xref: /dragonfly/sys/kern/kern_exec.c (revision 2d8a3be7)
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  * $DragonFly: src/sys/kern/kern_exec.c,v 1.12 2003/09/23 05:03:51 dillon Exp $
28  */
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/sysproto.h>
33 #include <sys/kernel.h>
34 #include <sys/mount.h>
35 #include <sys/filedesc.h>
36 #include <sys/fcntl.h>
37 #include <sys/acct.h>
38 #include <sys/exec.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/wait.h>
42 #include <sys/malloc.h>
43 #include <sys/proc.h>
44 #include <sys/signalvar.h>
45 #include <sys/pioctl.h>
46 #include <sys/namei.h>
47 #include <sys/sysent.h>
48 #include <sys/shm.h>
49 #include <sys/sysctl.h>
50 #include <sys/vnode.h>
51 #include <sys/aio.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_param.h>
55 #include <sys/lock.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_kern.h>
60 #include <vm/vm_extern.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_pager.h>
63 
64 #include <sys/user.h>
65 #include <machine/reg.h>
66 
67 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
68 
69 static register_t *exec_copyout_strings (struct image_params *);
70 
71 /* XXX This should be vm_size_t. */
72 static u_long ps_strings = PS_STRINGS;
73 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
74 
75 /* XXX This should be vm_size_t. */
76 static u_long usrstack = USRSTACK;
77 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
78 
79 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
80 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
81     &ps_arg_cache_limit, 0, "");
82 
83 int ps_argsopen = 1;
84 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
85 
86 /*
87  * Each of the items is a pointer to a `const struct execsw', hence the
88  * double pointer here.
89  */
90 static const struct execsw **execsw;
91 
92 /*
93  * execve() system call.
94  */
95 int
96 execve(struct execve_args *uap)
97 {
98 	struct thread *td = curthread;
99 	struct proc *p = td->td_proc;
100 	struct nameidata nd, *ndp;
101 	register_t *stack_base;
102 	int error, len, i;
103 	struct image_params image_params, *imgp;
104 	struct vattr attr;
105 	int (*img_first) (struct image_params *);
106 
107 	KKASSERT(p);
108 	imgp = &image_params;
109 
110 	/*
111 	 * Lock the process and set the P_INEXEC flag to indicate that
112 	 * it should be left alone until we're done here.  This is
113 	 * necessary to avoid race conditions - e.g. in ptrace() -
114 	 * that might allow a local user to illicitly obtain elevated
115 	 * privileges.
116 	 */
117 	p->p_flag |= P_INEXEC;
118 
119 	/*
120 	 * Initialize part of the common data
121 	 */
122 	imgp->proc = p;
123 	imgp->uap = uap;
124 	imgp->attr = &attr;
125 	imgp->argc = imgp->envc = 0;
126 	imgp->argv0 = NULL;
127 	imgp->entry_addr = 0;
128 	imgp->vmspace_destroyed = 0;
129 	imgp->interpreted = 0;
130 	imgp->interpreter_name[0] = '\0';
131 	imgp->auxargs = NULL;
132 	imgp->vp = NULL;
133 	imgp->firstpage = NULL;
134 	imgp->ps_strings = 0;
135 
136 	/*
137 	 * Allocate temporary demand zeroed space for argument and
138 	 *	environment strings
139 	 */
140 	imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX + PAGE_SIZE);
141 	if (imgp->stringbase == NULL) {
142 		error = ENOMEM;
143 		goto exec_fail;
144 	}
145 	imgp->stringp = imgp->stringbase;
146 	imgp->stringspace = ARG_MAX;
147 	imgp->image_header = imgp->stringbase + ARG_MAX;
148 
149 	/*
150 	 * Translate the file name. namei() returns a vnode pointer
151 	 *	in ni_vp amoung other things.
