xref: /freebsd/sys/kern/kern_exec.c (revision 729362425c09cf6b362366aabc6fb547eee8035a)
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 procsig *oldprocsig, *newprocsig;
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) {
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 	mp_fixme("procsig needs a lock");
413 	if (p->p_procsig->ps_refcnt > 1) {
414 		oldprocsig = p->p_procsig;
415 		PROC_UNLOCK(p);
416 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
417 		    M_SUBPROC, M_WAITOK);
418 		bcopy(oldprocsig, newprocsig, sizeof(*newprocsig));
419 		newprocsig->ps_refcnt = 1;
420 		oldprocsig->ps_refcnt--;
421 		PROC_LOCK(p);
422 		p->p_procsig = newprocsig;
423 		if (p->p_sigacts == &p->p_uarea->u_sigacts)
424 			panic("shared procsig but private sigacts?");
425 
426 		p->p_uarea->u_sigacts = *p->p_sigacts;
427 		p->p_sigacts = &p->p_uarea->u_sigacts;
428 	}
429 	/* Stop profiling */
430 	stopprofclock(p);
431 
432 	/* reset caught signals */
433 	execsigs(p);
434 
435 	/* name this process - nameiexec(p, ndp) */
436 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
437 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
438 	p->p_comm[len] = 0;
439 
440 	/*
441 	 * mark as execed, wakeup the process that vforked (if any) and tell
442 	 * it that it now has its own resources back
443 	 */
444 	p->p_flag |= P_EXEC;
445 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
446 		p->p_flag &= ~P_PPWAIT;
447 		wakeup(p->p_pptr);
448 	}
449 
450 	/*
451 	 * Implement image setuid/setgid.
452 	 *
453 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
454 	 * the process is being traced.
455 	 *
456 	 * XXXMAC: For the time being, use NOSUID to also prohibit
457 	 * transitions on the file system.
458 	 */
459 	oldcred = p->p_ucred;
460 	credential_changing = 0;
461 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
462 	    attr.va_uid;
463 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
464 	    attr.va_gid;
465 #ifdef MAC
466 	will_transition = mac_execve_will_transition(oldcred, imgp->vp,
467 	    interplabelvalid ? &interplabel : NULL, imgp);
468 	credential_changing |= will_transition;
469 #endif
470 
471 	if (credential_changing &&
472 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
473 	    (p->p_flag & P_TRACED) == 0) {
474 		/*
475 		 * Turn off syscall tracing for set-id programs, except for
476 		 * root.  Record any set-id flags first to make sure that
477 		 * we do not regain any tracing during a possible block.
478 		 */
479 		setsugid(p);
480 #ifdef KTRACE
481 		if (p->p_tracevp != NULL && suser_cred(oldcred, PRISON_ROOT)) {
482 			mtx_lock(&ktrace_mtx);
483 			p->p_traceflag = 0;
484 			tracevp = p->p_tracevp;
485 			p->p_tracevp = NULL;
486 			tracecred = p->p_tracecred;
487 			p->p_tracecred = NULL;
488 			mtx_unlock(&ktrace_mtx);
489 		}
490 #endif
491 		/*
492 		 * Close any file descriptors 0..2 that reference procfs,
493 		 * then make sure file descriptors 0..2 are in use.
494 		 *
495 		 * setugidsafety() may call closef() and then pfind()
496 		 * which may grab the process lock.
497 		 * fdcheckstd() may call falloc() which may block to
498 		 * allocate memory, so temporarily drop the process lock.
499 		 */
500 		PROC_UNLOCK(p);
501 		setugidsafety(td);
502 		error = fdcheckstd(td);
503 		if (error != 0)
504 			goto done1;
505 		PROC_LOCK(p);
506 		/*
507 		 * Set the new credentials.
508 		 */
509 		crcopy(newcred, oldcred);
510 		if (attr.va_mode & VSUID)
511 			change_euid(newcred, euip);
512 		if (attr.va_mode & VSGID)
513 			change_egid(newcred, attr.va_gid);
514 #ifdef MAC
515 		if (will_transition) {
516 			mac_execve_transition(oldcred, newcred, imgp->vp,
517 			    interplabelvalid ? &interplabel : NULL, imgp);
518 		}
519 #endif
520 		/*
521 		 * Implement correct POSIX saved-id behavior.
