xref: /dflybsd-src/sys/kern/kern_exec.c (revision ef6d73ea92a7fb40e70d16c80591e0001890e238)
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 #include <vm/vm_page2.h>
77 
78 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
79 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments");
80 
81 static register_t *exec_copyout_strings (struct image_params *);
82 
83 /* XXX This should be vm_size_t. */
84 static u_long ps_strings = PS_STRINGS;
85 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, "");
86 
87 /* XXX This should be vm_size_t. */
88 static u_long usrstack = USRSTACK;
89 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, "");
90 
91 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
92 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
93     &ps_arg_cache_limit, 0, "");
94 
95 int ps_argsopen = 1;
96 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, "");
97 
98 static int ktrace_suid = 0;
99 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, "");
100 
101 void print_execve_args(struct image_args *args);
102 int debug_execve_args = 0;
103 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args,
104     0, "");
105 
106 /*
107  * Exec arguments object cache
108  */
109 static struct objcache *exec_objcache;
110 
111 static
112 void
113 exec_objcache_init(void *arg __unused)
114 {
115 	int cluster_limit;
116 	size_t limsize;
117 
118 	/*
119 	 * Maximum number of concurrent execs.  This can be limiting on
120 	 * systems with a lot of cpu cores but it also eats a significant
121 	 * amount of memory.
122 	 */
123 	cluster_limit = (ncpus < 16) ? 16 : ncpus;
124 	limsize = kmem_lim_size();
125 	if (limsize > 7 * 1024)
126 		cluster_limit *= 2;
127 	if (limsize > 15 * 1024)
128 		cluster_limit *= 2;
129 
130 	exec_objcache = objcache_create_mbacked(
131 					M_EXECARGS, PATH_MAX + ARG_MAX,
132 					cluster_limit, 8,
133 					NULL, NULL, NULL);
134 }
135 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
136 
137 /*
138  * stackgap_random specifies if the stackgap should have a random size added
139  * to it.  It must be a power of 2.  If non-zero, the stack gap will be
140  * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
141  */
142 static int stackgap_random = 1024;
143 static int
144 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
145 {
146 	int error, new_val;
147 	new_val = stackgap_random;
148 	error = sysctl_handle_int(oidp, &new_val, 0, req);
149 	if (error != 0 || req->newptr == NULL)
150 		return (error);
151 	if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val)))
152 		return (EINVAL);
153 	stackgap_random = new_val;
154 
155 	return(0);
156 }
157 
158 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT,
159 	0, 0, sysctl_kern_stackgap, "I",
160 	"Max random stack gap (power of 2), static gap if negative");
161 
162 void
163 print_execve_args(struct image_args *args)
164 {
165 	char *cp;
166 	int ndx;
167 
168 	cp = args->begin_argv;
169 	for (ndx = 0; ndx < args->argc; ndx++) {
170 		kprintf("\targv[%d]: %s\n", ndx, cp);
171 		while (*cp++ != '\0');
172 	}
173 	for (ndx = 0; ndx < args->envc; ndx++) {
174 		kprintf("\tenvv[%d]: %s\n", ndx, cp);
175 		while (*cp++ != '\0');
176 	}
177 }
178 
179 /*
180  * Each of the items is a pointer to a `const struct execsw', hence the
181  * double pointer here.
182  */
183 static const struct execsw **execsw;
184 
185 /*
186  * Replace current vmspace with a new binary.
187  * Returns 0 on success, > 0 on recoverable error (use as errno).
188  * Returns -1 on lethal error which demands killing of the current
189  * process!
190  */
191 int
192 kern_execve(struct nlookupdata *nd, struct image_args *args)
193 {
194 	struct thread *td = curthread;
195 	struct lwp *lp = td->td_lwp;
196 	struct proc *p = td->td_proc;
197 	struct vnode *ovp;
198 	register_t *stack_base;
199 	struct pargs *pa;
200 	struct sigacts *ops;
201 	struct sigacts *nps;
202 	int error, len, i;
203 	struct image_params image_params, *imgp;
204 	struct vattr attr;
205 	int (*img_first) (struct image_params *);
206 
207 	if (debug_execve_args) {
208 		kprintf("%s()\n", __func__);
209 		print_execve_args(args);
210 	}
211 
212 	KKASSERT(p);
213 	lwkt_gettoken(&p->p_token);
214 	imgp = &image_params;
215 
216 	/*
217 	 * NOTE: P_INEXEC is handled by exec_new_vmspace() now.  We make
218 	 * no modifications to the process at all until we get there.
