xref: /dflybsd-src/sys/kern/kern_exec.c (revision bc76a771df54af7e361532b257cecc26227736b4)
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 1.22 2004/04/11 00:10:30 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/aio.h>
54 
55 #include <vm/vm.h>
56 #include <vm/vm_param.h>
57 #include <sys/lock.h>
58 #include <vm/pmap.h>
59 #include <vm/vm_page.h>
60 #include <vm/vm_map.h>
61 #include <vm/vm_kern.h>
62 #include <vm/vm_extern.h>
63 #include <vm/vm_object.h>
64 #include <vm/vm_pager.h>
65 
66 #include <sys/user.h>
67 #include <machine/reg.h>
68 
69 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
70 
71 static register_t *exec_copyout_strings (struct image_params *);
72 
73 /* XXX This should be vm_size_t. */
74 static u_long ps_strings = PS_STRINGS;
75 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
76 
77 /* XXX This should be vm_size_t. */
78 static u_long usrstack = USRSTACK;
79 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
80 
81 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
82 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
83     &ps_arg_cache_limit, 0, "");
84 
85 int ps_argsopen = 1;
86 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
87 
88 void print_execve_args(struct image_args *args);
89 int debug_execve_args = 0;
90 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
91     0, "");
92 
93 void
94 print_execve_args(struct image_args *args)
95 {
96 	char *cp;
97 	int ndx;
98 
99 	cp = args->begin_argv;
100 	for (ndx = 0; ndx < args->argc; ndx++) {
101 		printf("\targv[%d]: %s\n", ndx, cp);
102 		while (*cp++ != '\0');
103 	}
104 	for (ndx = 0; ndx < args->envc; ndx++) {
105 		printf("\tenvv[%d]: %s\n", ndx, cp);
106 		while (*cp++ != '\0');
107 	}
108 }
109 
110 /*
111  * Each of the items is a pointer to a `const struct execsw', hence the
112  * double pointer here.
113  */
114 static const struct execsw **execsw;
115 
116 int
117 kern_execve(struct nameidata *ndp, struct image_args *args)
118 {
119 	struct thread *td = curthread;
120 	struct proc *p = td->td_proc;
121 	register_t *stack_base;
122 	int error, len, i;
123 	struct image_params image_params, *imgp;
124 	struct vattr attr;
125 	int (*img_first) (struct image_params *);
126 
127 	if (debug_execve_args) {
128 		printf("%s()\n", __func__);
129 		print_execve_args(args);
130 	}
131 
132 	KKASSERT(p);
133 	imgp = &image_params;
134 
135 	/*
136 	 * Lock the process and set the P_INEXEC flag to indicate that
137 	 * it should be left alone until we're done here.  This is
138 	 * necessary to avoid race conditions - e.g. in ptrace() -
139 	 * that might allow a local user to illicitly obtain elevated
140 	 * privileges.
141 	 */
142 	p->p_flag |= P_INEXEC;
143 
144 	/*
145 	 * Initialize part of the common data
146 	 */
147 	imgp->proc = p;
148 	imgp->args = args;
149 	imgp->attr = &attr;
150 	imgp->entry_addr = 0;
151 	imgp->resident = 0;
152 	imgp->vmspace_destroyed = 0;
153 	imgp->interpreted = 0;
154 	imgp->interpreter_name[0] = 0;
155 	imgp->auxargs = NULL;
156 	imgp->vp = NULL;
157 	imgp->firstpage = NULL;
158 	imgp->ps_strings = 0;
159 	imgp->image_header = NULL;
160 
161 interpret:
162 
163 	/*
164 	 * Translate the file name. namei() returns a vnode pointer
165 	 *	in ni_vp amoung other things.
166 	 */
167 	if ((error = namei(ndp)) != 0)
168 		goto exec_fail;
169 
170 	imgp->vp = ndp->ni_vp;
171 
172 	/*
173 	 * Check file permissions (also 'opens' file)
174 	 */
175 	error = exec_check_permissions(imgp);
176 	if (error) {
177 		VOP_UNLOCK(imgp->vp, NULL, 0, td);
178 		goto exec_fail_dealloc;
179 	}
180 
181 	error = exec_map_first_page(imgp);
182 	VOP_UNLOCK(imgp->vp, NULL, 0, td);
183 	if (error)
184 		goto exec_fail_dealloc;
185 
186 	if (debug_execve_args && imgp->interpreted) {
187 		printf("    target is interpreted -- recursive pass\n");
188 		printf("    interpreter: %s\n", imgp->interpreter_name);
189 		print_execve_args(args);
190 	}
191 
192 	/*
193 	 *	If the current process has a special image activator it
194 	 *	wants to try first, call it.   For example, emulating shell
195 	 *	scripts differently.
