xref: /dragonfly/sys/kern/kern_exec.c (revision 73e0051e)
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 1.64 2008/10/26 04:29:19 sephe 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/priv.h>
46 #include <sys/ktrace.h>
47 #include <sys/signalvar.h>
48 #include <sys/pioctl.h>
49 #include <sys/nlookup.h>
50 #include <sys/sfbuf.h>
51 #include <sys/sysent.h>
52 #include <sys/shm.h>
53 #include <sys/sysctl.h>
54 #include <sys/vnode.h>
55 #include <sys/vmmeter.h>
56 #include <sys/aio.h>
57 #include <sys/libkern.h>
58 
59 #include <vm/vm.h>
60 #include <vm/vm_param.h>
61 #include <sys/lock.h>
62 #include <vm/pmap.h>
63 #include <vm/vm_page.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_kern.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
68 #include <vm/vnode_pager.h>
69 #include <vm/vm_pager.h>
70 
71 #include <sys/user.h>
72 #include <sys/reg.h>
73 
74 #include <sys/thread2.h>
75 
76 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
77 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
78 
79 static register_t *exec_copyout_strings (struct image_params *);
80 
81 /* XXX This should be vm_size_t. */
82 static u_long ps_strings = PS_STRINGS;
83 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
84 
85 /* XXX This should be vm_size_t. */
86 static u_long usrstack = USRSTACK;
87 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
88 
89 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
90 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
91     &ps_arg_cache_limit, 0, "");
92 
93 int ps_argsopen = 1;
94 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
95 
96 static int ktrace_suid = 0;
97 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
98 
99 void print_execve_args(struct image_args *args);
100 int debug_execve_args = 0;
101 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
102     0, "");
103 
104 /*
105  * Exec arguments object cache
106  */
107 static struct objcache *exec_objcache;
108 
109 static
110 void
111 exec_objcache_init(void *arg __unused)
112 {
113 	int cluster_limit;
114 
115 	cluster_limit = 16;	/* up to this many objects */
116 	exec_objcache = objcache_create_mbacked(
117 					M_EXECARGS, PATH_MAX + ARG_MAX,
118 					&cluster_limit,
119 					2,	/* minimal magazine capacity */
120 					NULL, NULL, NULL);
121 }
122 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
123 
124 /*
125  * stackgap_random specifies if the stackgap should have a random size added
126  * to it.  It must be a power of 2.  If non-zero, the stack gap will be
127  * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
128  */
129 static int stackgap_random = 1024;
130 static int
131 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
132 {
133 	int error, new_val;
134 	new_val = stackgap_random;
135 	error = sysctl_handle_int(oidp, &new_val, 0, req);
136 	if (error != 0 || req->newptr == NULL)
137 		return (error);
138 	if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
139 		return (EINVAL);
140 	stackgap_random = new_val;
141 
142 	return(0);
143 }
144 
145 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
146 	0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
147 
148 void
149 print_execve_args(struct image_args *args)
150 {
151 	char *cp;
152 	int ndx;
153 
154 	cp = args->begin_argv;
155 	for (ndx = 0; ndx < args->argc; ndx++) {
156 		kprintf("\targv[%d]: %s\n", ndx, cp);
157 		while (*cp++ != '\0');
158 	}
159 	for (ndx = 0; ndx < args->envc; ndx++) {
160 		kprintf("\tenvv[%d]: %s\n", ndx, cp);
161 		while (*cp++ != '\0');
162 	}
163 }
164 
165 /*
166  * Each of the items is a pointer to a `const struct execsw', hence the
167  * double pointer here.
168  */
169 static const struct execsw **execsw;
170 
171 /*
172  * Replace current vmspace with a new binary.
173  * Returns 0 on success, > 0 on recoverable error (use as errno).
174  * Returns -1 on lethal error which demands killing of the current
175  * process!
