xref: /dragonfly/sys/kern/kern_exec.c (revision 984263bc)
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: src/sys/kern/kern_exec.c,v 1.107.2.15 2002/07/30 15:40:46 nectar Exp $
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysproto.h>
32 #include <sys/kernel.h>
33 #include <sys/mount.h>
34 #include <sys/filedesc.h>
35 #include <sys/fcntl.h>
36 #include <sys/acct.h>
37 #include <sys/exec.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_elf.h>
40 #include <sys/wait.h>
41 #include <sys/malloc.h>
42 #include <sys/proc.h>
43 #include <sys/signalvar.h>
44 #include <sys/pioctl.h>
45 #include <sys/namei.h>
46 #include <sys/sysent.h>
47 #include <sys/shm.h>
48 #include <sys/sysctl.h>
49 #include <sys/vnode.h>
50 #include <sys/aio.h>
51 
52 #include <vm/vm.h>
53 #include <vm/vm_param.h>
54 #include <sys/lock.h>
55 #include <vm/pmap.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_map.h>
58 #include <vm/vm_kern.h>
59 #include <vm/vm_extern.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_pager.h>
62 
63 #include <sys/user.h>
64 #include <machine/reg.h>
65 
66 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
67 
68 static register_t *exec_copyout_strings __P((struct image_params *));
69 
70 /* XXX This should be vm_size_t. */
71 static u_long ps_strings = PS_STRINGS;
72 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
73 
74 /* XXX This should be vm_size_t. */
75 static u_long usrstack = USRSTACK;
76 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
77 
78 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
79 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
80     &ps_arg_cache_limit, 0, "");
81 
82 int ps_argsopen = 1;
83 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
84 
85 /*
86  * Each of the items is a pointer to a `const struct execsw', hence the
87  * double pointer here.
88  */
89 static const struct execsw **execsw;
90 
91 #ifndef _SYS_SYSPROTO_H_
92 struct execve_args {
93         char    *fname;
94         char    **argv;
95         char    **envv;
96 };
97 #endif
98 
99 /*
100  * execve() system call.
101  */
102 int
103 execve(p, uap)
104 	struct proc *p;
105 	register struct execve_args *uap;
106 {
107 	struct nameidata nd, *ndp;
108 	register_t *stack_base;
109 	int error, len, i;
110 	struct image_params image_params, *imgp;
111 	struct vattr attr;
112 	int (*img_first) __P((struct image_params *));
113 
114 	imgp = &image_params;
115 
116 	/*
117 	 * Lock the process and set the P_INEXEC flag to indicate that
118 	 * it should be left alone until we're done here.  This is
119 	 * necessary to avoid race conditions - e.g. in ptrace() -
120 	 * that might allow a local user to illicitly obtain elevated
121 	 * privileges.
122 	 */
123 	p->p_flag |= P_INEXEC;
124 
125 	/*
126 	 * Initialize part of the common data
127 	 */
128 	imgp->proc = p;
129 	imgp->uap = uap;
130 	imgp->attr = &attr;
131 	imgp->argc = imgp->envc = 0;
132 	imgp->argv0 = NULL;
133 	imgp->entry_addr = 0;
134 	imgp->vmspace_destroyed = 0;
135 	imgp->interpreted = 0;
136 	imgp->interpreter_name[0] = '\0';
137 	imgp->auxargs = NULL;
138 	imgp->vp = NULL;
139 	imgp->firstpage = NULL;
140 	imgp->ps_strings = 0;
141 
142 	/*
143 	 * Allocate temporary demand zeroed space for argument and
144 	 *	environment strings
145 	 */
146 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX + PAGE_SIZE);
147 	if (imgp->stringbase == NULL) {
148 		error = ENOMEM;
149 		goto exec_fail;
150 	}
151 	imgp->stringp = imgp->stringbase;
152 	imgp->stringspace = ARG_MAX;
153 	imgp->image_header = imgp->stringbase + ARG_MAX;
154 
155 	/*
156 	 * Translate the file name. namei() returns a vnode pointer
157 	 *	in ni_vp amoung other things.
