xref: /dragonfly/sys/kern/kern_exec.c (revision 95d468db)
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 	nd->nl_flags |= NLC_SHAREDLOCK;
250 	if ((error = nlookup(nd)) != 0)
251 		goto exec_fail;
252 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &imgp->vp);
253 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
254 	nd->nl_flags &= ~NLC_NCPISLOCKED;
255 	cache_unlock(&nd->nl_nch);
256 	if (error)
257 		goto exec_fail;
258 
259 	/*
260 	 * Check file permissions (also 'opens' file).
261 	 * Include also the top level mount in the check.
262 	 */
263 	error = exec_check_permissions(imgp, nd->nl_nch.mount);
264 	if (error) {
265 		vn_unlock(imgp->vp);
266 		goto exec_fail_dealloc;
267 	}
268 
269 	error = exec_map_first_page(imgp);
270 	vn_unlock(imgp->vp);
271 	if (error)
272 		goto exec_fail_dealloc;
273 
274 	imgp->proc->p_osrel = 0;
275 
276 	if (debug_execve_args && imgp->interpreted) {
277 		kprintf("    target is interpreted -- recursive pass\n");
278 		kprintf("    interpreter: %s\n", imgp->interpreter_name);
279 		print_execve_args(args);
280 	}
281 
282 	/*
283 	 *	If the current process has a special image activator it
284 	 *	wants to try first, call it.   For example, emulating shell
285 	 *	scripts differently.
286 	 */
287 	error = -1;
288 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
289 		error = img_first(imgp);
290 
291 	/*
292 	 *	If the vnode has a registered vmspace, exec the vmspace
293 	 */
294 	if (error == -1 && imgp->vp->v_resident) {
295 		error = exec_resident_imgact(imgp);
296 	}
297 
298 	/*
299 	 *	Loop through the list of image activators, calling each one.
300 	 *	An activator returns -1 if there is no match, 0 on success,
301 	 *	and an error otherwise.
302 	 */
303 	for (i = 0; error == -1 && execsw[i]; ++i) {
304 		if (execsw[i]->ex_imgact == NULL ||
305 		    execsw[i]->ex_imgact == img_first) {
306 			continue;
307 		}
308 		error = (*execsw[i]->ex_imgact)(imgp);
309 	}
310 
311 	if (error) {
312 		if (error == -1)
313 			error = ENOEXEC;
314 		goto exec_fail_dealloc;
315 	}
316 
317 	/*
318 	 * Special interpreter operation, cleanup and loop up to try to
319 	 * activate the interpreter.
320 	 */
321 	if (imgp->interpreted) {
322 		exec_unmap_first_page(imgp);
323 		nlookup_done(nd);
324 		vrele(imgp->vp);
325 		imgp->vp = NULL;
326 		error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
327 					NLC_FOLLOW);
328 		if (error)
329 			goto exec_fail;
330 		goto interpret;
331 	}
332 
333 	/*
334 	 * Do the best to calculate the full path to the image file
335 	 */
336 	if (imgp->auxargs != NULL &&
337 	   ((args->fname != NULL && args->fname[0] == '/') ||
338 	    vn_fullpath(imgp->proc,
339 			imgp->vp,
340 			&imgp->execpath,
341 			&imgp->freepath,
342 			0) != 0))
343 		imgp->execpath = args->fname;
344 
345 	/*
346 	 * Copy out strings (args and env) and initialize stack base
347 	 */
348 	stack_base = exec_copyout_strings(imgp);
349 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
350 
351 	/*
352 	 * If custom stack fixup routine present for this process
353 	 * let it do the stack setup.  If we are running a resident
354 	 * image there is no auxinfo or other image activator context
355 	 * so don't try to add fixups to the stack.
356 	 *
357 	 * Else stuff argument count as first item on stack
358 	 */
359 	if (p->p_sysent->sv_fixup && imgp->resident == 0)
360 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
361 	else
362 		suword(--stack_base, imgp->args->argc);
363 
364 	/*
365 	 * For security and other reasons, the file descriptor table cannot
366 	 * be shared after an exec.
