xref: /dflybsd-src/sys/kern/kern_exec.c (revision 57fed2afae86702adfd8bc0f2b73e76280fa6847)
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 1.28 2004/10/12 19:20:46 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/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/sfbuf.h>
49 #include <sys/sysent.h>
50 #include <sys/shm.h>
51 #include <sys/sysctl.h>
52 #include <sys/vnode.h>
53 #include <sys/vmmeter.h>
54 #include <sys/aio.h>
55 
56 #include <vm/vm.h>
57 #include <vm/vm_param.h>
58 #include <sys/lock.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_page.h>
61 #include <vm/vm_map.h>
62 #include <vm/vm_kern.h>
63 #include <vm/vm_extern.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_pager.h>
66 
67 #include <sys/user.h>
68 #include <machine/reg.h>
69 
70 #include <sys/thread2.h>
71 
72 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
73 
74 static register_t *exec_copyout_strings (struct image_params *);
75 
76 /* XXX This should be vm_size_t. */
77 static u_long ps_strings = PS_STRINGS;
78 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
79 
80 /* XXX This should be vm_size_t. */
81 static u_long usrstack = USRSTACK;
82 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
83 
84 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
85 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
86     &ps_arg_cache_limit, 0, "");
87 
88 int ps_argsopen = 1;
89 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
90 
91 void print_execve_args(struct image_args *args);
92 int debug_execve_args = 0;
93 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
94     0, "");
95 
96 void
97 print_execve_args(struct image_args *args)
98 {
99 	char *cp;
100 	int ndx;
101 
102 	cp = args->begin_argv;
103 	for (ndx = 0; ndx < args->argc; ndx++) {
104 		printf("\targv[%d]: %s\n", ndx, cp);
105 		while (*cp++ != '\0');
106 	}
107 	for (ndx = 0; ndx < args->envc; ndx++) {
108 		printf("\tenvv[%d]: %s\n", ndx, cp);
109 		while (*cp++ != '\0');
110 	}
111 }
112 
113 /*
114  * Each of the items is a pointer to a `const struct execsw', hence the
115  * double pointer here.
116  */
117 static const struct execsw **execsw;
118 
119 int
120 kern_execve(struct nameidata *ndp, struct image_args *args)
121 {
122 	struct thread *td = curthread;
123 	struct proc *p = td->td_proc;
124 	register_t *stack_base;
125 	int error, len, i;
126 	struct image_params image_params, *imgp;
127 	struct vattr attr;
128 	int (*img_first) (struct image_params *);
129 
130 	if (debug_execve_args) {
131 		printf("%s()\n", __func__);
132 		print_execve_args(args);
133 	}
134 
135 	KKASSERT(p);
136 	imgp = &image_params;
137 
138 	/*
139 	 * Lock the process and set the P_INEXEC flag to indicate that
140 	 * it should be left alone until we're done here.  This is
141 	 * necessary to avoid race conditions - e.g. in ptrace() -
142 	 * that might allow a local user to illicitly obtain elevated
143 	 * privileges.
144 	 */
145 	p->p_flag |= P_INEXEC;
146 
147 	/*
148 	 * Initialize part of the common data
149 	 */
150 	imgp->proc = p;
151 	imgp->args = args;
152 	imgp->attr = &attr;
153 	imgp->entry_addr = 0;
154 	imgp->resident = 0;
155 	imgp->vmspace_destroyed = 0;
156 	imgp->interpreted = 0;
157 	imgp->interpreter_name[0] = 0;
158 	imgp->auxargs = NULL;
159 	imgp->vp = NULL;
160 	imgp->firstpage = NULL;
161 	imgp->ps_strings = 0;
162 	imgp->image_header = NULL;
163 
164 interpret:
165 
166 	/*
167 	 * Translate the file name. namei() returns a vnode pointer
168 	 *	in ni_vp amoung other things.
