
382 		fdfree(td);
383 		p->p_fd = tmp;
384 	} else
385 		FILEDESC_UNLOCK(p->p_fd);
386 
387 	/*
388 	 * Malloc things before we need locks.
389 	 */
390 	newcred = crget();
391 	euip = uifind(attr.va_uid);
392 	i = imgp->endargs - imgp->stringbase;
393 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
394 		newargs = pargs_alloc(i);
395 
396 	/* close files on exec */
397 	fdcloseexec(td);
398 
399 	/* Get a reference to the vnode prior to locking the proc */
400 	VREF(ndp->ni_vp);
401 
402 	/*
403 	 * For security and other reasons, signal handlers cannot
404 	 * be shared after an exec. The new process gets a copy of the old
405 	 * handlers. In execsigs(), the new process will have its signals
406 	 * reset.
407 	 */
408 	PROC_LOCK(p);
409 	mp_fixme("procsig needs a lock");
410 	if (p->p_procsig->ps_refcnt > 1) {
411 		oldprocsig = p->p_procsig;
412 		PROC_UNLOCK(p);
413 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
414 		    M_SUBPROC, M_WAITOK);
415 		bcopy(oldprocsig, newprocsig, sizeof(*newprocsig));
416 		newprocsig->ps_refcnt = 1;
417 		oldprocsig->ps_refcnt--;
418 		PROC_LOCK(p);
419 		p->p_procsig = newprocsig;
420 		if (p->p_sigacts == &p->p_uarea->u_sigacts)
421 			panic("shared procsig but private sigacts?");
422 
423 		p->p_uarea->u_sigacts = *p->p_sigacts;
424 		p->p_sigacts = &p->p_uarea->u_sigacts;
425 	}
426 	/* Stop profiling */
427 	stopprofclock(p);
428 
429 	/* reset caught signals */
430 	execsigs(p);
431 
432 	/* name this process - nameiexec(p, ndp) */
433 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
434 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
435 	p->p_comm[len] = 0;
436 
437 	/*
438 	 * mark as execed, wakeup the process that vforked (if any) and tell
439 	 * it that it now has its own resources back
440 	 */
441 	p->p_flag |= P_EXEC;
442 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
443 		p->p_flag &= ~P_PPWAIT;
444 		wakeup(p->p_pptr);
445 	}
446 
447 	/*
448 	 * Implement image setuid/setgid.
449 	 *
450 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
451 	 * the process is being traced.
452 	 *
453 	 * XXXMAC: For the time being, use NOSUID to also prohibit
454 	 * transitions on the file system.
455 	 */
456 	oldcred = p->p_ucred;
457 	credential_changing = 0;
458 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
459 	    attr.va_uid;
460 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
461 	    attr.va_gid;
462 #ifdef MAC
463 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
464 	    interplabelvalid ? &interplabel : NULL, imgp);
465 	credential_changing |= will_transition;
466 #endif
467 
468 	if (credential_changing &&
469 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
470 	    (p->p_flag & P_TRACED) == 0) {
471 		/*
472 		 * Turn off syscall tracing for set-id programs, except for
473 		 * root.  Record any set-id flags first to make sure that
474 		 * we do not regain any tracing during a possible block.
475 		 */
476 		setsugid(p);
477 #ifdef KTRACE
478 		if (p->p_tracep && suser_cred(oldcred, PRISON_ROOT)) {
479 			mtx_lock(&ktrace_mtx);
480 			p->p_traceflag = 0;
481 			tracevp = p->p_tracep;
482 			p->p_tracep = NULL;
483 			mtx_unlock(&ktrace_mtx);
484 		}
485 #endif
486 		/*
487 		 * Close any file descriptors 0..2 that reference procfs,
488 		 * then make sure file descriptors 0..2 are in use.
489 		 *
490 		 * setugidsafety() may call closef() and then pfind()
491 		 * which may grab the process lock.
