xref: /freebsd/sys/kern/kern_exec.c (revision ebbd4fa8c8427d3dd847ba33c45c996e0500e6ff)
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 /*
78  * callout list for things to do at exec time
79  */
80 struct execlist {
81 	execlist_fn function;
82 	TAILQ_ENTRY(execlist) next;
83 };
84 
85 TAILQ_HEAD(exec_list_head, execlist);
86 static struct exec_list_head exec_list = TAILQ_HEAD_INITIALIZER(exec_list);
87 
88 static register_t *exec_copyout_strings(struct image_params *);
89 
90 /* XXX This should be vm_size_t. */
91 static u_long ps_strings = PS_STRINGS;
92 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
93 
94 /* XXX This should be vm_size_t. */
95 static u_long usrstack = USRSTACK;
96 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
97 
98 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
99 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
100     &ps_arg_cache_limit, 0, "");
101 
102 int ps_argsopen = 1;
103 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
104 
105 #ifdef __ia64__
106 /* XXX HACK */
107 static int regstkpages = 256;
108 SYSCTL_INT(_machdep, OID_AUTO, regstkpages, CTLFLAG_RW, &regstkpages, 0, "");
109 #endif
110 
111 /*
112  * Each of the items is a pointer to a `const struct execsw', hence the
113  * double pointer here.
114  */
115 static const struct execsw **execsw;
116 
117 #ifndef _SYS_SYSPROTO_H_
118 struct execve_args {
119         char    *fname;
120         char    **argv;
121         char    **envv;
122 };
123 #endif
124 
125 /*
126  * execve() system call.
127  *
128  * MPSAFE
129  */
130 int
131 execve(td, uap)
132 	struct thread *td;
133 	register struct execve_args *uap;
134 {
135 	struct proc *p = td->td_proc;
136 	struct nameidata nd, *ndp;
137 	struct ucred *newcred = NULL, *oldcred;
138 	struct uidinfo *euip;
139 	register_t *stack_base;
140 	int error, len, i;
141 	struct image_params image_params, *imgp;
142 	struct vattr attr;
143 	int (*img_first)(struct image_params *);
144 	struct pargs *oldargs = NULL, *newargs = NULL;
145 	struct procsig *oldprocsig, *newprocsig;
146 #ifdef KTRACE
147 	struct vnode *tracevp = NULL;
148 #endif
149 	struct vnode *textvp = NULL;
150 	int credential_changing;
151 
152 	imgp = &image_params;
153 
154 	/*
155 	 * Lock the process and set the P_INEXEC flag to indicate that
156 	 * it should be left alone until we're done here.  This is
157 	 * necessary to avoid race conditions - e.g. in ptrace() -
158 	 * that might allow a local user to illicitly obtain elevated
159 	 * privileges.
160 	 */
161 	PROC_LOCK(p);
162 	KASSERT((p->p_flag & P_INEXEC) == 0,
163 	    ("%s(): process already has P_INEXEC flag", __func__));
164 	if ((p->p_flag & P_KSES) && thread_single(SNGLE_EXIT)) {
165 		PROC_UNLOCK(p);
166 		return (ERESTART);	/* Try again later. */
167 	}
168 	/* If we get here all other threads are dead. */
169 	p->p_flag |= P_INEXEC;
170 	PROC_UNLOCK(p);
171 
172 	/*
173 	 * Initialize part of the common data
174 	 */
175 	imgp->proc = p;
176 	imgp->uap = uap;
177 	imgp->attr = &attr;
178 	imgp->argc = imgp->envc = 0;
179 	imgp->argv0 = NULL;
180 	imgp->entry_addr = 0;
181 	imgp->vmspace_destroyed = 0;
182 	imgp->interpreted = 0;
183 	imgp->interpreter_name[0] = '\0';
184 	imgp->auxargs = NULL;
185 	imgp->vp = NULL;
186 	imgp->object = NULL;
187 	imgp->firstpage = NULL;
188 	imgp->ps_strings = 0;
189 	imgp->auxarg_size = 0;
190 
191 	/*
192 	 * Allocate temporary demand zeroed space for argument and
193 	 *	environment strings
194 	 */
195 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX + PAGE_SIZE);
196 	if (imgp->stringbase == NULL) {
197 		error = ENOMEM;
198 		mtx_lock(&Giant);
199 		goto exec_fail;
200 	}
201 	imgp->stringp = imgp->stringbase;
202 	imgp->stringspace = ARG_MAX;
203 	imgp->image_header = imgp->stringbase + ARG_MAX;
204 
205 	/*
206 	 * Translate the file name. namei() returns a vnode pointer
207 	 *	in ni_vp amoung other things.