367 	 */
368 	if (p->p_fd->fd_refcnt > 1) {
369 		struct filedesc *tmp;
370 
371 		error = fdcopy(p, &tmp);
372 		if (error != 0)
373 			goto exec_fail;
374 		fdfree(p, tmp);
375 	}
376 
377 	/*
378 	 * For security and other reasons, signal handlers cannot
379 	 * be shared after an exec. The new proces gets a copy of the old
380 	 * handlers. In execsigs(), the new process will have its signals
381 	 * reset.
382 	 */
383 	ops = p->p_sigacts;
384 	if (ops->ps_refcnt > 1) {
385 		nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK);
386 		bcopy(ops, nps, sizeof(*nps));
387 		refcount_init(&nps->ps_refcnt, 1);
388 		p->p_sigacts = nps;
389 		if (refcount_release(&ops->ps_refcnt)) {
390 			kfree(ops, M_SUBPROC);
391 			ops = NULL;
392 		}
393 	}
394 
395 	/*
396 	 * For security and other reasons virtual kernels cannot be
397 	 * inherited by an exec.  This also allows a virtual kernel
398 	 * to fork/exec unrelated applications.
399 	 */
400 	if (p->p_vkernel)
401 		vkernel_exit(p);
402 
403 	/* Stop profiling */
404 	stopprofclock(p);
405 
406 	/* close files on exec */
407 	fdcloseexec(p);
408 
409 	/* reset caught signals */
410 	execsigs(p);
411 
412 	/* name this process - nameiexec(p, ndp) */
413 	len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
414 	bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
415 	p->p_comm[len] = 0;
416 	bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
417 
418 	/*
419 	 * mark as execed, wakeup the process that vforked (if any) and tell
420 	 * it that it now has its own resources back
421 	 *
422 	 * We are using the P_PPWAIT as an interlock so an atomic op is
423 	 * necessary to synchronize with the parent's cpu.
424 	 */
425 	p->p_flags |= P_EXEC;
426 	if (p->p_pptr && (p->p_flags & P_PPWAIT)) {
427 		if (p->p_pptr->p_upmap)
428 			p->p_pptr->p_upmap->invfork = 0;
429 		atomic_clear_int(&p->p_flags, P_PPWAIT);
430 		wakeup(p->p_pptr);
431 	}
432 
433 	/*
434 	 * Implement image setuid/setgid.
435 	 *
436 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
437 	 * the process is being traced.
438 	 */
439 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
440 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
441 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
442 	    (p->p_flags & P_TRACED) == 0) {
443 		/*
444 		 * Turn off syscall tracing for set-id programs, except for
445 		 * root.  Record any set-id flags first to make sure that
446 		 * we do not regain any tracing during a possible block.
447 		 */
448 		setsugid();
449 		if (p->p_tracenode && ktrace_suid == 0 &&
450 		    priv_check(td, PRIV_ROOT) != 0) {
451 			ktrdestroy(&p->p_tracenode);
452 			p->p_traceflag = 0;
453 		}
454 		/* Close any file descriptors 0..2 that reference procfs */
455 		setugidsafety(p);
456 		/* Make sure file descriptors 0..2 are in use. */
457 		error = fdcheckstd(lp);
458 		if (error != 0)
459 			goto exec_fail_dealloc;
460 		/*
461 		 * Set the new credentials.
462 		 */
463 		cratom_proc(p);
464 		if (attr.va_mode & VSUID)
465 			change_euid(attr.va_uid);
466 		if (attr.va_mode & VSGID)
467 			p->p_ucred->cr_gid = attr.va_gid;
468 
469 		/*
470 		 * Clear local varsym variables
471 		 */
472 		varsymset_clean(&p->p_varsymset);
473 	} else {
474 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
475 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
476 			p->p_flags &= ~P_SUGID;
477 	}
478 
479 	/*
480 	 * Implement correct POSIX saved-id behavior.