492 		 * fdcheckstd() may call falloc() which may block to
493 		 * allocate memory, so temporarily drop the process lock.
494 		 */
495 		PROC_UNLOCK(p);
496 		setugidsafety(td);
497 		error = fdcheckstd(td);
498 		if (error != 0)
499 			goto done1;
500 		PROC_LOCK(p);
501 		/*
502 		 * Set the new credentials.
503 		 */
504 		crcopy(newcred, oldcred);
505 		if (attr.va_mode & VSUID)
506 			change_euid(newcred, euip);
507 		if (attr.va_mode & VSGID)
508 			change_egid(newcred, attr.va_gid);
509 #ifdef MAC
510 		if (will_transition) {
511 			mac_execve_transition(oldcred, newcred, imgp->vp,
512 			    interplabelvalid ? &interplabel : NULL, imgp);
513 		}
514 #endif
515 		/*
516 		 * Implement correct POSIX saved-id behavior.
517 		 *
518 		 * XXXMAC: Note that the current logic will save the
519 		 * uid and gid if a MAC domain transition occurs, even
520 		 * though maybe it shouldn't.
521 		 */
522 		change_svuid(newcred, newcred->cr_uid);
523 		change_svgid(newcred, newcred->cr_gid);
524 		p->p_ucred = newcred;
525 		newcred = NULL;
526 	} else {
527 		if (oldcred->cr_uid == oldcred->cr_ruid &&
528 		    oldcred->cr_gid == oldcred->cr_rgid)
529 			p->p_flag &= ~P_SUGID;
530 		/*
531 		 * Implement correct POSIX saved-id behavior.
532 		 *
533 		 * XXX: It's not clear that the existing behavior is
534 		 * POSIX-compliant.  A number of sources indicate that the
535 		 * saved uid/gid should only be updated if the new ruid is
536 		 * not equal to the old ruid, or the new euid is not equal
537 		 * to the old euid and the new euid is not equal to the old
538 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
539 		 * Also, this code uses the new (replaced) euid and egid as
540 		 * the source, which may or may not be the right ones to use.
541 		 */
542 		if (oldcred->cr_svuid != oldcred->cr_uid ||
543 		    oldcred->cr_svgid != oldcred->cr_gid) {
544 			crcopy(newcred, oldcred);
545 			change_svuid(newcred, newcred->cr_uid);
546 			change_svgid(newcred, newcred->cr_gid);
547 			p->p_ucred = newcred;
548 			newcred = NULL;
549 		}
550 	}
551 
552 	/*
553 	 * Store the vp for use in procfs.  This vnode was referenced prior
554 	 * to locking the proc lock.
555 	 */
556 	textvp = p->p_textvp;
557 	p->p_textvp = ndp->ni_vp;
558 
559 	/*
560 	 * Notify others that we exec'd, and clear the P_INEXEC flag
561 	 * as we're now a bona fide freshly-execed process.
562 	 */
563 	KNOTE(&p->p_klist, NOTE_EXEC);
564 	p->p_flag &= ~P_INEXEC;
565 
566 	/*
567 	 * If tracing the process, trap to debugger so breakpoints
568 	 * can be set before the program executes.
569 	 */
570 	if (p->p_flag & P_TRACED)
571 		psignal(p, SIGTRAP);
572 
573 	/* clear "fork but no exec" flag, as we _are_ execing */
574 	p->p_acflag &= ~AFORK;
575 
576 	/* Free any previous argument cache */
577 	oldargs = p->p_args;
578 	p->p_args = NULL;
579 
580 	/* Cache arguments if they fit inside our allowance */
581 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
582 		bcopy(imgp->stringbase, newargs->ar_args, i);
583 		p->p_args = newargs;
584 		newargs = NULL;
585 	}
586 	PROC_UNLOCK(p);
587 
588 	/* Set values passed into the program in registers. */
589 	if (p->p_sysent->sv_setregs)
590 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
591 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
592 	else
593 		exec_setregs(td, imgp->entry_addr,
594 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
595 
596 done1:
597 	/*
598 	 * Free any resources malloc'd earlier that we didn't use.
