xref: /freebsd/sys/kern/kern_exec.c (revision 5215b1872feaad7ecf7cb1234749ecf04071deef)
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 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
80 static int kern_execve(struct thread *td, char *fname, char **argv,
81 	char **envv, struct mac *mac_p);
82 
83 /*
84  * callout list for things to do at exec time
85  */
86 struct execlist {
87 	execlist_fn function;
88 	TAILQ_ENTRY(execlist) next;
89 };
90 
91 TAILQ_HEAD(exec_list_head, execlist);
92 static struct exec_list_head exec_list = TAILQ_HEAD_INITIALIZER(exec_list);
93 
94 /* XXX This should be vm_size_t. */
95 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
96     NULL, 0, sysctl_kern_ps_strings, "LU", "");
97 
98 /* XXX This should be vm_size_t. */
99 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
100     NULL, 0, sysctl_kern_usrstack, "LU", "");
101 
102 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
103     NULL, 0, sysctl_kern_stackprot, "I", "");
104 
105 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
106 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
107     &ps_arg_cache_limit, 0, "");
108 
109 int ps_argsopen = 1;
110 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
111 
112 #ifdef __ia64__
113 /* XXX HACK */
114 static int regstkpages = 256;
115 SYSCTL_INT(_machdep, OID_AUTO, regstkpages, CTLFLAG_RW, &regstkpages, 0, "");
116 #endif
117 
118 static int
119 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
120 {
121 	struct proc *p;
122 
123 	p = curproc;
124 	return (SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
125 	   sizeof(p->p_sysent->sv_psstrings)));
126 }
127 
128 static int
129 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
130 {
131 	struct proc *p;
132 
133 	p = curproc;
134 	return (SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
135 	    sizeof(p->p_sysent->sv_usrstack)));
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 /*
155  * In-kernel implementation of execve().  All arguments are assumed to be
156  * userspace pointers from the passed thread.
157  *
158  * MPSAFE
159  */
160 static int
161 kern_execve(td, fname, argv, envv, mac_p)
162 	struct thread *td;
163 	char *fname;
164 	char **argv;
165 	char **envv;
166 	struct mac *mac_p;
167 {
168 	struct proc *p = td->td_proc;
169 	struct nameidata nd, *ndp;
170 	struct ucred *newcred = NULL, *oldcred;
171 	struct uidinfo *euip;
172 	register_t *stack_base;
173 	int error, len, i;
174 	struct image_params image_params, *imgp;
175 	struct vattr attr;
176 	int (*img_first)(struct image_params *);
177 	struct pargs *oldargs = NULL, *newargs = NULL;
178 	struct procsig *oldprocsig, *newprocsig;
179 #ifdef KTRACE
180 	struct vnode *tracevp = NULL;
181 #endif
182 	struct vnode *textvp = NULL;
183 	int credential_changing;
184 	int textset;
185 #ifdef MAC
186 	struct label interplabel;	/* label of the interpreted vnode */
187 	struct label execlabel;		/* optional label argument */
188 	int will_transition, interplabelvalid = 0;
189 #endif
190 
191 	imgp = &image_params;
192 
193 	/*
194 	 * Lock the process and set the P_INEXEC flag to indicate that
195 	 * it should be left alone until we're done here.  This is
196 	 * necessary to avoid race conditions - e.g. in ptrace() -
197 	 * that might allow a local user to illicitly obtain elevated
198 	 * privileges.
199 	 */
200 	PROC_LOCK(p);
201 	KASSERT((p->p_flag & P_INEXEC) == 0,
202 	    ("%s(): process already has P_INEXEC flag", __func__));
203 	if (p->p_flag & P_KSES) {
204 		if (thread_single(SINGLE_EXIT)) {
205 			PROC_UNLOCK(p);
206 			return (ERESTART);	/* Try again later. */
207 		}
208 		/*
209 		 * If we get here all other threads are dead,
210 		 * so unset the associated flags and lose KSE mode.