481 	 */
482 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
483 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
484 		cratom_proc(p);
485 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
486 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
487 	}
488 
489 	/*
490 	 * Store the vp for use in procfs.  Be sure to keep p_textvp
491 	 * consistent if we block during the switch-over.
492 	 */
493 	ovp = p->p_textvp;
494 	vref(imgp->vp);			/* ref new vp */
495 	p->p_textvp = imgp->vp;
496 	if (ovp)			/* release old vp */
497 		vrele(ovp);
498 
499 	/* Release old namecache handle to text file */
500 	if (p->p_textnch.ncp)
501 		cache_drop(&p->p_textnch);
502 
503 	if (nd->nl_nch.mount)
504 		cache_copy(&nd->nl_nch, &p->p_textnch);
505 
506         /*
507          * Notify others that we exec'd, and clear the P_INEXEC flag
508          * as we're now a bona fide freshly-execed process.
509          */
510 	KNOTE(&p->p_klist, NOTE_EXEC);
511 	p->p_flags &= ~P_INEXEC;
512 	if (p->p_stops)
513 		wakeup(&p->p_stype);
514 
515 	/*
516 	 * If tracing the process, trap to debugger so breakpoints
517 	 * 	can be set before the program executes.
518 	 */
519 	STOPEVENT(p, S_EXEC, 0);
520 
521 	if (p->p_flags & P_TRACED)
522 		ksignal(p, SIGTRAP);
523 
524 	/* clear "fork but no exec" flag, as we _are_ execing */
525 	p->p_acflag &= ~AFORK;
526 
527 	/* Set values passed into the program in registers. */
528 	exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
529 		     imgp->ps_strings);
530 
531 	/* Set the access time on the vnode */
532 	vn_mark_atime(imgp->vp, td);
533 
534 	/*
535 	 * Free any previous argument cache
536 	 */
537 	pa = p->p_args;
538 	p->p_args = NULL;
539 	if (pa && refcount_release(&pa->ar_ref)) {
540 		kfree(pa, M_PARGS);
541 		pa = NULL;
542 	}
543 
544 	/*
545 	 * Cache arguments if they fit inside our allowance
546 	 */
547 	i = imgp->args->begin_envv - imgp->args->begin_argv;
548 	if (sizeof(struct pargs) + i <= ps_arg_cache_limit) {
549 		pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK);
550 		refcount_init(&pa->ar_ref, 1);
551 		pa->ar_length = i;
552 		bcopy(imgp->args->begin_argv, pa->ar_args, i);
553 		KKASSERT(p->p_args == NULL);
554 		p->p_args = pa;
555 	}
556 
557 exec_fail_dealloc:
558 
559 	/*
560 	 * free various allocated resources
561 	 */
562 	if (imgp->firstpage)
563 		exec_unmap_first_page(imgp);
564 
565 	if (imgp->vp) {
566 		vrele(imgp->vp);
567 		imgp->vp = NULL;
568 	}
569 
570 	if (imgp->freepath)
571 		kfree(imgp->freepath, M_TEMP);
572 
573 	if (error == 0) {
574 		++mycpu->gd_cnt.v_exec;
575 		lwkt_reltoken(&p->p_token);
576 		return (0);
577 	}
578 
579 exec_fail:
580 	/*
581 	 * we're done here, clear P_INEXEC if we were the ones that
582 	 * set it.  Otherwise if vmspace_destroyed is still set we
583 	 * raced another thread and that thread is responsible for
584 	 * clearing it.
585 	 */
586 	if (imgp->vmspace_destroyed & 2) {
587 		p->p_flags &= ~P_INEXEC;
588 		if (p->p_stops)
589 			wakeup(&p->p_stype);
590 	}
591 	lwkt_reltoken(&p->p_token);
592 	if (imgp->vmspace_destroyed) {
593 		/*
594 		 * Sorry, no more process anymore. exit gracefully.