169 	 */
170 	if ((error = namei(ndp)) != 0)
171 		goto exec_fail;
172 
173 	imgp->vp = ndp->ni_vp;
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, td);
181 		goto exec_fail_dealloc;
182 	}
183 
184 	error = exec_map_first_page(imgp);
185 	VOP_UNLOCK(imgp->vp, 0, td);
186 	if (error)
187 		goto exec_fail_dealloc;
188 
189 	if (debug_execve_args && imgp->interpreted) {
190 		printf("    target is interpreted -- recursive pass\n");
191 		printf("    interpreter: %s\n", imgp->interpreter_name);
192 		print_execve_args(args);
193 	}
194 
195 	/*
196 	 *	If the current process has a special image activator it
197 	 *	wants to try first, call it.   For example, emulating shell
198 	 *	scripts differently.
199 	 */
200 	error = -1;
201 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
202 		error = img_first(imgp);
203 
204 	/*
205 	 *	If the vnode has a registered vmspace, exec the vmspace
206 	 */
207 	if (error == -1 && imgp->vp->v_resident) {
208 		error = exec_resident_imgact(imgp);
209 	}
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.  If we are running a resident
255 	 * image there is no auxinfo or other image activator context
256 	 * so don't try to add fixups to the stack.
257 	 *
258 	 * Else stuff argument count as first item on stack
259 	 */
260 	if (p->p_sysent->sv_fixup && imgp->resident == 0)
261 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
262 	else
263 		suword(--stack_base, imgp->args->argc);
264 
265 	/*
266 	 * For security and other reasons, the file descriptor table cannot
267 	 * be shared after an exec.
268 	 */
269 	if (p->p_fd->fd_refcnt > 1) {
270 		struct filedesc *tmp;
271 
272 		tmp = fdcopy(p);
273 		fdfree(p);
274 		p->p_fd = tmp;
275 	}
276 
277 	/*
278 	 * For security and other reasons, signal handlers cannot
279 	 * be shared after an exec. The new proces gets a copy of the old
280 	 * handlers. In execsigs(), the new process will have its signals
281 	 * reset.
282 	 */
283 	if (p->p_procsig->ps_refcnt > 1) {
284 		struct procsig *newprocsig;
285 
286 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
287 		       M_SUBPROC, M_WAITOK);
288 		bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig));
289 		p->p_procsig->ps_refcnt--;
290 		p->p_procsig = newprocsig;
291 		p->p_procsig->ps_refcnt = 1;
292 		if (p->p_sigacts == &p->p_addr->u_sigacts)
293 			panic("shared procsig but private sigacts?");
294 
295 		p->p_addr->u_sigacts = *p->p_sigacts;
296 		p->p_sigacts = &p->p_addr->u_sigacts;
297 	}
298 
299 	/* Stop profiling */
300 	stopprofclock(p);
301 
302 	/* close files on exec */
303 	fdcloseexec(p);
304 
305 	/* reset caught signals */
306 	execsigs(p);
307 
308 	/* name this process - nameiexec(p, ndp) */
309 	len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
310 	bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
311 	p->p_comm[len] = 0;
312 
313 	/*
314 	 * mark as execed, wakeup the process that vforked (if any) and tell
315 	 * it that it now has its own resources back
316 	 */
317 	p->p_flag |= P_EXEC;
318 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
319 		p->p_flag &= ~P_PPWAIT;
320 		wakeup((caddr_t)p->p_pptr);
321 	}
322 
323 	/*
324 	 * Implement image setuid/setgid.
325 	 *
326 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
327 	 * the process is being traced.
328 	 */
329 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
330 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
331 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
332 	    (p->p_flag & P_TRACED) == 0) {
333 		/*
334 		 * Turn off syscall tracing for set-id programs, except for
335 		 * root.  Record any set-id flags first to make sure that
336 		 * we do not regain any tracing during a possible block.