219 	 *
220 	 * Note that multiple threads may be trying to exec at the same
221 	 * time.  exec_new_vmspace() handles that too.
222 	 */
223 
224 	/*
225 	 * Initialize part of the common data
226 	 */
227 	imgp->proc = p;
228 	imgp->args = args;
229 	imgp->attr = &attr;
230 	imgp->entry_addr = 0;
231 	imgp->resident = 0;
232 	imgp->vmspace_destroyed = 0;
233 	imgp->interpreted = 0;
234 	imgp->interpreter_name[0] = 0;
235 	imgp->auxargs = NULL;
236 	imgp->vp = NULL;
237 	imgp->firstpage = NULL;
238 	imgp->ps_strings = 0;
239 	imgp->execpath = imgp->freepath = NULL;
240 	imgp->execpathp = 0;
241 	imgp->image_header = NULL;
242 
243 interpret:
244 
245 	/*
246 	 * Translate the file name to a vnode.  Unlock the cache entry to
247 	 * improve parallelism for programs exec'd in parallel.
248 	 */
249 	if ((error = nlookup(nd)) != 0)
250 		goto exec_fail;
251 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
252 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
253 	nd->nl_flags &= ~NLC_NCPISLOCKED;
254 	cache_unlock(&nd->nl_nch);
255 	if (error)
256 		goto exec_fail;
257 
258 	/*
259 	 * Check file permissions (also 'opens' file).
260 	 * Include also the top level mount in the check.
261 	 */
262 	error = exec_check_permissions(imgp, nd->nl_nch.mount);
263 	if (error) {
264 		vn_unlock(imgp->vp);
265 		goto exec_fail_dealloc;
266 	}
267 
268 	error = exec_map_first_page(imgp);
269 	vn_unlock(imgp->vp);
270 	if (error)
271 		goto exec_fail_dealloc;
272 
273 	imgp->proc->p_osrel = 0;
274 
275 	if (debug_execve_args && imgp->interpreted) {
276 		kprintf("    target is interpreted -- recursive pass\n");
277 		kprintf("    interpreter: %s\n", imgp->interpreter_name);
278 		print_execve_args(args);
279 	}
280 
281 	/*
282 	 *	If the current process has a special image activator it
283 	 *	wants to try first, call it.   For example, emulating shell
284 	 *	scripts differently.
285 	 */
286 	error = -1;
287 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
288 		error = img_first(imgp);
289 
290 	/*
291 	 *	If the vnode has a registered vmspace, exec the vmspace
292 	 */
293 	if (error == -1 && imgp->vp->v_resident) {
294 		error = exec_resident_imgact(imgp);
295 	}
296 
297 	/*
298 	 *	Loop through the list of image activators, calling each one.
299 	 *	An activator returns -1 if there is no match, 0 on success,
300 	 *	and an error otherwise.
301 	 */
302 	for (i = 0; error == -1 && execsw[i]; ++i) {
303 		if (execsw[i]->ex_imgact == NULL ||
304 		    execsw[i]->ex_imgact == img_first) {
305 			continue;
306 		}
307 		error = (*execsw[i]->ex_imgact)(imgp);
308 	}
309 
310 	if (error) {
311 		if (error == -1)
312 			error = ENOEXEC;
313 		goto exec_fail_dealloc;
314 	}
315 
316 	/*
317 	 * Special interpreter operation, cleanup and loop up to try to
318 	 * activate the interpreter.
319 	 */
320 	if (imgp->interpreted) {
321 		exec_unmap_first_page(imgp);
322 		nlookup_done(nd);
323 		vrele(imgp->vp);
324 		imgp->vp = NULL;
325 		error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
326 					NLC_FOLLOW);
327 		if (error)
328 			goto exec_fail;
329 		goto interpret;
330 	}
331 
332 	/*
333 	 * Do the best to calculate the full path to the image file
334 	 */
335 	if (imgp->auxargs != NULL &&
336 	   ((args->fname != NULL && args->fname[0] == '/') ||
337 	    vn_fullpath(imgp->proc,
338 			imgp->vp,
339 			&imgp->execpath,
340 			&imgp->freepath,
341 			0) != 0))
342 		imgp->execpath = args->fname;
343 
344 	/*
345 	 * Copy out strings (args and env) and initialize stack base
346 	 */
347 	stack_base = exec_copyout_strings(imgp);
348 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
349 
350 	/*
351 	 * If custom stack fixup routine present for this process
352 	 * let it do the stack setup.  If we are running a resident
353 	 * image there is no auxinfo or other image activator context
354 	 * so don't try to add fixups to the stack.
