xref: /dflybsd-src/sys/kern/kern_exec.c (revision 41871674d0079dec70d55eb824f39d07dc7b3310)
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 1.36 2006/03/29 18:44:50 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/nlookup.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 #include <sys/libkern.h>
56 
57 #include <vm/vm.h>
58 #include <vm/vm_param.h>
59 #include <sys/lock.h>
60 #include <vm/pmap.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_map.h>
63 #include <vm/vm_kern.h>
64 #include <vm/vm_extern.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_pager.h>
67 
68 #include <sys/user.h>
69 #include <machine/reg.h>
70 
71 #include <sys/thread2.h>
72 
73 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
74 
75 static register_t *exec_copyout_strings (struct image_params *);
76 
77 /* XXX This should be vm_size_t. */
78 static u_long ps_strings = PS_STRINGS;
79 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
80 
81 /* XXX This should be vm_size_t. */
82 static u_long usrstack = USRSTACK;
83 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
84 
85 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
86 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
87     &ps_arg_cache_limit, 0, "");
88 
89 int ps_argsopen = 1;
90 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
91 
92 void print_execve_args(struct image_args *args);
93 int debug_execve_args = 0;
94 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
95     0, "");
96 
97 /*
98  * stackgap_random specifies if the stackgap should have a random size added
99  * to it.  It must be a power of 2.  If non-zero, the stack gap will be
100  * calculated as: ALIGN(arc4random() & (stackgap_random - 1)).
101  */
102 static int stackgap_random = 1024;
103 static int
104 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
105 {
106 	int error, new_val;
107 	new_val = stackgap_random;
108 	error = sysctl_handle_int(oidp, &new_val, 0, req);
109 	if (error != 0 || req->newptr == NULL)
110 		return (error);
111 	if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
112 		return (EINVAL);
113 	stackgap_random = new_val;
114 
115 	return(0);
116 }
117 
118 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
119 	0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
120 
121 void
122 print_execve_args(struct image_args *args)
123 {
124 	char *cp;
125 	int ndx;
126 
127 	cp = args->begin_argv;
128 	for (ndx = 0; ndx < args->argc; ndx++) {
129 		printf("\targv[%d]: %s\n", ndx, cp);
130 		while (*cp++ != '\0');
131 	}
132 	for (ndx = 0; ndx < args->envc; ndx++) {
133 		printf("\tenvv[%d]: %s\n", ndx, cp);
134 		while (*cp++ != '\0');
135 	}
136 }
137 
138 /*
139  * Each of the items is a pointer to a `const struct execsw', hence the
140  * double pointer here.
141  */
142 static const struct execsw **execsw;
143 
144 int
145 kern_execve(struct nlookupdata *nd, struct image_args *args)
146 {
147 	struct thread *td = curthread;
148 	struct proc *p = td->td_proc;
149 	register_t *stack_base;
150 	int error, len, i;
151 	struct image_params image_params, *imgp;
152 	struct vattr attr;
153 	int (*img_first) (struct image_params *);
154 
155 	if (debug_execve_args) {
156 		printf("%s()\n", __func__);
157 		print_execve_args(args);
158 	}
159 
160 	KKASSERT(p);
161 	imgp = &image_params;
162 
163 	/*
164 	 * Lock the process and set the P_INEXEC flag to indicate that
165 	 * it should be left alone until we're done here.  This is
166 	 * necessary to avoid race conditions - e.g. in ptrace() -
167 	 * that might allow a local user to illicitly obtain elevated
168 	 * privileges.
169 	 */
170 	p->p_flag |= P_INEXEC;
171 
172 	/*
173 	 * Initialize part of the common data
174 	 */
175 	imgp->proc = p;
176 	imgp->args = args;
177 	imgp->attr = &attr;
178 	imgp->entry_addr = 0;
179 	imgp->resident = 0;
180 	imgp->vmspace_destroyed = 0;
181 	imgp->interpreted = 0;
182 	imgp->interpreter_name[0] = 0;
183 	imgp->auxargs = NULL;
184 	imgp->vp = NULL;
185 	imgp->firstpage = NULL;
186 	imgp->ps_strings = 0;
187 	imgp->image_header = NULL;
188 
189 interpret:
190 
191 	/*
192 	 * Translate the file name to a vnode.  Unlock the cache entry to
193 	 * improve parallelism for programs exec'd in parallel.