176  */
177 int
178 kern_execve(struct nlookupdata *nd, struct image_args *args)
179 {
180 	struct thread *td = curthread;
181 	struct lwp *lp = td->td_lwp;
182 	struct proc *p = td->td_proc;
183 	register_t *stack_base;
184 	int error, len, i;
185 	struct image_params image_params, *imgp;
186 	struct vattr attr;
187 	int (*img_first) (struct image_params *);
188 
189 	if (debug_execve_args) {
190 		kprintf("%s()\n", __func__);
191 		print_execve_args(args);
192 	}
193 
194 	KKASSERT(p);
195 	imgp = &image_params;
196 
197 	/*
198 	 * NOTE: P_INEXEC is handled by exec_new_vmspace() now.  We make
199 	 * no modifications to the process at all until we get there.
200 	 *
201 	 * Note that multiple threads may be trying to exec at the same
202 	 * time.  exec_new_vmspace() handles that too.
203 	 */
204 
205 	/*
206 	 * Initialize part of the common data
207 	 */
208 	imgp->proc = p;
209 	imgp->args = args;
210 	imgp->attr = &attr;
211 	imgp->entry_addr = 0;
212 	imgp->resident = 0;
213 	imgp->vmspace_destroyed = 0;
214 	imgp->interpreted = 0;
215 	imgp->interpreter_name[0] = 0;
216 	imgp->auxargs = NULL;
217 	imgp->vp = NULL;
218 	imgp->firstpage = NULL;
219 	imgp->ps_strings = 0;
220 	imgp->image_header = NULL;
221 
222 interpret:
223 
224 	/*
225 	 * Translate the file name to a vnode.  Unlock the cache entry to
226 	 * improve parallelism for programs exec'd in parallel.
227 	 */
228 	if ((error = nlookup(nd)) != 0)
229 		goto exec_fail;
230 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
231 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
232 	nd->nl_flags &= ~NLC_NCPISLOCKED;
233 	cache_unlock(&nd->nl_nch);
234 	if (error)
235 		goto exec_fail;
236 
237 	/*
238 	 * Check file permissions (also 'opens' file)
239 	 */
240 	error = exec_check_permissions(imgp);
241 	if (error) {
242 		vn_unlock(imgp->vp);
243 		goto exec_fail_dealloc;
244 	}
245 
246 	error = exec_map_first_page(imgp);
247 	vn_unlock(imgp->vp);
248 	if (error)
249 		goto exec_fail_dealloc;
250 
251 	if (debug_execve_args && imgp->interpreted) {
252 		kprintf("    target is interpreted -- recursive pass\n");
253 		kprintf("    interpreter: %s\n", imgp->interpreter_name);
254 		print_execve_args(args);
255 	}
256 
257 	/*
258 	 *	If the current process has a special image activator it
259 	 *	wants to try first, call it.   For example, emulating shell
260 	 *	scripts differently.
261 	 */
262 	error = -1;
263 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
264 		error = img_first(imgp);
265 
266 	/*
267 	 *	If the vnode has a registered vmspace, exec the vmspace
268 	 */
269 	if (error == -1 && imgp->vp->v_resident) {
270 		error = exec_resident_imgact(imgp);
271 	}
272 
273 	/*
274 	 *	Loop through the list of image activators, calling each one.
275 	 *	An activator returns -1 if there is no match, 0 on success,
276 	 *	and an error otherwise.
277 	 */
278 	for (i = 0; error == -1 && execsw[i]; ++i) {
279 		if (execsw[i]->ex_imgact == NULL ||
280 		    execsw[i]->ex_imgact == img_first) {
281 			continue;
282 		}
283 		error = (*execsw[i]->ex_imgact)(imgp);
284 	}
285 
286 	if (error) {
287 		if (error == -1)
288 			error = ENOEXEC;
289 		goto exec_fail_dealloc;
290 	}
291 
292 	/*
293 	 * Special interpreter operation, cleanup and loop up to try to
294 	 * activate the interpreter.
