xref: /dflybsd-src/sys/kern/kern_exec.c (revision c0d274d062fd959993bf623f25f7cb6a8a676c4e)
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  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysproto.h>
32 #include <sys/kernel.h>
33 #include <sys/mount.h>
34 #include <sys/filedesc.h>
35 #include <sys/fcntl.h>
36 #include <sys/acct.h>
37 #include <sys/exec.h>
38 #include <sys/imgact.h>
39 #include <sys/imgact_elf.h>
40 #include <sys/kern_syscall.h>
41 #include <sys/wait.h>
42 #include <sys/malloc.h>
43 #include <sys/proc.h>
44 #include <sys/priv.h>
45 #include <sys/ktrace.h>
46 #include <sys/signalvar.h>
47 #include <sys/pioctl.h>
48 #include <sys/nlookup.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/libkern.h>
55 
56 #include <cpu/lwbuf.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_param.h>
60 #include <sys/lock.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_page.h>
63 #include <vm/vm_map.h>
64 #include <vm/vm_kern.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_object.h>
67 #include <vm/vnode_pager.h>
68 #include <vm/vm_pager.h>
69 
70 #include <sys/user.h>
71 #include <sys/reg.h>
72 
73 #include <sys/refcount.h>
74 #include <sys/thread2.h>
75 #include <sys/mplock2.h>
76 
77 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
78 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
79 
80 static register_t *exec_copyout_strings (struct image_params *);
81 
82 /* XXX This should be vm_size_t. */
83 static u_long ps_strings = PS_STRINGS;
84 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
85 
86 /* XXX This should be vm_size_t. */
87 static u_long usrstack = USRSTACK;
88 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
89 
90 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
91 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
92     &ps_arg_cache_limit, 0, "");
93 
94 int ps_argsopen = 1;
95 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
96 
97 static int ktrace_suid = 0;
98 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
99 
100 void print_execve_args(struct image_args *args);
101 int debug_execve_args = 0;
102 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
103     0, "");
104 
105 /*
106  * Exec arguments object cache
107  */
108 static struct objcache *exec_objcache;
109 
110 static
111 void
112 exec_objcache_init(void *arg __unused)
113 {
114 	int cluster_limit;
115 
116 	cluster_limit = 16;	/* up to this many objects */
117 	exec_objcache = objcache_create_mbacked(
118 					M_EXECARGS, PATH_MAX + ARG_MAX,
119 					&cluster_limit, 8,
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 	struct pargs *pa;
185 	struct sigacts *ops;
186 	struct sigacts *nps;
187 	int error, len, i;
188 	struct image_params image_params, *imgp;
189 	struct vattr attr;
190 	int (*img_first) (struct image_params *);
191 
192 	if (debug_execve_args) {
193 		kprintf("%s()\n", __func__);
194 		print_execve_args(args);
195 	}
196 
197 	KKASSERT(p);
198 	lwkt_gettoken(&p->p_token);
199 	imgp = &image_params;
200 
201 	/*
202 	 * NOTE: P_INEXEC is handled by exec_new_vmspace() now.  We make
203 	 * no modifications to the process at all until we get there.
204 	 *
205 	 * Note that multiple threads may be trying to exec at the same
206 	 * time.  exec_new_vmspace() handles that too.
207 	 */
208 
209 	/*
210 	 * Initialize part of the common data
211 	 */
212 	imgp->proc = p;
213 	imgp->args = args;
214 	imgp->attr = &attr;
215 	imgp->entry_addr = 0;
216 	imgp->resident = 0;
217 	imgp->vmspace_destroyed = 0;
218 	imgp->interpreted = 0;
219 	imgp->interpreter_name[0] = 0;
220 	imgp->auxargs = NULL;
221 	imgp->vp = NULL;
222 	imgp->firstpage = NULL;
223 	imgp->ps_strings = 0;
224 	imgp->execpath = imgp->freepath = NULL;
225 	imgp->execpathp = 0;
226 	imgp->image_header = NULL;
227 
228 interpret:
229 
230 	/*
231 	 * Translate the file name to a vnode.  Unlock the cache entry to
232 	 * improve parallelism for programs exec'd in parallel.