211 		 */
212 		p->p_flag &= ~P_KSES;
213 		td->td_mailbox = NULL;
214 		thread_single_end();
215 	}
216 	p->p_flag |= P_INEXEC;
217 	PROC_UNLOCK(p);
218 
219 	/*
220 	 * Initialize part of the common data
221 	 */
222 	imgp->proc = p;
223 	imgp->userspace_argv = argv;
224 	imgp->userspace_envv = envv;
225 	imgp->execlabel = NULL;
226 	imgp->attr = &attr;
227 	imgp->argc = imgp->envc = 0;
228 	imgp->argv0 = NULL;
229 	imgp->entry_addr = 0;
230 	imgp->vmspace_destroyed = 0;
231 	imgp->interpreted = 0;
232 	imgp->interpreter_name[0] = '\0';
233 	imgp->auxargs = NULL;
234 	imgp->vp = NULL;
235 	imgp->object = NULL;
236 	imgp->firstpage = NULL;
237 	imgp->ps_strings = 0;
238 	imgp->auxarg_size = 0;
239 
240 #ifdef MAC
241 	error = mac_execve_enter(imgp, mac_p, &execlabel);
242 	if (error) {
243 		mtx_lock(&Giant);
244 		goto exec_fail;
245 	}
246 #endif
247 
248 	/*
249 	 * Allocate temporary demand zeroed space for argument and
250 	 *	environment strings
251 	 */
252 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX +
253 	    PAGE_SIZE);
254 	if (imgp->stringbase == NULL) {
255 		error = ENOMEM;
256 		mtx_lock(&Giant);
257 		goto exec_fail;
258 	}
259 	imgp->stringp = imgp->stringbase;
260 	imgp->stringspace = ARG_MAX;
261 	imgp->image_header = imgp->stringbase + ARG_MAX;
262 
263 	/*
264 	 * Translate the file name. namei() returns a vnode pointer
265 	 *	in ni_vp amoung other things.
266 	 */
267 	ndp = &nd;
268 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
269 	    UIO_USERSPACE, fname, td);
270 
271 	mtx_lock(&Giant);
272 interpret:
273 
274 	error = namei(ndp);
275 	if (error) {
276 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
277 		    ARG_MAX + PAGE_SIZE);
278 		goto exec_fail;
279 	}
280 
281 	imgp->vp = ndp->ni_vp;
282 	imgp->fname = fname;
283 
284 	/*
285 	 * Check file permissions (also 'opens' file)
286 	 */
287 	error = exec_check_permissions(imgp);
288 	if (error)
289 		goto exec_fail_dealloc;
290 
291 	if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0)
292 		vm_object_reference(imgp->object);
293 
294 	/*
295 	 * Set VV_TEXT now so no one can write to the executable while we're
296 	 * activating it.
297 	 *
298 	 * Remember if this was set before and unset it in case this is not
299 	 * actually an executable image.
300 	 */
301 	textset = imgp->vp->v_vflag & VV_TEXT;
302 	imgp->vp->v_vflag |= VV_TEXT;
303 
304 	error = exec_map_first_page(imgp);
305 	if (error)
306 		goto exec_fail_dealloc;
307 
308 	/*
309 	 *	If the current process has a special image activator it
310 	 *	wants to try first, call it.   For example, emulating shell
311 	 *	scripts differently.
312 	 */
313 	error = -1;
314 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
315 		error = img_first(imgp);
316 
317 	/*
318 	 *	Loop through the list of image activators, calling each one.
319 	 *	An activator returns -1 if there is no match, 0 on success,
320 	 *	and an error otherwise.
321 	 */
322 	for (i = 0; error == -1 && execsw[i]; ++i) {
323 		if (execsw[i]->ex_imgact == NULL ||
324 		    execsw[i]->ex_imgact == img_first) {
325 			continue;
326 		}
327 		error = (*execsw[i]->ex_imgact)(imgp);
328 	}
329 
330 	if (error) {
331 		if (error == -1) {
332 			if (textset == 0)
333 				imgp->vp->v_vflag &= ~VV_TEXT;
334 			error = ENOEXEC;
335 		}
336 		goto exec_fail_dealloc;
337 	}
338 
339 	/*
340 	 * Special interpreter operation, cleanup and loop up to try to
341 	 * activate the interpreter.
342 	 */
343 	if (imgp->interpreted) {
344 		exec_unmap_first_page(imgp);
345 		/*
346 		 * VV_TEXT needs to be unset for scripts.  There is a short
347 		 * period before we determine that something is a script where
348 		 * VV_TEXT will be set. The vnode lock is held over this
349 		 * entire period so nothing should illegitimately be blocked.
350 		 */
351 		imgp->vp->v_vflag &= ~VV_TEXT;
352 		/* free name buffer and old vnode */
353 		NDFREE(ndp, NDF_ONLY_PNBUF);
354 #ifdef MAC
355 		mac_init_vnode_label(&interplabel);
356 		mac_copy_vnode_label(&ndp->ni_vp->v_label, &interplabel);
357 		interplabelvalid = 1;
358 #endif
359 		vput(ndp->ni_vp);
360 		vm_object_deallocate(imgp->object);
361 		imgp->object = NULL;
362 		/* set new name to that of the interpreter */
363 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
364 		    UIO_SYSSPACE, imgp->interpreter_name, td);
365 		goto interpret;
366 	}
367 
368 	/*
369 	 * Copy out strings (args and env) and initialize stack base
370 	 */
371 	if (p->p_sysent->sv_copyout_strings)
372 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
373 	else
374 		stack_base = exec_copyout_strings(imgp);
375 
376 	/*
377 	 * If custom stack fixup routine present for this process
378 	 * let it do the stack setup.