295 	 */
296 	if (imgp->interpreted) {
297 		exec_unmap_first_page(imgp);
298 		nlookup_done(nd);
299 		vrele(imgp->vp);
300 		imgp->vp = NULL;
301 		error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
302 					NLC_FOLLOW);
303 		if (error)
304 			goto exec_fail;
305 		goto interpret;
306 	}
307 
308 	/*
309 	 * Copy out strings (args and env) and initialize stack base
310 	 */
311 	stack_base = exec_copyout_strings(imgp);
312 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
313 
314 	/*
315 	 * If custom stack fixup routine present for this process
316 	 * let it do the stack setup.  If we are running a resident
317 	 * image there is no auxinfo or other image activator context
318 	 * so don't try to add fixups to the stack.
319 	 *
320 	 * Else stuff argument count as first item on stack
321 	 */
322 	if (p->p_sysent->sv_fixup && imgp->resident == 0)
323 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
324 	else
325 		suword(--stack_base, imgp->args->argc);
326 
327 	/*
328 	 * For security and other reasons, the file descriptor table cannot
329 	 * be shared after an exec.
330 	 */
331 	if (p->p_fd->fd_refcnt > 1) {
332 		struct filedesc *tmp;
333 
334 		tmp = fdcopy(p);
335 		fdfree(p, tmp);
336 	}
337 
338 	/*
339 	 * For security and other reasons, signal handlers cannot
340 	 * be shared after an exec. The new proces gets a copy of the old
341 	 * handlers. In execsigs(), the new process will have its signals
342 	 * reset.
343 	 */
344 	if (p->p_sigacts->ps_refcnt > 1) {
345 		struct sigacts *newsigacts;
346 
347 		newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts),
348 		       M_SUBPROC, M_WAITOK);
349 		bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts));
350 		p->p_sigacts->ps_refcnt--;
351 		p->p_sigacts = newsigacts;
352 		p->p_sigacts->ps_refcnt = 1;
353 	}
354 
355 	/*
356 	 * For security and other reasons virtual kernels cannot be
357 	 * inherited by an exec.  This also allows a virtual kernel
358 	 * to fork/exec unrelated applications.
359 	 */
360 	if (p->p_vkernel)
361 		vkernel_exit(p);
362 
363 	/* Stop profiling */
364 	stopprofclock(p);
365 
366 	/* close files on exec */
367 	fdcloseexec(p);
368 
369 	/* reset caught signals */
370 	execsigs(p);
371 
372 	/* name this process - nameiexec(p, ndp) */
373 	len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
374 	bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
375 	p->p_comm[len] = 0;
376 	bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
377 
378 	/*
379 	 * mark as execed, wakeup the process that vforked (if any) and tell
380 	 * it that it now has its own resources back
381 	 */
382 	p->p_flag |= P_EXEC;
383 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
384 		p->p_flag &= ~P_PPWAIT;
385 		wakeup((caddr_t)p->p_pptr);
386 	}
387 
388 	/*
389 	 * Implement image setuid/setgid.
390 	 *
391 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
392 	 * the process is being traced.
393 	 */
394 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
395 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
396 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
397 	    (p->p_flag & P_TRACED) == 0) {
398 		/*
399 		 * Turn off syscall tracing for set-id programs, except for
400 		 * root.  Record any set-id flags first to make sure that
401 		 * we do not regain any tracing during a possible block.
402 		 */
403 		setsugid();
404 		if (p->p_tracenode && ktrace_suid == 0 &&
405 		    priv_check(td, PRIV_ROOT) != 0) {
406 			ktrdestroy(&p->p_tracenode);
407 			p->p_traceflag = 0;
408 		}
409 		/* Close any file descriptors 0..2 that reference procfs */
410 		setugidsafety(p);
411 		/* Make sure file descriptors 0..2 are in use. */
412 		error = fdcheckstd(p);
413 		if (error != 0)
414 			goto exec_fail_dealloc;
415 		/*
416 		 * Set the new credentials.
417 		 */
418 		cratom(&p->p_ucred);
419 		if (attr.va_mode & VSUID)
420 			change_euid(attr.va_uid);
421 		if (attr.va_mode & VSGID)
422 			p->p_ucred->cr_gid = attr.va_gid;
423 
424 		/*
425 		 * Clear local varsym variables
426 		 */
427 		varsymset_clean(&p->p_varsymset);
428 	} else {
429 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
430 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
431 			p->p_flag &= ~P_SUGID;
432 	}
433 
434 	/*
435 	 * Implement correct POSIX saved-id behavior.