196 	 */
197 	error = -1;
198 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
199 		error = img_first(imgp);
200 
201 	/*
202 	 *	If the vnode has a registered vmspace, exec the vmspace
203 	 */
204 	if (error == -1 && imgp->vp->v_resident) {
205 		error = exec_resident_imgact(imgp);
206 	}
207 
208 	/*
209 	 *	Loop through the list of image activators, calling each one.
210 	 *	An activator returns -1 if there is no match, 0 on success,
211 	 *	and an error otherwise.
212 	 */
213 	for (i = 0; error == -1 && execsw[i]; ++i) {
214 		if (execsw[i]->ex_imgact == NULL ||
215 		    execsw[i]->ex_imgact == img_first) {
216 			continue;
217 		}
218 		error = (*execsw[i]->ex_imgact)(imgp);
219 	}
220 
221 	if (error) {
222 		if (error == -1)
223 			error = ENOEXEC;
224 		goto exec_fail_dealloc;
225 	}
226 
227 	/*
228 	 * Special interpreter operation, cleanup and loop up to try to
229 	 * activate the interpreter.
230 	 */
231 	if (imgp->interpreted) {
232 		exec_unmap_first_page(imgp);
233 		/* free name buffer and old vnode */
234 		NDFREE(ndp, NDF_ONLY_PNBUF);
235 		vrele(ndp->ni_vp);
236 		/* set new name to that of the interpreter */
237 		NDINIT(ndp, NAMEI_LOOKUP,
238 		    CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
239 		    UIO_SYSSPACE, imgp->interpreter_name, td);
240 		goto interpret;
241 	}
242 
243 	/*
244 	 * Copy out strings (args and env) and initialize stack base
245 	 */
246 	stack_base = exec_copyout_strings(imgp);
247 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
248 
249 	/*
250 	 * If custom stack fixup routine present for this process
251 	 * let it do the stack setup.  If we are running a resident
252 	 * image there is no auxinfo or other image activator context
253 	 * so don't try to add fixups to the stack.
254 	 *
255 	 * Else stuff argument count as first item on stack
256 	 */
257 	if (p->p_sysent->sv_fixup && imgp->resident == 0)
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->vp) {
435 		NDFREE(ndp, NDF_ONLY_PNBUF);
436 		vrele(imgp->vp);
437 	}
438 
439 	if (error == 0)
440 		return (0);
441 
442 exec_fail:
443 	/* we're done here, clear P_INEXEC */
444 	p->p_flag &= ~P_INEXEC;
445 	if (imgp->vmspace_destroyed) {
446 		/* sorry, no more process anymore. exit gracefully */
447 		exit1(W_EXITCODE(0, SIGABRT));
448 		/* NOT REACHED */
449 		return(0);
450 	} else {
451 		return(error);
452 	}
453 }
454 
455 /*
456  * execve() system call.
457  */
458 int
459 execve(struct execve_args *uap)
460 {
461 	struct thread *td = curthread;
462 	struct nameidata nd;
463 	struct image_args args;
464 	int error;
465 
466 	NDINIT(&nd, NAMEI_LOOKUP, CNP_LOCKLEAF | CNP_FOLLOW | CNP_SAVENAME,
467 	    UIO_USERSPACE, uap->fname, td);
468 
469 	error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
470 				uap->argv, uap->envv);
471 	if (error == 0)
472 		error = kern_execve(&nd, &args);
473 
474 	exec_free_args(&args);
475 
476 	/*
477 	 * The syscall result is returned in registers to the new program.