152 	 */
153 	ndp = &nd;
154 	NDINIT(ndp, NAMEI_LOOKUP, CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
155 	    UIO_USERSPACE, uap->fname, td);
156 
157 interpret:
158 
159 	error = namei(ndp);
160 	if (error) {
161 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
162 			ARG_MAX + PAGE_SIZE);
163 		goto exec_fail;
164 	}
165 
166 	imgp->vp = ndp->ni_vp;
167 	imgp->fname = uap->fname;
168 
169 	/*
170 	 * Check file permissions (also 'opens' file)
171 	 */
172 	error = exec_check_permissions(imgp);
173 	if (error) {
174 		VOP_UNLOCK(imgp->vp, 0, td);
175 		goto exec_fail_dealloc;
176 	}
177 
178 	error = exec_map_first_page(imgp);
179 	VOP_UNLOCK(imgp->vp, 0, td);
180 	if (error)
181 		goto exec_fail_dealloc;
182 
183 	/*
184 	 *	If the current process has a special image activator it
185 	 *	wants to try first, call it.   For example, emulating shell
186 	 *	scripts differently.
187 	 */
188 	error = -1;
189 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
190 		error = img_first(imgp);
191 
192 	/*
193 	 *	Loop through the list of image activators, calling each one.
194 	 *	An activator returns -1 if there is no match, 0 on success,
195 	 *	and an error otherwise.
196 	 */
197 	for (i = 0; error == -1 && execsw[i]; ++i) {
198 		if (execsw[i]->ex_imgact == NULL ||
199 		    execsw[i]->ex_imgact == img_first) {
200 			continue;
201 		}
202 		error = (*execsw[i]->ex_imgact)(imgp);
203 	}
204 
205 	if (error) {
206 		if (error == -1)
207 			error = ENOEXEC;
208 		goto exec_fail_dealloc;
209 	}
210 
211 	/*
212 	 * Special interpreter operation, cleanup and loop up to try to
213 	 * activate the interpreter.
214 	 */
215 	if (imgp->interpreted) {
216 		exec_unmap_first_page(imgp);
217 		/* free name buffer and old vnode */
218 		NDFREE(ndp, NDF_ONLY_PNBUF);
219 		vrele(ndp->ni_vp);
220 		/* set new name to that of the interpreter */
221 		NDINIT(ndp, NAMEI_LOOKUP,
222 		    CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
223 		    UIO_SYSSPACE, imgp->interpreter_name, td);
224 		goto interpret;
225 	}
226 
227 	/*
228 	 * Copy out strings (args and env) and initialize stack base
229 	 */
230 	stack_base = exec_copyout_strings(imgp);
231 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
232 
233 	/*
234 	 * If custom stack fixup routine present for this process
235 	 * let it do the stack setup.
236 	 * Else stuff argument count as first item on stack
237 	 */
238 	if (p->p_sysent->sv_fixup)
239 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
240 	else
241 		suword(--stack_base, imgp->argc);
242 
243 	/*
244 	 * For security and other reasons, the file descriptor table cannot
245 	 * be shared after an exec.
246 	 */
247 	if (p->p_fd->fd_refcnt > 1) {
248 		struct filedesc *tmp;
249 
250 		tmp = fdcopy(p);
251 		fdfree(p);
252 		p->p_fd = tmp;
253 	}
254 
255 	/*
256 	 * For security and other reasons, signal handlers cannot
257 	 * be shared after an exec. The new proces gets a copy of the old
258 	 * handlers. In execsigs(), the new process will have its signals
259 	 * reset.
260 	 */
261 	if (p->p_procsig->ps_refcnt > 1) {
262 		struct procsig *newprocsig;
263 
264 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
265 		       M_SUBPROC, M_WAITOK);
266 		bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig));
267 		p->p_procsig->ps_refcnt--;
268 		p->p_procsig = newprocsig;
269 		p->p_procsig->ps_refcnt = 1;
270 		if (p->p_sigacts == &p->p_addr->u_sigacts)
271 			panic("shared procsig but private sigacts?");
272 
273 		p->p_addr->u_sigacts = *p->p_sigacts;
274 		p->p_sigacts = &p->p_addr->u_sigacts;
275 	}
276 
277 	/* Stop profiling */
278 	stopprofclock(p);
279 
280 	/* close files on exec */
281 	fdcloseexec(p);
282 
283 	/* reset caught signals */
284 	execsigs(p);
285 
286 	/* name this process - nameiexec(p, ndp) */
287 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
288 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
289 	p->p_comm[len] = 0;
290 
291 	/*
292 	 * mark as execed, wakeup the process that vforked (if any) and tell
293 	 * it that it now has its own resources back
294 	 */
295 	p->p_flag |= P_EXEC;
296 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
297 		p->p_flag &= ~P_PPWAIT;
298 		wakeup((caddr_t)p->p_pptr);
299 	}
300 
301 	/*
302 	 * Implement image setuid/setgid.