208 	 */
209 	ndp = &nd;
210 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
211 	    UIO_USERSPACE, uap->fname, td);
212 
213 	mtx_lock(&Giant);
214 interpret:
215 
216 	error = namei(ndp);
217 	if (error) {
218 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
219 			ARG_MAX + PAGE_SIZE);
220 		goto exec_fail;
221 	}
222 
223 	imgp->vp = ndp->ni_vp;
224 	imgp->fname = uap->fname;
225 
226 	/*
227 	 * Check file permissions (also 'opens' file)
228 	 */
229 	error = exec_check_permissions(imgp);
230 	if (error) {
231 		VOP_UNLOCK(imgp->vp, 0, td);
232 		goto exec_fail_dealloc;
233 	}
234 	VOP_GETVOBJECT(imgp->vp, &imgp->object);
235 	vm_object_reference(imgp->object);
236 
237 	error = exec_map_first_page(imgp);
238 	VOP_UNLOCK(imgp->vp, 0, td);
239 	if (error)
240 		goto exec_fail_dealloc;
241 
242 	/*
243 	 *	If the current process has a special image activator it
244 	 *	wants to try first, call it.   For example, emulating shell
245 	 *	scripts differently.
246 	 */
247 	error = -1;
248 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
249 		error = img_first(imgp);
250 
251 	/*
252 	 *	Loop through the list of image activators, calling each one.
253 	 *	An activator returns -1 if there is no match, 0 on success,
254 	 *	and an error otherwise.
255 	 */
256 	for (i = 0; error == -1 && execsw[i]; ++i) {
257 		if (execsw[i]->ex_imgact == NULL ||
258 		    execsw[i]->ex_imgact == img_first) {
259 			continue;
260 		}
261 		error = (*execsw[i]->ex_imgact)(imgp);
262 	}
263 
264 	if (error) {
265 		if (error == -1)
266 			error = ENOEXEC;
267 		goto exec_fail_dealloc;
268 	}
269 
270 	/*
271 	 * Special interpreter operation, cleanup and loop up to try to
272 	 * activate the interpreter.
273 	 */
274 	if (imgp->interpreted) {
275 		exec_unmap_first_page(imgp);
276 		/* free name buffer and old vnode */
277 		NDFREE(ndp, NDF_ONLY_PNBUF);
278 		vrele(ndp->ni_vp);
279 		vm_object_deallocate(imgp->object);
280 		imgp->object = NULL;
281 		/* set new name to that of the interpreter */
282 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
283 		    UIO_SYSSPACE, imgp->interpreter_name, td);
284 		goto interpret;
285 	}
286 
287 	/*
288 	 * Copy out strings (args and env) and initialize stack base
289 	 */
290 	if (p->p_sysent->sv_copyout_strings)
291 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
292 	else
293 		stack_base = exec_copyout_strings(imgp);
294 
295 	/*
296 	 * If custom stack fixup routine present for this process
297 	 * let it do the stack setup.
298 	 * Else stuff argument count as first item on stack
299 	 */
300 	if (p->p_sysent->sv_fixup)
301 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
302 	else
303 		suword(--stack_base, imgp->argc);
304 
305 	/*
306 	 * For security and other reasons, the file descriptor table cannot
307 	 * be shared after an exec.