436 	 */
437 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
438 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
439 		cratom(&p->p_ucred);
440 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
441 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
442 	}
443 
444 	/*
445 	 * Store the vp for use in procfs
446 	 */
447 	if (p->p_textvp)		/* release old reference */
448 		vrele(p->p_textvp);
449 	p->p_textvp = imgp->vp;
450 	vref(p->p_textvp);
451 
452         /*
453          * Notify others that we exec'd, and clear the P_INEXEC flag
454          * as we're now a bona fide freshly-execed process.
455          */
456 	KNOTE(&p->p_klist, NOTE_EXEC);
457 	p->p_flag &= ~P_INEXEC;
458 
459 	/*
460 	 * If tracing the process, trap to debugger so breakpoints
461 	 * 	can be set before the program executes.
462 	 */
463 	STOPEVENT(p, S_EXEC, 0);
464 
465 	if (p->p_flag & P_TRACED)
466 		ksignal(p, SIGTRAP);
467 
468 	/* clear "fork but no exec" flag, as we _are_ execing */
469 	p->p_acflag &= ~AFORK;
470 
471 	/* Set values passed into the program in registers. */
472 	exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
473 	    imgp->ps_strings);
474 
475 	/* Set the access time on the vnode */
476 	vn_mark_atime(imgp->vp, td);
477 
478 	/* Free any previous argument cache */
479 	if (p->p_args && --p->p_args->ar_ref == 0)
480 		FREE(p->p_args, M_PARGS);
481 	p->p_args = NULL;
482 
483 	/* Cache arguments if they fit inside our allowance */
484 	i = imgp->args->begin_envv - imgp->args->begin_argv;
485 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
486 		MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
487 		    M_PARGS, M_WAITOK);
488 		p->p_args->ar_ref = 1;
489 		p->p_args->ar_length = i;
490 		bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
491 	}
492 
493 exec_fail_dealloc:
494 
495 	/*
496 	 * free various allocated resources
497 	 */
498 	if (imgp->firstpage)
499 		exec_unmap_first_page(imgp);
500 
501 	if (imgp->vp) {
502 		vrele(imgp->vp);
503 		imgp->vp = NULL;
504 	}
505 
506 	if (error == 0) {
507 		++mycpu->gd_cnt.v_exec;
508 		return (0);
509 	}
510 
511 exec_fail:
512 	/*
513 	 * we're done here, clear P_INEXEC if we were the ones that
514 	 * set it.  Otherwise if vmspace_destroyed is still set we
515 	 * raced another thread and that thread is responsible for
516 	 * clearing it.
517 	 */
518 	if (imgp->vmspace_destroyed & 2)
519 		p->p_flag &= ~P_INEXEC;
520 	if (imgp->vmspace_destroyed) {
521 		/*
522 		 * Sorry, no more process anymore. exit gracefully.
523 		 * However we can't die right here, because our
524 		 * caller might have to clean up, so indicate a
525 		 * lethal error by returning -1.
526 		 */
527 		return(-1);
528 	} else {
529 		return(error);
530 	}
531 }
532 
533 /*
534  * execve() system call.
535  */
536 int
537 sys_execve(struct execve_args *uap)
538 {
539 	struct nlookupdata nd;
540 	struct image_args args;
541 	int error;
542 
543 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
544 	bzero(&args, sizeof(args));
545 	if (error == 0) {
546 		error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
547 					uap->argv, uap->envv);
548 	}
549 	if (error == 0)
550 		error = kern_execve(&nd, &args);
551 	nlookup_done(&nd);
552 	exec_free_args(&args);
553 
554 	if (error < 0) {
555 		/* We hit a lethal error condition.  Let's die now. */
556 		exit1(W_EXITCODE(0, SIGABRT));
557 		/* NOTREACHED */
558 	}
559 
560 	/*
561 	 * The syscall result is returned in registers to the new program.