233 	 */
234 	if ((error = nlookup(nd)) != 0)
235 		goto exec_fail;
236 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
237 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
238 	nd->nl_flags &= ~NLC_NCPISLOCKED;
239 	cache_unlock(&nd->nl_nch);
240 	if (error)
241 		goto exec_fail;
242 
243 	/*
244 	 * Check file permissions (also 'opens' file).
245 	 * Include also the top level mount in the check.
246 	 */
247 	error = exec_check_permissions(imgp, nd->nl_nch.mount);
248 	if (error) {
249 		vn_unlock(imgp->vp);
250 		goto exec_fail_dealloc;
251 	}
252 
253 	error = exec_map_first_page(imgp);
254 	vn_unlock(imgp->vp);
255 	if (error)
256 		goto exec_fail_dealloc;
257 
258 	if (debug_execve_args && imgp->interpreted) {
259 		kprintf("    target is interpreted -- recursive pass\n");
260 		kprintf("    interpreter: %s\n", imgp->interpreter_name);
261 		print_execve_args(args);
262 	}
263 
264 	/*
265 	 *	If the current process has a special image activator it
266 	 *	wants to try first, call it.   For example, emulating shell
267 	 *	scripts differently.
268 	 */
269 	error = -1;
270 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
271 		error = img_first(imgp);
272 
273 	/*
274 	 *	If the vnode has a registered vmspace, exec the vmspace
275 	 */
276 	if (error == -1 && imgp->vp->v_resident) {
277 		error = exec_resident_imgact(imgp);
278 	}
279 
280 	/*
281 	 *	Loop through the list of image activators, calling each one.
282 	 *	An activator returns -1 if there is no match, 0 on success,
283 	 *	and an error otherwise.
284 	 */
285 	for (i = 0; error == -1 && execsw[i]; ++i) {
286 		if (execsw[i]->ex_imgact == NULL ||
287 		    execsw[i]->ex_imgact == img_first) {
288 			continue;
289 		}
290 		error = (*execsw[i]->ex_imgact)(imgp);
291 	}
292 
293 	if (error) {
294 		if (error == -1)
295 			error = ENOEXEC;
296 		goto exec_fail_dealloc;
297 	}
298 
299 	/*
300 	 * Special interpreter operation, cleanup and loop up to try to
301 	 * activate the interpreter.
302 	 */
303 	if (imgp->interpreted) {
304 		exec_unmap_first_page(imgp);
305 		nlookup_done(nd);
306 		vrele(imgp->vp);
307 		imgp->vp = NULL;
308 		error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
309 					NLC_FOLLOW);
310 		if (error)
311 			goto exec_fail;
312 		goto interpret;
313 	}
314 
315 	/*
316 	 * Do the best to calculate the full path to the image file
317 	 */
318 	if (imgp->auxargs != NULL &&
319 	   ((args->fname != NULL && args->fname[0] == '/') ||
320 	    vn_fullpath(imgp->proc,
321 			imgp->vp,
322 			&imgp->execpath,
323 			&imgp->freepath,
324 			0) != 0))
325 		imgp->execpath = args->fname;
326 
327 	/*
328 	 * Copy out strings (args and env) and initialize stack base
329 	 */
330 	stack_base = exec_copyout_strings(imgp);
331 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
332 
333 	/*
334 	 * If custom stack fixup routine present for this process
335 	 * let it do the stack setup.  If we are running a resident
336 	 * image there is no auxinfo or other image activator context
337 	 * so don't try to add fixups to the stack.
338 	 *
339 	 * Else stuff argument count as first item on stack
340 	 */
341 	if (p->p_sysent->sv_fixup && imgp->resident == 0)
342 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
343 	else
344 		suword(--stack_base, imgp->args->argc);
345 
346 	/*
347 	 * For security and other reasons, the file descriptor table cannot
348 	 * be shared after an exec.
