xref: /dflybsd-src/sys/kern/kern_exec.c (revision e90a7c45c3303ed54c0fde732b2ba32dc80ffd9b)
1 /*
2  * Copyright (c) 1993, David Greenman
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD: src/sys/kern/kern_exec.c,v 1.107.2.15 2002/07/30 15:40:46 nectar Exp $
27  * $DragonFly: src/sys/kern/kern_exec.c,v 1.64 2008/10/26 04:29:19 sephe Exp $
28  */
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/sysproto.h>
33 #include <sys/kernel.h>
34 #include <sys/mount.h>
35 #include <sys/filedesc.h>
36 #include <sys/fcntl.h>
37 #include <sys/acct.h>
38 #include <sys/exec.h>
39 #include <sys/imgact.h>
40 #include <sys/imgact_elf.h>
41 #include <sys/kern_syscall.h>
42 #include <sys/wait.h>
43 #include <sys/malloc.h>
44 #include <sys/proc.h>
45 #include <sys/priv.h>
46 #include <sys/ktrace.h>
47 #include <sys/signalvar.h>
48 #include <sys/pioctl.h>
49 #include <sys/nlookup.h>
50 #include <sys/sysent.h>
51 #include <sys/shm.h>
52 #include <sys/sysctl.h>
53 #include <sys/vnode.h>
54 #include <sys/vmmeter.h>
55 #include <sys/aio.h>
56 #include <sys/libkern.h>
57 
58 #include <cpu/lwbuf.h>
59 
60 #include <vm/vm.h>
61 #include <vm/vm_param.h>
62 #include <sys/lock.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_page.h>
65 #include <vm/vm_map.h>
66 #include <vm/vm_kern.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69 #include <vm/vnode_pager.h>
70 #include <vm/vm_pager.h>
71 
72 #include <sys/user.h>
73 #include <sys/reg.h>
74 
75 #include <sys/thread2.h>
76 #include <sys/mplock2.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 
117 	cluster_limit = 16;	/* up to this many objects */
118 	exec_objcache = objcache_create_mbacked(
119 					M_EXECARGS, PATH_MAX + ARG_MAX,
120 					&cluster_limit,
121 					2,	/* minimal magazine capacity */
122 					NULL, NULL, NULL);
123 }
124 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0);
125 
126 /*
127  * stackgap_random specifies if the stackgap should have a random size added
128  * to it.  It must be a power of 2.  If non-zero, the stack gap will be
129  * calculated as: ALIGN(karc4random() & (stackgap_random - 1)).
130  */
131 static int stackgap_random = 1024;
132 static int
133 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS)
134 {
135 	int error, new_val;
136 	new_val = stackgap_random;
137 	error = sysctl_handle_int(oidp, &new_val, 0, req);
138 	if (error != 0 || req->newptr == NULL)
139 		return (error);
140 	if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val))
141 		return (EINVAL);
142 	stackgap_random = new_val;
143 
144 	return(0);
145 }
146 
147 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT,
148 	0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)");
149 
150 void
151 print_execve_args(struct image_args *args)
152 {
153 	char *cp;
154 	int ndx;
155 
156 	cp = args->begin_argv;
157 	for (ndx = 0; ndx < args->argc; ndx++) {
158 		kprintf("\targv[%d]: %s\n", ndx, cp);
159 		while (*cp++ != '\0');
160 	}
161 	for (ndx = 0; ndx < args->envc; ndx++) {
162 		kprintf("\tenvv[%d]: %s\n", ndx, cp);
163 		while (*cp++ != '\0');
164 	}
165 }
166 
167 /*
168  * Each of the items is a pointer to a `const struct execsw', hence the
169  * double pointer here.
170  */
171 static const struct execsw **execsw;
172 
173 /*
174  * Replace current vmspace with a new binary.
175  * Returns 0 on success, > 0 on recoverable error (use as errno).
176  * Returns -1 on lethal error which demands killing of the current
177  * process!