355 	 *
356 	 * Else stuff argument count as first item on stack
357 	 */
358 	if (p->p_sysent->sv_fixup && imgp->resident == 0)
359 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
360 	else
361 		suword(--stack_base, imgp->args->argc);
362 
363 	/*
364 	 * For security and other reasons, the file descriptor table cannot
365 	 * be shared after an exec.
366 	 */
367 	if (p->p_fd->fd_refcnt > 1) {
368 		struct filedesc *tmp;
369 
370 		error = fdcopy(p, &tmp);
371 		if (error != 0)
372 			goto exec_fail;
373 		fdfree(p, tmp);
374 	}
375 
376 	/*
377 	 * For security and other reasons, signal handlers cannot
378 	 * be shared after an exec. The new proces gets a copy of the old
379 	 * handlers. In execsigs(), the new process will have its signals
380 	 * reset.
381 	 */
382 	ops = p->p_sigacts;
383 	if (ops->ps_refcnt > 1) {
384 		nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
385 		bcopy(ops, nps, sizeof(*nps));
386 		refcount_init(&nps->ps_refcnt, 1);
387 		p->p_sigacts = nps;
388 		if (refcount_release(&ops->ps_refcnt)) {
389 			kfree(ops, M_SUBPROC);
390 			ops = NULL;
391 		}
392 	}
393 
394 	/*
395 	 * For security and other reasons virtual kernels cannot be
396 	 * inherited by an exec.  This also allows a virtual kernel
397 	 * to fork/exec unrelated applications.
398 	 */
399 	if (p->p_vkernel)
400 		vkernel_exit(p);
401 
402 	/* Stop profiling */
403 	stopprofclock(p);
404 
405 	/* close files on exec */
406 	fdcloseexec(p);
407 
408 	/* reset caught signals */
409 	execsigs(p);
410 
411 	/* name this process - nameiexec(p, ndp) */
412 	len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
413 	bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
414 	p->p_comm[len] = 0;
415 	bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
416 
417 	/*
418 	 * mark as execed, wakeup the process that vforked (if any) and tell
419 	 * it that it now has its own resources back
420 	 */
421 	p->p_flags |= P_EXEC;
422 	if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
423 		p->p_flags &= ~P_PPWAIT;
424 		wakeup((caddr_t)p->p_pptr);
425 	}
426 
427 	/*
428 	 * Implement image setuid/setgid.
429 	 *
430 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
431 	 * the process is being traced.
432 	 */
433 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
434 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
435 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
436 	    (p->p_flags & P_TRACED) == 0) {
437 		/*
438 		 * Turn off syscall tracing for set-id programs, except for
439 		 * root.  Record any set-id flags first to make sure that
440 		 * we do not regain any tracing during a possible block.
441 		 */
442 		setsugid();
443 		if (p->p_tracenode && ktrace_suid == 0 &&
444 		    priv_check(td, PRIV_ROOT) != 0) {
445 			ktrdestroy(&p->p_tracenode);
446 			p->p_traceflag = 0;
447 		}
448 		/* Close any file descriptors 0..2 that reference procfs */
449 		setugidsafety(p);
450 		/* Make sure file descriptors 0..2 are in use. */
451 		error = fdcheckstd(lp);
452 		if (error != 0)
453 			goto exec_fail_dealloc;
454 		/*
455 		 * Set the new credentials.
456 		 */
457 		cratom(&p->p_ucred);
458 		if (attr.va_mode & VSUID)
459 			change_euid(attr.va_uid);
460 		if (attr.va_mode & VSGID)
461 			p->p_ucred->cr_gid = attr.va_gid;
462 
463 		/*
464 		 * Clear local varsym variables
465 		 */
466 		varsymset_clean(&p->p_varsymset);
467 	} else {
468 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
469 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
470 			p->p_flags &= ~P_SUGID;
471 	}
472 
473 	/*
474 	 * Implement correct POSIX saved-id behavior.
475 	 */
476 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
477 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
478 		cratom(&p->p_ucred);
479 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
480 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
481 	}
482 
483 	/*
484 	 * Store the vp for use in procfs.  Be sure to keep p_textvp
485 	 * consistent if we block during the switch-over.