308 	 */
309 	FILEDESC_LOCK(p->p_fd);
310 	if (p->p_fd->fd_refcnt > 1) {
311 		struct filedesc *tmp;
312 
313 		tmp = fdcopy(td);
314 		FILEDESC_UNLOCK(p->p_fd);
315 		fdfree(td);
316 		p->p_fd = tmp;
317 	} else
318 		FILEDESC_UNLOCK(p->p_fd);
319 
320 	/*
321 	 * Malloc things before we need locks.
322 	 */
323 	newcred = crget();
324 	euip = uifind(attr.va_uid);
325 	i = imgp->endargs - imgp->stringbase;
326 	if (ps_arg_cache_limit >= i + sizeof(struct pargs))
327 		newargs = pargs_alloc(i);
328 
329 	/* close files on exec */
330 	fdcloseexec(td);
331 
332 	/*
333 	 * For security and other reasons, signal handlers cannot
334 	 * be shared after an exec. The new process gets a copy of the old
335 	 * handlers. In execsigs(), the new process will have its signals
336 	 * reset.
337 	 */
338 	PROC_LOCK(p);
339 	mp_fixme("procsig needs a lock");
340 	if (p->p_procsig->ps_refcnt > 1) {
341 		oldprocsig = p->p_procsig;
342 		PROC_UNLOCK(p);
343 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
344 		    M_SUBPROC, M_WAITOK);
345 		bcopy(oldprocsig, newprocsig, sizeof(*newprocsig));
346 		newprocsig->ps_refcnt = 1;
347 		oldprocsig->ps_refcnt--;
348 		PROC_LOCK(p);
349 		p->p_procsig = newprocsig;
350 		if (p->p_sigacts == &p->p_uarea->u_sigacts)
351 			panic("shared procsig but private sigacts?");
352 
353 		p->p_uarea->u_sigacts = *p->p_sigacts;
354 		p->p_sigacts = &p->p_uarea->u_sigacts;
355 	}
356 	/* Stop profiling */
357 	stopprofclock(p);
358 
359 	/* reset caught signals */
360 	execsigs(p);
361 
362 	/* name this process - nameiexec(p, ndp) */
363 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
364 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
365 	p->p_comm[len] = 0;
366 
367 	/*
368 	 * mark as execed, wakeup the process that vforked (if any) and tell
369 	 * it that it now has its own resources back
370 	 */
371 	p->p_flag |= P_EXEC;
372 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
373 		p->p_flag &= ~P_PPWAIT;
374 		wakeup(p->p_pptr);
375 	}
376 
377 	/*
378 	 * Implement image setuid/setgid.
379 	 *
380 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
381 	 * the process is being traced.
382 	 */
383 	oldcred = p->p_ucred;
384 	credential_changing = 0;
385 	credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
386 	    attr.va_uid;
387 	credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
388 	    attr.va_gid;
389 
390 	if (credential_changing &&
391 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
392 	    (p->p_flag & P_TRACED) == 0) {
393 		/*
394 		 * Turn off syscall tracing for set-id programs, except for
395 		 * root.  Record any set-id flags first to make sure that
396 		 * we do not regain any tracing during a possible block.
397 		 */
398 		setsugid(p);
399 #ifdef KTRACE
400 		if (p->p_tracep && suser_cred(oldcred, PRISON_ROOT)) {
401 			mtx_lock(&ktrace_mtx);
402 			p->p_traceflag = 0;
403 			tracevp = p->p_tracep;
404 			p->p_tracep = NULL;
405 			mtx_unlock(&ktrace_mtx);
406 		}
407 #endif
408 		/* Close any file descriptors 0..2 that reference procfs */
409 		setugidsafety(td);
410 		/* Make sure file descriptors 0..2 are in use.  */
411 		error = fdcheckstd(td);
412 		if (error != 0)
413 			goto done1;
414 		/*
415 		 * Set the new credentials.
416 		 */
417 		crcopy(newcred, oldcred);
418 		if (attr.va_mode & VSUID)
419 			change_euid(newcred, euip);
420 		if (attr.va_mode & VSGID)
421 			change_egid(newcred, attr.va_gid);
422 		/*
423 		 * Implement correct POSIX saved-id behavior.