178  */
179 int
180 kern_execve(struct nlookupdata *nd, struct image_args *args)
181 {
182 	struct thread *td = curthread;
183 	struct lwp *lp = td->td_lwp;
184 	struct proc *p = td->td_proc;
185 	register_t *stack_base;
186 	int error, len, i;
187 	struct image_params image_params, *imgp;
188 	struct vattr attr;
189 	int (*img_first) (struct image_params *);
190 
191 	if (debug_execve_args) {
192 		kprintf("%s()\n", __func__);
193 		print_execve_args(args);
194 	}
195 
196 	KKASSERT(p);
197 	imgp = &image_params;
198 
199 	/*
200 	 * NOTE: P_INEXEC is handled by exec_new_vmspace() now.  We make
201 	 * no modifications to the process at all until we get there.
202 	 *
203 	 * Note that multiple threads may be trying to exec at the same
204 	 * time.  exec_new_vmspace() handles that too.
205 	 */
206 
207 	/*
208 	 * Initialize part of the common data
209 	 */
210 	imgp->proc = p;
211 	imgp->args = args;
212 	imgp->attr = &attr;
213 	imgp->entry_addr = 0;
214 	imgp->resident = 0;
215 	imgp->vmspace_destroyed = 0;
216 	imgp->interpreted = 0;
217 	imgp->interpreter_name[0] = 0;
218 	imgp->auxargs = NULL;
219 	imgp->vp = NULL;
220 	imgp->firstpage = NULL;
221 	imgp->ps_strings = 0;
222 	imgp->image_header = NULL;
223 
224 interpret:
225 
226 	/*
227 	 * Translate the file name to a vnode.  Unlock the cache entry to
228 	 * improve parallelism for programs exec'd in parallel.
229 	 */
230 	if ((error = nlookup(nd)) != 0)
231 		goto exec_fail;
232 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp);
233 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
234 	nd->nl_flags &= ~NLC_NCPISLOCKED;
235 	cache_unlock(&nd->nl_nch);
236 	if (error)
237 		goto exec_fail;
238 
239 	/*
240 	 * Check file permissions (also 'opens' file).
241 	 * Include also the top level mount in the check.
242 	 */
243 	error = exec_check_permissions(imgp, nd->nl_nch.mount);
244 	if (error) {
245 		vn_unlock(imgp->vp);
246 		goto exec_fail_dealloc;
247 	}
248 
249 	error = exec_map_first_page(imgp);
250 	vn_unlock(imgp->vp);
251 	if (error)
252 		goto exec_fail_dealloc;
253 
254 	if (debug_execve_args && imgp->interpreted) {
255 		kprintf("    target is interpreted -- recursive pass\n");
256 		kprintf("    interpreter: %s\n", imgp->interpreter_name);
257 		print_execve_args(args);
258 	}
259 
260 	/*
261 	 *	If the current process has a special image activator it
262 	 *	wants to try first, call it.   For example, emulating shell
263 	 *	scripts differently.
264 	 */
265 	error = -1;
266 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
267 		error = img_first(imgp);
268 
269 	/*
270 	 *	If the vnode has a registered vmspace, exec the vmspace
271 	 */
272 	if (error == -1 && imgp->vp->v_resident) {
273 		error = exec_resident_imgact(imgp);
274 	}
275 
276 	/*
277 	 *	Loop through the list of image activators, calling each one.
278 	 *	An activator returns -1 if there is no match, 0 on success,
279 	 *	and an error otherwise.
280 	 */
281 	for (i = 0; error == -1 && execsw[i]; ++i) {
282 		if (execsw[i]->ex_imgact == NULL ||
283 		    execsw[i]->ex_imgact == img_first) {
284 			continue;
285 		}
286 		error = (*execsw[i]->ex_imgact)(imgp);
287 	}
288 
289 	if (error) {
290 		if (error == -1)
291 			error = ENOEXEC;
292 		goto exec_fail_dealloc;
293 	}
294 
295 	/*
296 	 * Special interpreter operation, cleanup and loop up to try to
297 	 * activate the interpreter.
298 	 */
299 	if (imgp->interpreted) {
300 		exec_unmap_first_page(imgp);
301 		nlookup_done(nd);
302 		vrele(imgp->vp);
303 		imgp->vp = NULL;
304 		error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE,
305 					NLC_FOLLOW);
306 		if (error)
307 			goto exec_fail;
308 		goto interpret;
309 	}
310 
311 	/*
312 	 * Copy out strings (args and env) and initialize stack base
313 	 */
314 	stack_base = exec_copyout_strings(imgp);
315 	p->p_vmspace->vm_minsaddr = (char *)stack_base;
316 
317 	/*
318 	 * If custom stack fixup routine present for this process
319 	 * let it do the stack setup.  If we are running a resident
320 	 * image there is no auxinfo or other image activator context
321 	 * so don't try to add fixups to the stack.