599 	 */
600 	uifree(euip);
601 	if (newcred == NULL)
602 		crfree(oldcred);
603 	else
604 		crfree(newcred);
605 	/*
606 	 * Handle deferred decrement of ref counts.
607 	 */
608 	if (textvp != NULL)
609 		vrele(textvp);
610 	if (ndp->ni_vp && error != 0)
611 		vrele(ndp->ni_vp);
612 #ifdef KTRACE
613 	if (tracevp != NULL)
614 		vrele(tracevp);
615 #endif
616 	if (oldargs != NULL)
617 		pargs_drop(oldargs);
618 	if (newargs != NULL)
619 		pargs_drop(newargs);
620 
621 exec_fail_dealloc:
622 
623 	/*
624 	 * free various allocated resources
625 	 */
626 	if (imgp->firstpage)
627 		exec_unmap_first_page(imgp);
628 
629 	if (imgp->vp) {
630 		NDFREE(ndp, NDF_ONLY_PNBUF);
631 		vput(imgp->vp);
632 	}
633 
634 	if (imgp->stringbase != NULL)
635 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
636 		    ARG_MAX + PAGE_SIZE);
637 
638 	if (imgp->object)
639 		vm_object_deallocate(imgp->object);
640 
641 	if (error == 0) {
642 		/*
643 		 * Stop the process here if its stop event mask has
644 		 * the S_EXEC bit set.
645 		 */
646 		STOPEVENT(p, S_EXEC, 0);
647 		goto done2;
648 	}
649 
650 exec_fail:
651 	/* we're done here, clear P_INEXEC */
652 	PROC_LOCK(p);
653 	p->p_flag &= ~P_INEXEC;
654 	PROC_UNLOCK(p);
655 
656 	if (imgp->vmspace_destroyed) {
657 		/* sorry, no more process anymore. exit gracefully */
658 #ifdef MAC
659 		mac_execve_exit(imgp);
660 		if (interplabelvalid)
661 			mac_destroy_vnode_label(&interplabel);
662 #endif
663 		exit1(td, W_EXITCODE(0, SIGABRT));
664 		/* NOT REACHED */
665 		error = 0;
666 	}
667 done2:
668 #ifdef MAC
669 	mac_execve_exit(imgp);
670 	if (interplabelvalid)
671 		mac_destroy_vnode_label(&interplabel);
672 #endif
673 	mtx_unlock(&Giant);
674 	return (error);
675 }
676 
677 #ifndef _SYS_SYSPROTO_H_
678 struct execve_args {
679         char    *fname;
680         char    **argv;
681         char    **envv;
682 };
683 #endif
684 
685 /*
686  * MPSAFE
687  */
688 int
689 execve(td, uap)
690 	struct thread *td;
691 	struct execve_args /* {
692 		syscallarg(char *) fname;
693 		syscallarg(char **) argv;
694 		syscallarg(char **) envv;
695 	} */ *uap;
696 {
697 
698 	return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
699 }
700 
701 #ifndef _SYS_SYSPROTO_H_
702 struct __mac_execve_args {
703 	char	*fname;
704 	char	**argv;
705 	char	**envv;
706 	struct mac	*mac_p;
707 };
708 #endif
709 
710 /*
711  * MPSAFE
712  */
713 int
714 __mac_execve(td, uap)
715 	struct thread *td;
716 	struct __mac_execve_args /* {
717 		syscallarg(char *) fname;
718 		syscallarg(char **) argv;
719 		syscallarg(char **) envv;
720 		syscallarg(struct mac *) mac_p;
721 	} */ *uap;
722 {
723 
724 #ifdef MAC
725 	return (kern_execve(td, uap->fname, uap->argv, uap->envv,
726 	    uap->mac_p));
727 #else
728 	return (ENOSYS);
729 #endif
730 }
731 
732 int
733 exec_map_first_page(imgp)
734 	struct image_params *imgp;
735 {
736 	int rv, i;
737 	int initial_pagein;
738 	vm_page_t ma[VM_INITIAL_PAGEIN];
739 	vm_object_t object;
740 
741 	GIANT_REQUIRED;
742 
743 	if (imgp->firstpage) {
744 		exec_unmap_first_page(imgp);
745 	}
746 
747 	VOP_GETVOBJECT(imgp->vp, &object);
748 
749 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
750 
751 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