522 		 *
523 		 * XXXMAC: Note that the current logic will save the
524 		 * uid and gid if a MAC domain transition occurs, even
525 		 * though maybe it shouldn't.
526 		 */
527 		change_svuid(newcred, newcred->cr_uid);
528 		change_svgid(newcred, newcred->cr_gid);
529 		p->p_ucred = newcred;
530 		newcred = NULL;
531 	} else {
532 		if (oldcred->cr_uid == oldcred->cr_ruid &&
533 		    oldcred->cr_gid == oldcred->cr_rgid)
534 			p->p_flag &= ~P_SUGID;
535 		/*
536 		 * Implement correct POSIX saved-id behavior.
537 		 *
538 		 * XXX: It's not clear that the existing behavior is
539 		 * POSIX-compliant.  A number of sources indicate that the
540 		 * saved uid/gid should only be updated if the new ruid is
541 		 * not equal to the old ruid, or the new euid is not equal
542 		 * to the old euid and the new euid is not equal to the old
543 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
544 		 * Also, this code uses the new (replaced) euid and egid as
545 		 * the source, which may or may not be the right ones to use.
546 		 */
547 		if (oldcred->cr_svuid != oldcred->cr_uid ||
548 		    oldcred->cr_svgid != oldcred->cr_gid) {
549 			crcopy(newcred, oldcred);
550 			change_svuid(newcred, newcred->cr_uid);
551 			change_svgid(newcred, newcred->cr_gid);
552 			p->p_ucred = newcred;
553 			newcred = NULL;
554 		}
555 	}
556 
557 	/*
558 	 * Store the vp for use in procfs.  This vnode was referenced prior
559 	 * to locking the proc lock.
560 	 */
561 	textvp = p->p_textvp;
562 	p->p_textvp = ndp->ni_vp;
563 
564 	/*
565 	 * Notify others that we exec'd, and clear the P_INEXEC flag
566 	 * as we're now a bona fide freshly-execed process.
567 	 */
568 	KNOTE(&p->p_klist, NOTE_EXEC);
569 	p->p_flag &= ~P_INEXEC;
570 
571 	/*
572 	 * If tracing the process, trap to debugger so breakpoints
573 	 * can be set before the program executes.
574 	 */
575 	if (p->p_flag & P_TRACED)
576 		psignal(p, SIGTRAP);
577 
578 	/* clear "fork but no exec" flag, as we _are_ execing */
579 	p->p_acflag &= ~AFORK;
580 
581 	/* Free any previous argument cache */
582 	oldargs = p->p_args;
583 	p->p_args = NULL;
584 
585 	/* Cache arguments if they fit inside our allowance */
586 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
587 		bcopy(imgp->stringbase, newargs->ar_args, i);
588 		p->p_args = newargs;
589 		newargs = NULL;
590 	}
591 	PROC_UNLOCK(p);
592 
593 	/* Set values passed into the program in registers. */
594 	if (p->p_sysent->sv_setregs)
595 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
596 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
597 	else
598 		exec_setregs(td, imgp->entry_addr,
599 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
600 
601 done1:
602 	/*
603 	 * Free any resources malloc'd earlier that we didn't use.
604 	 */
605 	uifree(euip);
606 	if (newcred == NULL)
607 		crfree(oldcred);
608 	else
609 		crfree(newcred);
610 	/*
611 	 * Handle deferred decrement of ref counts.
612 	 */
613 	if (textvp != NULL)
614 		vrele(textvp);
615 	if (ndp->ni_vp && error != 0)
616 		vrele(ndp->ni_vp);
617 #ifdef KTRACE
618 	if (tracevp != NULL)
619 		vrele(tracevp);
620 	if (tracecred != NULL)
621 		crfree(tracecred);
622 #endif
623 	if (oldargs != NULL)
624 		pargs_drop(oldargs);
625 	if (newargs != NULL)
626 		pargs_drop(newargs);
627 
628 exec_fail_dealloc:
629 
630 	/*
631 	 * free various allocated resources
632 	 */
633 	if (imgp->firstpage)
634 		exec_unmap_first_page(imgp);
635 
636 	if (imgp->vp) {
637 		NDFREE(ndp, NDF_ONLY_PNBUF);
638 		vput(imgp->vp);
639 	}
640 
641 	if (imgp->stringbase != NULL)
642 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
643 		    ARG_MAX + PAGE_SIZE);
644 
645 	if (imgp->object)
646 		vm_object_deallocate(imgp->object);
647 
648 	if (error == 0) {
649 		/*
650 		 * Stop the process here if its stop event mask has
651 		 * the S_EXEC bit set.