424 		 */
425 		change_svuid(newcred, newcred->cr_uid);
426 		change_svgid(newcred, newcred->cr_gid);
427 		p->p_ucred = newcred;
428 		newcred = NULL;
429 	} else {
430 		if (oldcred->cr_uid == oldcred->cr_ruid &&
431 		    oldcred->cr_gid == oldcred->cr_rgid)
432 			p->p_flag &= ~P_SUGID;
433 		/*
434 		 * Implement correct POSIX saved-id behavior.
435 		 *
436 		 * XXX: It's not clear that the existing behavior is
437 		 * POSIX-compliant.  A number of sources indicate that the
438 		 * saved uid/gid should only be updated if the new ruid is
439 		 * not equal to the old ruid, or the new euid is not equal
440 		 * to the old euid and the new euid is not equal to the old
441 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
442 		 * Also, this code uses the new (replaced) euid and egid as
443 		 * the source, which may or may not be the right ones to use.
444 		 */
445 		if (oldcred->cr_svuid != oldcred->cr_uid ||
446 		    oldcred->cr_svgid != oldcred->cr_gid) {
447 			crcopy(newcred, oldcred);
448 			change_svuid(newcred, newcred->cr_uid);
449 			change_svgid(newcred, newcred->cr_gid);
450 			p->p_ucred = newcred;
451 			newcred = NULL;
452 		}
453 	}
454 
455 	/*
456 	 * Store the vp for use in procfs
457 	 */
458 	textvp = p->p_textvp;
459 	VREF(ndp->ni_vp);
460 	p->p_textvp = ndp->ni_vp;
461 
462 	/*
463 	 * Notify others that we exec'd, and clear the P_INEXEC flag
464 	 * as we're now a bona fide freshly-execed process.
465 	 */
466 	KNOTE(&p->p_klist, NOTE_EXEC);
467 	p->p_flag &= ~P_INEXEC;
468 
469 	/*
470 	 * If tracing the process, trap to debugger so breakpoints
471 	 * can be set before the program executes.
472 	 */
473 	_STOPEVENT(p, S_EXEC, 0);
474 
475 	if (p->p_flag & P_TRACED)
476 		psignal(p, SIGTRAP);
477 
478 	/* clear "fork but no exec" flag, as we _are_ execing */
479 	p->p_acflag &= ~AFORK;
480 
481 	/* Free any previous argument cache */
482 	oldargs = p->p_args;
483 	p->p_args = NULL;
484 
485 	/* Set values passed into the program in registers. */
486 	if (p->p_sysent->sv_setregs)
487 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
488 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
489 	else
490 		setregs(td, imgp->entry_addr, (u_long)(uintptr_t)stack_base,
491 		    imgp->ps_strings);
492 
493 	/* Cache arguments if they fit inside our allowance */
494 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
495 		bcopy(imgp->stringbase, newargs->ar_args, i);
496 		p->p_args = newargs;
497 		newargs = NULL;
498 	}
499 done1:
500 	PROC_UNLOCK(p);
501 
502 	/*
503 	 * Free any resources malloc'd earlier that we didn't use.
504 	 */
505 	uifree(euip);
506 	if (newcred == NULL)
507 		crfree(oldcred);
508 	else
509 		crfree(newcred);
510 	/*
511 	 * Handle deferred decrement of ref counts.
