xref: /dflybsd-src/sys/kern/kern_exec.c (revision 6b08710ee11d58fba7e8982f046f86f888c48dfa)
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 1.14 2003/11/12 01:00:33 daver 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/kern_syscall.h>
42 #include <sys/wait.h>
43 #include <sys/malloc.h>
44 #include <sys/proc.h>
45 #include <sys/signalvar.h>
46 #include <sys/pioctl.h>
47 #include <sys/namei.h>
48 #include <sys/sysent.h>
49 #include <sys/shm.h>
50 #include <sys/sysctl.h>
51 #include <sys/vnode.h>
52 #include <sys/aio.h>
53 
54 #include <vm/vm.h>
55 #include <vm/vm_param.h>
56 #include <sys/lock.h>
57 #include <vm/pmap.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_map.h>
60 #include <vm/vm_kern.h>
61 #include <vm/vm_extern.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_pager.h>
64 
65 #include <sys/user.h>
66 #include <machine/reg.h>
67 
68 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
69 
70 static register_t *exec_copyout_strings (struct image_params *);
71 
72 /* XXX This should be vm_size_t. */
73 static u_long ps_strings = PS_STRINGS;
74 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
75 
76 /* XXX This should be vm_size_t. */
77 static u_long usrstack = USRSTACK;
78 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
79 
80 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
81 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
82     &ps_arg_cache_limit, 0, "");
83 
84 int ps_argsopen = 1;
85 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
86 
87 void print_execve_args(struct image_args *args);
88 int debug_execve_args = 0;
89 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
90     0, "");
91 
92 void
93 print_execve_args(struct image_args *args)
94 {
95 	char *cp;
96 	int ndx;
97 
98 	cp = args->begin_argv;
99 	for (ndx = 0; ndx < args->argc; ndx++) {
100 		printf("\targv[%d]: %s\n", ndx, cp);
101 		while (*cp++ != '\0');
102 	}
103 	for (ndx = 0; ndx < args->envc; ndx++) {
104 		printf("\tenvv[%d]: %s\n", ndx, cp);
105 		while (*cp++ != '\0');
106 	}
107 }
108 
109 /*
110  * Each of the items is a pointer to a `const struct execsw', hence the
111  * double pointer here.
112  */
113 static const struct execsw **execsw;
114 
115 int
116 kern_execve(struct nameidata *ndp, struct image_args *args)
117 {
118 	struct thread *td = curthread;
119 	struct proc *p = td->td_proc;
120 	register_t *stack_base;
121 	int error, len, i;
122 	struct image_params image_params, *imgp;
123 	struct vattr attr;
124 	int (*img_first) (struct image_params *);
125 
126 	if (debug_execve_args) {
127 		printf("%s()\n", __func__);
128 		print_execve_args(args);
129 	}
130 
131 	KKASSERT(p);
132 	imgp = &image_params;
133 
134 	/*
135 	 * Lock the process and set the P_INEXEC flag to indicate that
136 	 * it should be left alone until we're done here.  This is
137 	 * necessary to avoid race conditions - e.g. in ptrace() -
138 	 * that might allow a local user to illicitly obtain elevated
139 	 * privileges.
140 	 */
141 	p->p_flag |= P_INEXEC;
142 
143 	/*
144 	 * Initialize part of the common data
145 	 */
146 	imgp->proc = p;
147 	imgp->args = args;
148 	imgp->attr = &attr;
149 	imgp->entry_addr = 0;
150 	imgp->vmspace_destroyed = 0;
151 	imgp->interpreted = 0;
152 	imgp->auxargs = NULL;
153 	imgp->vp = NULL;
154 	imgp->firstpage = NULL;
155 	imgp->ps_strings = 0;
156 
157 	/*
158 	 * Allocate temporary demand zeroed space for argument and
159 	 *	environment strings
160 	 */
161 	imgp->image_header = (char *)kmem_alloc_wait(exec_map, PAGE_SIZE);
162 	if (imgp->image_header == NULL) {
163 		error = ENOMEM;
164 		goto exec_fail;
165 	}
166 
167 interpret:
168 
169 	/*
170 	 * Translate the file name. namei() returns a vnode pointer
171 	 *	in ni_vp amoung other things.