158 	 */
159 	ndp = &nd;
160 	NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
161 	    UIO_USERSPACE, uap->fname, p);
162 
163 interpret:
164 
165 	error = namei(ndp);
166 	if (error) {
167 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
168 			ARG_MAX + PAGE_SIZE);
169 		goto exec_fail;
170 	}
171 
172 	imgp->vp = ndp->ni_vp;
173 	imgp->fname = uap->fname;
174 
175 	/*
176 	 * Check file permissions (also 'opens' file)
177 	 */
178 	error = exec_check_permissions(imgp);
179 	if (error) {
180 		VOP_UNLOCK(imgp->vp, 0, p);
181 		goto exec_fail_dealloc;
182 	}
183 
184 	error = exec_map_first_page(imgp);
185 	VOP_UNLOCK(imgp->vp, 0, p);
186 	if (error)
187 		goto exec_fail_dealloc;
188 
189 	/*
190 	 *	If the current process has a special image activator it
191 	 *	wants to try first, call it.   For example, emulating shell
192 	 *	scripts differently.
193 	 */
194 	error = -1;
195 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
196 		error = img_first(imgp);
197 
198 	/*
199 	 *	Loop through the list of image activators, calling each one.
200 	 *	An activator returns -1 if there is no match, 0 on success,
201 	 *	and an error otherwise.
202 	 */
203 	for (i = 0; error == -1 && execsw[i]; ++i) {
204 		if (execsw[i]->ex_imgact == NULL ||
205 		    execsw[i]->ex_imgact == img_first) {
206 			continue;
207 		}
208 		error = (*execsw[i]->ex_imgact)(imgp);
209 	}
210 
211 	if (error) {
212 		if (error == -1)
213 			error = ENOEXEC;
214 		goto exec_fail_dealloc;
215 	}
216 
217 	/*
218 	 * Special interpreter operation, cleanup and loop up to try to
219 	 * activate the interpreter.
220 	 */
221 	if (imgp->interpreted) {
222 		exec_unmap_first_page(imgp);
223 		/* free name buffer and old vnode */
224 		NDFREE(ndp, NDF_ONLY_PNBUF);
225 		vrele(ndp->ni_vp);
226 		/* set new name to that of the interpreter */
227 		NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
228 		    UIO_SYSSPACE, imgp->interpreter_name, p);
229 		goto interpret;
230 	}
231 
232 	/*
233 	 * Copy out strings (args and env) and initialize stack base
234 	 */
235 	stack_base = exec_copyout_strings(imgp);
236 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
237 
238 	/*
239 	 * If custom stack fixup routine present for this process
240 	 * let it do the stack setup.
241 	 * Else stuff argument count as first item on stack
242 	 */
243 	if (p->p_sysent->sv_fixup)
244 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
245 	else
246 		suword(--stack_base, imgp->argc);
247 
248 	/*
249 	 * For security and other reasons, the file descriptor table cannot
250 	 * be shared after an exec.
251 	 */
252 	if (p->p_fd->fd_refcnt > 1) {
253 		struct filedesc *tmp;
254 
255 		tmp = fdcopy(p);
256 		fdfree(p);
257 		p->p_fd = tmp;
258 	}
259 
260 	/*
261 	 * For security and other reasons, signal handlers cannot
262 	 * be shared after an exec. The new proces gets a copy of the old
263 	 * handlers. In execsigs(), the new process will have its signals
264 	 * reset.
265 	 */
266 	if (p->p_procsig->ps_refcnt > 1) {
267 		struct procsig *newprocsig;
268 
269 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
270 		       M_SUBPROC, M_WAITOK);
271 		bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig));
272 		p->p_procsig->ps_refcnt--;
273 		p->p_procsig = newprocsig;
274 		p->p_procsig->ps_refcnt = 1;
275 		if (p->p_sigacts == &p->p_addr->u_sigacts)
276 			panic("shared procsig but private sigacts?");
277 
278 		p->p_addr->u_sigacts = *p->p_sigacts;
279 		p->p_sigacts = &p->p_addr->u_sigacts;
280 	}
281 
282 	/* Stop profiling */
283 	stopprofclock(p);
284 
285 	/* close files on exec */
286 	fdcloseexec(p);
287 
288 	/* reset caught signals */
289 	execsigs(p);
290 
291 	/* name this process - nameiexec(p, ndp) */
292 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
293 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
294 	p->p_comm[len] = 0;
295 
296 	/*
297 	 * mark as execed, wakeup the process that vforked (if any) and tell
298 	 * it that it now has its own resources back
299 	 */
300 	p->p_flag |= P_EXEC;
301 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
302 		p->p_flag &= ~P_PPWAIT;
303 		wakeup((caddr_t)p->p_pptr);
304 	}
305 
306 	/*
307 	 * Implement image setuid/setgid.