337 		 */
338 		setsugid();
339 		if (p->p_tracep && suser(td)) {
340 			struct vnode *vtmp;
341 
342 			if ((vtmp = p->p_tracep) != NULL) {
343 				p->p_tracep = NULL;
344 				p->p_traceflag = 0;
345 				vrele(vtmp);
346 			}
347 		}
348 		/* Close any file descriptors 0..2 that reference procfs */
349 		setugidsafety(p);
350 		/* Make sure file descriptors 0..2 are in use. */
351 		error = fdcheckstd(p);
352 		if (error != 0)
353 			goto exec_fail_dealloc;
354 		/*
355 		 * Set the new credentials.
356 		 */
357 		cratom(&p->p_ucred);
358 		if (attr.va_mode & VSUID)
359 			change_euid(attr.va_uid);
360 		if (attr.va_mode & VSGID)
361 			p->p_ucred->cr_gid = attr.va_gid;
362 
363 		/*
364 		 * Clear local varsym variables
365 		 */
366 		varsymset_clean(&p->p_varsymset);
367 	} else {
368 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
369 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
370 			p->p_flag &= ~P_SUGID;
371 	}
372 
373 	/*
374 	 * Implement correct POSIX saved-id behavior.
375 	 */
376 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
377 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
378 		cratom(&p->p_ucred);
379 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
380 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
381 	}
382 
383 	/*
384 	 * Store the vp for use in procfs
385 	 */
386 	if (p->p_textvp)		/* release old reference */
387 		vrele(p->p_textvp);
388 	vref(ndp->ni_vp);
389 	p->p_textvp = ndp->ni_vp;
390 
391         /*
392          * Notify others that we exec'd, and clear the P_INEXEC flag
393          * as we're now a bona fide freshly-execed process.
394          */
395 	KNOTE(&p->p_klist, NOTE_EXEC);
396 	p->p_flag &= ~P_INEXEC;
397 
398 	/*
399 	 * If tracing the process, trap to debugger so breakpoints
400 	 * 	can be set before the program executes.
401 	 */
402 	STOPEVENT(p, S_EXEC, 0);
403 
404 	if (p->p_flag & P_TRACED)
405 		psignal(p, SIGTRAP);
406 
407 	/* clear "fork but no exec" flag, as we _are_ execing */
408 	p->p_acflag &= ~AFORK;
409 
410 	/* Set values passed into the program in registers. */
411 	setregs(p, imgp->entry_addr, (u_long)(uintptr_t)stack_base,
412 	    imgp->ps_strings);
413 
414 	/* Free any previous argument cache */
415 	if (p->p_args && --p->p_args->ar_ref == 0)
416 		FREE(p->p_args, M_PARGS);
417 	p->p_args = NULL;
418 
419 	/* Cache arguments if they fit inside our allowance */
420 	i = imgp->args->begin_envv - imgp->args->begin_argv;
421 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
422 		MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
423 		    M_PARGS, M_WAITOK);
424 		p->p_args->ar_ref = 1;
425 		p->p_args->ar_length = i;
426 		bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
427 	}
428 
429 exec_fail_dealloc:
430 
431 	/*
432 	 * free various allocated resources
433 	 */
434 	if (imgp->firstpage)
435 		exec_unmap_first_page(imgp);
436 
437 	if (imgp->vp) {
438 		NDFREE(ndp, NDF_ONLY_PNBUF);
439 		vrele(imgp->vp);
440 	}
441 
442 	if (error == 0) {
443 		++mycpu->gd_cnt.v_exec;
444 		return (0);
445 	}
446 
447 exec_fail:
448 	/* we're done here, clear P_INEXEC */
449 	p->p_flag &= ~P_INEXEC;
450 	if (imgp->vmspace_destroyed) {
451 		/* sorry, no more process anymore. exit gracefully */
452 		exit1(W_EXITCODE(0, SIGABRT));
453 		/* NOT REACHED */
454 		return(0);
455 	} else {
456 		return(error);
457 	}
458 }
459 
460 /*
461  * execve() system call.