349 	 */
350 	if (p->p_fd->fd_refcnt > 1) {
351 		struct filedesc *tmp;
352 
353 		error = fdcopy(p, &tmp);
354 		if (error != 0)
355 			goto exec_fail;
356 		fdfree(p, tmp);
357 	}
358 
359 	/*
360 	 * For security and other reasons, signal handlers cannot
361 	 * be shared after an exec. The new proces gets a copy of the old
362 	 * handlers. In execsigs(), the new process will have its signals
363 	 * reset.
364 	 */
365 	ops = p->p_sigacts;
366 	if (ops->ps_refcnt > 1) {
367 		nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
368 		bcopy(ops, nps, sizeof(*nps));
369 		refcount_init(&nps->ps_refcnt, 1);
370 		p->p_sigacts = nps;
371 		if (refcount_release(&ops->ps_refcnt)) {
372 			kfree(ops, M_SUBPROC);
373 			ops = NULL;
374 		}
375 	}
376 
377 	/*
378 	 * For security and other reasons virtual kernels cannot be
379 	 * inherited by an exec.  This also allows a virtual kernel
380 	 * to fork/exec unrelated applications.
381 	 */
382 	if (p->p_vkernel)
383 		vkernel_exit(p);
384 
385 	/* Stop profiling */
386 	stopprofclock(p);
387 
388 	/* close files on exec */
389 	fdcloseexec(p);
390 
391 	/* reset caught signals */
392 	execsigs(p);
393 
394 	/* name this process - nameiexec(p, ndp) */
395 	len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
396 	bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
397 	p->p_comm[len] = 0;
398 	bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
399 
400 	/*
401 	 * mark as execed, wakeup the process that vforked (if any) and tell
402 	 * it that it now has its own resources back
403 	 */
404 	p->p_flag |= P_EXEC;
405 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
406 		p->p_flag &= ~P_PPWAIT;
407 		wakeup((caddr_t)p->p_pptr);
408 	}
409 
410 	/*
411 	 * Implement image setuid/setgid.
412 	 *
413 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
414 	 * the process is being traced.
415 	 */
416 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
417 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
418 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
419 	    (p->p_flag & P_TRACED) == 0) {
420 		/*
421 		 * Turn off syscall tracing for set-id programs, except for
422 		 * root.  Record any set-id flags first to make sure that
423 		 * we do not regain any tracing during a possible block.
424 		 */
425 		setsugid();
426 		if (p->p_tracenode && ktrace_suid == 0 &&
427 		    priv_check(td, PRIV_ROOT) != 0) {
428 			ktrdestroy(&p->p_tracenode);
429 			p->p_traceflag = 0;
430 		}
431 		/* Close any file descriptors 0..2 that reference procfs */
432 		setugidsafety(p);
433 		/* Make sure file descriptors 0..2 are in use. */
434 		error = fdcheckstd(lp);
435 		if (error != 0)
436 			goto exec_fail_dealloc;
437 		/*
438 		 * Set the new credentials.
439 		 */
440 		cratom(&p->p_ucred);
441 		if (attr.va_mode & VSUID)
442 			change_euid(attr.va_uid);
443 		if (attr.va_mode & VSGID)
444 			p->p_ucred->cr_gid = attr.va_gid;
445 
446 		/*
447 		 * Clear local varsym variables
448 		 */
449 		varsymset_clean(&p->p_varsymset);
450 	} else {
451 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
452 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
453 			p->p_flag &= ~P_SUGID;
454 	}
455 
456 	/*
457 	 * Implement correct POSIX saved-id behavior.
458 	 */
459 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
460 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
461 		cratom(&p->p_ucred);
462 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
463 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
464 	}
465 
466 	/*
467 	 * Store the vp for use in procfs
468 	 */
469 	if (p->p_textvp)		/* release old reference */
470 		vrele(p->p_textvp);
471 	p->p_textvp = imgp->vp;
472 	vref(p->p_textvp);
473 
474 	/* Release old namecache handle to text file */
475 	if (p->p_textnch.ncp)
476 		cache_drop(&p->p_textnch);
477 
478 	if (nd->nl_nch.mount)
479 		cache_copy(&nd->nl_nch, &p->p_textnch);
480 
481         /*
482          * Notify others that we exec'd, and clear the P_INEXEC flag
483          * as we're now a bona fide freshly-execed process.