379 	 * Else stuff argument count as first item on stack
380 	 */
381 	if (p->p_sysent->sv_fixup)
382 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
383 	else
384 		suword(--stack_base, imgp->argc);
385 
386 	/*
387 	 * For security and other reasons, the file descriptor table cannot
388 	 * be shared after an exec.
389 	 */
390 	FILEDESC_LOCK(p->p_fd);
391 	if (p->p_fd->fd_refcnt > 1) {
392 		struct filedesc *tmp;
393 
394 		tmp = fdcopy(td->td_proc->p_fd);
395 		FILEDESC_UNLOCK(p->p_fd);
396 		fdfree(td);
397 		p->p_fd = tmp;
398 	} else
399 		FILEDESC_UNLOCK(p->p_fd);
400 
401 	/*
402 	 * Malloc things before we need locks.
403 	 */
404 	newcred = crget();
405 	euip = uifind(attr.va_uid);
406 	i = imgp->endargs - imgp->stringbase;
407 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
408 		newargs = pargs_alloc(i);
409 
410 	/* close files on exec */
411 	fdcloseexec(td);
412 
413 	/* Get a reference to the vnode prior to locking the proc */
414 	VREF(ndp->ni_vp);
415 
416 	/*
417 	 * For security and other reasons, signal handlers cannot
418 	 * be shared after an exec. The new process gets a copy of the old
419 	 * handlers. In execsigs(), the new process will have its signals
420 	 * reset.
421 	 */
422 	PROC_LOCK(p);
423 	mp_fixme("procsig needs a lock");
424 	if (p->p_procsig->ps_refcnt > 1) {
425 		oldprocsig = p->p_procsig;
426 		PROC_UNLOCK(p);
427 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
428 		    M_SUBPROC, 0);
429 		bcopy(oldprocsig, newprocsig, sizeof(*newprocsig));
430 		newprocsig->ps_refcnt = 1;
431 		oldprocsig->ps_refcnt--;
432 		PROC_LOCK(p);
433 		p->p_procsig = newprocsig;
434 		if (p->p_sigacts == &p->p_uarea->u_sigacts)
435 			panic("shared procsig but private sigacts?");
436 
437 		p->p_uarea->u_sigacts = *p->p_sigacts;
438 		p->p_sigacts = &p->p_uarea->u_sigacts;
439 	}
440 	/* Stop profiling */
441 	stopprofclock(p);
442 
443 	/* reset caught signals */
444 	execsigs(p);
445 
446 	/* name this process - nameiexec(p, ndp) */
447 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
448 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
449 	p->p_comm[len] = 0;
450 
451 	/*
452 	 * mark as execed, wakeup the process that vforked (if any) and tell
453 	 * it that it now has its own resources back
454 	 */
455 	p->p_flag |= P_EXEC;
456 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
457 		p->p_flag &= ~P_PPWAIT;
458 		wakeup(p->p_pptr);
459 	}
460 
461 	/*
462 	 * Implement image setuid/setgid.
463 	 *
464 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
465 	 * the process is being traced.
466 	 *
467 	 * XXXMAC: For the time being, use NOSUID to also prohibit
468 	 * transitions on the file system.
469 	 */
470 	oldcred = p->p_ucred;
471 	credential_changing = 0;
472 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
473 	    attr.va_uid;
474 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
475 	    attr.va_gid;
476 #ifdef MAC
477 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
478 	    interplabelvalid ? &interplabel : NULL, imgp);
479 	credential_changing |= will_transition;
480 #endif
481 
482 	if (credential_changing &&
483 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
484 	    (p->p_flag & P_TRACED) == 0) {
485 		/*
486 		 * Turn off syscall tracing for set-id programs, except for
487 		 * root.  Record any set-id flags first to make sure that
488 		 * we do not regain any tracing during a possible block.
489 		 */
490 		setsugid(p);
491 #ifdef KTRACE
492 		if (p->p_tracep && suser_cred(oldcred, PRISON_ROOT)) {
493 			mtx_lock(&ktrace_mtx);
494 			p->p_traceflag = 0;
495 			tracevp = p->p_tracep;
496 			p->p_tracep = NULL;
497 			mtx_unlock(&ktrace_mtx);
498 		}
499 #endif
500 		/*
501 		 * Close any file descriptors 0..2 that reference procfs,
502 		 * then make sure file descriptors 0..2 are in use.
503 		 *
504 		 * setugidsafety() may call closef() and then pfind()
505 		 * which may grab the process lock.