562 	 * Linux will register %edx as an atexit function and we must be
563 	 * sure to set it to 0.  XXX
564 	 */
565 	if (error == 0)
566 		uap->sysmsg_result64 = 0;
567 
568 	return (error);
569 }
570 
571 int
572 exec_map_first_page(struct image_params *imgp)
573 {
574 	int rv, i;
575 	int initial_pagein;
576 	vm_page_t ma[VM_INITIAL_PAGEIN];
577 	vm_page_t m;
578 	vm_object_t object;
579 
580 	if (imgp->firstpage)
581 		exec_unmap_first_page(imgp);
582 
583 	/*
584 	 * The file has to be mappable.
585 	 */
586 	if ((object = imgp->vp->v_object) == NULL)
587 		return (EIO);
588 
589 	/*
590 	 * We shouldn't need protection for vm_page_grab() but we certainly
591 	 * need it for the lookup loop below (lookup/busy race), since
592 	 * an interrupt can unbusy and free the page before our busy check.
593 	 */
594 	crit_enter();
595 	m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
596 
597 	if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
598 		ma[0] = m;
599 		initial_pagein = VM_INITIAL_PAGEIN;
600 		if (initial_pagein > object->size)
601 			initial_pagein = object->size;
602 		for (i = 1; i < initial_pagein; i++) {
603 			if ((m = vm_page_lookup(object, i)) != NULL) {
604 				if ((m->flags & PG_BUSY) || m->busy)
605 					break;
606 				if (m->valid)
607 					break;
608 				vm_page_busy(m);
609 			} else {
610 				m = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
611 				if (m == NULL)
612 					break;
613 			}
614 			ma[i] = m;
615 		}
616 		initial_pagein = i;
617 
618 		/*
619 		 * get_pages unbusies all the requested pages except the
620 		 * primary page (at index 0 in this case).  The primary
621 		 * page may have been wired during the pagein (e.g. by
622 		 * the buffer cache) so vnode_pager_freepage() must be
623 		 * used to properly release it.
624 		 */
625 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
626 		m = vm_page_lookup(object, 0);
627 
628 		if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
629 			if (m) {
630 				vm_page_protect(m, VM_PROT_NONE);
631 				vnode_pager_freepage(m);
632 			}
633 			crit_exit();
634 			return EIO;
635 		}
636 	}
637 	vm_page_hold(m);
638 	vm_page_wakeup(m);	/* unbusy the page */
639 	crit_exit();
640 
641 	imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE);
642 	imgp->image_header = (void *)sf_buf_kva(imgp->firstpage);
643 
644 	return 0;
645 }
646 
647 void
648 exec_unmap_first_page(struct image_params *imgp)
649 {
650 	vm_page_t m;
651 
652 	crit_enter();
653 	if (imgp->firstpage != NULL) {
654 		m = sf_buf_page(imgp->firstpage);
655 		sf_buf_free(imgp->firstpage);
656 		imgp->firstpage = NULL;
657 		imgp->image_header = NULL;
658 		vm_page_unhold(m);
659 	}
660 	crit_exit();
661 }
662 
663 /*
664  * Destroy old address space, and allocate a new stack
665  *	The new stack is only SGROWSIZ large because it is grown
666  *	automatically in trap.c.
667  *
668  * This is the point of no return.
669  */
670 int
671 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
672 {
673 	struct vmspace *vmspace = imgp->proc->p_vmspace;
674 	vm_offset_t stack_addr = USRSTACK - maxssiz;
675 	struct proc *p;
676 	vm_map_t map;
677 	int error;
678 
679 	/*
680 	 * Indicate that we cannot gracefully error out any more, kill
681 	 * any other threads present, and set P_INEXEC to indicate that
682 	 * we are now messing with the process structure proper.
683 	 *
684 	 * If killalllwps() races return an error which coupled with
685 	 * vmspace_destroyed will cause us to exit.  This is what we
686 	 * want since another thread is patiently waiting for us to exit
687 	 * in that case.