484          */
485 	KNOTE(&p->p_klist, NOTE_EXEC);
486 	p->p_flag &= ~P_INEXEC;
487 
488 	/*
489 	 * If tracing the process, trap to debugger so breakpoints
490 	 * 	can be set before the program executes.
491 	 */
492 	STOPEVENT(p, S_EXEC, 0);
493 
494 	if (p->p_flag & P_TRACED)
495 		ksignal(p, SIGTRAP);
496 
497 	/* clear "fork but no exec" flag, as we _are_ execing */
498 	p->p_acflag &= ~AFORK;
499 
500 	/* Set values passed into the program in registers. */
501 	exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
502 	    imgp->ps_strings);
503 
504 	/* Set the access time on the vnode */
505 	vn_mark_atime(imgp->vp, td);
506 
507 	/*
508 	 * Free any previous argument cache
509 	 */
510 	pa = p->p_args;
511 	p->p_args = NULL;
512 	if (pa && refcount_release(&pa->ar_ref)) {
513 		kfree(pa, M_PARGS);
514 		pa = NULL;
515 	}
516 
517 	/*
518 	 * Cache arguments if they fit inside our allowance
519 	 */
520 	i = imgp->args->begin_envv - imgp->args->begin_argv;
521 	if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
522 		pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
523 		refcount_init(&pa->ar_ref, 1);
524 		pa->ar_length = i;
525 		bcopy(imgp->args->begin_argv, pa->ar_args, i);
526 		KKASSERT(p->p_args == NULL);
527 		p->p_args = pa;
528 	}
529 
530 exec_fail_dealloc:
531 
532 	/*
533 	 * free various allocated resources
534 	 */
535 	if (imgp->firstpage)
536 		exec_unmap_first_page(imgp);
537 
538 	if (imgp->vp) {
539 		vrele(imgp->vp);
540 		imgp->vp = NULL;
541 	}
542 
543 	if (error == 0) {
544 		++mycpu->gd_cnt.v_exec;
545 		lwkt_reltoken(&p->p_token);
546 		return (0);
547 	}
548 
549 	if (imgp->freepath)
550 		kfree(imgp->freepath, M_TEMP);
551 
552 exec_fail:
553 	/*
554 	 * we're done here, clear P_INEXEC if we were the ones that
555 	 * set it.  Otherwise if vmspace_destroyed is still set we
556 	 * raced another thread and that thread is responsible for
557 	 * clearing it.
558 	 */
559 	if (imgp->vmspace_destroyed & 2)
560 		p->p_flag &= ~P_INEXEC;
561 	lwkt_reltoken(&p->p_token);
562 	if (imgp->vmspace_destroyed) {
563 		/*
564 		 * Sorry, no more process anymore. exit gracefully.
565 		 * However we can't die right here, because our
566 		 * caller might have to clean up, so indicate a
567 		 * lethal error by returning -1.
568 		 */
569 		return(-1);
570 	} else {
571 		return(error);
572 	}
573 }
574 
575 /*
576  * execve() system call.
577  *
578  * MPALMOSTSAFE
579  */
580 int
581 sys_execve(struct execve_args *uap)
582 {
583 	struct nlookupdata nd;
584 	struct image_args args;
585 	int error;
586 
587 	bzero(&args, sizeof(args));
588 
589 	get_mplock();
590 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
591 	if (error == 0) {
592 		error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
593 					uap->argv, uap->envv);
594 	}
595 	if (error == 0)
596 		error = kern_execve(&nd, &args);
597 	nlookup_done(&nd);
598 	exec_free_args(&args);
599 
600 	if (error < 0) {
601 		/* We hit a lethal error condition.  Let's die now. */
602 		exit1(W_EXITCODE(0, SIGABRT));
603 		/* NOTREACHED */
604 	}
605 	rel_mplock();
606 
607 	/*
608 	 * The syscall result is returned in registers to the new program.