478 	 * Linux will register %edx as an atexit function and we must be
479 	 * sure to set it to 0.  XXX
480 	 */
481 	if (error == 0)
482 		uap->sysmsg_result64 = 0;
483 
484 	return (error);
485 }
486 
487 int
488 exec_map_first_page(struct image_params *imgp)
489 {
490 	int s, rv, i;
491 	int initial_pagein;
492 	vm_page_t ma[VM_INITIAL_PAGEIN];
493 	vm_object_t object;
494 
495 	if (imgp->firstpage)
496 		exec_unmap_first_page(imgp);
497 
498 	VOP_GETVOBJECT(imgp->vp, &object);
499 	s = splvm();
500 
501 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
502 
503 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
504 		initial_pagein = VM_INITIAL_PAGEIN;
505 		if (initial_pagein > object->size)
506 			initial_pagein = object->size;
507 		for (i = 1; i < initial_pagein; i++) {
508 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
509 				if ((ma[i]->flags & PG_BUSY) || ma[i]->busy)
510 					break;
511 				if (ma[i]->valid)
512 					break;
513 				vm_page_busy(ma[i]);
514 			} else {
515 				ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
516 				if (ma[i] == NULL)
517 					break;
518 			}
519 		}
520 		initial_pagein = i;
521 
522 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
523 		ma[0] = vm_page_lookup(object, 0);
524 
525 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || (ma[0]->valid == 0)) {
526 			if (ma[0]) {
527 				vm_page_protect(ma[0], VM_PROT_NONE);
528 				vm_page_free(ma[0]);
529 			}
530 			splx(s);
531 			return EIO;
532 		}
533 	}
534 
535 	vm_page_wire(ma[0]);
536 	vm_page_wakeup(ma[0]);
537 	splx(s);
538 
539 	imgp->firstpage = sf_buf_alloc(ma[0], SFBA_QUICK);
540 	imgp->image_header = (void *)sf_buf_kva(imgp->firstpage);
541 
542 	return 0;
543 }
544 
545 void
546 exec_unmap_first_page(imgp)
547 	struct image_params *imgp;
548 {
549 	vm_page_t m;
550 
551 	if (imgp->firstpage != NULL) {
552 		m = sf_buf_page(imgp->firstpage);
553 		sf_buf_free(imgp->firstpage);
554 		imgp->firstpage = NULL;
555 		imgp->image_header = NULL;
556 		vm_page_unwire(m, 1);
557 	}
558 }
559 
560 /*
561  * Destroy old address space, and allocate a new stack
562  *	The new stack is only SGROWSIZ large because it is grown
563  *	automatically in trap.c.
564  */
565 int
566 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
567 {
568 	int error;
569 	struct vmspace *vmspace = imgp->proc->p_vmspace;
570 	vm_offset_t stack_addr = USRSTACK - maxssiz;
571 	vm_map_t map;
572 
573 	imgp->vmspace_destroyed = 1;
574 
575 	/*
576 	 * Prevent a pending AIO from modifying the new address space.
577 	 */
578 	aio_proc_rundown(imgp->proc);
579 
580 	/*
581 	 * Blow away entire process VM, if address space not shared,
582 	 * otherwise, create a new VM space so that other threads are
583 	 * not disrupted.  If we are execing a resident vmspace we
584 	 * create a duplicate of it and remap the stack.
585 	 *
586 	 * The exitingcnt test is not strictly necessary but has been
587 	 * included for code sanity (to make the code more deterministic).
588 	 */
589 	map = &vmspace->vm_map;
590 	if (vmcopy) {
591 		vmspace_exec(imgp->proc, vmcopy);
592 		vmspace = imgp->proc->p_vmspace;
593 		pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
594 		map = &vmspace->vm_map;
595 	} else if (vmspace->vm_refcnt == 1 && vmspace->vm_exitingcnt == 0) {
596 		shmexit(vmspace);
597 		if (vmspace->vm_upcalls)
598 			upc_release(vmspace, imgp->proc);
599 		pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS);
600 		vm_map_remove(map, 0, VM_MAXUSER_ADDRESS);
601 	} else {
602 		vmspace_exec(imgp->proc, NULL);
603 		vmspace = imgp->proc->p_vmspace;
604 		map = &vmspace->vm_map;
605 	}
606 
607 	/* Allocate a new stack */
608 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
609 	    VM_PROT_ALL, VM_PROT_ALL, 0);
610 	if (error)
611 		return (error);
612 
613 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
614 	 * VM_STACK case, but they are still used to monitor the size of the
615 	 * process stack so we can check the stack rlimit.
616 	 */
617 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
618 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
619 
620 	return(0);
621 }
622 
623 /*
624  * Copy out argument and environment strings from the old process
625  *	address space into the temporary string buffer.
626  */
627 int
628 exec_copyin_args(struct image_args *args, char *fname,
629 		enum exec_path_segflg segflg, char **argv, char **envv)
630 {
631 	char	*argp, *envp;
632 	int	error = 0;
633 	size_t	length;
634 
635 	bzero(args, sizeof(*args));
636 	args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
637 	if (args->buf == NULL)
638 		return (ENOMEM);
639 	args->begin_argv = args->buf;
640 	args->endp = args->begin_argv;
641 	args->space = ARG_MAX;
642 
643 	args->fname = args->buf + ARG_MAX;
644 
645 	/*
646 	 * Copy the file name.