595 		 * However we can't die right here, because our
596 		 * caller might have to clean up, so indicate a
597 		 * lethal error by returning -1.
598 		 */
599 		return(-1);
600 	} else {
601 		return(error);
602 	}
603 }
604 
605 /*
606  * execve() system call.
607  */
608 int
609 sys_execve(struct execve_args *uap)
610 {
611 	struct nlookupdata nd;
612 	struct image_args args;
613 	int error;
614 
615 	bzero(&args, sizeof(args));
616 
617 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
618 	if (error == 0) {
619 		error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
620 					uap->argv, uap->envv);
621 	}
622 	if (error == 0)
623 		error = kern_execve(&nd, &args);
624 	nlookup_done(&nd);
625 	exec_free_args(&args);
626 
627 	if (error < 0) {
628 		/* We hit a lethal error condition.  Let's die now. */
629 		exit1(W_EXITCODE(0, SIGABRT));
630 		/* NOTREACHED */
631 	}
632 
633 	/*
634 	 * The syscall result is returned in registers to the new program.
635 	 * Linux will register %edx as an atexit function and we must be
636 	 * sure to set it to 0.  XXX
637 	 */
638 	if (error == 0)
639 		uap->sysmsg_result64 = 0;
640 
641 	return (error);
642 }
643 
644 int
645 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
646 	      struct lwbuf **plwb, const char **pdata)
647 {
648 	int rv;
649 	vm_page_t ma;
650 	vm_page_t m;
651 	vm_object_t object;
652 
653 	/*
654 	 * The file has to be mappable.
655 	 */
656 	if ((object = imgp->vp->v_object) == NULL)
657 		return (EIO);
658 
659 	if (pageno >= object->size)
660 		return (EIO);
661 
662 	/*
663 	 * Shortcut using shared locks, improve concurrent execs.
664 	 */
665 	vm_object_hold_shared(object);
666 	m = vm_page_lookup(object, pageno);
667 	if (m) {
668 		if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) {
669 			vm_page_hold(m);
670 			vm_page_sleep_busy(m, FALSE, "execpg");
671 			if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL &&
672 			    m->object == object && m->pindex == pageno) {
673 				vm_object_drop(object);
674 				goto done;
675 			}
676 			vm_page_unhold(m);
677 		}
678 	}
679 	vm_object_drop(object);
680 
681 	/*
682 	 * Do it the hard way
683 	 */
684 	vm_object_hold(object);
685 	m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
686 	while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
687 		ma = m;
688 
689 		/*
690 		 * get_pages unbusies all the requested pages except the
691 		 * primary page (at index 0 in this case).  The primary
692 		 * page may have been wired during the pagein (e.g. by
693 		 * the buffer cache) so vnode_pager_freepage() must be
694 		 * used to properly release it.
695 		 */
696 		rv = vm_pager_get_page(object, &ma, 1);
697 		m = vm_page_lookup(object, pageno);
698 
699 		if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
700 			if (m) {
701 				vm_page_protect(m, VM_PROT_NONE);
702 				vnode_pager_freepage(m);
703 			}
704 			vm_object_drop(object);
705 			return EIO;
706 		}
707 	}
708 	vm_page_hold(m);
709 	vm_page_wakeup(m);	/* unbusy the page */
710 	vm_object_drop(object);
711 
712 done:
713 	*plwb = lwbuf_alloc(m, *plwb);
714 	*pdata = (void *)lwbuf_kva(*plwb);
715 
716 	return (0);
717 }
718 
719 /*
720  * Map the first page of an executable image.
721  *
722  * NOTE: If the mapping fails we have to NULL-out firstpage which may
723  *	 still be pointing to our supplied lwp structure.