194 	 */
195 	if ((error = nlookup(nd)) != 0)
196 		goto exec_fail;
197 	error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
198 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
199 	nd->nl_flags &= ~NLC_NCPISLOCKED;
200 	cache_unlock(nd->nl_ncp);
201 	if (error)
202 		goto exec_fail;
203 
204 	/*
205 	 * Check file permissions (also 'opens' file)
206 	 */
207 	error = exec_check_permissions(imgp);
208 	if (error) {
209 		VOP_UNLOCK(imgp->vp, 0, td);
210 		goto exec_fail_dealloc;
211 	}
212 
213 	error = exec_map_first_page(imgp);
214 	VOP_UNLOCK(imgp->vp, 0, td);
215 	if (error)
216 		goto exec_fail_dealloc;
217 
218 	if (debug_execve_args && imgp->interpreted) {
219 		printf("    target is interpreted -- recursive pass\n");
220 		printf("    interpreter: %s\n", imgp->interpreter_name);
221 		print_execve_args(args);
222 	}
223 
224 	/*
225 	 *	If the current process has a special image activator it
226 	 *	wants to try first, call it.   For example, emulating shell
227 	 *	scripts differently.
228 	 */
229 	error = -1;
230 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
231 		error = img_first(imgp);
232 
233 	/*
234 	 *	If the vnode has a registered vmspace, exec the vmspace
235 	 */
236 	if (error == -1 && imgp->vp->v_resident) {
237 		error = exec_resident_imgact(imgp);
238 	}
239 
240 	/*
241 	 *	Loop through the list of image activators, calling each one.
242 	 *	An activator returns -1 if there is no match, 0 on success,
243 	 *	and an error otherwise.
244 	 */
245 	for (i = 0; error == -1 && execsw[i]; ++i) {
246 		if (execsw[i]->ex_imgact == NULL ||
247 		    execsw[i]->ex_imgact == img_first) {
248 			continue;
249 		}
250 		error = (*execsw[i]->ex_imgact)(imgp);
251 	}
252 
253 	if (error) {
254 		if (error == -1)
255 			error = ENOEXEC;
256 		goto exec_fail_dealloc;
257 	}
258 
259 	/*
260 	 * Special interpreter operation, cleanup and loop up to try to
261 	 * activate the interpreter.
262 	 */
263 	if (imgp->interpreted) {
264 		exec_unmap_first_page(imgp);
265 		nlookup_done(nd);
266 		vrele(imgp->vp);
267 		imgp->vp = NULL;
268 		error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
269 					NLC_FOLLOW);
270 		if (error)
271 			goto exec_fail;
272 		goto interpret;
273 	}
274 
275 	/*
276 	 * Copy out strings (args and env) and initialize stack base
277 	 */
278 	stack_base = exec_copyout_strings(imgp);
279 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
280 
281 	/*
282 	 * If custom stack fixup routine present for this process
283 	 * let it do the stack setup.  If we are running a resident
284 	 * image there is no auxinfo or other image activator context
285 	 * so don't try to add fixups to the stack.
286 	 *
287 	 * Else stuff argument count as first item on stack
288 	 */
289 	if (p->p_sysent->sv_fixup && imgp->resident == 0)
290 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
291 	else
292 		suword(--stack_base, imgp->args->argc);
293 
294 	/*
295 	 * For security and other reasons, the file descriptor table cannot
296 	 * be shared after an exec.
297 	 */
298 	if (p->p_fd->fd_refcnt > 1) {
299 		struct filedesc *tmp;
300 
301 		tmp = fdcopy(p);
302 		fdfree(p);
303 		p->p_fd = tmp;
304 	}
305 
306 	/*
307 	 * For security and other reasons, signal handlers cannot
308 	 * be shared after an exec. The new proces gets a copy of the old
309 	 * handlers. In execsigs(), the new process will have its signals
310 	 * reset.