512 	 */
513 	if (textvp != NULL)
514 		vrele(textvp);
515 #ifdef KTRACE
516 	if (tracevp != NULL)
517 		vrele(tracevp);
518 #endif
519 	if (oldargs != NULL)
520 		pargs_drop(oldargs);
521 	if (newargs != NULL)
522 		pargs_drop(newargs);
523 
524 exec_fail_dealloc:
525 
526 	/*
527 	 * free various allocated resources
528 	 */
529 	if (imgp->firstpage)
530 		exec_unmap_first_page(imgp);
531 
532 	if (imgp->stringbase != NULL)
533 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
534 			ARG_MAX + PAGE_SIZE);
535 
536 	if (imgp->vp) {
537 		NDFREE(ndp, NDF_ONLY_PNBUF);
538 		vrele(imgp->vp);
539 	}
540 
541 	if (imgp->object)
542 		vm_object_deallocate(imgp->object);
543 
544 	if (error == 0)
545 		goto done2;
546 
547 exec_fail:
548 	/* we're done here, clear P_INEXEC */
549 	PROC_LOCK(p);
550 	p->p_flag &= ~P_INEXEC;
551 	PROC_UNLOCK(p);
552 
553 	if (imgp->vmspace_destroyed) {
554 		/* sorry, no more process anymore. exit gracefully */
555 		exit1(td, W_EXITCODE(0, SIGABRT));
556 		/* NOT REACHED */
557 		error = 0;
558 	}
559 done2:
560 	mtx_unlock(&Giant);
561 	return (error);
562 }
563 
564 int
565 exec_map_first_page(imgp)
566 	struct image_params *imgp;
567 {
568 	int rv, i;
569 	int initial_pagein;
570 	vm_page_t ma[VM_INITIAL_PAGEIN];
571 	vm_object_t object;
572 
573 	GIANT_REQUIRED;
574 
575 	if (imgp->firstpage) {
576 		exec_unmap_first_page(imgp);
577 	}
578 
579 	VOP_GETVOBJECT(imgp->vp, &object);
580 
581 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
582 
583 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
584 		initial_pagein = VM_INITIAL_PAGEIN;
585 		if (initial_pagein > object->size)
586 			initial_pagein = object->size;
587 		for (i = 1; i < initial_pagein; i++) {
588 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
589 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
590 					break;
591 				if (ma[i]->valid)
592 					break;
593 				vm_page_busy(ma[i]);
594 			} else {
595 				ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
596 				if (ma[i] == NULL)
597 					break;
598 			}
599 		}
600 		initial_pagein = i;
601 
602 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
603 		ma[0] = vm_page_lookup(object, 0);
604 
605 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || (ma[0]->valid == 0)) {
606 			if (ma[0]) {
607 				vm_page_lock_queues();
608 				vm_page_protect(ma[0], VM_PROT_NONE);
609 				vm_page_free(ma[0]);
610 				vm_page_unlock_queues();
611 			}
612 			return EIO;
613 		}
614 	}
615 	vm_page_lock_queues();
616 	vm_page_wire(ma[0]);
617 	vm_page_wakeup(ma[0]);
618 	vm_page_unlock_queues();
619 
620 	pmap_qenter((vm_offset_t)imgp->image_header, ma, 1);
621 	imgp->firstpage = ma[0];
622 
623 	return 0;
624 }
625 
626 void
627 exec_unmap_first_page(imgp)
628 	struct image_params *imgp;
629 {
630 	GIANT_REQUIRED;
631 
632 	if (imgp->firstpage) {
633 		pmap_qremove((vm_offset_t)imgp->image_header, 1);
634 		vm_page_lock_queues();
635 		vm_page_unwire(imgp->firstpage, 1);
636 		vm_page_unlock_queues();
637 		imgp->firstpage = NULL;
638 	}
639 }
640 
641 /*
642  * Destroy old address space, and allocate a new stack
643  *	The new stack is only SGROWSIZ large because it is grown
644  *	automatically in trap.c.
645  */
646 int
647 exec_new_vmspace(imgp, minuser, maxuser, stack_addr)
648 	struct image_params *imgp;
649 	vm_offset_t minuser, maxuser, stack_addr;
650 {
651 	int error;
652 	struct execlist *ep;
653 	struct proc *p = imgp->proc;
654 	struct vmspace *vmspace = p->p_vmspace;
655 
656 	GIANT_REQUIRED;
657 
658 	stack_addr = stack_addr - maxssiz;
659 
660 	imgp->vmspace_destroyed = 1;
661 
662 	/*
663 	 * Perform functions registered with at_exec().