172 	 */
173 	error = namei(ndp);
174 	if (error) {
175 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
176 		    PAGE_SIZE);
177 		goto exec_fail;
178 	}
179 
180 	imgp->vp = ndp->ni_vp;
181 
182 	/*
183 	 * Check file permissions (also 'opens' file)
184 	 */
185 	error = exec_check_permissions(imgp);
186 	if (error) {
187 		VOP_UNLOCK(imgp->vp, 0, td);
188 		goto exec_fail_dealloc;
189 	}
190 
191 	error = exec_map_first_page(imgp);
192 	VOP_UNLOCK(imgp->vp, 0, td);
193 	if (error)
194 		goto exec_fail_dealloc;
195 
196 	if (debug_execve_args && imgp->interpreted) {
197 		printf("    target is interpreted -- recursive pass\n");
198 		printf("    interpreter: %s\n", imgp->interpreter_name);
199 		print_execve_args(args);
200 	}
201 
202 	/*
203 	 *	If the current process has a special image activator it
204 	 *	wants to try first, call it.   For example, emulating shell
205 	 *	scripts differently.
206 	 */
207 	error = -1;
208 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
209 		error = img_first(imgp);
210 
211 	/*
212 	 *	Loop through the list of image activators, calling each one.
213 	 *	An activator returns -1 if there is no match, 0 on success,
214 	 *	and an error otherwise.
215 	 */
216 	for (i = 0; error == -1 && execsw[i]; ++i) {
217 		if (execsw[i]->ex_imgact == NULL ||
218 		    execsw[i]->ex_imgact == img_first) {
219 			continue;
220 		}
221 		error = (*execsw[i]->ex_imgact)(imgp);
222 	}
223 
224 	if (error) {
225 		if (error == -1)
226 			error = ENOEXEC;
227 		goto exec_fail_dealloc;
228 	}
229 
230 	/*
231 	 * Special interpreter operation, cleanup and loop up to try to
232 	 * activate the interpreter.
233 	 */
234 	if (imgp->interpreted) {
235 		exec_unmap_first_page(imgp);
236 		/* free name buffer and old vnode */
237 		NDFREE(ndp, NDF_ONLY_PNBUF);
238 		vrele(ndp->ni_vp);
239 		/* set new name to that of the interpreter */
240 		NDINIT(ndp, NAMEI_LOOKUP,
241 		    CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
242 		    UIO_SYSSPACE, imgp->interpreter_name, td);
243 		goto interpret;
244 	}
245 
246 	/*
247 	 * Copy out strings (args and env) and initialize stack base
248 	 */
249 	stack_base = exec_copyout_strings(imgp);
250 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
251 
252 	/*
253 	 * If custom stack fixup routine present for this process
254 	 * let it do the stack setup.
255 	 * Else stuff argument count as first item on stack
256 	 */
257 	if (p->p_sysent->sv_fixup)
258 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
259 	else
260 		suword(--stack_base, imgp->args->argc);
261 
262 	/*
263 	 * For security and other reasons, the file descriptor table cannot
264 	 * be shared after an exec.
265 	 */
266 	if (p->p_fd->fd_refcnt > 1) {
267 		struct filedesc *tmp;
268 
269 		tmp = fdcopy(p);
270 		fdfree(p);
271 		p->p_fd = tmp;
272 	}
273 
274 	/*
275 	 * For security and other reasons, signal handlers cannot
276 	 * be shared after an exec. The new proces gets a copy of the old
277 	 * handlers. In execsigs(), the new process will have its signals
278 	 * reset.