724  */
725 int
726 exec_map_first_page(struct image_params *imgp)
727 {
728 	int err;
729 
730 	if (imgp->firstpage)
731 		exec_unmap_first_page(imgp);
732 
733 	imgp->firstpage = &imgp->firstpage_cache;
734 	err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
735 
736 	if (err) {
737 		imgp->firstpage = NULL;
738 		return err;
739 	}
740 
741 	return 0;
742 }
743 
744 void
745 exec_unmap_page(struct lwbuf *lwb)
746 {
747 	vm_page_t m;
748 
749 	crit_enter();
750 	if (lwb != NULL) {
751 		m = lwbuf_page(lwb);
752 		lwbuf_free(lwb);
753 		vm_page_unhold(m);
754 	}
755 	crit_exit();
756 }
757 
758 void
759 exec_unmap_first_page(struct image_params *imgp)
760 {
761 	exec_unmap_page(imgp->firstpage);
762 	imgp->firstpage = NULL;
763 	imgp->image_header = NULL;
764 }
765 
766 /*
767  * Destroy old address space, and allocate a new stack
768  *	The new stack is only SGROWSIZ large because it is grown
769  *	automatically in trap.c.
770  *
771  * This is the point of no return.
772  */
773 int
774 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
775 {
776 	struct vmspace *vmspace = imgp->proc->p_vmspace;
777 	vm_offset_t stack_addr = USRSTACK - maxssiz;
778 	struct proc *p;
779 	vm_map_t map;
780 	int error;
781 
782 	/*
783 	 * Indicate that we cannot gracefully error out any more, kill
784 	 * any other threads present, and set P_INEXEC to indicate that
785 	 * we are now messing with the process structure proper.
786 	 *
787 	 * If killalllwps() races return an error which coupled with
788 	 * vmspace_destroyed will cause us to exit.  This is what we
789 	 * want since another thread is patiently waiting for us to exit
790 	 * in that case.
791 	 */
792 	p = curproc;
793 	imgp->vmspace_destroyed = 1;
794 
795 	if (curthread->td_proc->p_nthreads > 1) {
796 		error = killalllwps(1);
797 		if (error)
798 			return (error);
799 	}
800 	imgp->vmspace_destroyed |= 2;	/* we are responsible for P_INEXEC */
801 	p->p_flags |= P_INEXEC;
802 
803 	/*
804 	 * Tell procfs to release its hold on the process.  It
805 	 * will return EAGAIN.
806 	 */
807 	if (p->p_stops)
808 		wakeup(&p->p_stype);
809 
810 	/*
811 	 * After setting P_INEXEC wait for any remaining references to
812 	 * the process (p) to go away.
813 	 *
814 	 * In particular, a vfork/exec sequence will replace p->p_vmspace
815 	 * and we must interlock anyone trying to access the space (aka
816 	 * procfs or sys_process.c calling procfs_domem()).
817 	 *
818 	 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us.
819 	 */
820 	PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0));
821 
822 	/*
823 	 * Blow away entire process VM, if address space not shared,
824 	 * otherwise, create a new VM space so that other threads are
825 	 * not disrupted.  If we are execing a resident vmspace we
826 	 * create a duplicate of it and remap the stack.
827 	 */
828 	map = &vmspace->vm_map;
829 	if (vmcopy) {
830 		vmspace_exec(imgp->proc, vmcopy);
831 		vmspace = imgp->proc->p_vmspace;
832 		pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
833 		map = &vmspace->vm_map;
834 	} else if (vmspace_getrefs(vmspace) == 1) {
835 		shmexit(vmspace);
836 		pmap_remove_pages(vmspace_pmap(vmspace),
837 				  0, VM_MAX_USER_ADDRESS);
838 		vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
839 	} else {
840 		vmspace_exec(imgp->proc, NULL);
841 		vmspace = imgp->proc->p_vmspace;
842 		map = &vmspace->vm_map;
843 	}
844 
845 	/* Allocate a new stack */
846 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
847 			     0, VM_PROT_ALL, VM_PROT_ALL, 0);
848 	if (error)
849 		return (error);
850 
851 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
852 	 * VM_STACK case, but they are still used to monitor the size of the
853 	 * process stack so we can check the stack rlimit.