303 	 *
304 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
305 	 * the process is being traced.
306 	 */
307 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
308 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
309 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
310 	    (p->p_flag & P_TRACED) == 0) {
311 		/*
312 		 * Turn off syscall tracing for set-id programs, except for
313 		 * root.  Record any set-id flags first to make sure that
314 		 * we do not regain any tracing during a possible block.
315 		 */
316 		setsugid();
317 		if (p->p_tracep && suser(td)) {
318 			struct vnode *vtmp;
319 
320 			if ((vtmp = p->p_tracep) != NULL) {
321 				p->p_tracep = NULL;
322 				p->p_traceflag = 0;
323 				vrele(vtmp);
324 			}
325 		}
326 		/* Close any file descriptors 0..2 that reference procfs */
327 		setugidsafety(p);
328 		/* Make sure file descriptors 0..2 are in use. */
329 		error = fdcheckstd(p);
330 		if (error != 0)
331 			goto exec_fail_dealloc;
332 		/*
333 		 * Set the new credentials.
334 		 */
335 		cratom(&p->p_ucred);
336 		if (attr.va_mode & VSUID)
337 			change_euid(attr.va_uid);
338 		if (attr.va_mode & VSGID)
339 			p->p_ucred->cr_gid = attr.va_gid;
340 	} else {
341 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
342 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
343 			p->p_flag &= ~P_SUGID;
344 	}
345 
346 	/*
347 	 * Implement correct POSIX saved-id behavior.
348 	 */
349 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
350 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
351 		cratom(&p->p_ucred);
352 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
353 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
354 	}
355 
356 	/*
357 	 * Store the vp for use in procfs
358 	 */
359 	if (p->p_textvp)		/* release old reference */
360 		vrele(p->p_textvp);
361 	VREF(ndp->ni_vp);
362 	p->p_textvp = ndp->ni_vp;
363 
364         /*
365          * Notify others that we exec'd, and clear the P_INEXEC flag
366          * as we're now a bona fide freshly-execed process.
367          */
368 	KNOTE(&p->p_klist, NOTE_EXEC);
369 	p->p_flag &= ~P_INEXEC;
370 
371 	/*
372 	 * If tracing the process, trap to debugger so breakpoints
373 	 * 	can be set before the program executes.
374 	 */
375 	STOPEVENT(p, S_EXEC, 0);
376 
377 	if (p->p_flag & P_TRACED)
378 		psignal(p, SIGTRAP);
379 
380 	/* clear "fork but no exec" flag, as we _are_ execing */
381 	p->p_acflag &= ~AFORK;
382 
383 	/* Set values passed into the program in registers. */
384 	setregs(p, imgp->entry_addr, (u_long)(uintptr_t)stack_base,
385 	    imgp->ps_strings);
386 
387 	/*
388 	 * The syscall result is returned in registers to the new program.
389 	 * Linux will register %edx as an atexit function and we must be
390 	 * sure to set it to 0.  XXX
391 	 */
392 	uap->sysmsg_result64 = 0;
393 
394 	/* Free any previous argument cache */
395 	if (p->p_args && --p->p_args->ar_ref == 0)
396 		FREE(p->p_args, M_PARGS);
397 	p->p_args = NULL;
398 
399 	/* Cache arguments if they fit inside our allowance */
400 	i = imgp->endargs - imgp->stringbase;
401 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
402 		MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
403 		    M_PARGS, M_WAITOK);
404 		p->p_args->ar_ref = 1;
405 		p->p_args->ar_length = i;
406 		bcopy(imgp->stringbase, p->p_args->ar_args, i);
407 	}
408 
409 exec_fail_dealloc:
410 
411 	/*
412 	 * free various allocated resources
413 	 */
414 	if (imgp->firstpage)
415 		exec_unmap_first_page(imgp);
416 
417 	if (imgp->stringbase != NULL)
418 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
419 			ARG_MAX + PAGE_SIZE);
420 
421 	if (imgp->vp) {
422 		NDFREE(ndp, NDF_ONLY_PNBUF);
423 		vrele(imgp->vp);
424 	}
425 
426 	if (error == 0)
427 		return (0);