688 	 */
689 	p = curproc;
690 	imgp->vmspace_destroyed = 1;
691 
692 	if (curthread->td_proc->p_nthreads > 1) {
693 		error = killalllwps(1);
694 		if (error)
695 			return (error);
696 	}
697 	imgp->vmspace_destroyed |= 2;	/* we are responsible for P_INEXEC */
698 	p->p_flag |= P_INEXEC;
699 
700 	/*
701 	 * Prevent a pending AIO from modifying the new address space.
702 	 */
703 	aio_proc_rundown(imgp->proc);
704 
705 	/*
706 	 * Blow away entire process VM, if address space not shared,
707 	 * otherwise, create a new VM space so that other threads are
708 	 * not disrupted.  If we are execing a resident vmspace we
709 	 * create a duplicate of it and remap the stack.
710 	 *
711 	 * The exitingcnt test is not strictly necessary but has been
712 	 * included for code sanity (to make the code more deterministic).
713 	 */
714 	map = &vmspace->vm_map;
715 	if (vmcopy) {
716 		vmspace_exec(imgp->proc, vmcopy);
717 		vmspace = imgp->proc->p_vmspace;
718 		pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
719 		map = &vmspace->vm_map;
720 	} else if (vmspace->vm_sysref.refcnt == 1 &&
721 		   vmspace->vm_exitingcnt == 0) {
722 		shmexit(vmspace);
723 		if (vmspace->vm_upcalls)
724 			upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
725 		pmap_remove_pages(vmspace_pmap(vmspace),
726 			0, VM_MAX_USER_ADDRESS);
727 		vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
728 	} else {
729 		vmspace_exec(imgp->proc, NULL);
730 		vmspace = imgp->proc->p_vmspace;
731 		map = &vmspace->vm_map;
732 	}
733 
734 	/* Allocate a new stack */
735 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
736 			     0, VM_PROT_ALL, VM_PROT_ALL, 0);
737 	if (error)
738 		return (error);
739 
740 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
741 	 * VM_STACK case, but they are still used to monitor the size of the
742 	 * process stack so we can check the stack rlimit.
743 	 */
744 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
745 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
746 
747 	return(0);
748 }
749 
750 /*
751  * Copy out argument and environment strings from the old process
752  *	address space into the temporary string buffer.
753  */
754 int
755 exec_copyin_args(struct image_args *args, char *fname,
756 		enum exec_path_segflg segflg, char **argv, char **envv)
757 {
758 	char	*argp, *envp;
759 	int	error = 0;
760 	size_t	length;
761 
762 	args->buf = objcache_get(exec_objcache, M_WAITOK);
763 	if (args->buf == NULL)
764 		return (ENOMEM);
765 	args->begin_argv = args->buf;
766 	args->endp = args->begin_argv;
767 	args->space = ARG_MAX;
768 
769 	args->fname = args->buf + ARG_MAX;
770 
771 	/*
772 	 * Copy the file name.
773 	 */
774 	if (segflg == PATH_SYSSPACE) {
775 		error = copystr(fname, args->fname, PATH_MAX, &length);
776 	} else if (segflg == PATH_USERSPACE) {
777 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
778 	}
779 
780 	/*
781 	 * Extract argument strings.  argv may not be NULL.  The argv
782 	 * array is terminated by a NULL entry.  We special-case the
783 	 * situation where argv[0] is NULL by passing { filename, NULL }
784 	 * to the new program to guarentee that the interpreter knows what
785 	 * file to open in case we exec an interpreted file.   Note that
786 	 * a NULL argv[0] terminates the argv[] array.
787 	 *
788 	 * XXX the special-casing of argv[0] is historical and needs to be
789 	 * revisited.
790 	 */
791 	if (argv == NULL)
792 		error = EFAULT;
793 	if (error == 0) {
794 		while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
795 			if (argp == (caddr_t)-1) {
796 				error = EFAULT;
797 				break;
798 			}
799 			error = copyinstr(argp, args->endp,
800 					    args->space, &length);
801 			if (error) {
802 				if (error == ENAMETOOLONG)
803 					error = E2BIG;
804 				break;
805 			}
806 			args->space -= length;
807 			args->endp += length;
808 			args->argc++;
809 		}
810 		if (args->argc == 0 && error == 0) {
811 			length = strlen(args->fname) + 1;
812 			if (length > args->space) {
813 				error = E2BIG;
814 			} else {
815 				bcopy(args->fname, args->endp, length);
816 				args->space -= length;
817 				args->endp += length;
818 				args->argc++;
819 			}
820 		}
821 	}
822 
823 	args->begin_envv = args->endp;
824 
825 	/*
826 	 * extract environment strings.  envv may be NULL.