854 	 */
855 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
856 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
857 
858 	return(0);
859 }
860 
861 /*
862  * Copy out argument and environment strings from the old process
863  *	address space into the temporary string buffer.
864  */
865 int
866 exec_copyin_args(struct image_args *args, char *fname,
867 		enum exec_path_segflg segflg, char **argv, char **envv)
868 {
869 	char	*argp, *envp;
870 	int	error = 0;
871 	size_t	length;
872 
873 	args->buf = objcache_get(exec_objcache, M_WAITOK);
874 	if (args->buf == NULL)
875 		return (ENOMEM);
876 	args->begin_argv = args->buf;
877 	args->endp = args->begin_argv;
878 	args->space = ARG_MAX;
879 
880 	args->fname = args->buf + ARG_MAX;
881 
882 	/*
883 	 * Copy the file name.
884 	 */
885 	if (segflg == PATH_SYSSPACE) {
886 		error = copystr(fname, args->fname, PATH_MAX, &length);
887 	} else if (segflg == PATH_USERSPACE) {
888 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
889 	}
890 
891 	/*
892 	 * Extract argument strings.  argv may not be NULL.  The argv
893 	 * array is terminated by a NULL entry.  We special-case the
894 	 * situation where argv[0] is NULL by passing { filename, NULL }
895 	 * to the new program to guarentee that the interpreter knows what
896 	 * file to open in case we exec an interpreted file.   Note that
897 	 * a NULL argv[0] terminates the argv[] array.
898 	 *
899 	 * XXX the special-casing of argv[0] is historical and needs to be
900 	 * revisited.
901 	 */
902 	if (argv == NULL)
903 		error = EFAULT;
904 	if (error == 0) {
905 		while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
906 			if (argp == (caddr_t)-1) {
907 				error = EFAULT;
908 				break;
909 			}
910 			error = copyinstr(argp, args->endp,
911 					  args->space, &length);
912 			if (error) {
913 				if (error == ENAMETOOLONG)
914 					error = E2BIG;
915 				break;
916 			}
917 			args->space -= length;
918 			args->endp += length;
919 			args->argc++;
920 		}
921 		if (args->argc == 0 && error == 0) {
922 			length = strlen(args->fname) + 1;
923 			if (length > args->space) {
924 				error = E2BIG;
925 			} else {
926 				bcopy(args->fname, args->endp, length);
927 				args->space -= length;
928 				args->endp += length;
929 				args->argc++;
930 			}
931 		}
932 	}
933 
934 	args->begin_envv = args->endp;
935 
936 	/*
937 	 * extract environment strings.  envv may be NULL.
938 	 */
939 	if (envv && error == 0) {
940 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
941 			if (envp == (caddr_t) -1) {
942 				error = EFAULT;
943 				break;
944 			}
945 			error = copyinstr(envp, args->endp,
946 					  args->space, &length);
947 			if (error) {
948 				if (error == ENAMETOOLONG)
949 					error = E2BIG;
950 				break;
951 			}
952 			args->space -= length;
953 			args->endp += length;
954 			args->envc++;
955 		}
956 	}
957 	return (error);
958 }
959 
960 void
961 exec_free_args(struct image_args *args)
962 {
963 	if (args->buf) {
964 		objcache_put(exec_objcache, args->buf);
965 		args->buf = NULL;
966 	}
967 }
968 
969 /*
970  * Copy strings out to the new process address space, constructing
971  * new arg and env vector tables. Return a pointer to the base
972  * so that it can be used as the initial stack pointer.