647 	 */
648 	if (segflg == PATH_SYSSPACE) {
649 		error = copystr(fname, args->fname, PATH_MAX, &length);
650 	} else if (segflg == PATH_USERSPACE) {
651 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
652 	}
653 
654 	/*
655 	 * extract argument strings
656 	 */
657 
658 	if (argv && error == 0) {
659 		/*
660 		 * The argv0 argument for execv() is allowed to be NULL,
661 		 * in which case we use our filename as argv[0].
662 		 * This guarantees that
663 		 * the interpreter knows what file to open in the case
664 		 * that we exec an interpreted file.
665 		 */
666 		argp = (caddr_t) (intptr_t) fuword(argv);
667 		if (argp == NULL) {
668 			length = strlen(args->fname) + 1;
669 			KKASSERT(length <= args->space);
670 			bcopy(args->fname, args->endp, length);
671 			args->space -= length;
672 			args->endp += length;
673 			args->argc++;
674 			argv++;
675 		}
676 		while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
677 			if (argp == (caddr_t) -1) {
678 				error = EFAULT;
679 				goto cleanup;
680 			}
681 			error = copyinstr(argp, args->endp,
682 					    args->space, &length);
683 			if (error == ENAMETOOLONG)
684 				error = E2BIG;
685 			if (error)
686 				goto cleanup;
687 			args->space -= length;
688 			args->endp += length;
689 			args->argc++;
690 		}
691 	}
692 
693 	args->begin_envv = args->endp;
694 
695 	/*
696 	 * extract environment strings
697 	 */
698 	if (envv && error == 0) {
699 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
700 			if (envp == (caddr_t) -1) {
701 				error = EFAULT;
702 				goto cleanup;
703 			}
704 			error = copyinstr(envp, args->endp, args->space,
705 			    &length);
706 			if (error == ENAMETOOLONG)
707 				error = E2BIG;
708 			if (error)
709 				goto cleanup;
710 			args->space -= length;
711 			args->endp += length;
712 			args->envc++;
713 		}
714 	}
715 
716 cleanup:
717 	return (error);
718 }
719 
720 void
721 exec_free_args(struct image_args *args)
722 {
723 	if (args->buf) {
724 		kmem_free_wakeup(exec_map,
725 				(vm_offset_t)args->buf, PATH_MAX + ARG_MAX);
726 		args->buf = NULL;
727 	}
728 }
729 
730 /*
731  * Copy strings out to the new process address space, constructing
732  *	new arg and env vector tables. Return a pointer to the base
733  *	so that it can be used as the initial stack pointer.
734  */
735 register_t *
736 exec_copyout_strings(struct image_params *imgp)
737 {
738 	int argc, envc;
739 	char **vectp;
740 	char *stringp, *destp;
741 	register_t *stack_base;
742 	struct ps_strings *arginfo;
743 	int szsigcode;
744 
745 	/*
746 	 * Calculate string base and vector table pointers.
747 	 * Also deal with signal trampoline code for this exec type.
748 	 */
749 	arginfo = (struct ps_strings *)PS_STRINGS;
750 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
751 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
752 	    roundup((ARG_MAX - imgp->args->space), sizeof(char *));
753 
754 	/*
755 	 * install sigcode
756 	 */
757 	if (szsigcode)
758 		copyout(imgp->proc->p_sysent->sv_sigcode,
759 		    ((caddr_t)arginfo - szsigcode), szsigcode);
760 
761 	/*
762 	 * If we have a valid auxargs ptr, prepare some room
763 	 * on the stack.
764 	 *
765 	 * The '+ 2' is for the null pointers at the end of each of the
766 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
767 	 * ELF Auxargs data.
768 	 */
769 	if (imgp->auxargs) {
770 		vectp = (char **)(destp - (imgp->args->argc +
771 			imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
772 	} else {
773 		vectp = (char **)(destp - (imgp->args->argc +
774 			imgp->args->envc + 2) * sizeof(char*));
775 	}
776 
777 	/*
778 	 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
779 	 * be done in crt1.o.  Note that GCC 3.x aligns the stack in main.
780 	 */
781 
782 	/*
783 	 * vectp also becomes our initial stack base
784 	 */
785 	stack_base = (register_t *)vectp;
786 
787 	stringp = imgp->args->begin_argv;
788 	argc = imgp->args->argc;
789 	envc = imgp->args->envc;
790 
791 	/*
792 	 * Copy out strings - arguments and environment.
793 	 */
794 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
795 
796 	/*
797 	 * Fill in "ps_strings" struct for ps, w, etc.