279 	 */
280 	if (p->p_procsig->ps_refcnt > 1) {
281 		struct procsig *newprocsig;
282 
283 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
284 		       M_SUBPROC, M_WAITOK);
285 		bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig));
286 		p->p_procsig->ps_refcnt--;
287 		p->p_procsig = newprocsig;
288 		p->p_procsig->ps_refcnt = 1;
289 		if (p->p_sigacts == &p->p_addr->u_sigacts)
290 			panic("shared procsig but private sigacts?");
291 
292 		p->p_addr->u_sigacts = *p->p_sigacts;
293 		p->p_sigacts = &p->p_addr->u_sigacts;
294 	}
295 
296 	/* Stop profiling */
297 	stopprofclock(p);
298 
299 	/* close files on exec */
300 	fdcloseexec(p);
301 
302 	/* reset caught signals */
303 	execsigs(p);
304 
305 	/* name this process - nameiexec(p, ndp) */
306 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
307 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
308 	p->p_comm[len] = 0;
309 
310 	/*
311 	 * mark as execed, wakeup the process that vforked (if any) and tell
312 	 * it that it now has its own resources back
313 	 */
314 	p->p_flag |= P_EXEC;
315 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
316 		p->p_flag &= ~P_PPWAIT;
317 		wakeup((caddr_t)p->p_pptr);
318 	}
319 
320 	/*
321 	 * Implement image setuid/setgid.
322 	 *
323 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
324 	 * the process is being traced.
325 	 */
326 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
327 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
328 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
329 	    (p->p_flag & P_TRACED) == 0) {
330 		/*
331 		 * Turn off syscall tracing for set-id programs, except for
332 		 * root.  Record any set-id flags first to make sure that
333 		 * we do not regain any tracing during a possible block.
334 		 */
335 		setsugid();
336 		if (p->p_tracep && suser(td)) {
337 			struct vnode *vtmp;
338 
339 			if ((vtmp = p->p_tracep) != NULL) {
340 				p->p_tracep = NULL;
341 				p->p_traceflag = 0;
342 				vrele(vtmp);
343 			}
344 		}
345 		/* Close any file descriptors 0..2 that reference procfs */
346 		setugidsafety(p);
347 		/* Make sure file descriptors 0..2 are in use. */
348 		error = fdcheckstd(p);
349 		if (error != 0)
350 			goto exec_fail_dealloc;
351 		/*
352 		 * Set the new credentials.
353 		 */
354 		cratom(&p->p_ucred);
355 		if (attr.va_mode & VSUID)
356 			change_euid(attr.va_uid);
357 		if (attr.va_mode & VSGID)
358 			p->p_ucred->cr_gid = attr.va_gid;
359 
360 		/*
361 		 * Clear local varsym variables
362 		 */
363 		varsymset_clean(&p->p_varsymset);
364 	} else {
365 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
366 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
367 			p->p_flag &= ~P_SUGID;
368 	}
369 
370 	/*
371 	 * Implement correct POSIX saved-id behavior.
372 	 */
373 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
374 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
375 		cratom(&p->p_ucred);
376 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
377 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
378 	}
379 
380 	/*
381 	 * Store the vp for use in procfs
382 	 */
383 	if (p->p_textvp)		/* release old reference */
384 		vrele(p->p_textvp);
385 	VREF(ndp->ni_vp);
386 	p->p_textvp = ndp->ni_vp;
387 
388         /*
389          * Notify others that we exec'd, and clear the P_INEXEC flag
390          * as we're now a bona fide freshly-execed process.
391          */
392 	KNOTE(&p->p_klist, NOTE_EXEC);
393 	p->p_flag &= ~P_INEXEC;
394 
395 	/*
396 	 * If tracing the process, trap to debugger so breakpoints
397 	 * 	can be set before the program executes.