486 	 */
487 	ovp = p->p_textvp;
488 	vref(imgp->vp);			/* ref new vp */
489 	p->p_textvp = imgp->vp;
490 	if (ovp)			/* release old vp */
491 		vrele(ovp);
492 
493 	/* Release old namecache handle to text file */
494 	if (p->p_textnch.ncp)
495 		cache_drop(&p->p_textnch);
496 
497 	if (nd->nl_nch.mount)
498 		cache_copy(&nd->nl_nch, &p->p_textnch);
499 
500         /*
501          * Notify others that we exec'd, and clear the P_INEXEC flag
502          * as we're now a bona fide freshly-execed process.
503          */
504 	KNOTE(&p->p_klist, NOTE_EXEC);
505 	p->p_flags &= ~P_INEXEC;
506 	if (p->p_stops)
507 		wakeup(&p->p_stype);
508 
509 	/*
510 	 * If tracing the process, trap to debugger so breakpoints
511 	 * 	can be set before the program executes.
512 	 */
513 	STOPEVENT(p, S_EXEC, 0);
514 
515 	if (p->p_flags & P_TRACED)
516 		ksignal(p, SIGTRAP);
517 
518 	/* clear "fork but no exec" flag, as we _are_ execing */
519 	p->p_acflag &= ~AFORK;
520 
521 	/* Set values passed into the program in registers. */
522 	exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
523 		     imgp->ps_strings);
524 
525 	/* Set the access time on the vnode */
526 	vn_mark_atime(imgp->vp, td);
527 
528 	/*
529 	 * Free any previous argument cache
530 	 */
531 	pa = p->p_args;
532 	p->p_args = NULL;
533 	if (pa && refcount_release(&pa->ar_ref)) {
534 		kfree(pa, M_PARGS);
535 		pa = NULL;
536 	}
537 
538 	/*
539 	 * Cache arguments if they fit inside our allowance
540 	 */
541 	i = imgp->args->begin_envv - imgp->args->begin_argv;
542 	if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
543 		pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
544 		refcount_init(&pa->ar_ref, 1);
545 		pa->ar_length = i;
546 		bcopy(imgp->args->begin_argv, pa->ar_args, i);
547 		KKASSERT(p->p_args == NULL);
548 		p->p_args = pa;
549 	}
550 
551 exec_fail_dealloc:
552 
553 	/*
554 	 * free various allocated resources
555 	 */
556 	if (imgp->firstpage)
557 		exec_unmap_first_page(imgp);
558 
559 	if (imgp->vp) {
560 		vrele(imgp->vp);
561 		imgp->vp = NULL;
562 	}
563 
564 	if (imgp->freepath)
565 		kfree(imgp->freepath, M_TEMP);
566 
567 	if (error == 0) {
568 		++mycpu->gd_cnt.v_exec;
569 		lwkt_reltoken(&p->p_token);
570 		return (0);
571 	}
572 
573 exec_fail:
574 	/*
575 	 * we're done here, clear P_INEXEC if we were the ones that
576 	 * set it.  Otherwise if vmspace_destroyed is still set we
577 	 * raced another thread and that thread is responsible for
578 	 * clearing it.
579 	 */
580 	if (imgp->vmspace_destroyed & 2) {
581 		p->p_flags &= ~P_INEXEC;
582 		if (p->p_stops)
583 			wakeup(&p->p_stype);
584 	}
585 	lwkt_reltoken(&p->p_token);
586 	if (imgp->vmspace_destroyed) {
587 		/*
588 		 * Sorry, no more process anymore. exit gracefully.
589 		 * However we can't die right here, because our
590 		 * caller might have to clean up, so indicate a
591 		 * lethal error by returning -1.
592 		 */
593 		return(-1);
594 	} else {
595 		return(error);
596 	}
597 }
598 
599 /*
600  * execve() system call.
601  */
602 int
603 sys_execve(struct execve_args *uap)
604 {
605 	struct nlookupdata nd;
606 	struct image_args args;
607 	int error;
608 
609 	bzero(&args, sizeof(args));
610 
611 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
612 	if (error == 0) {
613 		error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
614 					uap->argv, uap->envv);
615 	}
616 	if (error == 0)
617 		error = kern_execve(&nd, &args);
618 	nlookup_done(&nd);
619 	exec_free_args(&args);
620 
621 	if (error < 0) {
622 		/* We hit a lethal error condition.  Let's die now. */
623 		exit1(W_EXITCODE(0, SIGABRT));
624 		/* NOTREACHED */
625 	}
626 
627 	/*
628 	 * The syscall result is returned in registers to the new program.