752 		initial_pagein = VM_INITIAL_PAGEIN;
753 		if (initial_pagein > object->size)
754 			initial_pagein = object->size;
755 		for (i = 1; i < initial_pagein; i++) {
756 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
757 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
758 					break;
759 				if (ma[i]->valid)
760 					break;
761 				vm_page_busy(ma[i]);
762 			} else {
763 				ma[i] = vm_page_alloc(object, i,
764 				    VM_ALLOC_NORMAL);
765 				if (ma[i] == NULL)
766 					break;
767 			}
768 		}
769 		initial_pagein = i;
770 
771 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
772 		ma[0] = vm_page_lookup(object, 0);
773 
774 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
775 		    (ma[0]->valid == 0)) {
776 			if (ma[0]) {
777 				vm_page_lock_queues();
778 				pmap_remove_all(ma[0]);
779 				vm_page_free(ma[0]);
780 				vm_page_unlock_queues();
781 			}
782 			return (EIO);
783 		}
784 	}
785 	vm_page_lock_queues();
786 	vm_page_wire(ma[0]);
787 	vm_page_wakeup(ma[0]);
788 	vm_page_unlock_queues();
789 
790 	pmap_qenter((vm_offset_t)imgp->image_header, ma, 1);
791 	imgp->firstpage = ma[0];
792 
793 	return (0);
794 }
795 
796 void
797 exec_unmap_first_page(imgp)
798 	struct image_params *imgp;
799 {
800 	GIANT_REQUIRED;
801 
802 	if (imgp->firstpage) {
803 		pmap_qremove((vm_offset_t)imgp->image_header, 1);
804 		vm_page_lock_queues();
805 		vm_page_unwire(imgp->firstpage, 1);
806 		vm_page_unlock_queues();
807 		imgp->firstpage = NULL;
808 	}
809 }
810 
811 /*
812  * Destroy old address space, and allocate a new stack
813  *	The new stack is only SGROWSIZ large because it is grown
814  *	automatically in trap.c.
815  */
816 int
817 exec_new_vmspace(imgp, sv)
818 	struct image_params *imgp;
819 	struct sysentvec *sv;
820 {
821 	int error;
822 	struct execlist *ep;
823 	struct proc *p = imgp->proc;
824 	struct vmspace *vmspace = p->p_vmspace;
825 	vm_offset_t stack_addr;
826 	vm_map_t map;
827 
828 	GIANT_REQUIRED;
829 
830 	stack_addr = sv->sv_usrstack - maxssiz;
831 
832 	imgp->vmspace_destroyed = 1;
833 
834 	/*
835 	 * Perform functions registered with at_exec().
836 	 */
837 	TAILQ_FOREACH(ep, &exec_list, next)
838 		(*ep->function)(p);
839 
840 	/*
841 	 * Blow away entire process VM, if address space not shared,
842 	 * otherwise, create a new VM space so that other threads are
843 	 * not disrupted
844 	 */
845 	map = &vmspace->vm_map;
846 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
847 	    vm_map_max(map) == sv->sv_maxuser) {
848 		if (vmspace->vm_shm)
849 			shmexit(p);
850 		vm_page_lock_queues();
851 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
852 		    vm_map_max(map));
853 		vm_page_unlock_queues();
854 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
855 	} else {
856 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
857 		vmspace = p->p_vmspace;
858 		map = &vmspace->vm_map;
859 	}
860 
861 	/* Allocate a new stack */
862 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
863 	    sv->sv_stackprot, VM_PROT_ALL, 0);
864 	if (error)
865 		return (error);
866 
867 #ifdef __ia64__
868 	{
869 		/*
870 		 * Allocate backing store. We really need something
871 		 * similar to vm_map_stack which can allow the backing
872 		 * store to grow upwards. This will do for now.