428 
429 exec_fail:
430 	/* we're done here, clear P_INEXEC */
431 	p->p_flag &= ~P_INEXEC;
432 	if (imgp->vmspace_destroyed) {
433 		/* sorry, no more process anymore. exit gracefully */
434 		exit1(W_EXITCODE(0, SIGABRT));
435 		/* NOT REACHED */
436 		return(0);
437 	} else {
438 		return(error);
439 	}
440 }
441 
442 int
443 exec_map_first_page(struct image_params *imgp)
444 {
445 	int s, rv, i;
446 	int initial_pagein;
447 	vm_page_t ma[VM_INITIAL_PAGEIN];
448 	vm_object_t object;
449 
450 
451 	if (imgp->firstpage) {
452 		exec_unmap_first_page(imgp);
453 	}
454 
455 	VOP_GETVOBJECT(imgp->vp, &object);
456 	s = splvm();
457 
458 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
459 
460 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
461 		initial_pagein = VM_INITIAL_PAGEIN;
462 		if (initial_pagein > object->size)
463 			initial_pagein = object->size;
464 		for (i = 1; i < initial_pagein; i++) {
465 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
466 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
467 					break;
468 				if (ma[i]->valid)
469 					break;
470 				vm_page_busy(ma[i]);
471 			} else {
472 				ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
473 				if (ma[i] == NULL)
474 					break;
475 			}
476 		}
477 		initial_pagein = i;
478 
479 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
480 		ma[0] = vm_page_lookup(object, 0);
481 
482 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || (ma[0]->valid == 0)) {
483 			if (ma[0]) {
484 				vm_page_protect(ma[0], VM_PROT_NONE);
485 				vm_page_free(ma[0]);
486 			}
487 			splx(s);
488 			return EIO;
489 		}
490 	}
491 
492 	vm_page_wire(ma[0]);
493 	vm_page_wakeup(ma[0]);
494 	splx(s);
495 
496 	pmap_kenter((vm_offset_t) imgp->image_header, VM_PAGE_TO_PHYS(ma[0]));
497 	imgp->firstpage = ma[0];
498 
499 	return 0;
500 }
501 
502 void
503 exec_unmap_first_page(imgp)
504 	struct image_params *imgp;
505 {
506 	if (imgp->firstpage) {
507 		pmap_kremove((vm_offset_t) imgp->image_header);
508 		vm_page_unwire(imgp->firstpage, 1);
509 		imgp->firstpage = NULL;
510 	}
511 }
512 
513 /*
514  * Destroy old address space, and allocate a new stack
515  *	The new stack is only SGROWSIZ large because it is grown
516  *	automatically in trap.c.
517  */
518 int
519 exec_new_vmspace(imgp)
520 	struct image_params *imgp;
521 {
522 	int error;
523 	struct vmspace *vmspace = imgp->proc->p_vmspace;
524 	vm_offset_t stack_addr = USRSTACK - maxssiz;
525 	vm_map_t map = &vmspace->vm_map;
526 
527 	imgp->vmspace_destroyed = 1;
528 
529 	/*
530 	 * Prevent a pending AIO from modifying the new address space.
531 	 */
532 	aio_proc_rundown(imgp->proc);
533 
534 	/*
535 	 * Blow away entire process VM, if address space not shared,
536 	 * otherwise, create a new VM space so that other threads are
537 	 * not disrupted
538 	 */
539 	if (vmspace->vm_refcnt == 1) {
540 		shmexit(vmspace);
541 		pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS);
542 		vm_map_remove(map, 0, VM_MAXUSER_ADDRESS);
543 	} else {
544 		vmspace_exec(imgp->proc);
545 		vmspace = imgp->proc->p_vmspace;
546 		map = &vmspace->vm_map;
547 	}
548 
549 	/* Allocate a new stack */
550 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
551 	    VM_PROT_ALL, VM_PROT_ALL, 0);
552 	if (error)
553 		return (error);
554 
555 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
556 	 * VM_STACK case, but they are still used to monitor the size of the
557 	 * process stack so we can check the stack rlimit.
558 	 */
559 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
560 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
561 
562 	return(0);
563 }
564 
565 /*
566  * Copy out argument and environment strings from the old process
567  *	address space into the temporary string buffer.