462  */
463 int
464 execve(struct execve_args *uap)
465 {
466 	struct thread *td = curthread;
467 	struct nameidata nd;
468 	struct image_args args;
469 	int error;
470 
471 	NDINIT(&nd, NAMEI_LOOKUP, CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
472 	    UIO_USERSPACE, uap->fname, td);
473 
474 	error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
475 				uap->argv, uap->envv);
476 	if (error == 0)
477 		error = kern_execve(&nd, &args);
478 
479 	exec_free_args(&args);
480 
481 	/*
482 	 * The syscall result is returned in registers to the new program.
483 	 * Linux will register %edx as an atexit function and we must be
484 	 * sure to set it to 0.  XXX
485 	 */
486 	if (error == 0)
487 		uap->sysmsg_result64 = 0;
488 
489 	return (error);
490 }
491 
492 int
493 exec_map_first_page(struct image_params *imgp)
494 {
495 	int rv, i;
496 	int initial_pagein;
497 	vm_page_t ma[VM_INITIAL_PAGEIN];
498 	vm_page_t m;
499 	vm_object_t object;
500 
501 	if (imgp->firstpage)
502 		exec_unmap_first_page(imgp);
503 
504 	VOP_GETVOBJECT(imgp->vp, &object);
505 
506 	/*
507 	 * We shouldn't need protection for vm_page_grab() but we certainly
508 	 * need it for the lookup loop below (lookup/busy race), since
509 	 * an interrupt can unbusy and free the page before our busy check.
510 	 */
511 	crit_enter();
512 	m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
513 
514 	if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
515 		ma[0] = m;
516 		initial_pagein = VM_INITIAL_PAGEIN;
517 		if (initial_pagein > object->size)
518 			initial_pagein = object->size;
519 		for (i = 1; i < initial_pagein; i++) {
520 			if ((m = vm_page_lookup(object, i)) != NULL) {
521 				if ((m->flags & PG_BUSY) || m->busy)
522 					break;
523 				if (m->valid)
524 					break;
525 				vm_page_busy(m);
526 			} else {
527 				m = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
528 				if (m == NULL)
529 					break;
530 			}
531 			ma[i] = m;
532 		}
533 		initial_pagein = i;
534 
535 		/*
536 		 * get_pages unbusies all the requested pages except the
537 		 * primary page (at index 0 in this case).
538 		 */
539 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
540 		m = vm_page_lookup(object, 0);
541 
542 		if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
543 			if (m) {
544 				vm_page_protect(m, VM_PROT_NONE);
545 				vm_page_free(m);
546 			}
547 			crit_exit();
548 			return EIO;
549 		}
550 	}
551 	vm_page_hold(m);
552 	vm_page_wakeup(m);	/* unbusy the page */
553 	crit_exit();
554 
555 	imgp->firstpage = sf_buf_alloc(m, SFBA_QUICK);
556 	imgp->image_header = (void *)sf_buf_kva(imgp->firstpage);
557 
558 	return 0;
559 }
560 
561 void
562 exec_unmap_first_page(imgp)
563 	struct image_params *imgp;
564 {
565 	vm_page_t m;
566 
567 	crit_enter();
568 	if (imgp->firstpage != NULL) {
569 		m = sf_buf_page(imgp->firstpage);
570 		sf_buf_free(imgp->firstpage);
571 		imgp->firstpage = NULL;
572 		imgp->image_header = NULL;
573 		vm_page_unhold(m);
574 	}
575 	crit_exit();
576 }
577 
578 /*
579  * Destroy old address space, and allocate a new stack
580  *	The new stack is only SGROWSIZ large because it is grown
581  *	automatically in trap.c.
582  */
583 int
584 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
585 {
586 	int error;
587 	struct vmspace *vmspace = imgp->proc->p_vmspace;
588 	vm_offset_t stack_addr = USRSTACK - maxssiz;
589 	vm_map_t map;
590 
591 	imgp->vmspace_destroyed = 1;
592 
593 	/*
594 	 * Prevent a pending AIO from modifying the new address space.