398 	 */
399 	STOPEVENT(p, S_EXEC, 0);
400 
401 	if (p->p_flag & P_TRACED)
402 		psignal(p, SIGTRAP);
403 
404 	/* clear "fork but no exec" flag, as we _are_ execing */
405 	p->p_acflag &= ~AFORK;
406 
407 	/* Set values passed into the program in registers. */
408 	setregs(p, imgp->entry_addr, (u_long)(uintptr_t)stack_base,
409 	    imgp->ps_strings);
410 
411 	/* Free any previous argument cache */
412 	if (p->p_args && --p->p_args->ar_ref == 0)
413 		FREE(p->p_args, M_PARGS);
414 	p->p_args = NULL;
415 
416 	/* Cache arguments if they fit inside our allowance */
417 	i = imgp->args->begin_envv - imgp->args->begin_argv;
418 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
419 		MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
420 		    M_PARGS, M_WAITOK);
421 		p->p_args->ar_ref = 1;
422 		p->p_args->ar_length = i;
423 		bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
424 	}
425 
426 exec_fail_dealloc:
427 
428 	/*
429 	 * free various allocated resources
430 	 */
431 	if (imgp->firstpage)
432 		exec_unmap_first_page(imgp);
433 
434 	if (imgp->image_header != NULL)
435 		kmem_free_wakeup(exec_map, (vm_offset_t)imgp->image_header,
436 		    PAGE_SIZE);
437 
438 	if (imgp->vp) {
439 		NDFREE(ndp, NDF_ONLY_PNBUF);
440 		vrele(imgp->vp);
441 	}
442 
443 	if (error == 0)
444 		return (0);
445 
446 exec_fail:
447 	/* we're done here, clear P_INEXEC */
448 	p->p_flag &= ~P_INEXEC;
449 	if (imgp->vmspace_destroyed) {
450 		/* sorry, no more process anymore. exit gracefully */
451 		exit1(W_EXITCODE(0, SIGABRT));
452 		/* NOT REACHED */
453 		return(0);
454 	} else {
455 		return(error);
456 	}
457 }
458 
459 /*
460  * execve() system call.
461  */
462 int
463 execve(struct execve_args *uap)
464 {
465 	struct thread *td = curthread;
466 	struct nameidata nd;
467 	struct image_args args;
468 	int error;
469 
470 	NDINIT(&nd, NAMEI_LOOKUP, CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
471 	    UIO_USERSPACE, uap->fname, td);
472 
473 	error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
474 	    uap->argv, uap->envv);
475 	if (error)
476 		return (error);
477 
478 	error = kern_execve(&nd, &args);
479 
480 	exec_free_args(&args);
481 
482 	/*
483 	 * The syscall result is returned in registers to the new program.
484 	 * Linux will register %edx as an atexit function and we must be
485 	 * sure to set it to 0.  XXX
486 	 */
487 	if (error == 0)
488 		uap->sysmsg_result64 = 0;
489 
490 	return (error);
491 }
492 
493 int
494 exec_map_first_page(struct image_params *imgp)
495 {
496 	int s, rv, i;
497 	int initial_pagein;
498 	vm_page_t ma[VM_INITIAL_PAGEIN];
499 	vm_object_t object;
500 
501 
502 	if (imgp->firstpage) {
503 		exec_unmap_first_page(imgp);
504 	}
505 
506 	VOP_GETVOBJECT(imgp->vp, &object);
507 	s = splvm();
508 
509 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
510 
511 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
512 		initial_pagein = VM_INITIAL_PAGEIN;
513 		if (initial_pagein > object->size)
514 			initial_pagein = object->size;
515 		for (i = 1; i < initial_pagein; i++) {
516 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
517 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
518 					break;
519 				if (ma[i]->valid)
520 					break;
521 				vm_page_busy(ma[i]);
522 			} else {
523 				ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
524 				if (ma[i] == NULL)
525 					break;
526 			}
527 		}
528 		initial_pagein = i;
529 
530 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
531 		ma[0] = vm_page_lookup(object, 0);
532 
533 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || (ma[0]->valid == 0)) {
534 			if (ma[0]) {
535 				vm_page_protect(ma[0], VM_PROT_NONE);
536 				vm_page_free(ma[0]);
537 			}
538 			splx(s);
539 			return EIO;
540 		}
541 	}
542 
543 	vm_page_wire(ma[0]);
544 	vm_page_wakeup(ma[0]);
545 	splx(s);
546 
547 	pmap_kenter((vm_offset_t) imgp->image_header, VM_PAGE_TO_PHYS(ma[0]));
548 	imgp->firstpage = ma[0];
549 
550 	return 0;
551 }
552 
553 void
554 exec_unmap_first_page(imgp)
555 	struct image_params *imgp;
556 {
557 	if (imgp->firstpage) {
558 		pmap_kremove((vm_offset_t) imgp->image_header);
559 		vm_page_unwire(imgp->firstpage, 1);
560 		imgp->firstpage = NULL;
561 	}
562 }
563 
564 /*
565  * Destroy old address space, and allocate a new stack
566  *	The new stack is only SGROWSIZ large because it is grown
567  *	automatically in trap.c.