311 	 */
312 	if (p->p_procsig->ps_refcnt > 1) {
313 		struct procsig *newprocsig;
314 
315 		MALLOC(newprocsig, struct procsig *, sizeof(struct procsig),
316 		       M_SUBPROC, M_WAITOK);
317 		bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig));
318 		p->p_procsig->ps_refcnt--;
319 		p->p_procsig = newprocsig;
320 		p->p_procsig->ps_refcnt = 1;
321 		if (p->p_sigacts == &p->p_addr->u_sigacts)
322 			panic("shared procsig but private sigacts?");
323 
324 		p->p_addr->u_sigacts = *p->p_sigacts;
325 		p->p_sigacts = &p->p_addr->u_sigacts;
326 	}
327 
328 	/* Stop profiling */
329 	stopprofclock(p);
330 
331 	/* close files on exec */
332 	fdcloseexec(p);
333 
334 	/* reset caught signals */
335 	execsigs(p);
336 
337 	/* name this process - nameiexec(p, ndp) */
338 	len = min(nd->nl_ncp->nc_nlen, MAXCOMLEN);
339 	bcopy(nd->nl_ncp->nc_name, p->p_comm, len);
340 	p->p_comm[len] = 0;
341 	bcopy(p->p_comm, p->p_lwp.lwp_thread->td_comm, MAXCOMLEN+1);
342 
343 	/*
344 	 * mark as execed, wakeup the process that vforked (if any) and tell
345 	 * it that it now has its own resources back
346 	 */
347 	p->p_flag |= P_EXEC;
348 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
349 		p->p_flag &= ~P_PPWAIT;
350 		wakeup((caddr_t)p->p_pptr);
351 	}
352 
353 	/*
354 	 * Implement image setuid/setgid.
355 	 *
356 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
357 	 * the process is being traced.
358 	 */
359 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
360 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
361 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
362 	    (p->p_flag & P_TRACED) == 0) {
363 		/*
364 		 * Turn off syscall tracing for set-id programs, except for
365 		 * root.  Record any set-id flags first to make sure that
366 		 * we do not regain any tracing during a possible block.
367 		 */
368 		setsugid();
369 		if (p->p_tracep && suser(td)) {
370 			struct vnode *vtmp;
371 
372 			if ((vtmp = p->p_tracep) != NULL) {
373 				p->p_tracep = NULL;
374 				p->p_traceflag = 0;
375 				vrele(vtmp);
376 			}
377 		}
378 		/* Close any file descriptors 0..2 that reference procfs */
379 		setugidsafety(p);
380 		/* Make sure file descriptors 0..2 are in use. */
381 		error = fdcheckstd(p);
382 		if (error != 0)
383 			goto exec_fail_dealloc;
384 		/*
385 		 * Set the new credentials.
386 		 */
387 		cratom(&p->p_ucred);
388 		if (attr.va_mode & VSUID)
389 			change_euid(attr.va_uid);
390 		if (attr.va_mode & VSGID)
391 			p->p_ucred->cr_gid = attr.va_gid;
392 
393 		/*
394 		 * Clear local varsym variables
395 		 */
396 		varsymset_clean(&p->p_varsymset);
397 	} else {
398 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
399 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
400 			p->p_flag &= ~P_SUGID;
401 	}
402 
403 	/*
404 	 * Implement correct POSIX saved-id behavior.
405 	 */
406 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
407 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
408 		cratom(&p->p_ucred);
409 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
410 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
411 	}
412 
413 	/*
414 	 * Store the vp for use in procfs
415 	 */
416 	if (p->p_textvp)		/* release old reference */
417 		vrele(p->p_textvp);
418 	p->p_textvp = imgp->vp;
419 	vref(p->p_textvp);
420 
421         /*
422          * Notify others that we exec'd, and clear the P_INEXEC flag
423          * as we're now a bona fide freshly-execed process.
424          */
425 	KNOTE(&p->p_klist, NOTE_EXEC);
426 	p->p_flag &= ~P_INEXEC;
427 
428 	/*
429 	 * If tracing the process, trap to debugger so breakpoints
430 	 * 	can be set before the program executes.