873 		 */
874 		vm_offset_t bsaddr;
875 		bsaddr = p->p_sysent->sv_usrstack - 2 * maxssiz;
876 		error = vm_map_find(map, 0, 0, &bsaddr,
877 		    regstkpages * PAGE_SIZE, 0, VM_PROT_ALL, VM_PROT_ALL, 0);
878 		FIRST_THREAD_IN_PROC(p)->td_md.md_bspstore = bsaddr;
879 	}
880 #endif
881 
882 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
883 	 * VM_STACK case, but they are still used to monitor the size of the
884 	 * process stack so we can check the stack rlimit.
885 	 */
886 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
887 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
888 
889 	return (0);
890 }
891 
892 /*
893  * Copy out argument and environment strings from the old process
894  *	address space into the temporary string buffer.
895  */
896 int
897 exec_extract_strings(imgp)
898 	struct image_params *imgp;
899 {
900 	char	**argv, **envv;
901 	char	*argp, *envp;
902 	int	error;
903 	size_t	length;
904 
905 	/*
906 	 * extract arguments first
907 	 */
908 
909 	argv = imgp->userspace_argv;
910 
911 	if (argv) {
912 		argp = (caddr_t)(intptr_t)fuword(argv);
913 		if (argp == (caddr_t)-1)
914 			return (EFAULT);
915 		if (argp)
916 			argv++;
917 		if (imgp->argv0)
918 			argp = imgp->argv0;
919 		if (argp) {
920 			do {
921 				if (argp == (caddr_t)-1)
922 					return (EFAULT);
923 				if ((error = copyinstr(argp, imgp->stringp,
924 				    imgp->stringspace, &length))) {
925 					if (error == ENAMETOOLONG)
926 						return (E2BIG);
927 					return (error);
928 				}
929 				imgp->stringspace -= length;
930 				imgp->stringp += length;
931 				imgp->argc++;
932 			} while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
933 		}
934 	}
935 
936 	imgp->endargs = imgp->stringp;
937 
938 	/*
939 	 * extract environment strings
940 	 */
941 
942 	envv = imgp->userspace_envv;
943 
944 	if (envv) {
945 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
946 			if (envp == (caddr_t)-1)
947 				return (EFAULT);
948 			if ((error = copyinstr(envp, imgp->stringp,
949 			    imgp->stringspace, &length))) {
950 				if (error == ENAMETOOLONG)
951 					return (E2BIG);
952 				return (error);
953 			}
954 			imgp->stringspace -= length;
955 			imgp->stringp += length;
956 			imgp->envc++;
957 		}
958 	}
959 
960 	return (0);
961 }
962 
963 /*
964  * Copy strings out to the new process address space, constructing
965  *	new arg and env vector tables. Return a pointer to the base
966  *	so that it can be used as the initial stack pointer.
967  */
968 register_t *
969 exec_copyout_strings(imgp)
970 	struct image_params *imgp;
971 {
972 	int argc, envc;
973 	char **vectp;
974 	char *stringp, *destp;
975 	register_t *stack_base;
976 	struct ps_strings *arginfo;
977 	struct proc *p;
978 	int szsigcode;
979 
980 	/*
981 	 * Calculate string base and vector table pointers.
982 	 * Also deal with signal trampoline code for this exec type.