308 	 *
309 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
310 	 * the process is being traced.
311 	 */
312 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
313 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
314 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
315 	    (p->p_flag & P_TRACED) == 0) {
316 		/*
317 		 * Turn off syscall tracing for set-id programs, except for
318 		 * root.  Record any set-id flags first to make sure that
319 		 * we do not regain any tracing during a possible block.
320 		 */
321 		setsugid(p);
322 		if (p->p_tracep && suser(p)) {
323 			struct vnode *vtmp;
324 
325 			if ((vtmp = p->p_tracep) != NULL) {
326 				p->p_tracep = NULL;
327 				p->p_traceflag = 0;
328 				vrele(vtmp);
329 			}
330 		}
331 		/* Close any file descriptors 0..2 that reference procfs */
332 		setugidsafety(p);
333 		/* Make sure file descriptors 0..2 are in use. */
334 		error = fdcheckstd(p);
335 		if (error != 0)
336 			goto exec_fail_dealloc;
337 		/*
338 		 * Set the new credentials.
339 		 */
340 		p->p_ucred = crcopy(p->p_ucred);
341 		if (attr.va_mode & VSUID)
342 			change_euid(p, attr.va_uid);
343 		if (attr.va_mode & VSGID)
344 			p->p_ucred->cr_gid = attr.va_gid;
345 	} else {
346 		if (p->p_ucred->cr_uid == p->p_cred->p_ruid &&
347 		    p->p_ucred->cr_gid == p->p_cred->p_rgid)
348 			p->p_flag &= ~P_SUGID;
349 	}
350 
351 	/*
352 	 * Implement correct POSIX saved-id behavior.
353 	 */
354 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
355 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
356 
357 	/*
358 	 * Store the vp for use in procfs
359 	 */
360 	if (p->p_textvp)		/* release old reference */
361 		vrele(p->p_textvp);
362 	VREF(ndp->ni_vp);
363 	p->p_textvp = ndp->ni_vp;
364 
365         /*
366          * Notify others that we exec'd, and clear the P_INEXEC flag
367          * as we're now a bona fide freshly-execed process.
368          */
369 	KNOTE(&p->p_klist, NOTE_EXEC);
370 	p->p_flag &= ~P_INEXEC;
371 
372 	/*
373 	 * If tracing the process, trap to debugger so breakpoints
374 	 * 	can be set before the program executes.
375 	 */
376 	STOPEVENT(p, S_EXEC, 0);
377 
378 	if (p->p_flag & P_TRACED)
379 		psignal(p, SIGTRAP);
380 
381 	/* clear "fork but no exec" flag, as we _are_ execing */
382 	p->p_acflag &= ~AFORK;
383 
384 	/* Set values passed into the program in registers. */
385 	setregs(p, imgp->entry_addr, (u_long)(uintptr_t)stack_base,
386 	    imgp->ps_strings);
387 
388 	/* Free any previous argument cache */
389 	if (p->p_args && --p->p_args->ar_ref == 0)
390 		FREE(p->p_args, M_PARGS);
391 	p->p_args = NULL;
392 
393 	/* Cache arguments if they fit inside our allowance */
394 	i = imgp->endargs - imgp->stringbase;
395 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
396 		MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
397 		    M_PARGS, M_WAITOK);
398 		p->p_args->ar_ref = 1;
399 		p->p_args->ar_length = i;
400 		bcopy(imgp->stringbase, p->p_args->ar_args, i);
401 	}
402 
403 exec_fail_dealloc:
404 
405 	/*
406 	 * free various allocated resources
407 	 */
408 	if (imgp->firstpage)
409 		exec_unmap_first_page(imgp);
410 
411 	if (imgp->stringbase != NULL)
412 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
413 			ARG_MAX + PAGE_SIZE);
414 
415 	if (imgp->vp) {
416 		NDFREE(ndp, NDF_ONLY_PNBUF);
417 		vrele(imgp->vp);
418 	}
419 
420 	if (error == 0)
421 		return (0);
422 
423 exec_fail:
424 	/* we're done here, clear P_INEXEC */
425 	p->p_flag &= ~P_INEXEC;
426 	if (imgp->vmspace_destroyed) {
427 		/* sorry, no more process anymore. exit gracefully */
428 		exit1(p, W_EXITCODE(0, SIGABRT));