664 	 */
665 	TAILQ_FOREACH(ep, &exec_list, next)
666 		(*ep->function)(p);
667 
668 	/*
669 	 * Blow away entire process VM, if address space not shared,
670 	 * otherwise, create a new VM space so that other threads are
671 	 * not disrupted
672 	 */
673 	if (vmspace->vm_refcnt == 1
674 	    && vm_map_min(&vmspace->vm_map) == minuser
675 	    && vm_map_max(&vmspace->vm_map) == maxuser) {
676 		if (vmspace->vm_shm)
677 			shmexit(p);
678 		pmap_remove_pages(vmspace_pmap(vmspace), minuser, maxuser);
679 		vm_map_remove(&vmspace->vm_map, minuser, maxuser);
680 	} else {
681 		vmspace_exec(p, minuser, maxuser);
682 		vmspace = p->p_vmspace;
683 	}
684 
685 	/* Allocate a new stack */
686 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
687 	    VM_PROT_ALL, VM_PROT_ALL, 0);
688 	if (error)
689 		return (error);
690 
691 #ifdef __ia64__
692 	{
693 		/*
694 		 * Allocate backing store. We really need something
695 		 * similar to vm_map_stack which can allow the backing
696 		 * store to grow upwards. This will do for now.
697 		 */
698 		vm_offset_t bsaddr;
699 		bsaddr = USRSTACK - 2*maxssiz;
700 		error = vm_map_find(&vmspace->vm_map, 0, 0, &bsaddr,
701 				    regstkpages * PAGE_SIZE, 0,
702 				    VM_PROT_ALL, VM_PROT_ALL, 0);
703 		FIRST_THREAD_IN_PROC(p)->td_md.md_bspstore = bsaddr;
704 	}
705 #endif
706 
707 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
708 	 * VM_STACK case, but they are still used to monitor the size of the
709 	 * process stack so we can check the stack rlimit.
710 	 */
711 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
712 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
713 
714 	return(0);
715 }
716 
717 /*
718  * Copy out argument and environment strings from the old process
719  *	address space into the temporary string buffer.
720  */
721 int
722 exec_extract_strings(imgp)
723 	struct image_params *imgp;
724 {
725 	char	**argv, **envv;
726 	char	*argp, *envp;
727 	int	error;
728 	size_t	length;
729 
730 	/*
731 	 * extract arguments first
732 	 */
733 
734 	argv = imgp->uap->argv;
735 
736 	if (argv) {
737 		argp = (caddr_t) (intptr_t) fuword(argv);
738 		if (argp == (caddr_t) -1)
739 			return (EFAULT);
740 		if (argp)
741 			argv++;
742 		if (imgp->argv0)
743 			argp = imgp->argv0;
744 		if (argp) {
745 			do {
746 				if (argp == (caddr_t) -1)
747 					return (EFAULT);
748 				if ((error = copyinstr(argp, imgp->stringp,
749 				    imgp->stringspace, &length))) {
750 					if (error == ENAMETOOLONG)
751 						return(E2BIG);
752 					return (error);
753 				}
754 				imgp->stringspace -= length;
755 				imgp->stringp += length;
756 				imgp->argc++;
757 			} while ((argp = (caddr_t) (intptr_t) fuword(argv++)));
758 		}
759 	}
760 
761 	imgp->endargs = imgp->stringp;
762 
763 	/*
764 	 * extract environment strings
765 	 */
766 
767 	envv = imgp->uap->envv;
768 
769 	if (envv) {
770 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
771 			if (envp == (caddr_t) -1)
772 				return (EFAULT);
773 			if ((error = copyinstr(envp, imgp->stringp,
774 			    imgp->stringspace, &length))) {
775 				if (error == ENAMETOOLONG)
776 					return(E2BIG);
777 				return (error);
778 			}
779 			imgp->stringspace -= length;
780 			imgp->stringp += length;
781 			imgp->envc++;
782 		}
783 	}
784 
785 	return (0);
786 }
787 
788 /*
789  * Copy strings out to the new process address space, constructing
790  *	new arg and env vector tables. Return a pointer to the base
791  *	so that it can be used as the initial stack pointer.