827 	 */
828 	if (envv && error == 0) {
829 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
830 			if (envp == (caddr_t) -1) {
831 				error = EFAULT;
832 				break;
833 			}
834 			error = copyinstr(envp, args->endp, args->space,
835 			    &length);
836 			if (error) {
837 				if (error == ENAMETOOLONG)
838 					error = E2BIG;
839 				break;
840 			}
841 			args->space -= length;
842 			args->endp += length;
843 			args->envc++;
844 		}
845 	}
846 	return (error);
847 }
848 
849 void
850 exec_free_args(struct image_args *args)
851 {
852 	if (args->buf) {
853 		objcache_put(exec_objcache, args->buf);
854 		args->buf = NULL;
855 	}
856 }
857 
858 /*
859  * Copy strings out to the new process address space, constructing
860  *	new arg and env vector tables. Return a pointer to the base
861  *	so that it can be used as the initial stack pointer.
862  */
863 register_t *
864 exec_copyout_strings(struct image_params *imgp)
865 {
866 	int argc, envc, sgap;
867 	char **vectp;
868 	char *stringp, *destp;
869 	register_t *stack_base;
870 	struct ps_strings *arginfo;
871 	int szsigcode;
872 
873 	/*
874 	 * Calculate string base and vector table pointers.
875 	 * Also deal with signal trampoline code for this exec type.
876 	 */
877 	arginfo = (struct ps_strings *)PS_STRINGS;
878 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
879 	if (stackgap_random != 0)
880 		sgap = ALIGN(karc4random() & (stackgap_random - 1));
881 	else
882 		sgap = 0;
883 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
884 	    roundup((ARG_MAX - imgp->args->space), sizeof(char *));
885 
886 	/*
887 	 * install sigcode
888 	 */
889 	if (szsigcode)
890 		copyout(imgp->proc->p_sysent->sv_sigcode,
891 		    ((caddr_t)arginfo - szsigcode), szsigcode);
892 
893 	/*
894 	 * If we have a valid auxargs ptr, prepare some room
895 	 * on the stack.
896 	 *
897 	 * The '+ 2' is for the null pointers at the end of each of the
898 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
899 	 * ELF Auxargs data.
900 	 */
901 	if (imgp->auxargs) {
902 		vectp = (char **)(destp - (imgp->args->argc +
903 			imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
904 	} else {
905 		vectp = (char **)(destp - (imgp->args->argc +
906 			imgp->args->envc + 2) * sizeof(char*));
907 	}
908 
909 	/*
910 	 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
911 	 * be done in crt1.o.  Note that GCC 3.x aligns the stack in main.
912 	 */
913 
914 	/*
915 	 * vectp also becomes our initial stack base
916 	 */
917 	stack_base = (register_t *)vectp;
918 
919 	stringp = imgp->args->begin_argv;
920 	argc = imgp->args->argc;
921 	envc = imgp->args->envc;
922 
923 	/*
924 	 * Copy out strings - arguments and environment.
925 	 */
926 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
927 
928 	/*
929 	 * Fill in "ps_strings" struct for ps, w, etc.
930 	 */
931 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
932 	suword(&arginfo->ps_nargvstr, argc);
933 
934 	/*
935 	 * Fill in argument portion of vector table.
936 	 */
937 	for (; argc > 0; --argc) {
938 		suword(vectp++, (long)(intptr_t)destp);
939 		while (*stringp++ != 0)
940 			destp++;
941 		destp++;
942 	}
943 
944 	/* a null vector table pointer separates the argp's from the envp's */
945 	suword(vectp++, 0);
946 
947 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
948 	suword(&arginfo->ps_nenvstr, envc);
949 
950 	/*
951 	 * Fill in environment portion of vector table.