429 		/* NOT REACHED */
430 		return(0);
431 	} else {
432 		return(error);
433 	}
434 }
435 
436 int
437 exec_map_first_page(imgp)
438 	struct image_params *imgp;
439 {
440 	int s, rv, i;
441 	int initial_pagein;
442 	vm_page_t ma[VM_INITIAL_PAGEIN];
443 	vm_object_t object;
444 
445 
446 	if (imgp->firstpage) {
447 		exec_unmap_first_page(imgp);
448 	}
449 
450 	VOP_GETVOBJECT(imgp->vp, &object);
451 	s = splvm();
452 
453 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
454 
455 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
456 		initial_pagein = VM_INITIAL_PAGEIN;
457 		if (initial_pagein > object->size)
458 			initial_pagein = object->size;
459 		for (i = 1; i < initial_pagein; i++) {
460 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
461 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
462 					break;
463 				if (ma[i]->valid)
464 					break;
465 				vm_page_busy(ma[i]);
466 			} else {
467 				ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
468 				if (ma[i] == NULL)
469 					break;
470 			}
471 		}
472 		initial_pagein = i;
473 
474 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
475 		ma[0] = vm_page_lookup(object, 0);
476 
477 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || (ma[0]->valid == 0)) {
478 			if (ma[0]) {
479 				vm_page_protect(ma[0], VM_PROT_NONE);
480 				vm_page_free(ma[0]);
481 			}
482 			splx(s);
483 			return EIO;
484 		}
485 	}
486 
487 	vm_page_wire(ma[0]);
488 	vm_page_wakeup(ma[0]);
489 	splx(s);
490 
491 	pmap_kenter((vm_offset_t) imgp->image_header, VM_PAGE_TO_PHYS(ma[0]));
492 	imgp->firstpage = ma[0];
493 
494 	return 0;
495 }
496 
497 void
498 exec_unmap_first_page(imgp)
499 	struct image_params *imgp;
500 {
501 	if (imgp->firstpage) {
502 		pmap_kremove((vm_offset_t) imgp->image_header);
503 		vm_page_unwire(imgp->firstpage, 1);
504 		imgp->firstpage = NULL;
505 	}
506 }
507 
508 /*
509  * Destroy old address space, and allocate a new stack
510  *	The new stack is only SGROWSIZ large because it is grown
511  *	automatically in trap.c.
512  */
513 int
514 exec_new_vmspace(imgp)
515 	struct image_params *imgp;
516 {
517 	int error;
518 	struct vmspace *vmspace = imgp->proc->p_vmspace;
519 	vm_offset_t stack_addr = USRSTACK - maxssiz;
520 	vm_map_t map = &vmspace->vm_map;
521 
522 	imgp->vmspace_destroyed = 1;
523 
524 	/*
525 	 * Prevent a pending AIO from modifying the new address space.
526 	 */
527 	aio_proc_rundown(imgp->proc);
528 
529 	/*
530 	 * Blow away entire process VM, if address space not shared,
531 	 * otherwise, create a new VM space so that other threads are
532 	 * not disrupted
533 	 */
534 	if (vmspace->vm_refcnt == 1) {
535 		if (vmspace->vm_shm)
536 			shmexit(imgp->proc);
537 		pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS);
538 		vm_map_remove(map, 0, VM_MAXUSER_ADDRESS);
539 	} else {
540 		vmspace_exec(imgp->proc);
541 		vmspace = imgp->proc->p_vmspace;
542 		map = &vmspace->vm_map;
543 	}
544 
545 	/* Allocate a new stack */
546 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
547 	    VM_PROT_ALL, VM_PROT_ALL, 0);
548 	if (error)
549 		return (error);
550 
551 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
552 	 * VM_STACK case, but they are still used to monitor the size of the
553 	 * process stack so we can check the stack rlimit.
554 	 */
555 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
556 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
557 
558 	return(0);
559 }
560 
561 /*
562  * Copy out argument and environment strings from the old process
563  *	address space into the temporary string buffer.