595 	 */
596 	aio_proc_rundown(imgp->proc);
597 
598 	/*
599 	 * Blow away entire process VM, if address space not shared,
600 	 * otherwise, create a new VM space so that other threads are
601 	 * not disrupted.  If we are execing a resident vmspace we
602 	 * create a duplicate of it and remap the stack.
603 	 *
604 	 * The exitingcnt test is not strictly necessary but has been
605 	 * included for code sanity (to make the code more deterministic).
606 	 */
607 	map = &vmspace->vm_map;
608 	if (vmcopy) {
609 		vmspace_exec(imgp->proc, vmcopy);
610 		vmspace = imgp->proc->p_vmspace;
611 		pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
612 		map = &vmspace->vm_map;
613 	} else if (vmspace->vm_refcnt == 1 && vmspace->vm_exitingcnt == 0) {
614 		shmexit(vmspace);
615 		if (vmspace->vm_upcalls)
616 			upc_release(vmspace, imgp->proc);
617 		pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS);
618 		vm_map_remove(map, 0, VM_MAXUSER_ADDRESS);
619 	} else {
620 		vmspace_exec(imgp->proc, NULL);
621 		vmspace = imgp->proc->p_vmspace;
622 		map = &vmspace->vm_map;
623 	}
624 
625 	/* Allocate a new stack */
626 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
627 	    VM_PROT_ALL, VM_PROT_ALL, 0);
628 	if (error)
629 		return (error);
630 
631 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
632 	 * VM_STACK case, but they are still used to monitor the size of the
633 	 * process stack so we can check the stack rlimit.
634 	 */
635 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
636 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
637 
638 	return(0);
639 }
640 
641 /*
642  * Copy out argument and environment strings from the old process
643  *	address space into the temporary string buffer.
644  */
645 int
646 exec_copyin_args(struct image_args *args, char *fname,
647 		enum exec_path_segflg segflg, char **argv, char **envv)
648 {
649 	char	*argp, *envp;
650 	int	error = 0;
651 	size_t	length;
652 
653 	bzero(args, sizeof(*args));
654 	args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
655 	if (args->buf == NULL)
656 		return (ENOMEM);
657 	args->begin_argv = args->buf;
658 	args->endp = args->begin_argv;
659 	args->space = ARG_MAX;
660 
661 	args->fname = args->buf + ARG_MAX;
662 
663 	/*
664 	 * Copy the file name.
665 	 */
666 	if (segflg == PATH_SYSSPACE) {
667 		error = copystr(fname, args->fname, PATH_MAX, &length);
668 	} else if (segflg == PATH_USERSPACE) {
669 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
670 	}
671 
672 	/*
673 	 * extract argument strings
674 	 */
675 
676 	if (argv && error == 0) {
677 		/*
678 		 * The argv0 argument for execv() is allowed to be NULL,
679 		 * in which case we use our filename as argv[0].
680 		 * This guarantees that
681 		 * the interpreter knows what file to open in the case
682 		 * that we exec an interpreted file.