568  */
569 int
570 exec_new_vmspace(imgp)
571 	struct image_params *imgp;
572 {
573 	int error;
574 	struct vmspace *vmspace = imgp->proc->p_vmspace;
575 	vm_offset_t stack_addr = USRSTACK - maxssiz;
576 	vm_map_t map = &vmspace->vm_map;
577 
578 	imgp->vmspace_destroyed = 1;
579 
580 	/*
581 	 * Prevent a pending AIO from modifying the new address space.
582 	 */
583 	aio_proc_rundown(imgp->proc);
584 
585 	/*
586 	 * Blow away entire process VM, if address space not shared,
587 	 * otherwise, create a new VM space so that other threads are
588 	 * not disrupted
589 	 */
590 	if (vmspace->vm_refcnt == 1) {
591 		shmexit(vmspace);
592 		pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS);
593 		vm_map_remove(map, 0, VM_MAXUSER_ADDRESS);
594 	} else {
595 		vmspace_exec(imgp->proc);
596 		vmspace = imgp->proc->p_vmspace;
597 		map = &vmspace->vm_map;
598 	}
599 
600 	/* Allocate a new stack */
601 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
602 	    VM_PROT_ALL, VM_PROT_ALL, 0);
603 	if (error)
604 		return (error);
605 
606 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
607 	 * VM_STACK case, but they are still used to monitor the size of the
608 	 * process stack so we can check the stack rlimit.
609 	 */
610 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
611 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
612 
613 	return(0);
614 }
615 
616 /*
617  * Copy out argument and environment strings from the old process
618  *	address space into the temporary string buffer.
619  */
620 int
621 exec_copyin_args(struct image_args *args, char *fname,
622     enum exec_path_segflg segflg, char **argv, char **envv)
623 {
624 	char	*argp, *envp;
625 	int	error = 0;
626 	size_t	length;
627 
628 	args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
629 	if (args->buf == NULL)
630 		return (ENOMEM);
631 	args->begin_argv = args->buf;
632 	args->endp = args->begin_argv;
633 	args->space = ARG_MAX;
634 	args->argc = 0;
635 	args->envc = 0;
636 
637 	args->fname = args->buf + ARG_MAX;
638 
639 	/*
640 	 * Copy the file name.
641 	 */
642 	if (segflg == PATH_SYSSPACE) {
643 		error = copystr(fname, args->fname, PATH_MAX, &length);
644 	} else if (segflg == PATH_USERSPACE) {
645 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
646 	}
647 
648 	/*
649 	 * extract argument strings
650 	 */
651 
652 	if (argv) {
653 		argp = (caddr_t) (intptr_t) fuword(argv);
654 		/*
655 		 * First argument is a special case.  If it is a null
656 		 * string, we must copy the file name into the string
657 		 * buffer as the first argument.  This guarantees that
658 		 * the interpreter knows what file to open in the case
659 		 * that we exec an interpreted file.