629 	 * Linux will register %edx as an atexit function and we must be
630 	 * sure to set it to 0.  XXX
631 	 */
632 	if (error == 0)
633 		uap->sysmsg_result64 = 0;
634 
635 	return (error);
636 }
637 
638 int
639 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
640 	      struct lwbuf **plwb, const char **pdata)
641 {
642 	int rv;
643 	vm_page_t ma;
644 	vm_page_t m;
645 	vm_object_t object;
646 
647 	/*
648 	 * The file has to be mappable.
649 	 */
650 	if ((object = imgp->vp->v_object) == NULL)
651 		return (EIO);
652 
653 	if (pageno >= object->size)
654 		return (EIO);
655 
656 	vm_object_hold(object);
657 	m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
658 	while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
659 		ma = m;
660 
661 		/*
662 		 * get_pages unbusies all the requested pages except the
663 		 * primary page (at index 0 in this case).  The primary
664 		 * page may have been wired during the pagein (e.g. by
665 		 * the buffer cache) so vnode_pager_freepage() must be
666 		 * used to properly release it.
667 		 */
668 		rv = vm_pager_get_page(object, &ma, 1);
669 		m = vm_page_lookup(object, pageno);
670 
671 		if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
672 			if (m) {
673 				vm_page_protect(m, VM_PROT_NONE);
674 				vnode_pager_freepage(m);
675 			}
676 			vm_object_drop(object);
677 			return EIO;
678 		}
679 	}
680 	vm_page_hold(m);
681 	vm_page_wakeup(m);	/* unbusy the page */
682 	vm_object_drop(object);
683 
684 	*plwb = lwbuf_alloc(m, *plwb);
685 	*pdata = (void *)lwbuf_kva(*plwb);
686 
687 	return (0);
688 }
689 
690 /*
691  * Map the first page of an executable image.
692  *
693  * NOTE: If the mapping fails we have to NULL-out firstpage which may
694  *	 still be pointing to our supplied lwp structure.
695  */
696 int
697 exec_map_first_page(struct image_params *imgp)
698 {
699 	int err;
700 
701 	if (imgp->firstpage)
702 		exec_unmap_first_page(imgp);
703 
704 	imgp->firstpage = &imgp->firstpage_cache;
705 	err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
706 
707 	if (err) {
708 		imgp->firstpage = NULL;
709 		return err;
710 	}
711 
712 	return 0;
713 }
714 
715 void
716 exec_unmap_page(struct lwbuf *lwb)
717 {
718 	vm_page_t m;
719 
720 	crit_enter();
721 	if (lwb != NULL) {
722 		m = lwbuf_page(lwb);
723 		lwbuf_free(lwb);
724 		vm_page_unhold(m);
725 	}
726 	crit_exit();
727 }
728 
729 void
730 exec_unmap_first_page(struct image_params *imgp)
731 {
732 	exec_unmap_page(imgp->firstpage);
733 	imgp->firstpage = NULL;
734 	imgp->image_header = NULL;
735 }
736 
737 /*
738  * Destroy old address space, and allocate a new stack
739  *	The new stack is only SGROWSIZ large because it is grown
740  *	automatically in trap.c.
741  *
742  * This is the point of no return.
743  */
744 int
745 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
746 {
747 	struct vmspace *vmspace = imgp->proc->p_vmspace;
748 	vm_offset_t stack_addr = USRSTACK - maxssiz;
749 	struct proc *p;
750 	vm_map_t map;
751 	int error;
752 
753 	/*
754 	 * Indicate that we cannot gracefully error out any more, kill
755 	 * any other threads present, and set P_INEXEC to indicate that
756 	 * we are now messing with the process structure proper.
757 	 *
758 	 * If killalllwps() races return an error which coupled with
759 	 * vmspace_destroyed will cause us to exit.  This is what we
760 	 * want since another thread is patiently waiting for us to exit
761 	 * in that case.
762 	 */
763 	p = curproc;
764 	imgp->vmspace_destroyed = 1;
765 
766 	if (curthread->td_proc->p_nthreads > 1) {
767 		error = killalllwps(1);
768 		if (error)
769 			return (error);
770 	}
771 	imgp->vmspace_destroyed |= 2;	/* we are responsible for P_INEXEC */
772 	p->p_flags |= P_INEXEC;
773 
774 	/*
775 	 * Tell procfs to release its hold on the process.  It
776 	 * will return EAGAIN.