506 		 * fdcheckstd() may call falloc() which may block to
507 		 * allocate memory, so temporarily drop the process lock.
508 		 */
509 		PROC_UNLOCK(p);
510 		setugidsafety(td);
511 		error = fdcheckstd(td);
512 		if (error != 0)
513 			goto done1;
514 		PROC_LOCK(p);
515 		/*
516 		 * Set the new credentials.
517 		 */
518 		crcopy(newcred, oldcred);
519 		if (attr.va_mode & VSUID)
520 			change_euid(newcred, euip);
521 		if (attr.va_mode & VSGID)
522 			change_egid(newcred, attr.va_gid);
523 #ifdef MAC
524 		if (will_transition) {
525 			mac_execve_transition(oldcred, newcred, imgp->vp,
526 			    interplabelvalid ? &interplabel : NULL, imgp);
527 		}
528 #endif
529 		/*
530 		 * Implement correct POSIX saved-id behavior.
531 		 *
532 		 * XXXMAC: Note that the current logic will save the
533 		 * uid and gid if a MAC domain transition occurs, even
534 		 * though maybe it shouldn't.
535 		 */
536 		change_svuid(newcred, newcred->cr_uid);
537 		change_svgid(newcred, newcred->cr_gid);
538 		p->p_ucred = newcred;
539 		newcred = NULL;
540 	} else {
541 		if (oldcred->cr_uid == oldcred->cr_ruid &&
542 		    oldcred->cr_gid == oldcred->cr_rgid)
543 			p->p_flag &= ~P_SUGID;
544 		/*
545 		 * Implement correct POSIX saved-id behavior.
546 		 *
547 		 * XXX: It's not clear that the existing behavior is
548 		 * POSIX-compliant.  A number of sources indicate that the
549 		 * saved uid/gid should only be updated if the new ruid is
550 		 * not equal to the old ruid, or the new euid is not equal
551 		 * to the old euid and the new euid is not equal to the old
552 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
553 		 * Also, this code uses the new (replaced) euid and egid as
554 		 * the source, which may or may not be the right ones to use.
555 		 */
556 		if (oldcred->cr_svuid != oldcred->cr_uid ||
557 		    oldcred->cr_svgid != oldcred->cr_gid) {
558 			crcopy(newcred, oldcred);
559 			change_svuid(newcred, newcred->cr_uid);
560 			change_svgid(newcred, newcred->cr_gid);
561 			p->p_ucred = newcred;
562 			newcred = NULL;
563 		}
564 	}
565 
566 	/*
567 	 * Store the vp for use in procfs.  This vnode was referenced prior
568 	 * to locking the proc lock.
569 	 */
570 	textvp = p->p_textvp;
571 	p->p_textvp = ndp->ni_vp;
572 
573 	/*
574 	 * Notify others that we exec'd, and clear the P_INEXEC flag
575 	 * as we're now a bona fide freshly-execed process.
576 	 */
577 	KNOTE(&p->p_klist, NOTE_EXEC);
578 	p->p_flag &= ~P_INEXEC;
579 
580 	/*
581 	 * If tracing the process, trap to debugger so breakpoints
582 	 * can be set before the program executes.
583 	 */
584 	if (p->p_flag & P_TRACED)
585 		psignal(p, SIGTRAP);
586 
587 	/* clear "fork but no exec" flag, as we _are_ execing */
588 	p->p_acflag &= ~AFORK;
589 
590 	/* Free any previous argument cache */
591 	oldargs = p->p_args;
592 	p->p_args = NULL;
593 
594 	/* Cache arguments if they fit inside our allowance */
595 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
596 		bcopy(imgp->stringbase, newargs->ar_args, i);
597 		p->p_args = newargs;
598 		newargs = NULL;
599 	}
600 	PROC_UNLOCK(p);
601 
602 	/* Set values passed into the program in registers. */
603 	if (p->p_sysent->sv_setregs)
604 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
605 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
606 	else
607 		exec_setregs(td, imgp->entry_addr,
608 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
609 
610 done1:
611 	/*
612 	 * Free any resources malloc'd earlier that we didn't use.
613 	 */
614 	uifree(euip);
615 	if (newcred == NULL)
616 		crfree(oldcred);
617 	else
618 		crfree(newcred);
619 	/*
620 	 * Handle deferred decrement of ref counts.