973  *
974  * The format is, roughly:
975  *
976  *	[argv[]]			<-- vectp
977  *	[envp[]]
978  *	[ELF_Auxargs]
979  *
980  *	[args & env]			<-- destp
981  *	[sgap]
982  *	[SPARE_USRSPACE]
983  *	[execpath]
984  *	[szsigcode]
985  *	[ps_strings]			top of user stack
986  *
987  */
988 register_t *
989 exec_copyout_strings(struct image_params *imgp)
990 {
991 	int argc, envc, sgap;
992 	int gap;
993 	int argsenvspace;
994 	char **vectp;
995 	char *stringp, *destp;
996 	register_t *stack_base;
997 	struct ps_strings *arginfo;
998 	size_t execpath_len;
999 	int szsigcode;
1000 
1001 	/*
1002 	 * Calculate string base and vector table pointers.
1003 	 * Also deal with signal trampoline code for this exec type.
1004 	 */
1005 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1006 		execpath_len = strlen(imgp->execpath) + 1;
1007 	else
1008 		execpath_len = 0;
1009 	arginfo = (struct ps_strings *)PS_STRINGS;
1010 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
1011 
1012 	argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *));
1013 	gap = stackgap_random;
1014 	cpu_ccfence();
1015 	if (gap != 0) {
1016 		if (gap < 0)
1017 			sgap = ALIGN(-gap);
1018 		else
1019 			sgap = ALIGN(karc4random() & (gap - 1));
1020 	} else {
1021 		sgap = 0;
1022 	}
1023 
1024 	/*
1025 	 * Calculate destp, which points to [args & env] and above.
1026 	 */
1027 	destp = (caddr_t)arginfo -
1028 		szsigcode -
1029 		roundup(execpath_len, sizeof(char *)) -
1030 		SPARE_USRSPACE -
1031 		sgap -
1032 		argsenvspace;
1033 
1034 	/*
1035 	 * install sigcode
1036 	 */
1037 	if (szsigcode) {
1038 		copyout(imgp->proc->p_sysent->sv_sigcode,
1039 			((caddr_t)arginfo - szsigcode), szsigcode);
1040 	}
1041 
1042 	/*
1043 	 * Copy the image path for the rtld
1044 	 */
1045 	if (execpath_len) {
1046 		imgp->execpathp = (uintptr_t)arginfo
1047 				  - szsigcode
1048 				  - roundup(execpath_len, sizeof(char *));
1049 		copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
1050 	}
1051 
1052 	/*
1053 	 * Calculate base for argv[], envp[], and ELF_Auxargs.
1054 	 */
1055 	vectp = (char **)destp - (AT_COUNT * 2);
1056 	vectp -= imgp->args->argc + imgp->args->envc + 2;
1057 
1058 	stack_base = (register_t *)vectp;
1059 
1060 	stringp = imgp->args->begin_argv;
1061 	argc = imgp->args->argc;
1062 	envc = imgp->args->envc;
1063 
1064 	/*
1065 	 * Copy out strings - arguments and environment (at destp)
1066 	 */
1067 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
1068 
1069 	/*
1070 	 * Fill in "ps_strings" struct for ps, w, etc.
1071 	 */
1072 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1073 	suword32(&arginfo->ps_nargvstr, argc);
1074 
1075 	/*
1076 	 * Fill in argument portion of vector table.
1077 	 */
1078 	for (; argc > 0; --argc) {
1079 		suword(vectp++, (long)(intptr_t)destp);
1080 		while (*stringp++ != 0)
1081 			destp++;
1082 		destp++;
1083 	}
1084 
1085 	/* a null vector table pointer separates the argp's from the envp's */
1086 	suword(vectp++, 0);
1087 
1088 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1089 	suword32(&arginfo->ps_nenvstr, envc);
1090 
1091 	/*
1092 	 * Fill in environment portion of vector table.