609 	 * Linux will register %edx as an atexit function and we must be
610 	 * sure to set it to 0.  XXX
611 	 */
612 	if (error == 0)
613 		uap->sysmsg_result64 = 0;
614 
615 	return (error);
616 }
617 
618 int
619 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
620 	      struct lwbuf **plwb, const char **pdata)
621 {
622 	int rv;
623 	vm_page_t ma;
624 	vm_page_t m;
625 	vm_object_t object;
626 
627 	/*
628 	 * The file has to be mappable.
629 	 */
630 	if ((object = imgp->vp->v_object) == NULL)
631 		return (EIO);
632 
633 	if (pageno >= object->size)
634 		return (EIO);
635 
636 	m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
637 
638 	lwkt_gettoken(&vm_token);
639 	while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
640 		ma = m;
641 
642 		/*
643 		 * get_pages unbusies all the requested pages except the
644 		 * primary page (at index 0 in this case).  The primary
645 		 * page may have been wired during the pagein (e.g. by
646 		 * the buffer cache) so vnode_pager_freepage() must be
647 		 * used to properly release it.
648 		 */
649 		rv = vm_pager_get_page(object, &ma, 1);
650 		m = vm_page_lookup(object, pageno);
651 
652 		if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
653 			if (m) {
654 				vm_page_protect(m, VM_PROT_NONE);
655 				vnode_pager_freepage(m);
656 			}
657 			lwkt_reltoken(&vm_token);
658 			return EIO;
659 		}
660 	}
661 	vm_page_hold(m);	/* requires vm_token to be held */
662 	vm_page_wakeup(m);	/* unbusy the page */
663 	lwkt_reltoken(&vm_token);
664 
665 	*plwb = lwbuf_alloc(m, *plwb);
666 	*pdata = (void *)lwbuf_kva(*plwb);
667 
668 	return (0);
669 }
670 
671 int
672 exec_map_first_page(struct image_params *imgp)
673 {
674 	int err;
675 
676 	if (imgp->firstpage)
677 		exec_unmap_first_page(imgp);
678 
679 	imgp->firstpage = &imgp->firstpage_cache;
680 	err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
681 
682 	if (err)
683 		return err;
684 
685 	return 0;
686 }
687 
688 void
689 exec_unmap_page(struct lwbuf *lwb)
690 {
691 	vm_page_t m;
692 
693 	crit_enter();
694 	if (lwb != NULL) {
695 		m = lwbuf_page(lwb);
696 		lwbuf_free(lwb);
697 		vm_page_unhold(m);
698 	}
699 	crit_exit();
700 }
701 
702 void
703 exec_unmap_first_page(struct image_params *imgp)
704 {
705 	exec_unmap_page(imgp->firstpage);
706 	imgp->firstpage = NULL;
707 	imgp->image_header = NULL;
708 }
709 
710 /*
711  * Destroy old address space, and allocate a new stack
712  *	The new stack is only SGROWSIZ large because it is grown
713  *	automatically in trap.c.
714  *
715  * This is the point of no return.
716  */
717 int
718 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
719 {
720 	struct vmspace *vmspace = imgp->proc->p_vmspace;
721 	vm_offset_t stack_addr = USRSTACK - maxssiz;
722 	struct proc *p;
723 	vm_map_t map;
724 	int error;
725 
726 	/*
727 	 * Indicate that we cannot gracefully error out any more, kill
728 	 * any other threads present, and set P_INEXEC to indicate that
729 	 * we are now messing with the process structure proper.
730 	 *
731 	 * If killalllwps() races return an error which coupled with
732 	 * vmspace_destroyed will cause us to exit.  This is what we
733 	 * want since another thread is patiently waiting for us to exit
734 	 * in that case.
735 	 */
736 	p = curproc;
737 	imgp->vmspace_destroyed = 1;
738 
739 	if (curthread->td_proc->p_nthreads > 1) {
740 		error = killalllwps(1);
741 		if (error)
742 			return (error);
743 	}
744 	imgp->vmspace_destroyed |= 2;	/* we are responsible for P_INEXEC */
745 	p->p_flag |= P_INEXEC;
746 
747 	/*
748 	 * Blow away entire process VM, if address space not shared,
749 	 * otherwise, create a new VM space so that other threads are
750 	 * not disrupted.  If we are execing a resident vmspace we
751 	 * create a duplicate of it and remap the stack.