322 	 *
323 	 * Else stuff argument count as first item on stack
324 	 */
325 	if (p->p_sysent->sv_fixup && imgp->resident == 0)
326 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
327 	else
328 		suword(--stack_base, imgp->args->argc);
329 
330 	/*
331 	 * For security and other reasons, the file descriptor table cannot
332 	 * be shared after an exec.
333 	 */
334 	if (p->p_fd->fd_refcnt > 1) {
335 		struct filedesc *tmp;
336 
337 		tmp = fdcopy(p);
338 		fdfree(p, tmp);
339 	}
340 
341 	/*
342 	 * For security and other reasons, signal handlers cannot
343 	 * be shared after an exec. The new proces gets a copy of the old
344 	 * handlers. In execsigs(), the new process will have its signals
345 	 * reset.
346 	 */
347 	if (p->p_sigacts->ps_refcnt > 1) {
348 		struct sigacts *newsigacts;
349 
350 		newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts),
351 		       M_SUBPROC, M_WAITOK);
352 		bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts));
353 		p->p_sigacts->ps_refcnt--;
354 		p->p_sigacts = newsigacts;
355 		p->p_sigacts->ps_refcnt = 1;
356 	}
357 
358 	/*
359 	 * For security and other reasons virtual kernels cannot be
360 	 * inherited by an exec.  This also allows a virtual kernel
361 	 * to fork/exec unrelated applications.
362 	 */
363 	if (p->p_vkernel)
364 		vkernel_exit(p);
365 
366 	/* Stop profiling */
367 	stopprofclock(p);
368 
369 	/* close files on exec */
370 	fdcloseexec(p);
371 
372 	/* reset caught signals */
373 	execsigs(p);
374 
375 	/* name this process - nameiexec(p, ndp) */
376 	len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN);
377 	bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len);
378 	p->p_comm[len] = 0;
379 	bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1);
380 
381 	/*
382 	 * mark as execed, wakeup the process that vforked (if any) and tell
383 	 * it that it now has its own resources back
384 	 */
385 	p->p_flag |= P_EXEC;
386 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
387 		p->p_flag &= ~P_PPWAIT;
388 		wakeup((caddr_t)p->p_pptr);
389 	}
390 
391 	/*
392 	 * Implement image setuid/setgid.
393 	 *
394 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
395 	 * the process is being traced.
396 	 */
397 	if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) ||
398 	     ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) &&
399 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
400 	    (p->p_flag & P_TRACED) == 0) {
401 		/*
402 		 * Turn off syscall tracing for set-id programs, except for
403 		 * root.  Record any set-id flags first to make sure that
404 		 * we do not regain any tracing during a possible block.
405 		 */
406 		setsugid();
407 		if (p->p_tracenode && ktrace_suid == 0 &&
408 		    priv_check(td, PRIV_ROOT) != 0) {
409 			ktrdestroy(&p->p_tracenode);
410 			p->p_traceflag = 0;
411 		}
412 		/* Close any file descriptors 0..2 that reference procfs */
413 		setugidsafety(p);
414 		/* Make sure file descriptors 0..2 are in use. */
415 		error = fdcheckstd(lp);
416 		if (error != 0)
417 			goto exec_fail_dealloc;
418 		/*
419 		 * Set the new credentials.
420 		 */
421 		cratom(&p->p_ucred);
422 		if (attr.va_mode & VSUID)
423 			change_euid(attr.va_uid);
424 		if (attr.va_mode & VSGID)
425 			p->p_ucred->cr_gid = attr.va_gid;
426 
427 		/*
428 		 * Clear local varsym variables
429 		 */
430 		varsymset_clean(&p->p_varsymset);
431 	} else {
432 		if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid &&
433 		    p->p_ucred->cr_gid == p->p_ucred->cr_rgid)
434 			p->p_flag &= ~P_SUGID;
435 	}
436 
437 	/*
438 	 * Implement correct POSIX saved-id behavior.