621 	 */
622 	if (textvp != NULL)
623 		vrele(textvp);
624 	if (ndp->ni_vp && error != 0)
625 		vrele(ndp->ni_vp);
626 #ifdef KTRACE
627 	if (tracevp != NULL)
628 		vrele(tracevp);
629 #endif
630 	if (oldargs != NULL)
631 		pargs_drop(oldargs);
632 	if (newargs != NULL)
633 		pargs_drop(newargs);
634 
635 exec_fail_dealloc:
636 
637 	/*
638 	 * free various allocated resources
639 	 */
640 	if (imgp->firstpage)
641 		exec_unmap_first_page(imgp);
642 
643 	if (imgp->vp) {
644 		NDFREE(ndp, NDF_ONLY_PNBUF);
645 		vput(imgp->vp);
646 	}
647 
648 	if (imgp->stringbase != NULL)
649 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
650 		    ARG_MAX + PAGE_SIZE);
651 
652 	if (imgp->object)
653 		vm_object_deallocate(imgp->object);
654 
655 	if (error == 0) {
656 		/*
657 		 * Stop the process here if its stop event mask has
658 		 * the S_EXEC bit set.
659 		 */
660 		STOPEVENT(p, S_EXEC, 0);
661 		goto done2;
662 	}
663 
664 exec_fail:
665 	/* we're done here, clear P_INEXEC */
666 	PROC_LOCK(p);
667 	p->p_flag &= ~P_INEXEC;
668 	PROC_UNLOCK(p);
669 
670 	if (imgp->vmspace_destroyed) {
671 		/* sorry, no more process anymore. exit gracefully */
672 #ifdef MAC
673 		mac_execve_exit(imgp);
674 		if (interplabelvalid)
675 			mac_destroy_vnode_label(&interplabel);
676 #endif
677 		exit1(td, W_EXITCODE(0, SIGABRT));
678 		/* NOT REACHED */
679 		error = 0;
680 	}
681 done2:
682 #ifdef MAC
683 	mac_execve_exit(imgp);
684 	if (interplabelvalid)
685 		mac_destroy_vnode_label(&interplabel);
686 #endif
687 	mtx_unlock(&Giant);
688 	return (error);
689 }
690 
691 #ifndef _SYS_SYSPROTO_H_
692 struct execve_args {
693         char    *fname;
694         char    **argv;
695         char    **envv;
696 };
697 #endif
698 
699 /*
700  * MPSAFE
701  */
702 int
703 execve(td, uap)
704 	struct thread *td;
705 	struct execve_args /* {
706 		char *fname;
707 		char **argv;
708 		char **envv;
709 	} */ *uap;
710 {
711 
712 	return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
713 }
714 
715 #ifndef _SYS_SYSPROTO_H_
716 struct __mac_execve_args {
717 	char	*fname;
718 	char	**argv;
719 	char	**envv;
720 	struct mac	*mac_p;
721 };
722 #endif
723 
724 /*
725  * MPSAFE
726  */
727 int
728 __mac_execve(td, uap)
729 	struct thread *td;
730 	struct __mac_execve_args /* {
731 		char *fname;
732 		char **argv;
733 		char **envv;
734 		struct mac *mac_p;
735 	} */ *uap;
736 {
737 
738 #ifdef MAC
739 	return (kern_execve(td, uap->fname, uap->argv, uap->envv,
740 	    uap->mac_p));
741 #else
742 	return (ENOSYS);
743 #endif
744 }
745 
746 int
747 exec_map_first_page(imgp)
748 	struct image_params *imgp;
749 {
750 	int rv, i;
751 	int initial_pagein;
752 	vm_page_t ma[VM_INITIAL_PAGEIN];
753 	vm_object_t object;
754 
755 	GIANT_REQUIRED;
756 
757 	if (imgp->firstpage) {
758 		exec_unmap_first_page(imgp);
759 	}
760 
761 	VOP_GETVOBJECT(imgp->vp, &object);
762 
763 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
764 
765 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
766 		initial_pagein = VM_INITIAL_PAGEIN;
767 		if (initial_pagein > object->size)
768 			initial_pagein = object->size;
769 		for (i = 1; i < initial_pagein; i++) {
770 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
771 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
772 					break;
773 				if (ma[i]->valid)
774 					break;
775 				vm_page_lock_queues();
776 				vm_page_busy(ma[i]);
777 				vm_page_unlock_queues();
778 			} else {
779 				ma[i] = vm_page_alloc(object, i,
780 				    VM_ALLOC_NORMAL);
781 				if (ma[i] == NULL)
782 					break;
783 			}
784 		}
785 		initial_pagein = i;
786 
787 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
788 		ma[0] = vm_page_lookup(object, 0);
789 
790 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
791 		    (ma[0]->valid == 0)) {
792 			if (ma[0]) {
793 				vm_page_lock_queues();
794 				pmap_remove_all(ma[0]);
795 				vm_page_free(ma[0]);
796 				vm_page_unlock_queues();
797 			}
798 			return (EIO);
799 		}
800 	}
801 	vm_page_lock_queues();
802 	vm_page_wire(ma[0]);
803 	vm_page_wakeup(ma[0]);
804 	vm_page_unlock_queues();
805 
806 	pmap_qenter((vm_offset_t)imgp->image_header, ma, 1);
807 	imgp->firstpage = ma[0];
808 
809 	return (0);
810 }
811 
812 void
813 exec_unmap_first_page(imgp)
814 	struct image_params *imgp;
815 {
816 	GIANT_REQUIRED;
817 
818 	if (imgp->firstpage) {
819 		pmap_qremove((vm_offset_t)imgp->image_header, 1);
820 		vm_page_lock_queues();
821 		vm_page_unwire(imgp->firstpage, 1);
822 		vm_page_unlock_queues();
823 		imgp->firstpage = NULL;
824 	}
825 }
826 
827 /*
828  * Destroy old address space, and allocate a new stack
829  *	The new stack is only SGROWSIZ large because it is grown
830  *	automatically in trap.c.