752 	 *
753 	 * The exitingcnt test is not strictly necessary but has been
754 	 * included for code sanity (to make the code more deterministic).
755 	 */
756 	map = &vmspace->vm_map;
757 	if (vmcopy) {
758 		vmspace_exec(imgp->proc, vmcopy);
759 		vmspace = imgp->proc->p_vmspace;
760 		pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
761 		map = &vmspace->vm_map;
762 	} else if (vmspace->vm_sysref.refcnt == 1 &&
763 		   vmspace->vm_exitingcnt == 0) {
764 		shmexit(vmspace);
765 		if (vmspace->vm_upcalls)
766 			upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
767 		pmap_remove_pages(vmspace_pmap(vmspace),
768 			0, VM_MAX_USER_ADDRESS);
769 		vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
770 	} else {
771 		vmspace_exec(imgp->proc, NULL);
772 		vmspace = imgp->proc->p_vmspace;
773 		map = &vmspace->vm_map;
774 	}
775 
776 	/* Allocate a new stack */
777 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
778 			     0, VM_PROT_ALL, VM_PROT_ALL, 0);
779 	if (error)
780 		return (error);
781 
782 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
783 	 * VM_STACK case, but they are still used to monitor the size of the
784 	 * process stack so we can check the stack rlimit.
785 	 */
786 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
787 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
788 
789 	return(0);
790 }
791 
792 /*
793  * Copy out argument and environment strings from the old process
794  *	address space into the temporary string buffer.
795  */
796 int
797 exec_copyin_args(struct image_args *args, char *fname,
798 		enum exec_path_segflg segflg, char **argv, char **envv)
799 {
800 	char	*argp, *envp;
801 	int	error = 0;
802 	size_t	length;
803 
804 	args->buf = objcache_get(exec_objcache, M_WAITOK);
805 	if (args->buf == NULL)
806 		return (ENOMEM);
807 	args->begin_argv = args->buf;
808 	args->endp = args->begin_argv;
809 	args->space = ARG_MAX;
810 
811 	args->fname = args->buf + ARG_MAX;
812 
813 	/*
814 	 * Copy the file name.
815 	 */
816 	if (segflg == PATH_SYSSPACE) {
817 		error = copystr(fname, args->fname, PATH_MAX, &length);
818 	} else if (segflg == PATH_USERSPACE) {
819 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
820 	}
821 
822 	/*
823 	 * Extract argument strings.  argv may not be NULL.  The argv
824 	 * array is terminated by a NULL entry.  We special-case the
825 	 * situation where argv[0] is NULL by passing { filename, NULL }
826 	 * to the new program to guarentee that the interpreter knows what
827 	 * file to open in case we exec an interpreted file.   Note that
828 	 * a NULL argv[0] terminates the argv[] array.
829 	 *
830 	 * XXX the special-casing of argv[0] is historical and needs to be
831 	 * revisited.
832 	 */
833 	if (argv == NULL)
834 		error = EFAULT;
835 	if (error == 0) {
836 		while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
837 			if (argp == (caddr_t)-1) {
838 				error = EFAULT;
839 				break;
840 			}
841 			error = copyinstr(argp, args->endp,
842 					    args->space, &length);
843 			if (error) {
844 				if (error == ENAMETOOLONG)
845 					error = E2BIG;
846 				break;
847 			}
848 			args->space -= length;
849 			args->endp += length;
850 			args->argc++;
851 		}
852 		if (args->argc == 0 && error == 0) {
853 			length = strlen(args->fname) + 1;
854 			if (length > args->space) {
855 				error = E2BIG;
856 			} else {
857 				bcopy(args->fname, args->endp, length);
858 				args->space -= length;
859 				args->endp += length;
860 				args->argc++;
861 			}
862 		}
863 	}
864 
865 	args->begin_envv = args->endp;
866 
867 	/*
868 	 * extract environment strings.  envv may be NULL.