431 	 */
432 	STOPEVENT(p, S_EXEC, 0);
433 
434 	if (p->p_flag & P_TRACED)
435 		psignal(p, SIGTRAP);
436 
437 	/* clear "fork but no exec" flag, as we _are_ execing */
438 	p->p_acflag &= ~AFORK;
439 
440 	/* Set values passed into the program in registers. */
441 	setregs(p, imgp->entry_addr, (u_long)(uintptr_t)stack_base,
442 	    imgp->ps_strings);
443 
444 	/* Free any previous argument cache */
445 	if (p->p_args && --p->p_args->ar_ref == 0)
446 		FREE(p->p_args, M_PARGS);
447 	p->p_args = NULL;
448 
449 	/* Cache arguments if they fit inside our allowance */
450 	i = imgp->args->begin_envv - imgp->args->begin_argv;
451 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
452 		MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
453 		    M_PARGS, M_WAITOK);
454 		p->p_args->ar_ref = 1;
455 		p->p_args->ar_length = i;
456 		bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
457 	}
458 
459 exec_fail_dealloc:
460 
461 	/*
462 	 * free various allocated resources
463 	 */
464 	if (imgp->firstpage)
465 		exec_unmap_first_page(imgp);
466 
467 	if (imgp->vp) {
468 		vrele(imgp->vp);
469 		imgp->vp = NULL;
470 	}
471 
472 	if (error == 0) {
473 		++mycpu->gd_cnt.v_exec;
474 		return (0);
475 	}
476 
477 exec_fail:
478 	/* we're done here, clear P_INEXEC */
479 	p->p_flag &= ~P_INEXEC;
480 	if (imgp->vmspace_destroyed) {
481 		/* sorry, no more process anymore. exit gracefully */
482 		exit1(W_EXITCODE(0, SIGABRT));
483 		/* NOT REACHED */
484 		return(0);
485 	} else {
486 		return(error);
487 	}
488 }
489 
490 /*
491  * execve() system call.
492  */
493 int
494 execve(struct execve_args *uap)
495 {
496 	struct nlookupdata nd;
497 	struct image_args args;
498 	int error;
499 
500 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
501 	if (error == 0) {
502 		error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
503 					uap->argv, uap->envv);
504 	}
505 	if (error == 0)
506 		error = kern_execve(&nd, &args);
507 	nlookup_done(&nd);
508 	exec_free_args(&args);
509 
510 	/*
511 	 * The syscall result is returned in registers to the new program.
512 	 * Linux will register %edx as an atexit function and we must be
513 	 * sure to set it to 0.  XXX
514 	 */
515 	if (error == 0)
516 		uap->sysmsg_result64 = 0;
517 
518 	return (error);
519 }
520 
521 int
522 exec_map_first_page(struct image_params *imgp)
523 {
524 	int rv, i;
525 	int initial_pagein;
526 	vm_page_t ma[VM_INITIAL_PAGEIN];
527 	vm_page_t m;
528 	vm_object_t object;
529 
530 	if (imgp->firstpage)
531 		exec_unmap_first_page(imgp);
532 
533 	/*
534 	 * The file has to be mappable.
535 	 */
536 	if ((object = imgp->vp->v_object) == NULL)
537 		return (EIO);
538 
539 	/*
540 	 * We shouldn't need protection for vm_page_grab() but we certainly
541 	 * need it for the lookup loop below (lookup/busy race), since
542 	 * an interrupt can unbusy and free the page before our busy check.
543 	 */
544 	crit_enter();
545 	m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
546 
547 	if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
548 		ma[0] = m;
549 		initial_pagein = VM_INITIAL_PAGEIN;
550 		if (initial_pagein > object->size)
551 			initial_pagein = object->size;
552 		for (i = 1; i < initial_pagein; i++) {
553 			if ((m = vm_page_lookup(object, i)) != NULL) {
554 				if ((m->flags & PG_BUSY) || m->busy)
555 					break;
556 				if (m->valid)
557 					break;
558 				vm_page_busy(m);
559 			} else {
560 				m = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
561 				if (m == NULL)
562 					break;
563 			}
564 			ma[i] = m;
565 		}
566 		initial_pagein = i;
567 
568 		/*
569 		 * get_pages unbusies all the requested pages except the
570 		 * primary page (at index 0 in this case).
571 		 */
572 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
573 		m = vm_page_lookup(object, 0);
574 
575 		if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
576 			if (m) {
577 				vm_page_protect(m, VM_PROT_NONE);
578 				vm_page_free(m);
579 			}
580 			crit_exit();
581 			return EIO;
582 		}
583 	}
584 	vm_page_hold(m);
585 	vm_page_wakeup(m);	/* unbusy the page */
586 	crit_exit();
587 
588 	imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE);
589 	imgp->image_header = (void *)sf_buf_kva(imgp->firstpage);
590 
591 	return 0;
592 }
593 
594 void
595 exec_unmap_first_page(struct image_params *imgp)
596 {
597 	vm_page_t m;
598 
599 	crit_enter();
600 	if (imgp->firstpage != NULL) {
601 		m = sf_buf_page(imgp->firstpage);
602 		sf_buf_free(imgp->firstpage);
603 		imgp->firstpage = NULL;
604 		imgp->image_header = NULL;
605 		vm_page_unhold(m);
606 	}
607 	crit_exit();
608 }
609 
610 /*
611  * Destroy old address space, and allocate a new stack
612  *	The new stack is only SGROWSIZ large because it is grown
613  *	automatically in trap.c.