777 	 */
778 	if (p->p_stops)
779 		wakeup(&p->p_stype);
780 
781 	/*
782 	 * After setting P_INEXEC wait for any remaining references to
783 	 * the process (p) to go away.
784 	 *
785 	 * In particular, a vfork/exec sequence will replace p->p_vmspace
786 	 * and we must interlock anyone trying to access the space (aka
787 	 * procfs or sys_process.c calling procfs_domem()).
788 	 *
789 	 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
790 	 */
791 	PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
792 
793 	/*
794 	 * Blow away entire process VM, if address space not shared,
795 	 * otherwise, create a new VM space so that other threads are
796 	 * not disrupted.  If we are execing a resident vmspace we
797 	 * create a duplicate of it and remap the stack.
798 	 */
799 	map = &vmspace->vm_map;
800 	if (vmcopy) {
801 		vmspace_exec(imgp->proc, vmcopy);
802 		vmspace = imgp->proc->p_vmspace;
803 		pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
804 		map = &vmspace->vm_map;
805 	} else if (vmspace->vm_sysref.refcnt == 1) {
806 		shmexit(vmspace);
807 		pmap_remove_pages(vmspace_pmap(vmspace),
808 				  0, VM_MAX_USER_ADDRESS);
809 		vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
810 	} else {
811 		vmspace_exec(imgp->proc, NULL);
812 		vmspace = imgp->proc->p_vmspace;
813 		map = &vmspace->vm_map;
814 	}
815 
816 	/* Allocate a new stack */
817 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
818 			     0, VM_PROT_ALL, VM_PROT_ALL, 0);
819 	if (error)
820 		return (error);
821 
822 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
823 	 * VM_STACK case, but they are still used to monitor the size of the
824 	 * process stack so we can check the stack rlimit.
825 	 */
826 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
827 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
828 
829 	return(0);
830 }
831 
832 /*
833  * Copy out argument and environment strings from the old process
834  *	address space into the temporary string buffer.
835  */
836 int
837 exec_copyin_args(struct image_args *args, char *fname,
838 		enum exec_path_segflg segflg, char **argv, char **envv)
839 {
840 	char	*argp, *envp;
841 	int	error = 0;
842 	size_t	length;
843 
844 	args->buf = objcache_get(exec_objcache, M_WAITOK);
845 	if (args->buf == NULL)
846 		return (ENOMEM);
847 	args->begin_argv = args->buf;
848 	args->endp = args->begin_argv;
849 	args->space = ARG_MAX;
850 
851 	args->fname = args->buf + ARG_MAX;
852 
853 	/*
854 	 * Copy the file name.
855 	 */
856 	if (segflg == PATH_SYSSPACE) {
857 		error = copystr(fname, args->fname, PATH_MAX, &length);
858 	} else if (segflg == PATH_USERSPACE) {
859 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
860 	}
861 
862 	/*
863 	 * Extract argument strings.  argv may not be NULL.  The argv
864 	 * array is terminated by a NULL entry.  We special-case the
865 	 * situation where argv[0] is NULL by passing { filename, NULL }
866 	 * to the new program to guarentee that the interpreter knows what
867 	 * file to open in case we exec an interpreted file.   Note that
868 	 * a NULL argv[0] terminates the argv[] array.
869 	 *
870 	 * XXX the special-casing of argv[0] is historical and needs to be
871 	 * revisited.
872 	 */
873 	if (argv == NULL)
874 		error = EFAULT;
875 	if (error == 0) {
876 		while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
877 			if (argp == (caddr_t)-1) {
878 				error = EFAULT;
879 				break;
880 			}
881 			error = copyinstr(argp, args->endp,
882 					  args->space, &length);
883 			if (error) {
884 				if (error == ENAMETOOLONG)
885 					error = E2BIG;
886 				break;
887 			}
888 			args->space -= length;
889 			args->endp += length;
890 			args->argc++;
891 		}
892 		if (args->argc == 0 && error == 0) {
893 			length = strlen(args->fname) + 1;
894 			if (length > args->space) {
895 				error = E2BIG;
896 			} else {
897 				bcopy(args->fname, args->endp, length);
898 				args->space -= length;
899 				args->endp += length;
900 				args->argc++;
901 			}
902 		}
903 	}
904 
905 	args->begin_envv = args->endp;
906 
907 	/*
908 	 * extract environment strings.  envv may be NULL.