439 	 */
440 	if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid ||
441 	    p->p_ucred->cr_svgid != p->p_ucred->cr_gid) {
442 		cratom(&p->p_ucred);
443 		p->p_ucred->cr_svuid = p->p_ucred->cr_uid;
444 		p->p_ucred->cr_svgid = p->p_ucred->cr_gid;
445 	}
446 
447 	/*
448 	 * Store the vp for use in procfs
449 	 */
450 	if (p->p_textvp)		/* release old reference */
451 		vrele(p->p_textvp);
452 	p->p_textvp = imgp->vp;
453 	vref(p->p_textvp);
454 
455 	/* Release old namecache handle to text file */
456 	if (p->p_textnch.ncp)
457 		cache_drop(&p->p_textnch);
458 
459 	if (nd->nl_nch.mount)
460 		cache_copy(&nd->nl_nch, &p->p_textnch);
461 
462         /*
463          * Notify others that we exec'd, and clear the P_INEXEC flag
464          * as we're now a bona fide freshly-execed process.
465          */
466 	KNOTE(&p->p_klist, NOTE_EXEC);
467 	p->p_flag &= ~P_INEXEC;
468 
469 	/*
470 	 * If tracing the process, trap to debugger so breakpoints
471 	 * 	can be set before the program executes.
472 	 */
473 	STOPEVENT(p, S_EXEC, 0);
474 
475 	if (p->p_flag & P_TRACED)
476 		ksignal(p, SIGTRAP);
477 
478 	/* clear "fork but no exec" flag, as we _are_ execing */
479 	p->p_acflag &= ~AFORK;
480 
481 	/* Set values passed into the program in registers. */
482 	exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base,
483 	    imgp->ps_strings);
484 
485 	/* Set the access time on the vnode */
486 	vn_mark_atime(imgp->vp, td);
487 
488 	/* Free any previous argument cache */
489 	if (p->p_args && --p->p_args->ar_ref == 0)
490 		FREE(p->p_args, M_PARGS);
491 	p->p_args = NULL;
492 
493 	/* Cache arguments if they fit inside our allowance */
494 	i = imgp->args->begin_envv - imgp->args->begin_argv;
495 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
496 		MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i,
497 		    M_PARGS, M_WAITOK);
498 		p->p_args->ar_ref = 1;
499 		p->p_args->ar_length = i;
500 		bcopy(imgp->args->begin_argv, p->p_args->ar_args, i);
501 	}
502 
503 exec_fail_dealloc:
504 
505 	/*
506 	 * free various allocated resources
507 	 */
508 	if (imgp->firstpage)
509 		exec_unmap_first_page(imgp);
510 
511 	if (imgp->vp) {
512 		vrele(imgp->vp);
513 		imgp->vp = NULL;
514 	}
515 
516 	if (error == 0) {
517 		++mycpu->gd_cnt.v_exec;
518 		return (0);
519 	}
520 
521 exec_fail:
522 	/*
523 	 * we're done here, clear P_INEXEC if we were the ones that
524 	 * set it.  Otherwise if vmspace_destroyed is still set we
525 	 * raced another thread and that thread is responsible for
526 	 * clearing it.
527 	 */
528 	if (imgp->vmspace_destroyed & 2)
529 		p->p_flag &= ~P_INEXEC;
530 	if (imgp->vmspace_destroyed) {
531 		/*
532 		 * Sorry, no more process anymore. exit gracefully.
533 		 * However we can't die right here, because our
534 		 * caller might have to clean up, so indicate a
535 		 * lethal error by returning -1.
536 		 */
537 		return(-1);
538 	} else {
539 		return(error);
540 	}
541 }
542 
543 /*
544  * execve() system call.
545  *
546  * MPALMOSTSAFE
547  */
548 int
549 sys_execve(struct execve_args *uap)
550 {
551 	struct nlookupdata nd;
552 	struct image_args args;
553 	int error;
554 
555 	bzero(&args, sizeof(args));
556 
557 	get_mplock();
558 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
559 	if (error == 0) {
560 		error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE,
561 					uap->argv, uap->envv);
562 	}
563 	if (error == 0)
564 		error = kern_execve(&nd, &args);
565 	nlookup_done(&nd);
566 	exec_free_args(&args);
567 
568 	if (error < 0) {
569 		/* We hit a lethal error condition.  Let's die now. */
570 		exit1(W_EXITCODE(0, SIGABRT));
571 		/* NOTREACHED */
572 	}
573 	rel_mplock();
574 
575 	/*
576 	 * The syscall result is returned in registers to the new program.