614  */
615 int
616 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
617 {
618 	int error;
619 	struct vmspace *vmspace = imgp->proc->p_vmspace;
620 	vm_offset_t stack_addr = USRSTACK - maxssiz;
621 	vm_map_t map;
622 
623 	imgp->vmspace_destroyed = 1;
624 
625 	/*
626 	 * Prevent a pending AIO from modifying the new address space.
627 	 */
628 	aio_proc_rundown(imgp->proc);
629 
630 	/*
631 	 * Blow away entire process VM, if address space not shared,
632 	 * otherwise, create a new VM space so that other threads are
633 	 * not disrupted.  If we are execing a resident vmspace we
634 	 * create a duplicate of it and remap the stack.
635 	 *
636 	 * The exitingcnt test is not strictly necessary but has been
637 	 * included for code sanity (to make the code more deterministic).
638 	 */
639 	map = &vmspace->vm_map;
640 	if (vmcopy) {
641 		vmspace_exec(imgp->proc, vmcopy);
642 		vmspace = imgp->proc->p_vmspace;
643 		pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
644 		map = &vmspace->vm_map;
645 	} else if (vmspace->vm_refcnt == 1 && vmspace->vm_exitingcnt == 0) {
646 		shmexit(vmspace);
647 		if (vmspace->vm_upcalls)
648 			upc_release(vmspace, &imgp->proc->p_lwp);
649 		pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS);
650 		vm_map_remove(map, 0, VM_MAXUSER_ADDRESS);
651 	} else {
652 		vmspace_exec(imgp->proc, NULL);
653 		vmspace = imgp->proc->p_vmspace;
654 		map = &vmspace->vm_map;
655 	}
656 
657 	/* Allocate a new stack */
658 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
659 	    VM_PROT_ALL, VM_PROT_ALL, 0);
660 	if (error)
661 		return (error);
662 
663 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
664 	 * VM_STACK case, but they are still used to monitor the size of the
665 	 * process stack so we can check the stack rlimit.
666 	 */
667 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
668 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
669 
670 	return(0);
671 }
672 
673 /*
674  * Copy out argument and environment strings from the old process
675  *	address space into the temporary string buffer.
676  */
677 int
678 exec_copyin_args(struct image_args *args, char *fname,
679 		enum exec_path_segflg segflg, char **argv, char **envv)
680 {
681 	char	*argp, *envp;
682 	int	error = 0;
683 	size_t	length;
684 
685 	bzero(args, sizeof(*args));
686 	args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
687 	if (args->buf == NULL)
688 		return (ENOMEM);
689 	args->begin_argv = args->buf;
690 	args->endp = args->begin_argv;
691 	args->space = ARG_MAX;
692 
693 	args->fname = args->buf + ARG_MAX;
694 
695 	/*
696 	 * Copy the file name.
697 	 */
698 	if (segflg == PATH_SYSSPACE) {
699 		error = copystr(fname, args->fname, PATH_MAX, &length);
700 	} else if (segflg == PATH_USERSPACE) {
701 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
702 	}
703 
704 	/*
705 	 * Extract argument strings.  argv may not be NULL.  The argv
706 	 * array is terminated by a NULL entry.  We special-case the
707 	 * situation where argv[0] is NULL by passing { filename, NULL }
708 	 * to the new program to guarentee that the interpreter knows what
709 	 * file to open in case we exec an interpreted file.   Note that
710 	 * a NULL argv[0] terminates the argv[] array.
711 	 *
712 	 * XXX the special-casing of argv[0] is historical and needs to be
713 	 * revisited.