909 	 */
910 	if (envv && error == 0) {
911 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
912 			if (envp == (caddr_t) -1) {
913 				error = EFAULT;
914 				break;
915 			}
916 			error = copyinstr(envp, args->endp,
917 					  args->space, &length);
918 			if (error) {
919 				if (error == ENAMETOOLONG)
920 					error = E2BIG;
921 				break;
922 			}
923 			args->space -= length;
924 			args->endp += length;
925 			args->envc++;
926 		}
927 	}
928 	return (error);
929 }
930 
931 void
932 exec_free_args(struct image_args *args)
933 {
934 	if (args->buf) {
935 		objcache_put(exec_objcache, args->buf);
936 		args->buf = NULL;
937 	}
938 }
939 
940 /*
941  * Copy strings out to the new process address space, constructing
942  * new arg and env vector tables. Return a pointer to the base
943  * so that it can be used as the initial stack pointer.
944  *
945  * The format is, roughly:
946  *
947  *	[argv[]]			<-- vectp
948  *	[envp[]]
949  *	[ELF_Auxargs]
950  *
951  *	[args & env]			<-- destp
952  *	[sgap]
953  *	[SPARE_USRSPACE]
954  *	[execpath]
955  *	[szsigcode]
956  *	[ps_strings]			top of user stack
957  *
958  */
959 register_t *
960 exec_copyout_strings(struct image_params *imgp)
961 {
962 	int argc, envc, sgap;
963 	int gap;
964 	int argsenvspace;
965 	char **vectp;
966 	char *stringp, *destp;
967 	register_t *stack_base;
968 	struct ps_strings *arginfo;
969 	size_t execpath_len;
970 	int szsigcode;
971 
972 	/*
973 	 * Calculate string base and vector table pointers.
974 	 * Also deal with signal trampoline code for this exec type.
975 	 */
976 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
977 		execpath_len = strlen(imgp->execpath) + 1;
978 	else
979 		execpath_len = 0;
980 	arginfo = (struct ps_strings *)PS_STRINGS;
981 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
982 
983 	argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *));
984 	gap = stackgap_random;
985 	cpu_ccfence();
986 	if (gap != 0) {
987 		if (gap < 0)
988 			sgap = ALIGN(-gap);
989 		else
990 			sgap = ALIGN(karc4random() & (gap - 1));
991 	} else {
992 		sgap = 0;
993 	}
994 
995 	/*
996 	 * Calculate destp, which points to [args & env] and above.
997 	 */
998 	destp = (caddr_t)arginfo -
999 		szsigcode -
1000 		roundup(execpath_len, sizeof(char *)) -
1001 		SPARE_USRSPACE -
1002 		sgap -
1003 		argsenvspace;
1004 
1005 	/*
1006 	 * install sigcode
1007 	 */
1008 	if (szsigcode) {
1009 		copyout(imgp->proc->p_sysent->sv_sigcode,
1010 			((caddr_t)arginfo - szsigcode), szsigcode);
1011 	}
1012 
1013 	/*
1014 	 * Copy the image path for the rtld
1015 	 */
1016 	if (execpath_len) {
1017 		imgp->execpathp = (uintptr_t)arginfo
1018 				  - szsigcode
1019 				  - roundup(execpath_len, sizeof(char *));
1020 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
1021 	}
1022 
1023 	/*
1024 	 * Calculate base for argv[], envp[], and ELF_Auxargs.
1025 	 */
1026 	vectp = (char **)destp - (AT_COUNT * 2);
1027 	vectp -= imgp->args->argc + imgp->args->envc + 2;
1028 
1029 	stack_base = (register_t *)vectp;
1030 
1031 	stringp = imgp->args->begin_argv;
1032 	argc = imgp->args->argc;
1033 	envc = imgp->args->envc;
1034 
1035 	/*
1036 	 * Copy out strings - arguments and environment (at destp)
1037 	 */
1038 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
1039 
1040 	/*
1041 	 * Fill in "ps_strings" struct for ps, w, etc.
1042 	 */
1043 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1044 	suword32(&arginfo->ps_nargvstr, argc);
1045 
1046 	/*
1047 	 * Fill in argument portion of vector table.