831  */
832 int
833 exec_new_vmspace(imgp, sv)
834 	struct image_params *imgp;
835 	struct sysentvec *sv;
836 {
837 	int error;
838 	struct execlist *ep;
839 	struct proc *p = imgp->proc;
840 	struct vmspace *vmspace = p->p_vmspace;
841 	vm_offset_t stack_addr;
842 	vm_map_t map;
843 
844 	GIANT_REQUIRED;
845 
846 	stack_addr = sv->sv_usrstack - maxssiz;
847 
848 	imgp->vmspace_destroyed = 1;
849 
850 	/*
851 	 * Perform functions registered with at_exec().
852 	 */
853 	TAILQ_FOREACH(ep, &exec_list, next)
854 		(*ep->function)(p);
855 
856 	/*
857 	 * Blow away entire process VM, if address space not shared,
858 	 * otherwise, create a new VM space so that other threads are
859 	 * not disrupted
860 	 */
861 	map = &vmspace->vm_map;
862 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
863 	    vm_map_max(map) == sv->sv_maxuser) {
864 		shmexit(vmspace);
865 		vm_page_lock_queues();
866 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
867 		    vm_map_max(map));
868 		vm_page_unlock_queues();
869 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
870 	} else {
871 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
872 		vmspace = p->p_vmspace;
873 		map = &vmspace->vm_map;
874 	}
875 
876 	/* Allocate a new stack */
877 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
878 	    sv->sv_stackprot, VM_PROT_ALL, 0);
879 	if (error)
880 		return (error);
881 
882 #ifdef __ia64__
883 	{
884 		/*
885 		 * Allocate backing store. We really need something
886 		 * similar to vm_map_stack which can allow the backing
887 		 * store to grow upwards. This will do for now.
888 		 */
889 		vm_offset_t bsaddr;
890 		bsaddr = p->p_sysent->sv_usrstack - 2 * maxssiz;
891 		error = vm_map_find(map, 0, 0, &bsaddr,
892 		    regstkpages * PAGE_SIZE, 0, VM_PROT_ALL, VM_PROT_ALL, 0);
893 		FIRST_THREAD_IN_PROC(p)->td_md.md_bspstore = bsaddr;
894 	}
895 #endif
896 
897 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
898 	 * VM_STACK case, but they are still used to monitor the size of the
899 	 * process stack so we can check the stack rlimit.
900 	 */
901 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
902 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
903 
904 	return (0);
905 }
906 
907 /*
908  * Copy out argument and environment strings from the old process
909  *	address space into the temporary string buffer.
910  */
911 int
912 exec_extract_strings(imgp)
913 	struct image_params *imgp;
914 {
915 	char	**argv, **envv;
916 	char	*argp, *envp;
917 	int	error;
918 	size_t	length;
919 
920 	/*
921 	 * extract arguments first
922 	 */
923 
924 	argv = imgp->userspace_argv;
925 
926 	if (argv) {
927 		argp = (caddr_t)(intptr_t)fuword(argv);
928 		if (argp == (caddr_t)-1)
929 			return (EFAULT);
930 		if (argp)
931 			argv++;
932 		if (imgp->argv0)
933 			argp = imgp->argv0;
934 		if (argp) {
935 			do {
936 				if (argp == (caddr_t)-1)
937 					return (EFAULT);
938 				if ((error = copyinstr(argp, imgp->stringp,
939 				    imgp->stringspace, &length))) {
940 					if (error == ENAMETOOLONG)
941 						return (E2BIG);
942 					return (error);
943 				}
944 				imgp->stringspace -= length;
945 				imgp->stringp += length;
946 				imgp->argc++;
947 			} while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
948 		}
949 	}
950 
951 	imgp->endargs = imgp->stringp;
952 
953 	/*
954 	 * extract environment strings
955 	 */
956 
957 	envv = imgp->userspace_envv;
958 
959 	if (envv) {
960 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
961 			if (envp == (caddr_t)-1)
962 				return (EFAULT);
963 			if ((error = copyinstr(envp, imgp->stringp,
964 			    imgp->stringspace, &length))) {
965 				if (error == ENAMETOOLONG)
966 					return (E2BIG);
967 				return (error);
968 			}
969 			imgp->stringspace -= length;
970 			imgp->stringp += length;
971 			imgp->envc++;
972 		}
973 	}
974 
975 	return (0);
976 }
977 
978 /*
979  * Copy strings out to the new process address space, constructing
980  *	new arg and env vector tables. Return a pointer to the base
981  *	so that it can be used as the initial stack pointer.