714 	 */
715 	if (argv == NULL)
716 		error = EFAULT;
717 	if (error == 0) {
718 		while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
719 			if (argp == (caddr_t)-1) {
720 				error = EFAULT;
721 				break;
722 			}
723 			error = copyinstr(argp, args->endp,
724 					    args->space, &length);
725 			if (error) {
726 				if (error == ENAMETOOLONG)
727 					error = E2BIG;
728 				break;
729 			}
730 			args->space -= length;
731 			args->endp += length;
732 			args->argc++;
733 		}
734 		if (args->argc == 0 && error == 0) {
735 			length = strlen(args->fname) + 1;
736 			if (length > args->space) {
737 				error = E2BIG;
738 			} else {
739 				bcopy(args->fname, args->endp, length);
740 				args->space -= length;
741 				args->endp += length;
742 				args->argc++;
743 			}
744 		}
745 	}
746 
747 	args->begin_envv = args->endp;
748 
749 	/*
750 	 * extract environment strings.  envv may be NULL.
751 	 */
752 	if (envv && error == 0) {
753 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
754 			if (envp == (caddr_t) -1) {
755 				error = EFAULT;
756 				break;
757 			}
758 			error = copyinstr(envp, args->endp, args->space,
759 			    &length);
760 			if (error) {
761 				if (error == ENAMETOOLONG)
762 					error = E2BIG;
763 				break;
764 			}
765 			args->space -= length;
766 			args->endp += length;
767 			args->envc++;
768 		}
769 	}
770 	return (error);
771 }
772 
773 void
774 exec_free_args(struct image_args *args)
775 {
776 	if (args->buf) {
777 		kmem_free_wakeup(exec_map,
778 				(vm_offset_t)args->buf, PATH_MAX + ARG_MAX);
779 		args->buf = NULL;
780 	}
781 }
782 
783 /*
784  * Copy strings out to the new process address space, constructing
785  *	new arg and env vector tables. Return a pointer to the base
786  *	so that it can be used as the initial stack pointer.
787  */
788 register_t *
789 exec_copyout_strings(struct image_params *imgp)
790 {
791 	int argc, envc, sgap;
792 	char **vectp;
793 	char *stringp, *destp;
794 	register_t *stack_base;
795 	struct ps_strings *arginfo;
796 	int szsigcode;
797 
798 	/*
799 	 * Calculate string base and vector table pointers.
800 	 * Also deal with signal trampoline code for this exec type.
801 	 */
802 	arginfo = (struct ps_strings *)PS_STRINGS;
803 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
804 	if (stackgap_random != 0)
805 		sgap = ALIGN(arc4random() & (stackgap_random - 1));
806 	else
807 		sgap = 0;
808 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
809 	    roundup((ARG_MAX - imgp->args->space), sizeof(char *));
810 
811 	/*
812 	 * install sigcode
813 	 */
814 	if (szsigcode)
815 		copyout(imgp->proc->p_sysent->sv_sigcode,
816 		    ((caddr_t)arginfo - szsigcode), szsigcode);
817 
818 	/*
819 	 * If we have a valid auxargs ptr, prepare some room
820 	 * on the stack.
821 	 *
822 	 * The '+ 2' is for the null pointers at the end of each of the
823 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
824 	 * ELF Auxargs data.
825 	 */
826 	if (imgp->auxargs) {
827 		vectp = (char **)(destp - (imgp->args->argc +
828 			imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
829 	} else {
830 		vectp = (char **)(destp - (imgp->args->argc +
831 			imgp->args->envc + 2) * sizeof(char*));
832 	}
833 
834 	/*
835 	 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
836 	 * be done in crt1.o.  Note that GCC 3.x aligns the stack in main.
837 	 */
838 
839 	/*
840 	 * vectp also becomes our initial stack base
841 	 */
842 	stack_base = (register_t *)vectp;
843 
844 	stringp = imgp->args->begin_argv;
845 	argc = imgp->args->argc;
846 	envc = imgp->args->envc;
847 
848 	/*
849 	 * Copy out strings - arguments and environment.
850 	 */
851 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
852 
853 	/*
854 	 * Fill in "ps_strings" struct for ps, w, etc.
855 	 */
856 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
857 	suword(&arginfo->ps_nargvstr, argc);
858 
859 	/*
860 	 * Fill in argument portion of vector table.