652 		 */
653 		STOPEVENT(p, S_EXEC, 0);
654 		goto done2;
655 	}
656 
657 exec_fail:
658 	/* we're done here, clear P_INEXEC */
659 	PROC_LOCK(p);
660 	p->p_flag &= ~P_INEXEC;
661 	PROC_UNLOCK(p);
662 
663 	if (imgp->vmspace_destroyed) {
664 		/* sorry, no more process anymore. exit gracefully */
665 #ifdef MAC
666 		mac_execve_exit(imgp);
667 		if (interplabelvalid)
668 			mac_destroy_vnode_label(&interplabel);
669 #endif
670 		exit1(td, W_EXITCODE(0, SIGABRT));
671 		/* NOT REACHED */
672 		error = 0;
673 	}
674 done2:
675 #ifdef MAC
676 	mac_execve_exit(imgp);
677 	if (interplabelvalid)
678 		mac_destroy_vnode_label(&interplabel);
679 #endif
680 	mtx_unlock(&Giant);
681 	return (error);
682 }
683 
684 #ifndef _SYS_SYSPROTO_H_
685 struct execve_args {
686         char    *fname;
687         char    **argv;
688         char    **envv;
689 };
690 #endif
691 
692 /*
693  * MPSAFE
694  */
695 int
696 execve(td, uap)
697 	struct thread *td;
698 	struct execve_args /* {
699 		char *fname;
700 		char **argv;
701 		char **envv;
702 	} */ *uap;
703 {
704 
705 	return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
706 }
707 
708 #ifndef _SYS_SYSPROTO_H_
709 struct __mac_execve_args {
710 	char	*fname;
711 	char	**argv;
712 	char	**envv;
713 	struct mac	*mac_p;
714 };
715 #endif
716 
717 /*
718  * MPSAFE
719  */
720 int
721 __mac_execve(td, uap)
722 	struct thread *td;
723 	struct __mac_execve_args /* {
724 		char *fname;
725 		char **argv;
726 		char **envv;
727 		struct mac *mac_p;
728 	} */ *uap;
729 {
730 
731 #ifdef MAC
732 	return (kern_execve(td, uap->fname, uap->argv, uap->envv,
733 	    uap->mac_p));
734 #else
735 	return (ENOSYS);
736 #endif
737 }
738 
739 int
740 exec_map_first_page(imgp)
741 	struct image_params *imgp;
742 {
743 	int rv, i;
744 	int initial_pagein;
745 	vm_page_t ma[VM_INITIAL_PAGEIN];
746 	vm_object_t object;
747 
748 	GIANT_REQUIRED;
749 
750 	if (imgp->firstpage) {
751 		exec_unmap_first_page(imgp);
752 	}
753 
754 	VOP_GETVOBJECT(imgp->vp, &object);
755 
756 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
757 
758 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
759 		initial_pagein = VM_INITIAL_PAGEIN;
760 		if (initial_pagein > object->size)
761 			initial_pagein = object->size;
762 		for (i = 1; i < initial_pagein; i++) {
763 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
764 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
765 					break;
766 				if (ma[i]->valid)
767 					break;
768 				vm_page_lock_queues();
769 				vm_page_busy(ma[i]);
770 				vm_page_unlock_queues();
771 			} else {
772 				ma[i] = vm_page_alloc(object, i,
773 				    VM_ALLOC_NORMAL);
774 				if (ma[i] == NULL)
775 					break;
776 			}
777 		}
778 		initial_pagein = i;
779 
780 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
781 		ma[0] = vm_page_lookup(object, 0);
782 
783 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
784 		    (ma[0]->valid == 0)) {
785 			if (ma[0]) {
786 				vm_page_lock_queues();
787 				pmap_remove_all(ma[0]);
788 				vm_page_free(ma[0]);
789 				vm_page_unlock_queues();
790 			}
791 			return (EIO);
792 		}
793 	}
794 	vm_page_lock_queues();
795 	vm_page_wire(ma[0]);
796 	vm_page_wakeup(ma[0]);
797 	vm_page_unlock_queues();
798 
799 	pmap_qenter((vm_offset_t)imgp->image_header, ma, 1);
800 	imgp->firstpage = ma[0];
801 
802 	return (0);
803 }
804 
805 void
806 exec_unmap_first_page(imgp)
807 	struct image_params *imgp;
808 {
809 	GIANT_REQUIRED;
810 
811 	if (imgp->firstpage) {
812 		pmap_qremove((vm_offset_t)imgp->image_header, 1);
813 		vm_page_lock_queues();
814 		vm_page_unwire(imgp->firstpage, 1);
815 		vm_page_unlock_queues();
816 		imgp->firstpage = NULL;
817 	}
818 }
819 
820 /*
821  * Destroy old address space, and allocate a new stack
822  *	The new stack is only SGROWSIZ large because it is grown
823  *	automatically in trap.c.