660 		 */
661 		while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
662 			if (argp == (caddr_t) -1) {
663 				error = EFAULT;
664 				goto cleanup;
665 			}
666 			error = copyinstr(argp, args->endp,
667 			    args->space, &length);
668 			if (error == ENAMETOOLONG)
669 				error = E2BIG;
670 			if (error)
671 				goto cleanup;
672 			args->space -= length;
673 			args->endp += length;
674 			args->argc++;
675 		}
676 	}
677 
678 	args->begin_envv = args->endp;
679 
680 	/*
681 	 * extract environment strings
682 	 */
683 
684 	if (envv) {
685 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
686 			if (envp == (caddr_t) -1) {
687 				error = EFAULT;
688 				goto cleanup;
689 			}
690 			error = copyinstr(envp, args->endp, args->space,
691 			    &length);
692 			if (error == ENAMETOOLONG)
693 				error = E2BIG;
694 			if (error)
695 				goto cleanup;
696 			args->space -= length;
697 			args->endp += length;
698 			args->envc++;
699 		}
700 	}
701 
702 cleanup:
703 	if (error)
704 		exec_free_args(args);
705 	return (error);
706 }
707 
708 void
709 exec_free_args(struct image_args *args)
710 {
711 	kmem_free_wakeup(exec_map, (vm_offset_t)args->buf, PATH_MAX + ARG_MAX);
712 }
713 
714 /*
715  * Copy strings out to the new process address space, constructing
716  *	new arg and env vector tables. Return a pointer to the base
717  *	so that it can be used as the initial stack pointer.
718  */
719 register_t *
720 exec_copyout_strings(struct image_params *imgp)
721 {
722 	int argc, envc;
723 	char **vectp;
724 	char *stringp, *destp;
725 	register_t *stack_base;
726 	struct ps_strings *arginfo;
727 	int szsigcode;
728 
729 	/*
730 	 * Calculate string base and vector table pointers.
731 	 * Also deal with signal trampoline code for this exec type.
732 	 */
733 	arginfo = (struct ps_strings *)PS_STRINGS;
734 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
735 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
736 	    roundup((ARG_MAX - imgp->args->space), sizeof(char *));
737 
738 	/*
739 	 * install sigcode
740 	 */
741 	if (szsigcode)
742 		copyout(imgp->proc->p_sysent->sv_sigcode,
743 		    ((caddr_t)arginfo - szsigcode), szsigcode);
744 
745 	/*
746 	 * If we have a valid auxargs ptr, prepare some room
747 	 * on the stack.
748 	 */
749 	if (imgp->auxargs)
750 	/*
751 	 * The '+ 2' is for the null pointers at the end of each of the
752 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
753 	 * ELF Auxargs data.
754 	 */
755 		vectp = (char **)(destp - (imgp->args->argc +
756 		    imgp->args->envc + 2 + AT_COUNT*2) * sizeof(char*));
757 	else
758 	/*
759 	 * The '+ 2' is for the null pointers at the end of each of the
760 	 * arg and env vector sets
761 	 */
762 		vectp = (char **)
763 		    (destp - (imgp->args->argc + imgp->args->envc + 2) *
764 		    sizeof(char*));
765 
766 	/*
767 	 * vectp also becomes our initial stack base
768 	 */
769 	stack_base = (register_t *)vectp;
770 
771 	stringp = imgp->args->begin_argv;
772 	argc = imgp->args->argc;
773 	envc = imgp->args->envc;
774 
775 	/*
776 	 * Copy out strings - arguments and environment.
777 	 */
778 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
779 
780 	/*
781 	 * Fill in "ps_strings" struct for ps, w, etc.
782 	 */
783 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
784 	suword(&arginfo->ps_nargvstr, argc);
785 
786 	/*
787 	 * Fill in argument portion of vector table.
788 	 */
789 	for (; argc > 0; --argc) {
790 		suword(vectp++, (long)(intptr_t)destp);
791 		while (*stringp++ != 0)
792 			destp++;
793 		destp++;
794 	}
795 
796 	/* a null vector table pointer seperates the argp's from the envp's */
797 	suword(vectp++, 0);
798 
799 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
800 	suword(&arginfo->ps_nenvstr, envc);
801 
802 	/*
803 	 * Fill in environment portion of vector table.