869 	 */
870 	if (envv && error == 0) {
871 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
872 			if (envp == (caddr_t) -1) {
873 				error = EFAULT;
874 				break;
875 			}
876 			error = copyinstr(envp, args->endp, args->space,
877 			    &length);
878 			if (error) {
879 				if (error == ENAMETOOLONG)
880 					error = E2BIG;
881 				break;
882 			}
883 			args->space -= length;
884 			args->endp += length;
885 			args->envc++;
886 		}
887 	}
888 	return (error);
889 }
890 
891 void
892 exec_free_args(struct image_args *args)
893 {
894 	if (args->buf) {
895 		objcache_put(exec_objcache, args->buf);
896 		args->buf = NULL;
897 	}
898 }
899 
900 /*
901  * Copy strings out to the new process address space, constructing
902  *	new arg and env vector tables. Return a pointer to the base
903  *	so that it can be used as the initial stack pointer.
904  */
905 register_t *
906 exec_copyout_strings(struct image_params *imgp)
907 {
908 	int argc, envc, sgap;
909 	char **vectp;
910 	char *stringp, *destp;
911 	register_t *stack_base;
912 	struct ps_strings *arginfo;
913 	size_t execpath_len;
914 	int szsigcode;
915 
916 	/*
917 	 * Calculate string base and vector table pointers.
918 	 * Also deal with signal trampoline code for this exec type.
919 	 */
920 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
921 		execpath_len = strlen(imgp->execpath) + 1;
922 	else
923 		execpath_len = 0;
924 	arginfo = (struct ps_strings *)PS_STRINGS;
925 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
926 	if (stackgap_random != 0)
927 		sgap = ALIGN(karc4random() & (stackgap_random - 1));
928 	else
929 		sgap = 0;
930 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
931 	    roundup(execpath_len, sizeof(char *)) -
932 	    roundup((ARG_MAX - imgp->args->space), sizeof(char *));
933 
934 	/*
935 	 * install sigcode
936 	 */
937 	if (szsigcode)
938 		copyout(imgp->proc->p_sysent->sv_sigcode,
939 		    ((caddr_t)arginfo - szsigcode), szsigcode);
940 
941 	/*
942 	 * Copy the image path for the rtld
943 	 */
944 	if (execpath_len != 0) {
945 		imgp->execpathp = (uintptr_t)arginfo
946 				  - szsigcode
947 				  - execpath_len;
948 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
949 	}
950 
951 	/*
952 	 * If we have a valid auxargs ptr, prepare some room
953 	 * on the stack.
954 	 *
955 	 * The '+ 2' is for the null pointers at the end of each of the
956 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
957 	 * ELF Auxargs data.
958 	 */
959 	if (imgp->auxargs) {
960 		vectp = (char **)(destp - (imgp->args->argc +
961 			imgp->args->envc + 2 + (AT_COUNT * 2) + execpath_len) *
962 			sizeof(char*));
963 	} else {
964 		vectp = (char **)(destp - (imgp->args->argc +
965 			imgp->args->envc + 2) * sizeof(char*));
966 	}
967 
968 	/*
969 	 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
970 	 * be done in crt1.o.  Note that GCC 3.x aligns the stack in main.
971 	 */
972 
973 	/*
974 	 * vectp also becomes our initial stack base
975 	 */
976 	stack_base = (register_t *)vectp;
977 
978 	stringp = imgp->args->begin_argv;
979 	argc = imgp->args->argc;
980 	envc = imgp->args->envc;
981 
982 	/*
983 	 * Copy out strings - arguments and environment.
984 	 */
985 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
986 
987 	/*
988 	 * Fill in "ps_strings" struct for ps, w, etc.
989 	 */
990 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
991 	suword(&arginfo->ps_nargvstr, argc);
992 
993 	/*
994 	 * Fill in argument portion of vector table.
995 	 */
996 	for (; argc > 0; --argc) {
997 		suword(vectp++, (long)(intptr_t)destp);
998 		while (*stringp++ != 0)
999 			destp++;
1000 		destp++;
1001 	}
1002 
1003 	/* a null vector table pointer separates the argp's from the envp's */
1004 	suword(vectp++, 0);
1005 
1006 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1007 	suword(&arginfo->ps_nenvstr, envc);
1008 
1009 	/*
1010 	 * Fill in environment portion of vector table.