861 	 */
862 	for (; argc > 0; --argc) {
863 		suword(vectp++, (long)(intptr_t)destp);
864 		while (*stringp++ != 0)
865 			destp++;
866 		destp++;
867 	}
868 
869 	/* a null vector table pointer separates the argp's from the envp's */
870 	suword(vectp++, 0);
871 
872 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
873 	suword(&arginfo->ps_nenvstr, envc);
874 
875 	/*
876 	 * Fill in environment portion of vector table.
877 	 */
878 	for (; envc > 0; --envc) {
879 		suword(vectp++, (long)(intptr_t)destp);
880 		while (*stringp++ != 0)
881 			destp++;
882 		destp++;
883 	}
884 
885 	/* end of vector table is a null pointer */
886 	suword(vectp, 0);
887 
888 	return (stack_base);
889 }
890 
891 /*
892  * Check permissions of file to execute.
893  *	Return 0 for success or error code on failure.
894  */
895 int
896 exec_check_permissions(struct image_params *imgp)
897 {
898 	struct proc *p = imgp->proc;
899 	struct vnode *vp = imgp->vp;
900 	struct vattr *attr = imgp->attr;
901 	struct thread *td = p->p_thread;
902 	int error;
903 
904 	/* Get file attributes */
905 	error = VOP_GETATTR(vp, attr, td);
906 	if (error)
907 		return (error);
908 
909 	/*
910 	 * 1) Check if file execution is disabled for the filesystem that this
911 	 *	file resides on.
912 	 * 2) Insure that at least one execute bit is on - otherwise root
913 	 *	will always succeed, and we don't want to happen unless the
914 	 *	file really is executable.
915 	 * 3) Insure that the file is a regular file.
916 	 */
917 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
918 	    ((attr->va_mode & 0111) == 0) ||
919 	    (attr->va_type != VREG)) {
920 		return (EACCES);
921 	}
922 
923 	/*
924 	 * Zero length files can't be exec'd
925 	 */
926 	if (attr->va_size == 0)
927 		return (ENOEXEC);
928 
929 	/*
930 	 *  Check for execute permission to file based on current credentials.
931 	 */
932 	error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td);
933 	if (error)
934 		return (error);
935 
936 	/*
937 	 * Check number of open-for-writes on the file and deny execution
938 	 * if there are any.
939 	 */
940 	if (vp->v_writecount)
941 		return (ETXTBSY);
942 
943 	/*
944 	 * Call filesystem specific open routine, which allows us to read,
945 	 * write, and mmap the file.  Without the VOP_OPEN we can only
946 	 * stat the file.
947 	 */
948 	error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL, td);
949 	if (error)
950 		return (error);
951 
952 	return (0);
953 }
954 
955 /*
956  * Exec handler registration
957  */
958 int
959 exec_register(const struct execsw *execsw_arg)
960 {
961 	const struct execsw **es, **xs, **newexecsw;
962 	int count = 2;	/* New slot and trailing NULL */
963 
964 	if (execsw)
965 		for (es = execsw; *es; es++)
966 			count++;
967 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
968 	if (newexecsw == NULL)
969 		return ENOMEM;
970 	xs = newexecsw;
971 	if (execsw)
972 		for (es = execsw; *es; es++)
973 			*xs++ = *es;
974 	*xs++ = execsw_arg;
975 	*xs = NULL;
976 	if (execsw)
977 		free(execsw, M_TEMP);
978 	execsw = newexecsw;
979 	return 0;
980 }
981 
982 int
983 exec_unregister(const struct execsw *execsw_arg)
984 {
985 	const struct execsw **es, **xs, **newexecsw;
986 	int count = 1;
987 
988 	if (execsw == NULL)
989 		panic("unregister with no handlers left?");
990 
991 	for (es = execsw; *es; es++) {
992 		if (*es == execsw_arg)
993 			break;
994 	}
995 	if (*es == NULL)
996 		return ENOENT;
997 	for (es = execsw; *es; es++)
998 		if (*es != execsw_arg)
999 			count++;
1000 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1001 	if (newexecsw == NULL)
1002 		return ENOMEM;
1003 	xs = newexecsw;
1004 	for (es = execsw; *es; es++)
1005 		if (*es != execsw_arg)
1006 			*xs++ = *es;
1007 	*xs = NULL;
1008 	if (execsw)
1009 		free(execsw, M_TEMP);
1010 	execsw = newexecsw;
1011 	return 0;
1012 }
1013