683 		 */
684 		argp = (caddr_t) (intptr_t) fuword(argv);
685 		if (argp == NULL) {
686 			length = strlen(args->fname) + 1;
687 			KKASSERT(length <= args->space);
688 			bcopy(args->fname, args->endp, length);
689 			args->space -= length;
690 			args->endp += length;
691 			args->argc++;
692 			argv++;
693 		}
694 		while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
695 			if (argp == (caddr_t) -1) {
696 				error = EFAULT;
697 				goto cleanup;
698 			}
699 			error = copyinstr(argp, args->endp,
700 					    args->space, &length);
701 			if (error == ENAMETOOLONG)
702 				error = E2BIG;
703 			if (error)
704 				goto cleanup;
705 			args->space -= length;
706 			args->endp += length;
707 			args->argc++;
708 		}
709 	}
710 
711 	args->begin_envv = args->endp;
712 
713 	/*
714 	 * extract environment strings
715 	 */
716 	if (envv && error == 0) {
717 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
718 			if (envp == (caddr_t) -1) {
719 				error = EFAULT;
720 				goto cleanup;
721 			}
722 			error = copyinstr(envp, args->endp, args->space,
723 			    &length);
724 			if (error == ENAMETOOLONG)
725 				error = E2BIG;
726 			if (error)
727 				goto cleanup;
728 			args->space -= length;
729 			args->endp += length;
730 			args->envc++;
731 		}
732 	}
733 
734 cleanup:
735 	return (error);
736 }
737 
738 void
739 exec_free_args(struct image_args *args)
740 {
741 	if (args->buf) {
742 		kmem_free_wakeup(exec_map,
743 				(vm_offset_t)args->buf, PATH_MAX + ARG_MAX);
744 		args->buf = NULL;
745 	}
746 }
747 
748 /*
749  * Copy strings out to the new process address space, constructing
750  *	new arg and env vector tables. Return a pointer to the base
751  *	so that it can be used as the initial stack pointer.
752  */
753 register_t *
754 exec_copyout_strings(struct image_params *imgp)
755 {
756 	int argc, envc;
757 	char **vectp;
758 	char *stringp, *destp;
759 	register_t *stack_base;
760 	struct ps_strings *arginfo;
761 	int szsigcode;
762 
763 	/*
764 	 * Calculate string base and vector table pointers.
765 	 * Also deal with signal trampoline code for this exec type.
766 	 */
767 	arginfo = (struct ps_strings *)PS_STRINGS;
768 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
769 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
770 	    roundup((ARG_MAX - imgp->args->space), sizeof(char *));
771 
772 	/*
773 	 * install sigcode
774 	 */
775 	if (szsigcode)
776 		copyout(imgp->proc->p_sysent->sv_sigcode,
777 		    ((caddr_t)arginfo - szsigcode), szsigcode);
778 
779 	/*
780 	 * If we have a valid auxargs ptr, prepare some room
781 	 * on the stack.
782 	 *
783 	 * The '+ 2' is for the null pointers at the end of each of the
784 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
785 	 * ELF Auxargs data.
786 	 */
787 	if (imgp->auxargs) {
788 		vectp = (char **)(destp - (imgp->args->argc +
789 			imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
790 	} else {
791 		vectp = (char **)(destp - (imgp->args->argc +
792 			imgp->args->envc + 2) * sizeof(char*));
793 	}
794 
795 	/*
796 	 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
797 	 * be done in crt1.o.  Note that GCC 3.x aligns the stack in main.
798 	 */
799 
800 	/*
801 	 * vectp also becomes our initial stack base
802 	 */
803 	stack_base = (register_t *)vectp;
804 
805 	stringp = imgp->args->begin_argv;
806 	argc = imgp->args->argc;
807 	envc = imgp->args->envc;
808 
809 	/*
810 	 * Copy out strings - arguments and environment.
811 	 */
812 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
813 
814 	/*
815 	 * Fill in "ps_strings" struct for ps, w, etc.
816 	 */
817 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
818 	suword(&arginfo->ps_nargvstr, argc);
819 
820 	/*
821 	 * Fill in argument portion of vector table.
822 	 */
823 	for (; argc > 0; --argc) {
824 		suword(vectp++, (long)(intptr_t)destp);
825 		while (*stringp++ != 0)
826 			destp++;
827 		destp++;
828 	}
829 
830 	/* a null vector table pointer separates the argp's from the envp's */
831 	suword(vectp++, 0);
832 
833 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
834 	suword(&arginfo->ps_nenvstr, envc);
835 
836 	/*
837 	 * Fill in environment portion of vector table.