1093 	 */
1094 	for (; envc > 0; --envc) {
1095 		suword(vectp++, (long)(intptr_t)destp);
1096 		while (*stringp++ != 0)
1097 			destp++;
1098 		destp++;
1099 	}
1100 
1101 	/* end of vector table is a null pointer */
1102 	suword(vectp, 0);
1103 
1104 	return (stack_base);
1105 }
1106 
1107 /*
1108  * Check permissions of file to execute.
1109  *	Return 0 for success or error code on failure.
1110  */
1111 int
1112 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
1113 {
1114 	struct proc *p = imgp->proc;
1115 	struct vnode *vp = imgp->vp;
1116 	struct vattr *attr = imgp->attr;
1117 	int error;
1118 
1119 	/* Get file attributes */
1120 	error = VOP_GETATTR(vp, attr);
1121 	if (error)
1122 		return (error);
1123 
1124 	/*
1125 	 * 1) Check if file execution is disabled for the filesystem that this
1126 	 *	file resides on.
1127 	 * 2) Insure that at least one execute bit is on - otherwise root
1128 	 *	will always succeed, and we don't want to happen unless the
1129 	 *	file really is executable.
1130 	 * 3) Insure that the file is a regular file.
1131 	 */
1132 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1133 	    ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1134 	    ((attr->va_mode & 0111) == 0) ||
1135 	    (attr->va_type != VREG)) {
1136 		return (EACCES);
1137 	}
1138 
1139 	/*
1140 	 * Zero length files can't be exec'd
1141 	 */
1142 	if (attr->va_size == 0)
1143 		return (ENOEXEC);
1144 
1145 	/*
1146 	 *  Check for execute permission to file based on current credentials.
1147 	 */
1148 	error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1149 	if (error)
1150 		return (error);
1151 
1152 	/*
1153 	 * Check number of open-for-writes on the file and deny execution
1154 	 * if there are any.
1155 	 */
1156 	if (vp->v_writecount)
1157 		return (ETXTBSY);
1158 
1159 	/*
1160 	 * Call filesystem specific open routine, which allows us to read,
1161 	 * write, and mmap the file.  Without the VOP_OPEN we can only
1162 	 * stat the file.
1163 	 */
1164 	error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1165 	if (error)
1166 		return (error);
1167 
1168 	return (0);
1169 }
1170 
1171 /*
1172  * Exec handler registration
1173  */
1174 int
1175 exec_register(const struct execsw *execsw_arg)
1176 {
1177 	const struct execsw **es, **xs, **newexecsw;
1178 	int count = 2;	/* New slot and trailing NULL */
1179 
1180 	if (execsw)
1181 		for (es = execsw; *es; es++)
1182 			count++;
1183 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1184 	xs = newexecsw;
1185 	if (execsw)
1186 		for (es = execsw; *es; es++)
1187 			*xs++ = *es;
1188 	*xs++ = execsw_arg;
1189 	*xs = NULL;
1190 	if (execsw)
1191 		kfree(execsw, M_TEMP);
1192 	execsw = newexecsw;
1193 	return 0;
1194 }
1195 
1196 int
1197 exec_unregister(const struct execsw *execsw_arg)
1198 {
1199 	const struct execsw **es, **xs, **newexecsw;
1200 	int count = 1;
1201 
1202 	if (execsw == NULL)
1203 		panic("unregister with no handlers left?");
1204 
1205 	for (es = execsw; *es; es++) {
1206 		if (*es == execsw_arg)
1207 			break;
1208 	}
1209 	if (*es == NULL)
1210 		return ENOENT;
1211 	for (es = execsw; *es; es++)
1212 		if (*es != execsw_arg)
1213 			count++;
1214 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1215 	xs = newexecsw;
1216 	for (es = execsw; *es; es++)
1217 		if (*es != execsw_arg)
1218 			*xs++ = *es;
1219 	*xs = NULL;
1220 	if (execsw)
1221 		kfree(execsw, M_TEMP);
1222 	execsw = newexecsw;
1223 	return 0;
1224 }
1225