564  */
565 int
566 exec_extract_strings(imgp)
567 	struct image_params *imgp;
568 {
569 	char	**argv, **envv;
570 	char	*argp, *envp;
571 	int	error;
572 	size_t	length;
573 
574 	/*
575 	 * extract arguments first
576 	 */
577 
578 	argv = imgp->uap->argv;
579 
580 	if (argv) {
581 		argp = (caddr_t) (intptr_t) fuword(argv);
582 		if (argp == (caddr_t) -1)
583 			return (EFAULT);
584 		if (argp)
585 			argv++;
586 		if (imgp->argv0)
587 			argp = imgp->argv0;
588 		if (argp) {
589 			do {
590 				if (argp == (caddr_t) -1)
591 					return (EFAULT);
592 				if ((error = copyinstr(argp, imgp->stringp,
593 				    imgp->stringspace, &length))) {
594 					if (error == ENAMETOOLONG)
595 						return(E2BIG);
596 					return (error);
597 				}
598 				imgp->stringspace -= length;
599 				imgp->stringp += length;
600 				imgp->argc++;
601 			} while ((argp = (caddr_t) (intptr_t) fuword(argv++)));
602 		}
603 	}
604 
605 	imgp->endargs = imgp->stringp;
606 
607 	/*
608 	 * extract environment strings
609 	 */
610 
611 	envv = imgp->uap->envv;
612 
613 	if (envv) {
614 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
615 			if (envp == (caddr_t) -1)
616 				return (EFAULT);
617 			if ((error = copyinstr(envp, imgp->stringp,
618 			    imgp->stringspace, &length))) {
619 				if (error == ENAMETOOLONG)
620 					return(E2BIG);
621 				return (error);
622 			}
623 			imgp->stringspace -= length;
624 			imgp->stringp += length;
625 			imgp->envc++;
626 		}
627 	}
628 
629 	return (0);
630 }
631 
632 /*
633  * Copy strings out to the new process address space, constructing
634  *	new arg and env vector tables. Return a pointer to the base
635  *	so that it can be used as the initial stack pointer.
636  */
637 register_t *
638 exec_copyout_strings(imgp)
639 	struct image_params *imgp;
640 {
641 	int argc, envc;
642 	char **vectp;
643 	char *stringp, *destp;
644 	register_t *stack_base;
645 	struct ps_strings *arginfo;
646 	int szsigcode;
647 
648 	/*
649 	 * Calculate string base and vector table pointers.
650 	 * Also deal with signal trampoline code for this exec type.
651 	 */
652 	arginfo = (struct ps_strings *)PS_STRINGS;
653 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
654 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
655 		roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
656 
657 	/*
658 	 * install sigcode
659 	 */
660 	if (szsigcode)
661 		copyout(imgp->proc->p_sysent->sv_sigcode,
662 			((caddr_t)arginfo - szsigcode), szsigcode);
663 
664 	/*
665 	 * If we have a valid auxargs ptr, prepare some room
666 	 * on the stack.
667 	 */
668 	if (imgp->auxargs)
669 	/*
670 	 * The '+ 2' is for the null pointers at the end of each of the
671 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
672 	 * ELF Auxargs data.
673 	 */
674 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
675 				  AT_COUNT*2) * sizeof(char*));
676 	else
677 	/*
678 	 * The '+ 2' is for the null pointers at the end of each of the
679 	 * arg and env vector sets
680 	 */
681 		vectp = (char **)
682 			(destp - (imgp->argc + imgp->envc + 2) * sizeof(char*));
683 
684 	/*
685 	 * vectp also becomes our initial stack base
686 	 */
687 	stack_base = (register_t *)vectp;
688 
689 	stringp = imgp->stringbase;
690 	argc = imgp->argc;
691 	envc = imgp->envc;
692 
693 	/*
694 	 * Copy out strings - arguments and environment.
695 	 */
696 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
697 
698 	/*
699 	 * Fill in "ps_strings" struct for ps, w, etc.
700 	 */
701 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
702 	suword(&arginfo->ps_nargvstr, argc);
703 
704 	/*
705 	 * Fill in argument portion of vector table.