983 	 */
984 	p = imgp->proc;
985 	szsigcode = 0;
986 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
987 	if (p->p_sysent->sv_szsigcode != NULL)
988 		szsigcode = *(p->p_sysent->sv_szsigcode);
989 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
990 	    roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
991 
992 	/*
993 	 * install sigcode
994 	 */
995 	if (szsigcode)
996 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
997 		    szsigcode), szsigcode);
998 
999 	/*
1000 	 * If we have a valid auxargs ptr, prepare some room
1001 	 * on the stack.
1002 	 */
1003 	if (imgp->auxargs) {
1004 		/*
1005 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1006 		 * lower compatibility.
1007 		 */
1008 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1009 		    (AT_COUNT * 2);
1010 		/*
1011 		 * The '+ 2' is for the null pointers at the end of each of
1012 		 * the arg and env vector sets,and imgp->auxarg_size is room
1013 		 * for argument of Runtime loader.
1014 		 */
1015 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1016 		    imgp->auxarg_size) * sizeof(char *));
1017 
1018 	} else
1019 		/*
1020 		 * The '+ 2' is for the null pointers at the end of each of
1021 		 * the arg and env vector sets
1022 		 */
1023 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1024 		    sizeof(char *));
1025 
1026 	/*
1027 	 * vectp also becomes our initial stack base
1028 	 */
1029 	stack_base = (register_t *)vectp;
1030 
1031 	stringp = imgp->stringbase;
1032 	argc = imgp->argc;
1033 	envc = imgp->envc;
1034 
1035 	/*
1036 	 * Copy out strings - arguments and environment.
1037 	 */
1038 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1039 
1040 	/*
1041 	 * Fill in "ps_strings" struct for ps, w, etc.
1042 	 */
1043 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1044 	suword(&arginfo->ps_nargvstr, argc);
1045 
1046 	/*
1047 	 * Fill in argument portion of vector table.
1048 	 */
1049 	for (; argc > 0; --argc) {
1050 		suword(vectp++, (long)(intptr_t)destp);
1051 		while (*stringp++ != 0)
1052 			destp++;
1053 		destp++;
1054 	}
1055 
1056 	/* a null vector table pointer separates the argp's from the envp's */
1057 	suword(vectp++, 0);
1058 
1059 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1060 	suword(&arginfo->ps_nenvstr, envc);
1061 
1062 	/*
1063 	 * Fill in environment portion of vector table.
1064 	 */
1065 	for (; envc > 0; --envc) {
1066 		suword(vectp++, (long)(intptr_t)destp);
1067 		while (*stringp++ != 0)
1068 			destp++;
1069 		destp++;
1070 	}
1071 
1072 	/* end of vector table is a null pointer */
1073 	suword(vectp, 0);
1074 
1075 	return (stack_base);
1076 }
1077 
1078 /*
1079  * Check permissions of file to execute.
1080  *	Called with imgp->vp locked.
1081  *	Return 0 for success or error code on failure.
1082  */
1083 int
1084 exec_check_permissions(imgp)
1085 	struct image_params *imgp;
1086 {
1087 	struct vnode *vp = imgp->vp;
1088 	struct vattr *attr = imgp->attr;
1089 	struct thread *td;
1090 	int error;
1091 
1092 	td = curthread;			/* XXXKSE */
1093 
1094 #ifdef MAC
1095 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1096 	if (error)
1097 		return (error);
1098 #endif
1099 
1100 	/* Get file attributes */
1101 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1102 	if (error)
1103 		return (error);
1104 
1105 	/*
1106 	 * 1) Check if file execution is disabled for the filesystem that this
1107 	 *	file resides on.
1108 	 * 2) Insure that at least one execute bit is on - otherwise root
1109 	 *	will always succeed, and we don't want to happen unless the
1110 	 *	file really is executable.
1111 	 * 3) Insure that the file is a regular file.
1112 	 */
1113 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1114 	    ((attr->va_mode & 0111) == 0) ||
1115 	    (attr->va_type != VREG))
1116 		return (EACCES);
1117 
1118 	/*
1119 	 * Zero length files can't be exec'd
1120 	 */
1121 	if (attr->va_size == 0)
1122 		return (ENOEXEC);
1123 
1124 	/*
1125 	 *  Check for execute permission to file based on current credentials.