824  */
825 int
826 exec_new_vmspace(imgp, sv)
827 	struct image_params *imgp;
828 	struct sysentvec *sv;
829 {
830 	int error;
831 	struct proc *p = imgp->proc;
832 	struct vmspace *vmspace = p->p_vmspace;
833 	vm_offset_t stack_addr;
834 	vm_map_t map;
835 
836 	GIANT_REQUIRED;
837 
838 	stack_addr = sv->sv_usrstack - maxssiz;
839 
840 	imgp->vmspace_destroyed = 1;
841 
842 	EVENTHANDLER_INVOKE(process_exec, p);
843 
844 	/*
845 	 * Blow away entire process VM, if address space not shared,
846 	 * otherwise, create a new VM space so that other threads are
847 	 * not disrupted
848 	 */
849 	map = &vmspace->vm_map;
850 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
851 	    vm_map_max(map) == sv->sv_maxuser) {
852 		shmexit(vmspace);
853 		vm_page_lock_queues();
854 		pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
855 		    vm_map_max(map));
856 		vm_page_unlock_queues();
857 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
858 	} else {
859 		vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
860 		vmspace = p->p_vmspace;
861 		map = &vmspace->vm_map;
862 	}
863 
864 	/* Allocate a new stack */
865 	error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
866 	    sv->sv_stackprot, VM_PROT_ALL, 0);
867 	if (error)
868 		return (error);
869 
870 #ifdef __ia64__
871 	{
872 		/*
873 		 * Allocate backing store. We really need something
874 		 * similar to vm_map_stack which can allow the backing
875 		 * store to grow upwards. This will do for now.
876 		 */
877 		vm_offset_t bsaddr;
878 		bsaddr = p->p_sysent->sv_usrstack - 2 * maxssiz;
879 		error = vm_map_find(map, 0, 0, &bsaddr,
880 		    regstkpages * PAGE_SIZE, 0, VM_PROT_ALL, VM_PROT_ALL, 0);
881 		FIRST_THREAD_IN_PROC(p)->td_md.md_bspstore = bsaddr;
882 	}
883 #endif
884 
885 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
886 	 * VM_STACK case, but they are still used to monitor the size of the
887 	 * process stack so we can check the stack rlimit.
888 	 */
889 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
890 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
891 
892 	return (0);
893 }
894 
895 /*
896  * Copy out argument and environment strings from the old process
897  *	address space into the temporary string buffer.