798 	 */
799 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
800 	suword(&arginfo->ps_nargvstr, argc);
801 
802 	/*
803 	 * Fill in argument portion of vector table.
804 	 */
805 	for (; argc > 0; --argc) {
806 		suword(vectp++, (long)(intptr_t)destp);
807 		while (*stringp++ != 0)
808 			destp++;
809 		destp++;
810 	}
811 
812 	/* a null vector table pointer separates the argp's from the envp's */
813 	suword(vectp++, 0);
814 
815 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
816 	suword(&arginfo->ps_nenvstr, envc);
817 
818 	/*
819 	 * Fill in environment portion of vector table.
820 	 */
821 	for (; envc > 0; --envc) {
822 		suword(vectp++, (long)(intptr_t)destp);
823 		while (*stringp++ != 0)
824 			destp++;
825 		destp++;
826 	}
827 
828 	/* end of vector table is a null pointer */
829 	suword(vectp, 0);
830 
831 	return (stack_base);
832 }
833 
834 /*
835  * Check permissions of file to execute.
836  *	Return 0 for success or error code on failure.
837  */
838 int
839 exec_check_permissions(imgp)
840 	struct image_params *imgp;
841 {
842 	struct proc *p = imgp->proc;
843 	struct vnode *vp = imgp->vp;
844 	struct vattr *attr = imgp->attr;
845 	struct thread *td = p->p_thread;
846 	int error;
847 
848 	/* Get file attributes */
849 	error = VOP_GETATTR(vp, attr, td);
850 	if (error)
851 		return (error);
852 
853 	/*
854 	 * 1) Check if file execution is disabled for the filesystem that this
855 	 *	file resides on.
856 	 * 2) Insure that at least one execute bit is on - otherwise root
857 	 *	will always succeed, and we don't want to happen unless the
858 	 *	file really is executable.
859 	 * 3) Insure that the file is a regular file.
860 	 */
861 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
862 	    ((attr->va_mode & 0111) == 0) ||
863 	    (attr->va_type != VREG)) {
864 		return (EACCES);
865 	}
866 
867 	/*
868 	 * Zero length files can't be exec'd
869 	 */
870 	if (attr->va_size == 0)
871 		return (ENOEXEC);
872 
873 	/*
874 	 *  Check for execute permission to file based on current credentials.
875 	 */
876 	error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td);
877 	if (error)
878 		return (error);
879 
880 	/*
881 	 * Check number of open-for-writes on the file and deny execution
882 	 * if there are any.
883 	 */
884 	if (vp->v_writecount)
885 		return (ETXTBSY);
886 
887 	/*
888 	 * Call filesystem specific open routine (which does nothing in the
889 	 * general case).
890 	 */
891 	error = VOP_OPEN(vp, FREAD, p->p_ucred, td);
892 	if (error)
893 		return (error);
894 
895 	return (0);
896 }
897 
898 /*
899  * Exec handler registration
900  */
901 int
902 exec_register(execsw_arg)
903 	const struct execsw *execsw_arg;
904 {
905 	const struct execsw **es, **xs, **newexecsw;
906 	int count = 2;	/* New slot and trailing NULL */
907 
908 	if (execsw)
909 		for (es = execsw; *es; es++)
910 			count++;
911 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
912 	if (newexecsw == NULL)
913 		return ENOMEM;
914 	xs = newexecsw;
915 	if (execsw)
916 		for (es = execsw; *es; es++)
917 			*xs++ = *es;
918 	*xs++ = execsw_arg;
919 	*xs = NULL;
920 	if (execsw)
921 		free(execsw, M_TEMP);
922 	execsw = newexecsw;
923 	return 0;
924 }
925 
926 int
927 exec_unregister(execsw_arg)
928 	const struct execsw *execsw_arg;
929 {
930 	const struct execsw **es, **xs, **newexecsw;
931 	int count = 1;
932 
933 	if (execsw == NULL)
934 		panic("unregister with no handlers left?\n");
935 
936 	for (es = execsw; *es; es++) {
937 		if (*es == execsw_arg)
938 			break;
939 	}
940 	if (*es == NULL)
941 		return ENOENT;
942 	for (es = execsw; *es; es++)
943 		if (*es != execsw_arg)
944 			count++;
945 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
946 	if (newexecsw == NULL)
947 		return ENOMEM;
948 	xs = newexecsw;
949 	for (es = execsw; *es; es++)
950 		if (*es != execsw_arg)
951 			*xs++ = *es;
952 	*xs = NULL;
953 	if (execsw)
954 		free(execsw, M_TEMP);
955 	execsw = newexecsw;
956 	return 0;
957 }
958