1126 	 */
1127 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1128 	if (error)
1129 		return (error);
1130 
1131 	/*
1132 	 * Check number of open-for-writes on the file and deny execution
1133 	 * if there are any.
1134 	 */
1135 	if (vp->v_writecount)
1136 		return (ETXTBSY);
1137 
1138 	/*
1139 	 * Call filesystem specific open routine (which does nothing in the
1140 	 * general case).
1141 	 */
1142 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td);
1143 	return (error);
1144 }
1145 
1146 /*
1147  * Exec handler registration
1148  */
1149 int
1150 exec_register(execsw_arg)
1151 	const struct execsw *execsw_arg;
1152 {
1153 	const struct execsw **es, **xs, **newexecsw;
1154 	int count = 2;	/* New slot and trailing NULL */
1155 
1156 	if (execsw)
1157 		for (es = execsw; *es; es++)
1158 			count++;
1159 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1160 	if (newexecsw == NULL)
1161 		return (ENOMEM);
1162 	xs = newexecsw;
1163 	if (execsw)
1164 		for (es = execsw; *es; es++)
1165 			*xs++ = *es;
1166 	*xs++ = execsw_arg;
1167 	*xs = NULL;
1168 	if (execsw)
1169 		free(execsw, M_TEMP);
1170 	execsw = newexecsw;
1171 	return (0);
1172 }
1173 
1174 int
1175 exec_unregister(execsw_arg)
1176 	const struct execsw *execsw_arg;
1177 {
1178 	const struct execsw **es, **xs, **newexecsw;
1179 	int count = 1;
1180 
1181 	if (execsw == NULL)
1182 		panic("unregister with no handlers left?\n");
1183 
1184 	for (es = execsw; *es; es++) {
1185 		if (*es == execsw_arg)
1186 			break;
1187 	}
1188 	if (*es == NULL)
1189 		return (ENOENT);
1190 	for (es = execsw; *es; es++)
1191 		if (*es != execsw_arg)
1192 			count++;
1193 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1194 	if (newexecsw == NULL)
1195 		return (ENOMEM);
1196 	xs = newexecsw;
1197 	for (es = execsw; *es; es++)
1198 		if (*es != execsw_arg)
1199 			*xs++ = *es;
1200 	*xs = NULL;
1201 	if (execsw)
1202 		free(execsw, M_TEMP);
1203 	execsw = newexecsw;
1204 	return (0);
1205 }
1206 
1207 int
1208 at_exec(function)
1209 	execlist_fn function;
1210 {
1211 	struct execlist *ep;
1212 
1213 #ifdef INVARIANTS
1214 	/* Be noisy if the programmer has lost track of things */
1215 	if (rm_at_exec(function))
1216 		printf("WARNING: exec callout entry (%p) already present\n",
1217 		    function);
1218 #endif
1219 	ep = malloc(sizeof(*ep), M_ATEXEC, M_NOWAIT);
1220 	if (ep == NULL)
1221 		return (ENOMEM);
1222 	ep->function = function;
1223 	TAILQ_INSERT_TAIL(&exec_list, ep, next);
1224 	return (0);
1225 }
1226 
1227 /*
1228  * Scan the exec callout list for the given item and remove it.
1229  * Returns the number of items removed (0 or 1)
1230  */
1231 int
1232 rm_at_exec(function)
1233 	execlist_fn function;
1234 {
1235 	struct execlist *ep;
1236 
1237 	TAILQ_FOREACH(ep, &exec_list, next) {
1238 		if (ep->function == function) {
1239 			TAILQ_REMOVE(&exec_list, ep, next);
1240 			free(ep, M_ATEXEC);
1241 			return (1);
1242 		}
1243 	}
1244 	return (0);
1245 }
1246