898  */
899 int
900 exec_extract_strings(imgp)
901 	struct image_params *imgp;
902 {
903 	char	**argv, **envv;
904 	char	*argp, *envp;
905 	int	error;
906 	size_t	length;
907 
908 	/*
909 	 * extract arguments first
910 	 */
911 
912 	argv = imgp->userspace_argv;
913 
914 	if (argv) {
915 		argp = (caddr_t)(intptr_t)fuword(argv);
916 		if (argp == (caddr_t)-1)
917 			return (EFAULT);
918 		if (argp)
919 			argv++;
920 		if (imgp->argv0)
921 			argp = imgp->argv0;
922 		if (argp) {
923 			do {
924 				if (argp == (caddr_t)-1)
925 					return (EFAULT);
926 				if ((error = copyinstr(argp, imgp->stringp,
927 				    imgp->stringspace, &length))) {
928 					if (error == ENAMETOOLONG)
929 						return (E2BIG);
930 					return (error);
931 				}
932 				imgp->stringspace -= length;
933 				imgp->stringp += length;
934 				imgp->argc++;
935 			} while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
936 		}
937 	}
938 
939 	imgp->endargs = imgp->stringp;
940 
941 	/*
942 	 * extract environment strings
943 	 */
944 
945 	envv = imgp->userspace_envv;
946 
947 	if (envv) {
948 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
949 			if (envp == (caddr_t)-1)
950 				return (EFAULT);
951 			if ((error = copyinstr(envp, imgp->stringp,
952 			    imgp->stringspace, &length))) {
953 				if (error == ENAMETOOLONG)
954 					return (E2BIG);
955 				return (error);
956 			}
957 			imgp->stringspace -= length;
958 			imgp->stringp += length;
959 			imgp->envc++;
960 		}
961 	}
962 
963 	return (0);
964 }
965 
966 /*
967  * Copy strings out to the new process address space, constructing
968  *	new arg and env vector tables. Return a pointer to the base
969  *	so that it can be used as the initial stack pointer.
970  */
971 register_t *
972 exec_copyout_strings(imgp)
973 	struct image_params *imgp;
974 {
975 	int argc, envc;
976 	char **vectp;
977 	char *stringp, *destp;
978 	register_t *stack_base;
979 	struct ps_strings *arginfo;
980 	struct proc *p;
981 	int szsigcode;
982 
983 	/*
984 	 * Calculate string base and vector table pointers.
985 	 * Also deal with signal trampoline code for this exec type.
986 	 */
987 	p = imgp->proc;
988 	szsigcode = 0;
989 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
990 	if (p->p_sysent->sv_szsigcode != NULL)
991 		szsigcode = *(p->p_sysent->sv_szsigcode);
992 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
993 	    roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
994 
995 	/*
996 	 * install sigcode
997 	 */
998 	if (szsigcode)
999 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1000 		    szsigcode), szsigcode);
1001 
1002 	/*
1003 	 * If we have a valid auxargs ptr, prepare some room
1004 	 * on the stack.
1005 	 */
1006 	if (imgp->auxargs) {
1007 		/*
1008 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1009 		 * lower compatibility.
1010 		 */
1011 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1012 		    (AT_COUNT * 2);
1013 		/*
1014 		 * The '+ 2' is for the null pointers at the end of each of
1015 		 * the arg and env vector sets,and imgp->auxarg_size is room
1016 		 * for argument of Runtime loader.
1017 		 */
1018 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1019 		    imgp->auxarg_size) * sizeof(char *));
1020 
1021 	} else
1022 		/*
1023 		 * The '+ 2' is for the null pointers at the end of each of
1024 		 * the arg and env vector sets
1025 		 */
1026 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1027 		    sizeof(char *));
1028 
1029 	/*
1030 	 * vectp also becomes our initial stack base
1031 	 */
1032 	stack_base = (register_t *)vectp;
1033 
1034 	stringp = imgp->stringbase;
1035 	argc = imgp->argc;
1036 	envc = imgp->envc;
1037 
1038 	/*
1039 	 * Copy out strings - arguments and environment.
1040 	 */
1041 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1042 
1043 	/*
1044 	 * Fill in "ps_strings" struct for ps, w, etc.
1045 	 */
1046 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1047 	suword(&arginfo->ps_nargvstr, argc);
1048 
1049 	/*
1050 	 * Fill in argument portion of vector table.
1051 	 */
1052 	for (; argc > 0; --argc) {
1053 		suword(vectp++, (long)(intptr_t)destp);
1054 		while (*stringp++ != 0)
1055 			destp++;
1056 		destp++;
1057 	}
1058 
1059 	/* a null vector table pointer separates the argp's from the envp's */
1060 	suword(vectp++, 0);
1061 
1062 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1063 	suword(&arginfo->ps_nenvstr, envc);
1064 
1065 	/*
1066 	 * Fill in environment portion of vector table.