952 	 */
953 	for (; envc > 0; --envc) {
954 		suword(vectp++, (long)(intptr_t)destp);
955 		while (*stringp++ != 0)
956 			destp++;
957 		destp++;
958 	}
959 
960 	/* end of vector table is a null pointer */
961 	suword(vectp, 0);
962 
963 	return (stack_base);
964 }
965 
966 /*
967  * Check permissions of file to execute.
968  *	Return 0 for success or error code on failure.
969  */
970 int
971 exec_check_permissions(struct image_params *imgp)
972 {
973 	struct proc *p = imgp->proc;
974 	struct vnode *vp = imgp->vp;
975 	struct vattr *attr = imgp->attr;
976 	int error;
977 
978 	/* Get file attributes */
979 	error = VOP_GETATTR(vp, attr);
980 	if (error)
981 		return (error);
982 
983 	/*
984 	 * 1) Check if file execution is disabled for the filesystem that this
985 	 *	file resides on.
986 	 * 2) Insure that at least one execute bit is on - otherwise root
987 	 *	will always succeed, and we don't want to happen unless the
988 	 *	file really is executable.
989 	 * 3) Insure that the file is a regular file.
990 	 */
991 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
992 	    ((attr->va_mode & 0111) == 0) ||
993 	    (attr->va_type != VREG)) {
994 		return (EACCES);
995 	}
996 
997 	/*
998 	 * Zero length files can't be exec'd
999 	 */
1000 	if (attr->va_size == 0)
1001 		return (ENOEXEC);
1002 
1003 	/*
1004 	 *  Check for execute permission to file based on current credentials.
1005 	 */
1006 	error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1007 	if (error)
1008 		return (error);
1009 
1010 	/*
1011 	 * Check number of open-for-writes on the file and deny execution
1012 	 * if there are any.
1013 	 */
1014 	if (vp->v_writecount)
1015 		return (ETXTBSY);
1016 
1017 	/*
1018 	 * Call filesystem specific open routine, which allows us to read,
1019 	 * write, and mmap the file.  Without the VOP_OPEN we can only
1020 	 * stat the file.
1021 	 */
1022 	error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1023 	if (error)
1024 		return (error);
1025 
1026 	return (0);
1027 }
1028 
1029 /*
1030  * Exec handler registration
1031  */
1032 int
1033 exec_register(const struct execsw *execsw_arg)
1034 {
1035 	const struct execsw **es, **xs, **newexecsw;
1036 	int count = 2;	/* New slot and trailing NULL */
1037 
1038 	if (execsw)
1039 		for (es = execsw; *es; es++)
1040 			count++;
1041 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1042 	xs = newexecsw;
1043 	if (execsw)
1044 		for (es = execsw; *es; es++)
1045 			*xs++ = *es;
1046 	*xs++ = execsw_arg;
1047 	*xs = NULL;
1048 	if (execsw)
1049 		kfree(execsw, M_TEMP);
1050 	execsw = newexecsw;
1051 	return 0;
1052 }
1053 
1054 int
1055 exec_unregister(const struct execsw *execsw_arg)
1056 {
1057 	const struct execsw **es, **xs, **newexecsw;
1058 	int count = 1;
1059 
1060 	if (execsw == NULL)
1061 		panic("unregister with no handlers left?");
1062 
1063 	for (es = execsw; *es; es++) {
1064 		if (*es == execsw_arg)
1065 			break;
1066 	}
1067 	if (*es == NULL)
1068 		return ENOENT;
1069 	for (es = execsw; *es; es++)
1070 		if (*es != execsw_arg)
1071 			count++;
1072 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1073 	xs = newexecsw;
1074 	for (es = execsw; *es; es++)
1075 		if (*es != execsw_arg)
1076 			*xs++ = *es;
1077 	*xs = NULL;
1078 	if (execsw)
1079 		kfree(execsw, M_TEMP);
1080 	execsw = newexecsw;
1081 	return 0;
1082 }
1083