982  */
983 register_t *
984 exec_copyout_strings(imgp)
985 	struct image_params *imgp;
986 {
987 	int argc, envc;
988 	char **vectp;
989 	char *stringp, *destp;
990 	register_t *stack_base;
991 	struct ps_strings *arginfo;
992 	struct proc *p;
993 	int szsigcode;
994 
995 	/*
996 	 * Calculate string base and vector table pointers.
997 	 * Also deal with signal trampoline code for this exec type.
998 	 */
999 	p = imgp->proc;
1000 	szsigcode = 0;
1001 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1002 	if (p->p_sysent->sv_szsigcode != NULL)
1003 		szsigcode = *(p->p_sysent->sv_szsigcode);
1004 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1005 	    roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
1006 
1007 	/*
1008 	 * install sigcode
1009 	 */
1010 	if (szsigcode)
1011 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1012 		    szsigcode), szsigcode);
1013 
1014 	/*
1015 	 * If we have a valid auxargs ptr, prepare some room
1016 	 * on the stack.
1017 	 */
1018 	if (imgp->auxargs) {
1019 		/*
1020 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1021 		 * lower compatibility.
1022 		 */
1023 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1024 		    (AT_COUNT * 2);
1025 		/*
1026 		 * The '+ 2' is for the null pointers at the end of each of
1027 		 * the arg and env vector sets,and imgp->auxarg_size is room
1028 		 * for argument of Runtime loader.
1029 		 */
1030 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1031 		    imgp->auxarg_size) * sizeof(char *));
1032 
1033 	} else
1034 		/*
1035 		 * The '+ 2' is for the null pointers at the end of each of
1036 		 * the arg and env vector sets
1037 		 */
1038 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1039 		    sizeof(char *));
1040 
1041 	/*
1042 	 * vectp also becomes our initial stack base
1043 	 */
1044 	stack_base = (register_t *)vectp;
1045 
1046 	stringp = imgp->stringbase;
1047 	argc = imgp->argc;
1048 	envc = imgp->envc;
1049 
1050 	/*
1051 	 * Copy out strings - arguments and environment.
1052 	 */
1053 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1054 
1055 	/*
1056 	 * Fill in "ps_strings" struct for ps, w, etc.
1057 	 */
1058 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1059 	suword(&arginfo->ps_nargvstr, argc);
1060 
1061 	/*
1062 	 * Fill in argument portion of vector table.
1063 	 */
1064 	for (; argc > 0; --argc) {
1065 		suword(vectp++, (long)(intptr_t)destp);
1066 		while (*stringp++ != 0)
1067 			destp++;
1068 		destp++;
1069 	}
1070 
1071 	/* a null vector table pointer separates the argp's from the envp's */
1072 	suword(vectp++, 0);
1073 
1074 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1075 	suword(&arginfo->ps_nenvstr, envc);
1076 
1077 	/*
1078 	 * Fill in environment portion of vector table.
1079 	 */
1080 	for (; envc > 0; --envc) {
1081 		suword(vectp++, (long)(intptr_t)destp);
1082 		while (*stringp++ != 0)
1083 			destp++;
1084 		destp++;
1085 	}
1086 
1087 	/* end of vector table is a null pointer */
1088 	suword(vectp, 0);
1089 
1090 	return (stack_base);
1091 }
1092 
1093 /*
1094  * Check permissions of file to execute.
1095  *	Called with imgp->vp locked.
1096  *	Return 0 for success or error code on failure.
1097  */
1098 int
1099 exec_check_permissions(imgp)
1100 	struct image_params *imgp;
1101 {
1102 	struct vnode *vp = imgp->vp;
1103 	struct vattr *attr = imgp->attr;
1104 	struct thread *td;
1105 	int error;
1106 
1107 	td = curthread;			/* XXXKSE */
1108 
1109 	/* Get file attributes */
1110 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1111 	if (error)
1112 		return (error);
1113 
1114 #ifdef MAC
1115 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1116 	if (error)
1117 		return (error);
1118 #endif
1119 
1120 	/*
1121 	 * 1) Check if file execution is disabled for the filesystem that this
1122 	 *	file resides on.