1011 	 */
1012 	for (; envc > 0; --envc) {
1013 		suword(vectp++, (long)(intptr_t)destp);
1014 		while (*stringp++ != 0)
1015 			destp++;
1016 		destp++;
1017 	}
1018 
1019 	/* end of vector table is a null pointer */
1020 	suword(vectp, 0);
1021 
1022 	return (stack_base);
1023 }
1024 
1025 /*
1026  * Check permissions of file to execute.
1027  *	Return 0 for success or error code on failure.
1028  */
1029 int
1030 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
1031 {
1032 	struct proc *p = imgp->proc;
1033 	struct vnode *vp = imgp->vp;
1034 	struct vattr *attr = imgp->attr;
1035 	int error;
1036 
1037 	/* Get file attributes */
1038 	error = VOP_GETATTR(vp, attr);
1039 	if (error)
1040 		return (error);
1041 
1042 	/*
1043 	 * 1) Check if file execution is disabled for the filesystem that this
1044 	 *	file resides on.
1045 	 * 2) Insure that at least one execute bit is on - otherwise root
1046 	 *	will always succeed, and we don't want to happen unless the
1047 	 *	file really is executable.
1048 	 * 3) Insure that the file is a regular file.
1049 	 */
1050 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1051 	    ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1052 	    ((attr->va_mode & 0111) == 0) ||
1053 	    (attr->va_type != VREG)) {
1054 		return (EACCES);
1055 	}
1056 
1057 	/*
1058 	 * Zero length files can't be exec'd
1059 	 */
1060 	if (attr->va_size == 0)
1061 		return (ENOEXEC);
1062 
1063 	/*
1064 	 *  Check for execute permission to file based on current credentials.
1065 	 */
1066 	error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1067 	if (error)
1068 		return (error);
1069 
1070 	/*
1071 	 * Check number of open-for-writes on the file and deny execution
1072 	 * if there are any.
1073 	 */
1074 	if (vp->v_writecount)
1075 		return (ETXTBSY);
1076 
1077 	/*
1078 	 * Call filesystem specific open routine, which allows us to read,
1079 	 * write, and mmap the file.  Without the VOP_OPEN we can only
1080 	 * stat the file.
1081 	 */
1082 	error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1083 	if (error)
1084 		return (error);
1085 
1086 	return (0);
1087 }
1088 
1089 /*
1090  * Exec handler registration
1091  */
1092 int
1093 exec_register(const struct execsw *execsw_arg)
1094 {
1095 	const struct execsw **es, **xs, **newexecsw;
1096 	int count = 2;	/* New slot and trailing NULL */
1097 
1098 	if (execsw)
1099 		for (es = execsw; *es; es++)
1100 			count++;
1101 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1102 	xs = newexecsw;
1103 	if (execsw)
1104 		for (es = execsw; *es; es++)
1105 			*xs++ = *es;
1106 	*xs++ = execsw_arg;
1107 	*xs = NULL;
1108 	if (execsw)
1109 		kfree(execsw, M_TEMP);
1110 	execsw = newexecsw;
1111 	return 0;
1112 }
1113 
1114 int
1115 exec_unregister(const struct execsw *execsw_arg)
1116 {
1117 	const struct execsw **es, **xs, **newexecsw;
1118 	int count = 1;
1119 
1120 	if (execsw == NULL)
1121 		panic("unregister with no handlers left?");
1122 
1123 	for (es = execsw; *es; es++) {
1124 		if (*es == execsw_arg)
1125 			break;
1126 	}
1127 	if (*es == NULL)
1128 		return ENOENT;
1129 	for (es = execsw; *es; es++)
1130 		if (*es != execsw_arg)
1131 			count++;
1132 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1133 	xs = newexecsw;
1134 	for (es = execsw; *es; es++)
1135 		if (*es != execsw_arg)
1136 			*xs++ = *es;
1137 	*xs = NULL;
1138 	if (execsw)
1139 		kfree(execsw, M_TEMP);
1140 	execsw = newexecsw;
1141 	return 0;
1142 }
1143