577 	 * Linux will register %edx as an atexit function and we must be
578 	 * sure to set it to 0.  XXX
579 	 */
580 	if (error == 0)
581 		uap->sysmsg_result64 = 0;
582 
583 	return (error);
584 }
585 
586 int
587 exec_map_page(struct image_params *imgp, vm_pindex_t pageno,
588 	      struct lwbuf **plwb, const char **pdata)
589 {
590 	int rv;
591 	vm_page_t ma;
592 	vm_page_t m;
593 	vm_object_t object;
594 
595 	/*
596 	 * The file has to be mappable.
597 	 */
598 	if ((object = imgp->vp->v_object) == NULL)
599 		return (EIO);
600 
601 	if (pageno >= object->size)
602 		return (EIO);
603 
604 	m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
605 
606 	lwkt_gettoken(&vm_token);
607 	while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
608 		ma = m;
609 
610 		/*
611 		 * get_pages unbusies all the requested pages except the
612 		 * primary page (at index 0 in this case).  The primary
613 		 * page may have been wired during the pagein (e.g. by
614 		 * the buffer cache) so vnode_pager_freepage() must be
615 		 * used to properly release it.
616 		 */
617 		rv = vm_pager_get_page(object, &ma, 1);
618 		m = vm_page_lookup(object, pageno);
619 
620 		if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) {
621 			if (m) {
622 				vm_page_protect(m, VM_PROT_NONE);
623 				vnode_pager_freepage(m);
624 			}
625 			lwkt_reltoken(&vm_token);
626 			return EIO;
627 		}
628 	}
629 	vm_page_hold(m);	/* requires vm_token to be held */
630 	vm_page_wakeup(m);	/* unbusy the page */
631 	lwkt_reltoken(&vm_token);
632 
633 	*plwb = lwbuf_alloc(m, *plwb);
634 	*pdata = (void *)lwbuf_kva(*plwb);
635 
636 	return (0);
637 }
638 
639 int
640 exec_map_first_page(struct image_params *imgp)
641 {
642 	int err;
643 
644 	if (imgp->firstpage)
645 		exec_unmap_first_page(imgp);
646 
647 	imgp->firstpage = &imgp->firstpage_cache;
648 	err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header);
649 
650 	if (err)
651 		return err;
652 
653 	return 0;
654 }
655 
656 void
657 exec_unmap_page(struct lwbuf *lwb)
658 {
659 	vm_page_t m;
660 
661 	crit_enter();
662 	if (lwb != NULL) {
663 		m = lwbuf_page(lwb);
664 		lwbuf_free(lwb);
665 		vm_page_unhold(m);
666 	}
667 	crit_exit();
668 }
669 
670 void
671 exec_unmap_first_page(struct image_params *imgp)
672 {
673 	exec_unmap_page(imgp->firstpage);
674 	imgp->firstpage = NULL;
675 	imgp->image_header = NULL;
676 }
677 
678 /*
679  * Destroy old address space, and allocate a new stack
680  *	The new stack is only SGROWSIZ large because it is grown
681  *	automatically in trap.c.
682  *
683  * This is the point of no return.
684  */
685 int
686 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy)
687 {
688 	struct vmspace *vmspace = imgp->proc->p_vmspace;
689 	vm_offset_t stack_addr = USRSTACK - maxssiz;
690 	struct proc *p;
691 	vm_map_t map;
692 	int error;
693 
694 	/*
695 	 * Indicate that we cannot gracefully error out any more, kill
696 	 * any other threads present, and set P_INEXEC to indicate that
697 	 * we are now messing with the process structure proper.
698 	 *
699 	 * If killalllwps() races return an error which coupled with
700 	 * vmspace_destroyed will cause us to exit.  This is what we
701 	 * want since another thread is patiently waiting for us to exit
702 	 * in that case.
703 	 */
704 	p = curproc;
705 	imgp->vmspace_destroyed = 1;
706 
707 	if (curthread->td_proc->p_nthreads > 1) {
708 		error = killalllwps(1);
709 		if (error)
710 			return (error);
711 	}
712 	imgp->vmspace_destroyed |= 2;	/* we are responsible for P_INEXEC */
713 	p->p_flag |= P_INEXEC;
714 
715 	/*
716 	 * Prevent a pending AIO from modifying the new address space.