792  */
793 register_t *
794 exec_copyout_strings(imgp)
795 	struct image_params *imgp;
796 {
797 	int argc, envc;
798 	char **vectp;
799 	char *stringp, *destp;
800 	register_t *stack_base;
801 	struct ps_strings *arginfo;
802 	int szsigcode;
803 
804 	/*
805 	 * Calculate string base and vector table pointers.
806 	 * Also deal with signal trampoline code for this exec type.
807 	 */
808 	arginfo = (struct ps_strings *)PS_STRINGS;
809 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
810 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
811 		roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
812 
813 	/*
814 	 * install sigcode
815 	 */
816 	if (szsigcode)
817 		copyout(imgp->proc->p_sysent->sv_sigcode,
818 			((caddr_t)arginfo - szsigcode), szsigcode);
819 
820 	/*
821 	 * If we have a valid auxargs ptr, prepare some room
822 	 * on the stack.
823 	 */
824 	if (imgp->auxargs) {
825 		/*
826 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
827 		 * lower compatibility.
828 		 */
829 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size
830 			: (AT_COUNT * 2);
831 		/*
832 		 * The '+ 2' is for the null pointers at the end of each of
833 		 * the arg and env vector sets,and imgp->auxarg_size is room
834 		 * for argument of Runtime loader.
835 		 */
836 		vectp = (char **) (destp - (imgp->argc + imgp->envc + 2 +
837 				       imgp->auxarg_size) * sizeof(char *));
838 
839 	} else
840 		/*
841 		 * The '+ 2' is for the null pointers at the end of each of
842 		 * the arg and env vector sets
843 		 */
844 		vectp = (char **)
845 			(destp - (imgp->argc + imgp->envc + 2) * sizeof(char *));
846 
847 	/*
848 	 * vectp also becomes our initial stack base
849 	 */
850 	stack_base = (register_t *)vectp;
851 
852 	stringp = imgp->stringbase;
853 	argc = imgp->argc;
854 	envc = imgp->envc;
855 
856 	/*
857 	 * Copy out strings - arguments and environment.
858 	 */
859 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
860 
861 	/*
862 	 * Fill in "ps_strings" struct for ps, w, etc.
863 	 */
864 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
865 	suword(&arginfo->ps_nargvstr, argc);
866 
867 	/*
868 	 * Fill in argument portion of vector table.
869 	 */
870 	for (; argc > 0; --argc) {
871 		suword(vectp++, (long)(intptr_t)destp);
872 		while (*stringp++ != 0)
873 			destp++;
874 		destp++;
875 	}
876 
877 	/* a null vector table pointer separates the argp's from the envp's */
878 	suword(vectp++, 0);
879 
880 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
881 	suword(&arginfo->ps_nenvstr, envc);
882 
883 	/*
884 	 * Fill in environment portion of vector table.
885 	 */
886 	for (; envc > 0; --envc) {
887 		suword(vectp++, (long)(intptr_t)destp);
888 		while (*stringp++ != 0)
889 			destp++;
890 		destp++;
891 	}
892 
893 	/* end of vector table is a null pointer */
894 	suword(vectp, 0);
895 
896 	return (stack_base);
897 }
898 
899 /*
900  * Check permissions of file to execute.
901  *	Called with imgp->vp locked.
902  *	Return 0 for success or error code on failure.
903  */
904 int
905 exec_check_permissions(imgp)
906 	struct image_params *imgp;
907 {
908 	struct vnode *vp = imgp->vp;
909 	struct vattr *attr = imgp->attr;
910 	struct thread *td;
911 	int error;
912 
913 	td = curthread;			/* XXXKSE */
914 
915 #ifdef MAC
916 	error = mac_check_vnode_exec(td->td_ucred, imgp->vp);
917 	if (error)
918 		return (error);
919 #endif
920 
921 	/* Get file attributes */
922 	error = VOP_GETATTR(vp, attr, td->td_ucred, td);
923 	if (error)
924 		return (error);
925 
926 	/*
927 	 * 1) Check if file execution is disabled for the filesystem that this
928 	 *	file resides on.
929 	 * 2) Insure that at least one execute bit is on - otherwise root
930 	 *	will always succeed, and we don't want to happen unless the
931 	 *	file really is executable.