1048 	 */
1049 	for (; argc > 0; --argc) {
1050 		suword(vectp++, (long)(intptr_t)destp);
1051 		while (*stringp++ != 0)
1052 			destp++;
1053 		destp++;
1054 	}
1055 
1056 	/* a null vector table pointer separates the argp's from the envp's */
1057 	suword(vectp++, 0);
1058 
1059 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1060 	suword32(&arginfo->ps_nenvstr, envc);
1061 
1062 	/*
1063 	 * Fill in environment portion of vector table.
1064 	 */
1065 	for (; envc > 0; --envc) {
1066 		suword(vectp++, (long)(intptr_t)destp);
1067 		while (*stringp++ != 0)
1068 			destp++;
1069 		destp++;
1070 	}
1071 
1072 	/* end of vector table is a null pointer */
1073 	suword(vectp, 0);
1074 
1075 	return (stack_base);
1076 }
1077 
1078 /*
1079  * Check permissions of file to execute.
1080  *	Return 0 for success or error code on failure.
1081  */
1082 int
1083 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
1084 {
1085 	struct proc *p = imgp->proc;
1086 	struct vnode *vp = imgp->vp;
1087 	struct vattr *attr = imgp->attr;
1088 	int error;
1089 
1090 	/* Get file attributes */
1091 	error = VOP_GETATTR(vp, attr);
1092 	if (error)
1093 		return (error);
1094 
1095 	/*
1096 	 * 1) Check if file execution is disabled for the filesystem that this
1097 	 *	file resides on.
1098 	 * 2) Insure that at least one execute bit is on - otherwise root
1099 	 *	will always succeed, and we don't want to happen unless the
1100 	 *	file really is executable.
1101 	 * 3) Insure that the file is a regular file.
1102 	 */
1103 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1104 	    ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1105 	    ((attr->va_mode & 0111) == 0) ||
1106 	    (attr->va_type != VREG)) {
1107 		return (EACCES);
1108 	}
1109 
1110 	/*
1111 	 * Zero length files can't be exec'd
1112 	 */
1113 	if (attr->va_size == 0)
1114 		return (ENOEXEC);
1115 
1116 	/*
1117 	 *  Check for execute permission to file based on current credentials.
1118 	 */
1119 	error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1120 	if (error)
1121 		return (error);
1122 
1123 	/*
1124 	 * Check number of open-for-writes on the file and deny execution
1125 	 * if there are any.
1126 	 */
1127 	if (vp->v_writecount)
1128 		return (ETXTBSY);
1129 
1130 	/*
1131 	 * Call filesystem specific open routine, which allows us to read,
1132 	 * write, and mmap the file.  Without the VOP_OPEN we can only
1133 	 * stat the file.
1134 	 */
1135 	error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1136 	if (error)
1137 		return (error);
1138 
1139 	return (0);
1140 }
1141 
1142 /*
1143  * Exec handler registration
1144  */
1145 int
1146 exec_register(const struct execsw *execsw_arg)
1147 {
1148 	const struct execsw **es, **xs, **newexecsw;
1149 	int count = 2;	/* New slot and trailing NULL */
1150 
1151 	if (execsw)
1152 		for (es = execsw; *es; es++)
1153 			count++;
1154 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1155 	xs = newexecsw;
1156 	if (execsw)
1157 		for (es = execsw; *es; es++)
1158 			*xs++ = *es;
1159 	*xs++ = execsw_arg;
1160 	*xs = NULL;
1161 	if (execsw)
1162 		kfree(execsw, M_TEMP);
1163 	execsw = newexecsw;
1164 	return 0;
1165 }
1166 
1167 int
1168 exec_unregister(const struct execsw *execsw_arg)
1169 {
1170 	const struct execsw **es, **xs, **newexecsw;
1171 	int count = 1;
1172 
1173 	if (execsw == NULL)
1174 		panic("unregister with no handlers left?");
1175 
1176 	for (es = execsw; *es; es++) {
1177 		if (*es == execsw_arg)
1178 			break;
1179 	}
1180 	if (*es == NULL)
1181 		return ENOENT;
1182 	for (es = execsw; *es; es++)
1183 		if (*es != execsw_arg)
1184 			count++;
1185 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1186 	xs = newexecsw;
1187 	for (es = execsw; *es; es++)
1188 		if (*es != execsw_arg)
1189 			*xs++ = *es;
1190 	*xs = NULL;
1191 	if (execsw)
1192 		kfree(execsw, M_TEMP);
1193 	execsw = newexecsw;
1194 	return 0;
1195 }
1196