568  */
569 int
570 exec_extract_strings(imgp)
571 	struct image_params *imgp;
572 {
573 	char	**argv, **envv;
574 	char	*argp, *envp;
575 	int	error;
576 	size_t	length;
577 
578 	/*
579 	 * extract arguments first
580 	 */
581 
582 	argv = imgp->uap->argv;
583 
584 	if (argv) {
585 		argp = (caddr_t) (intptr_t) fuword(argv);
586 		if (argp == (caddr_t) -1)
587 			return (EFAULT);
588 		if (argp)
589 			argv++;
590 		if (imgp->argv0)
591 			argp = imgp->argv0;
592 		if (argp) {
593 			do {
594 				if (argp == (caddr_t) -1)
595 					return (EFAULT);
596 				if ((error = copyinstr(argp, imgp->stringp,
597 				    imgp->stringspace, &length))) {
598 					if (error == ENAMETOOLONG)
599 						return(E2BIG);
600 					return (error);
601 				}
602 				imgp->stringspace -= length;
603 				imgp->stringp += length;
604 				imgp->argc++;
605 			} while ((argp = (caddr_t) (intptr_t) fuword(argv++)));
606 		}
607 	}
608 
609 	imgp->endargs = imgp->stringp;
610 
611 	/*
612 	 * extract environment strings
613 	 */
614 
615 	envv = imgp->uap->envv;
616 
617 	if (envv) {
618 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
619 			if (envp == (caddr_t) -1)
620 				return (EFAULT);
621 			if ((error = copyinstr(envp, imgp->stringp,
622 			    imgp->stringspace, &length))) {
623 				if (error == ENAMETOOLONG)
624 					return(E2BIG);
625 				return (error);
626 			}
627 			imgp->stringspace -= length;
628 			imgp->stringp += length;
629 			imgp->envc++;
630 		}
631 	}
632 
633 	return (0);
634 }
635 
636 /*
637  * Copy strings out to the new process address space, constructing
638  *	new arg and env vector tables. Return a pointer to the base
639  *	so that it can be used as the initial stack pointer.
640  */
641 register_t *
642 exec_copyout_strings(imgp)
643 	struct image_params *imgp;
644 {
645 	int argc, envc;
646 	char **vectp;
647 	char *stringp, *destp;
648 	register_t *stack_base;
649 	struct ps_strings *arginfo;
650 	int szsigcode;
651 
652 	/*
653 	 * Calculate string base and vector table pointers.
654 	 * Also deal with signal trampoline code for this exec type.
655 	 */
656 	arginfo = (struct ps_strings *)PS_STRINGS;
657 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
658 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
659 		roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
660 
661 	/*
662 	 * install sigcode
663 	 */
664 	if (szsigcode)
665 		copyout(imgp->proc->p_sysent->sv_sigcode,
666 			((caddr_t)arginfo - szsigcode), szsigcode);
667 
668 	/*
669 	 * If we have a valid auxargs ptr, prepare some room
670 	 * on the stack.
671 	 */
672 	if (imgp->auxargs)
673 	/*
674 	 * The '+ 2' is for the null pointers at the end of each of the
675 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
676 	 * ELF Auxargs data.
677 	 */
678 		vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
679 				  AT_COUNT*2) * sizeof(char*));
680 	else
681 	/*
682 	 * The '+ 2' is for the null pointers at the end of each of the
683 	 * arg and env vector sets
684 	 */
685 		vectp = (char **)
686 			(destp - (imgp->argc + imgp->envc + 2) * sizeof(char*));
687 
688 	/*
689 	 * vectp also becomes our initial stack base
690 	 */
691 	stack_base = (register_t *)vectp;
692 
693 	stringp = imgp->stringbase;
694 	argc = imgp->argc;
695 	envc = imgp->envc;
696 
697 	/*
698 	 * Copy out strings - arguments and environment.
699 	 */
700 	copyout(stringp, destp, ARG_MAX - imgp->stringspace);
701 
702 	/*
703 	 * Fill in "ps_strings" struct for ps, w, etc.
704 	 */
705 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
706 	suword(&arginfo->ps_nargvstr, argc);
707 
708 	/*
709 	 * Fill in argument portion of vector table.