932 	 * 3) Insure that the file is a regular file.
933 	 */
934 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
935 	    ((attr->va_mode & 0111) == 0) ||
936 	    (attr->va_type != VREG))
937 		return (EACCES);
938 
939 	/*
940 	 * Zero length files can't be exec'd
941 	 */
942 	if (attr->va_size == 0)
943 		return (ENOEXEC);
944 
945 	/*
946 	 *  Check for execute permission to file based on current credentials.
947 	 */
948 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
949 	if (error)
950 		return (error);
951 
952 	/*
953 	 * Check number of open-for-writes on the file and deny execution
954 	 * if there are any.
955 	 */
956 	if (vp->v_writecount)
957 		return (ETXTBSY);
958 
959 	/*
960 	 * Call filesystem specific open routine (which does nothing in the
961 	 * general case).
962 	 */
963 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td);
964 	return (error);
965 }
966 
967 /*
968  * Exec handler registration
969  */
970 int
971 exec_register(execsw_arg)
972 	const struct execsw *execsw_arg;
973 {
974 	const struct execsw **es, **xs, **newexecsw;
975 	int count = 2;	/* New slot and trailing NULL */
976 
977 	if (execsw)
978 		for (es = execsw; *es; es++)
979 			count++;
980 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
981 	if (newexecsw == NULL)
982 		return ENOMEM;
983 	xs = newexecsw;
984 	if (execsw)
985 		for (es = execsw; *es; es++)
986 			*xs++ = *es;
987 	*xs++ = execsw_arg;
988 	*xs = NULL;
989 	if (execsw)
990 		free(execsw, M_TEMP);
991 	execsw = newexecsw;
992 	return 0;
993 }
994 
995 int
996 exec_unregister(execsw_arg)
997 	const struct execsw *execsw_arg;
998 {
999 	const struct execsw **es, **xs, **newexecsw;
1000 	int count = 1;
1001 
1002 	if (execsw == NULL)
1003 		panic("unregister with no handlers left?\n");
1004 
1005 	for (es = execsw; *es; es++) {
1006 		if (*es == execsw_arg)
1007 			break;
1008 	}
1009 	if (*es == NULL)
1010 		return ENOENT;
1011 	for (es = execsw; *es; es++)
1012 		if (*es != execsw_arg)
1013 			count++;
1014 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1015 	if (newexecsw == NULL)
1016 		return ENOMEM;
1017 	xs = newexecsw;
1018 	for (es = execsw; *es; es++)
1019 		if (*es != execsw_arg)
1020 			*xs++ = *es;
1021 	*xs = NULL;
1022 	if (execsw)
1023 		free(execsw, M_TEMP);
1024 	execsw = newexecsw;
1025 	return 0;
1026 }
1027 
1028 int
1029 at_exec(function)
1030 	execlist_fn function;
1031 {
1032 	struct execlist *ep;
1033 
1034 #ifdef INVARIANTS
1035 	/* Be noisy if the programmer has lost track of things */
1036 	if (rm_at_exec(function))
1037 		printf("WARNING: exec callout entry (%p) already present\n",
1038 		    function);
1039 #endif
1040 	ep = malloc(sizeof(*ep), M_ATEXEC, M_NOWAIT);
1041 	if (ep == NULL)
1042 		return (ENOMEM);
1043 	ep->function = function;
1044 	TAILQ_INSERT_TAIL(&exec_list, ep, next);
1045 	return (0);
1046 }
1047 
1048 /*
1049  * Scan the exec callout list for the given item and remove it.
1050  * Returns the number of items removed (0 or 1)
1051  */
1052 int
1053 rm_at_exec(function)
1054 	execlist_fn function;
1055 {
1056 	struct execlist *ep;
1057 
1058 	TAILQ_FOREACH(ep, &exec_list, next) {
1059 		if (ep->function == function) {
1060 			TAILQ_REMOVE(&exec_list, ep, next);
1061 			free(ep, M_ATEXEC);
1062 			return(1);
1063 		}
1064 	}
1065 	return (0);
1066 }
1067 
1068