1123 	 * 2) Insure that at least one execute bit is on - otherwise root
1124 	 *	will always succeed, and we don't want to happen unless the
1125 	 *	file really is executable.
1126 	 * 3) Insure that the file is a regular file.
1127 	 */
1128 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1129 	    ((attr->va_mode & 0111) == 0) ||
1130 	    (attr->va_type != VREG))
1131 		return (EACCES);
1132 
1133 	/*
1134 	 * Zero length files can't be exec'd
1135 	 */
1136 	if (attr->va_size == 0)
1137 		return (ENOEXEC);
1138 
1139 	/*
1140 	 *  Check for execute permission to file based on current credentials.
1141 	 */
1142 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1143 	if (error)
1144 		return (error);
1145 
1146 	/*
1147 	 * Check number of open-for-writes on the file and deny execution
1148 	 * if there are any.
1149 	 */
1150 	if (vp->v_writecount)
1151 		return (ETXTBSY);
1152 
1153 	/*
1154 	 * Call filesystem specific open routine (which does nothing in the
1155 	 * general case).
1156 	 */
1157 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td);
1158 	return (error);
1159 }
1160 
1161 /*
1162  * Exec handler registration
1163  */
1164 int
1165 exec_register(execsw_arg)
1166 	const struct execsw *execsw_arg;
1167 {
1168 	const struct execsw **es, **xs, **newexecsw;
1169 	int count = 2;	/* New slot and trailing NULL */
1170 
1171 	if (execsw)
1172 		for (es = execsw; *es; es++)
1173 			count++;
1174 	newexecsw = malloc(count * sizeof(*es), M_TEMP, 0);
1175 	if (newexecsw == NULL)
1176 		return (ENOMEM);
1177 	xs = newexecsw;
1178 	if (execsw)
1179 		for (es = execsw; *es; es++)
1180 			*xs++ = *es;
1181 	*xs++ = execsw_arg;
1182 	*xs = NULL;
1183 	if (execsw)
1184 		free(execsw, M_TEMP);
1185 	execsw = newexecsw;
1186 	return (0);
1187 }
1188 
1189 int
1190 exec_unregister(execsw_arg)
1191 	const struct execsw *execsw_arg;
1192 {
1193 	const struct execsw **es, **xs, **newexecsw;
1194 	int count = 1;
1195 
1196 	if (execsw == NULL)
1197 		panic("unregister with no handlers left?\n");
1198 
1199 	for (es = execsw; *es; es++) {
1200 		if (*es == execsw_arg)
1201 			break;
1202 	}
1203 	if (*es == NULL)
1204 		return (ENOENT);
1205 	for (es = execsw; *es; es++)
1206 		if (*es != execsw_arg)
1207 			count++;
1208 	newexecsw = malloc(count * sizeof(*es), M_TEMP, 0);
1209 	if (newexecsw == NULL)
1210 		return (ENOMEM);
1211 	xs = newexecsw;
1212 	for (es = execsw; *es; es++)
1213 		if (*es != execsw_arg)
1214 			*xs++ = *es;
1215 	*xs = NULL;
1216 	if (execsw)
1217 		free(execsw, M_TEMP);
1218 	execsw = newexecsw;
1219 	return (0);
1220 }
1221 
1222 int
1223 at_exec(function)
1224 	execlist_fn function;
1225 {
1226 	struct execlist *ep;
1227 
1228 #ifdef INVARIANTS
1229 	/* Be noisy if the programmer has lost track of things */
1230 	if (rm_at_exec(function))
1231 		printf("WARNING: exec callout entry (%p) already present\n",
1232 		    function);
1233 #endif
1234 	ep = malloc(sizeof(*ep), M_ATEXEC, M_NOWAIT);
1235 	if (ep == NULL)
1236 		return (ENOMEM);
1237 	ep->function = function;
1238 	TAILQ_INSERT_TAIL(&exec_list, ep, next);
1239 	return (0);
1240 }
1241 
1242 /*
1243  * Scan the exec callout list for the given item and remove it.
1244  * Returns the number of items removed (0 or 1)
1245  */
1246 int
1247 rm_at_exec(function)
1248 	execlist_fn function;
1249 {
1250 	struct execlist *ep;
1251 
1252 	TAILQ_FOREACH(ep, &exec_list, next) {
1253 		if (ep->function == function) {
1254 			TAILQ_REMOVE(&exec_list, ep, next);
1255 			free(ep, M_ATEXEC);
1256 			return (1);
1257 		}
1258 	}
1259 	return (0);
1260 }
1261