710 	 */
711 	for (; argc > 0; --argc) {
712 		suword(vectp++, (long)(intptr_t)destp);
713 		while (*stringp++ != 0)
714 			destp++;
715 		destp++;
716 	}
717 
718 	/* a null vector table pointer seperates the argp's from the envp's */
719 	suword(vectp++, 0);
720 
721 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
722 	suword(&arginfo->ps_nenvstr, envc);
723 
724 	/*
725 	 * Fill in environment portion of vector table.
726 	 */
727 	for (; envc > 0; --envc) {
728 		suword(vectp++, (long)(intptr_t)destp);
729 		while (*stringp++ != 0)
730 			destp++;
731 		destp++;
732 	}
733 
734 	/* end of vector table is a null pointer */
735 	suword(vectp, 0);
736 
737 	return (stack_base);
738 }
739 
740 /*
741  * Check permissions of file to execute.
742  *	Return 0 for success or error code on failure.
743  */
744 int
745 exec_check_permissions(imgp)
746 	struct image_params *imgp;
747 {
748 	struct proc *p = imgp->proc;
749 	struct vnode *vp = imgp->vp;
750 	struct vattr *attr = imgp->attr;
751 	struct thread *td = p->p_thread;
752 	int error;
753 
754 	/* Get file attributes */
755 	error = VOP_GETATTR(vp, attr, td);
756 	if (error)
757 		return (error);
758 
759 	/*
760 	 * 1) Check if file execution is disabled for the filesystem that this
761 	 *	file resides on.
762 	 * 2) Insure that at least one execute bit is on - otherwise root
763 	 *	will always succeed, and we don't want to happen unless the
764 	 *	file really is executable.
765 	 * 3) Insure that the file is a regular file.
766 	 */
767 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
768 	    ((attr->va_mode & 0111) == 0) ||
769 	    (attr->va_type != VREG)) {
770 		return (EACCES);
771 	}
772 
773 	/*
774 	 * Zero length files can't be exec'd
775 	 */
776 	if (attr->va_size == 0)
777 		return (ENOEXEC);
778 
779 	/*
780 	 *  Check for execute permission to file based on current credentials.
781 	 */
782 	error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td);
783 	if (error)
784 		return (error);
785 
786 	/*
787 	 * Check number of open-for-writes on the file and deny execution
788 	 * if there are any.
789 	 */
790 	if (vp->v_writecount)
791 		return (ETXTBSY);
792 
793 	/*
794 	 * Call filesystem specific open routine (which does nothing in the
795 	 * general case).
796 	 */
797 	error = VOP_OPEN(vp, FREAD, p->p_ucred, td);
798 	if (error)
799 		return (error);
800 
801 	return (0);
802 }
803 
804 /*
805  * Exec handler registration
806  */
807 int
808 exec_register(execsw_arg)
809 	const struct execsw *execsw_arg;
810 {
811 	const struct execsw **es, **xs, **newexecsw;
812 	int count = 2;	/* New slot and trailing NULL */
813 
814 	if (execsw)
815 		for (es = execsw; *es; es++)
816 			count++;
817 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
818 	if (newexecsw == NULL)
819 		return ENOMEM;
820 	xs = newexecsw;
821 	if (execsw)
822 		for (es = execsw; *es; es++)
823 			*xs++ = *es;
824 	*xs++ = execsw_arg;
825 	*xs = NULL;
826 	if (execsw)
827 		free(execsw, M_TEMP);
828 	execsw = newexecsw;
829 	return 0;
830 }
831 
832 int
833 exec_unregister(execsw_arg)
834 	const struct execsw *execsw_arg;
835 {
836 	const struct execsw **es, **xs, **newexecsw;
837 	int count = 1;
838 
839 	if (execsw == NULL)
840 		panic("unregister with no handlers left?\n");
841 
842 	for (es = execsw; *es; es++) {
843 		if (*es == execsw_arg)
844 			break;
845 	}
846 	if (*es == NULL)
847 		return ENOENT;
848 	for (es = execsw; *es; es++)
849 		if (*es != execsw_arg)
850 			count++;
851 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
852 	if (newexecsw == NULL)
853 		return ENOMEM;
854 	xs = newexecsw;
855 	for (es = execsw; *es; es++)
856 		if (*es != execsw_arg)
857 			*xs++ = *es;
858 	*xs = NULL;
859 	if (execsw)
860 		free(execsw, M_TEMP);
861 	execsw = newexecsw;
862 	return 0;
863 }
864