706 	 */
707 	for (; argc > 0; --argc) {
708 		suword(vectp++, (long)(intptr_t)destp);
709 		while (*stringp++ != 0)
710 			destp++;
711 		destp++;
712 	}
713 
714 	/* a null vector table pointer seperates the argp's from the envp's */
715 	suword(vectp++, 0);
716 
717 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
718 	suword(&arginfo->ps_nenvstr, envc);
719 
720 	/*
721 	 * Fill in environment portion of vector table.
722 	 */
723 	for (; envc > 0; --envc) {
724 		suword(vectp++, (long)(intptr_t)destp);
725 		while (*stringp++ != 0)
726 			destp++;
727 		destp++;
728 	}
729 
730 	/* end of vector table is a null pointer */
731 	suword(vectp, 0);
732 
733 	return (stack_base);
734 }
735 
736 /*
737  * Check permissions of file to execute.
738  *	Return 0 for success or error code on failure.
739  */
740 int
741 exec_check_permissions(imgp)
742 	struct image_params *imgp;
743 {
744 	struct proc *p = imgp->proc;
745 	struct vnode *vp = imgp->vp;
746 	struct vattr *attr = imgp->attr;
747 	int error;
748 
749 	/* Get file attributes */
750 	error = VOP_GETATTR(vp, attr, p->p_ucred, p);
751 	if (error)
752 		return (error);
753 
754 	/*
755 	 * 1) Check if file execution is disabled for the filesystem that this
756 	 *	file resides on.
757 	 * 2) Insure that at least one execute bit is on - otherwise root
758 	 *	will always succeed, and we don't want to happen unless the
759 	 *	file really is executable.
760 	 * 3) Insure that the file is a regular file.
761 	 */
762 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
763 	    ((attr->va_mode & 0111) == 0) ||
764 	    (attr->va_type != VREG)) {
765 		return (EACCES);
766 	}
767 
768 	/*
769 	 * Zero length files can't be exec'd
770 	 */
771 	if (attr->va_size == 0)
772 		return (ENOEXEC);
773 
774 	/*
775 	 *  Check for execute permission to file based on current credentials.
776 	 */
777 	error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p);
778 	if (error)
779 		return (error);
780 
781 	/*
782 	 * Check number of open-for-writes on the file and deny execution
783 	 * if there are any.
784 	 */
785 	if (vp->v_writecount)
786 		return (ETXTBSY);
787 
788 	/*
789 	 * Call filesystem specific open routine (which does nothing in the
790 	 * general case).
791 	 */
792 	error = VOP_OPEN(vp, FREAD, p->p_ucred, p);
793 	if (error)
794 		return (error);
795 
796 	return (0);
797 }
798 
799 /*
800  * Exec handler registration
801  */
802 int
803 exec_register(execsw_arg)
804 	const struct execsw *execsw_arg;
805 {
806 	const struct execsw **es, **xs, **newexecsw;
807 	int count = 2;	/* New slot and trailing NULL */
808 
809 	if (execsw)
810 		for (es = execsw; *es; es++)
811 			count++;
812 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
813 	if (newexecsw == NULL)
814 		return ENOMEM;
815 	xs = newexecsw;
816 	if (execsw)
817 		for (es = execsw; *es; es++)
818 			*xs++ = *es;
819 	*xs++ = execsw_arg;
820 	*xs = NULL;
821 	if (execsw)
822 		free(execsw, M_TEMP);
823 	execsw = newexecsw;
824 	return 0;
825 }
826 
827 int
828 exec_unregister(execsw_arg)
829 	const struct execsw *execsw_arg;
830 {
831 	const struct execsw **es, **xs, **newexecsw;
832 	int count = 1;
833 
834 	if (execsw == NULL)
835 		panic("unregister with no handlers left?\n");
836 
837 	for (es = execsw; *es; es++) {
838 		if (*es == execsw_arg)
839 			break;
840 	}
841 	if (*es == NULL)
842 		return ENOENT;
843 	for (es = execsw; *es; es++)
844 		if (*es != execsw_arg)
845 			count++;
846 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
847 	if (newexecsw == NULL)
848 		return ENOMEM;
849 	xs = newexecsw;
850 	for (es = execsw; *es; es++)
851 		if (*es != execsw_arg)
852 			*xs++ = *es;
853 	*xs = NULL;
854 	if (execsw)
855 		free(execsw, M_TEMP);
856 	execsw = newexecsw;
857 	return 0;
858 }
859