804 	 */
805 	for (; envc > 0; --envc) {
806 		suword(vectp++, (long)(intptr_t)destp);
807 		while (*stringp++ != 0)
808 			destp++;
809 		destp++;
810 	}
811 
812 	/* end of vector table is a null pointer */
813 	suword(vectp, 0);
814 
815 	return (stack_base);
816 }
817 
818 /*
819  * Check permissions of file to execute.
820  *	Return 0 for success or error code on failure.
821  */
822 int
823 exec_check_permissions(imgp)
824 	struct image_params *imgp;
825 {
826 	struct proc *p = imgp->proc;
827 	struct vnode *vp = imgp->vp;
828 	struct vattr *attr = imgp->attr;
829 	struct thread *td = p->p_thread;
830 	int error;
831 
832 	/* Get file attributes */
833 	error = VOP_GETATTR(vp, attr, td);
834 	if (error)
835 		return (error);
836 
837 	/*
838 	 * 1) Check if file execution is disabled for the filesystem that this
839 	 *	file resides on.
840 	 * 2) Insure that at least one execute bit is on - otherwise root
841 	 *	will always succeed, and we don't want to happen unless the
842 	 *	file really is executable.
843 	 * 3) Insure that the file is a regular file.
844 	 */
845 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
846 	    ((attr->va_mode & 0111) == 0) ||
847 	    (attr->va_type != VREG)) {
848 		return (EACCES);
849 	}
850 
851 	/*
852 	 * Zero length files can't be exec'd
853 	 */
854 	if (attr->va_size == 0)
855 		return (ENOEXEC);
856 
857 	/*
858 	 *  Check for execute permission to file based on current credentials.
859 	 */
860 	error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td);
861 	if (error)
862 		return (error);
863 
864 	/*
865 	 * Check number of open-for-writes on the file and deny execution
866 	 * if there are any.
867 	 */
868 	if (vp->v_writecount)
869 		return (ETXTBSY);
870 
871 	/*
872 	 * Call filesystem specific open routine (which does nothing in the
873 	 * general case).
874 	 */
875 	error = VOP_OPEN(vp, FREAD, p->p_ucred, td);
876 	if (error)
877 		return (error);
878 
879 	return (0);
880 }
881 
882 /*
883  * Exec handler registration
884  */
885 int
886 exec_register(execsw_arg)
887 	const struct execsw *execsw_arg;
888 {
889 	const struct execsw **es, **xs, **newexecsw;
890 	int count = 2;	/* New slot and trailing NULL */
891 
892 	if (execsw)
893 		for (es = execsw; *es; es++)
894 			count++;
895 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
896 	if (newexecsw == NULL)
897 		return ENOMEM;
898 	xs = newexecsw;
899 	if (execsw)
900 		for (es = execsw; *es; es++)
901 			*xs++ = *es;
902 	*xs++ = execsw_arg;
903 	*xs = NULL;
904 	if (execsw)
905 		free(execsw, M_TEMP);
906 	execsw = newexecsw;
907 	return 0;
908 }
909 
910 int
911 exec_unregister(execsw_arg)
912 	const struct execsw *execsw_arg;
913 {
914 	const struct execsw **es, **xs, **newexecsw;
915 	int count = 1;
916 
917 	if (execsw == NULL)
918 		panic("unregister with no handlers left?\n");
919 
920 	for (es = execsw; *es; es++) {
921 		if (*es == execsw_arg)
922 			break;
923 	}
924 	if (*es == NULL)
925 		return ENOENT;
926 	for (es = execsw; *es; es++)
927 		if (*es != execsw_arg)
928 			count++;
929 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
930 	if (newexecsw == NULL)
931 		return ENOMEM;
932 	xs = newexecsw;
933 	for (es = execsw; *es; es++)
934 		if (*es != execsw_arg)
935 			*xs++ = *es;
936 	*xs = NULL;
937 	if (execsw)
938 		free(execsw, M_TEMP);
939 	execsw = newexecsw;
940 	return 0;
941 }
942