838 	 */
839 	for (; envc > 0; --envc) {
840 		suword(vectp++, (long)(intptr_t)destp);
841 		while (*stringp++ != 0)
842 			destp++;
843 		destp++;
844 	}
845 
846 	/* end of vector table is a null pointer */
847 	suword(vectp, 0);
848 
849 	return (stack_base);
850 }
851 
852 /*
853  * Check permissions of file to execute.
854  *	Return 0 for success or error code on failure.
855  */
856 int
857 exec_check_permissions(imgp)
858 	struct image_params *imgp;
859 {
860 	struct proc *p = imgp->proc;
861 	struct vnode *vp = imgp->vp;
862 	struct vattr *attr = imgp->attr;
863 	struct thread *td = p->p_thread;
864 	int error;
865 
866 	/* Get file attributes */
867 	error = VOP_GETATTR(vp, attr, td);
868 	if (error)
869 		return (error);
870 
871 	/*
872 	 * 1) Check if file execution is disabled for the filesystem that this
873 	 *	file resides on.
874 	 * 2) Insure that at least one execute bit is on - otherwise root
875 	 *	will always succeed, and we don't want to happen unless the
876 	 *	file really is executable.
877 	 * 3) Insure that the file is a regular file.
878 	 */
879 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
880 	    ((attr->va_mode & 0111) == 0) ||
881 	    (attr->va_type != VREG)) {
882 		return (EACCES);
883 	}
884 
885 	/*
886 	 * Zero length files can't be exec'd
887 	 */
888 	if (attr->va_size == 0)
889 		return (ENOEXEC);
890 
891 	/*
892 	 *  Check for execute permission to file based on current credentials.
893 	 */
894 	error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td);
895 	if (error)
896 		return (error);
897 
898 	/*
899 	 * Check number of open-for-writes on the file and deny execution
900 	 * if there are any.
901 	 */
902 	if (vp->v_writecount)
903 		return (ETXTBSY);
904 
905 	/*
906 	 * Call filesystem specific open routine (which does nothing in the
907 	 * general case).
908 	 */
909 	error = VOP_OPEN(vp, FREAD, p->p_ucred, td);
910 	if (error)
911 		return (error);
912 
913 	return (0);
914 }
915 
916 /*
917  * Exec handler registration
918  */
919 int
920 exec_register(execsw_arg)
921 	const struct execsw *execsw_arg;
922 {
923 	const struct execsw **es, **xs, **newexecsw;
924 	int count = 2;	/* New slot and trailing NULL */
925 
926 	if (execsw)
927 		for (es = execsw; *es; es++)
928 			count++;
929 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
930 	if (newexecsw == NULL)
931 		return ENOMEM;
932 	xs = newexecsw;
933 	if (execsw)
934 		for (es = execsw; *es; es++)
935 			*xs++ = *es;
936 	*xs++ = execsw_arg;
937 	*xs = NULL;
938 	if (execsw)
939 		free(execsw, M_TEMP);
940 	execsw = newexecsw;
941 	return 0;
942 }
943 
944 int
945 exec_unregister(execsw_arg)
946 	const struct execsw *execsw_arg;
947 {
948 	const struct execsw **es, **xs, **newexecsw;
949 	int count = 1;
950 
951 	if (execsw == NULL)
952 		panic("unregister with no handlers left?");
953 
954 	for (es = execsw; *es; es++) {
955 		if (*es == execsw_arg)
956 			break;
957 	}
958 	if (*es == NULL)
959 		return ENOENT;
960 	for (es = execsw; *es; es++)
961 		if (*es != execsw_arg)
962 			count++;
963 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
964 	if (newexecsw == NULL)
965 		return ENOMEM;
966 	xs = newexecsw;
967 	for (es = execsw; *es; es++)
968 		if (*es != execsw_arg)
969 			*xs++ = *es;
970 	*xs = NULL;
971 	if (execsw)
972 		free(execsw, M_TEMP);
973 	execsw = newexecsw;
974 	return 0;
975 }
976