1067 	 */
1068 	for (; envc > 0; --envc) {
1069 		suword(vectp++, (long)(intptr_t)destp);
1070 		while (*stringp++ != 0)
1071 			destp++;
1072 		destp++;
1073 	}
1074 
1075 	/* end of vector table is a null pointer */
1076 	suword(vectp, 0);
1077 
1078 	return (stack_base);
1079 }
1080 
1081 /*
1082  * Check permissions of file to execute.
1083  *	Called with imgp->vp locked.
1084  *	Return 0 for success or error code on failure.
1085  */
1086 int
1087 exec_check_permissions(imgp)
1088 	struct image_params *imgp;
1089 {
1090 	struct vnode *vp = imgp->vp;
1091 	struct vattr *attr = imgp->attr;
1092 	struct thread *td;
1093 	int error;
1094 
1095 	td = curthread;			/* XXXKSE */
1096 
1097 	/* Get file attributes */
1098 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1099 	if (error)
1100 		return (error);
1101 
1102 #ifdef MAC
1103 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1104 	if (error)
1105 		return (error);
1106 #endif
1107 
1108 	/*
1109 	 * 1) Check if file execution is disabled for the filesystem that this
1110 	 *	file resides on.
1111 	 * 2) Insure that at least one execute bit is on - otherwise root
1112 	 *	will always succeed, and we don't want to happen unless the
1113 	 *	file really is executable.
1114 	 * 3) Insure that the file is a regular file.
1115 	 */
1116 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1117 	    ((attr->va_mode & 0111) == 0) ||
1118 	    (attr->va_type != VREG))
1119 		return (EACCES);
1120 
1121 	/*
1122 	 * Zero length files can't be exec'd
1123 	 */
1124 	if (attr->va_size == 0)
1125 		return (ENOEXEC);
1126 
1127 	/*
1128 	 *  Check for execute permission to file based on current credentials.
1129 	 */
1130 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1131 	if (error)
1132 		return (error);
1133 
1134 	/*
1135 	 * Check number of open-for-writes on the file and deny execution
1136 	 * if there are any.
1137 	 */
1138 	if (vp->v_writecount)
1139 		return (ETXTBSY);
1140 
1141 	/*
1142 	 * Call filesystem specific open routine (which does nothing in the
1143 	 * general case).
1144 	 */
1145 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td);
1146 	return (error);
1147 }
1148 
1149 /*
1150  * Exec handler registration
1151  */
1152 int
1153 exec_register(execsw_arg)
1154 	const struct execsw *execsw_arg;
1155 {
1156 	const struct execsw **es, **xs, **newexecsw;
1157 	int count = 2;	/* New slot and trailing NULL */
1158 
1159 	if (execsw)
1160 		for (es = execsw; *es; es++)
1161 			count++;
1162 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1163 	if (newexecsw == NULL)
1164 		return (ENOMEM);
1165 	xs = newexecsw;
1166 	if (execsw)
1167 		for (es = execsw; *es; es++)
1168 			*xs++ = *es;
1169 	*xs++ = execsw_arg;
1170 	*xs = NULL;
1171 	if (execsw)
1172 		free(execsw, M_TEMP);
1173 	execsw = newexecsw;
1174 	return (0);
1175 }
1176 
1177 int
1178 exec_unregister(execsw_arg)
1179 	const struct execsw *execsw_arg;
1180 {
1181 	const struct execsw **es, **xs, **newexecsw;
1182 	int count = 1;
1183 
1184 	if (execsw == NULL)
1185 		panic("unregister with no handlers left?\n");
1186 
1187 	for (es = execsw; *es; es++) {
1188 		if (*es == execsw_arg)
1189 			break;
1190 	}
1191 	if (*es == NULL)
1192 		return (ENOENT);
1193 	for (es = execsw; *es; es++)
1194 		if (*es != execsw_arg)
1195 			count++;
1196 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1197 	if (newexecsw == NULL)
1198 		return (ENOMEM);
1199 	xs = newexecsw;
1200 	for (es = execsw; *es; es++)
1201 		if (*es != execsw_arg)
1202 			*xs++ = *es;
1203 	*xs = NULL;
1204 	if (execsw)
1205 		free(execsw, M_TEMP);
1206 	execsw = newexecsw;
1207 	return (0);
1208 }
1209