717 	 */
718 	aio_proc_rundown(imgp->proc);
719 
720 	/*
721 	 * Blow away entire process VM, if address space not shared,
722 	 * otherwise, create a new VM space so that other threads are
723 	 * not disrupted.  If we are execing a resident vmspace we
724 	 * create a duplicate of it and remap the stack.
725 	 *
726 	 * The exitingcnt test is not strictly necessary but has been
727 	 * included for code sanity (to make the code more deterministic).
728 	 */
729 	map = &vmspace->vm_map;
730 	if (vmcopy) {
731 		vmspace_exec(imgp->proc, vmcopy);
732 		vmspace = imgp->proc->p_vmspace;
733 		pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK);
734 		map = &vmspace->vm_map;
735 	} else if (vmspace->vm_sysref.refcnt == 1 &&
736 		   vmspace->vm_exitingcnt == 0) {
737 		shmexit(vmspace);
738 		if (vmspace->vm_upcalls)
739 			upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc));
740 		pmap_remove_pages(vmspace_pmap(vmspace),
741 			0, VM_MAX_USER_ADDRESS);
742 		vm_map_remove(map, 0, VM_MAX_USER_ADDRESS);
743 	} else {
744 		vmspace_exec(imgp->proc, NULL);
745 		vmspace = imgp->proc->p_vmspace;
746 		map = &vmspace->vm_map;
747 	}
748 
749 	/* Allocate a new stack */
750 	error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz,
751 			     0, VM_PROT_ALL, VM_PROT_ALL, 0);
752 	if (error)
753 		return (error);
754 
755 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
756 	 * VM_STACK case, but they are still used to monitor the size of the
757 	 * process stack so we can check the stack rlimit.
758 	 */
759 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
760 	vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz;
761 
762 	return(0);
763 }
764 
765 /*
766  * Copy out argument and environment strings from the old process
767  *	address space into the temporary string buffer.
768  */
769 int
770 exec_copyin_args(struct image_args *args, char *fname,
771 		enum exec_path_segflg segflg, char **argv, char **envv)
772 {
773 	char	*argp, *envp;
774 	int	error = 0;
775 	size_t	length;
776 
777 	args->buf = objcache_get(exec_objcache, M_WAITOK);
778 	if (args->buf == NULL)
779 		return (ENOMEM);
780 	args->begin_argv = args->buf;
781 	args->endp = args->begin_argv;
782 	args->space = ARG_MAX;
783 
784 	args->fname = args->buf + ARG_MAX;
785 
786 	/*
787 	 * Copy the file name.
788 	 */
789 	if (segflg == PATH_SYSSPACE) {
790 		error = copystr(fname, args->fname, PATH_MAX, &length);
791 	} else if (segflg == PATH_USERSPACE) {
792 		error = copyinstr(fname, args->fname, PATH_MAX, &length);
793 	}
794 
795 	/*
796 	 * Extract argument strings.  argv may not be NULL.  The argv
797 	 * array is terminated by a NULL entry.  We special-case the
798 	 * situation where argv[0] is NULL by passing { filename, NULL }
799 	 * to the new program to guarentee that the interpreter knows what
800 	 * file to open in case we exec an interpreted file.   Note that
801 	 * a NULL argv[0] terminates the argv[] array.
802 	 *
803 	 * XXX the special-casing of argv[0] is historical and needs to be
804 	 * revisited.
805 	 */
806 	if (argv == NULL)
807 		error = EFAULT;
808 	if (error == 0) {
809 		while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) {
810 			if (argp == (caddr_t)-1) {
811 				error = EFAULT;
812 				break;
813 			}
814 			error = copyinstr(argp, args->endp,
815 					    args->space, &length);
816 			if (error) {
817 				if (error == ENAMETOOLONG)
818 					error = E2BIG;
819 				break;
820 			}
821 			args->space -= length;
822 			args->endp += length;
823 			args->argc++;
824 		}
825 		if (args->argc == 0 && error == 0) {
826 			length = strlen(args->fname) + 1;
827 			if (length > args->space) {
828 				error = E2BIG;
829 			} else {
830 				bcopy(args->fname, args->endp, length);
831 				args->space -= length;
832 				args->endp += length;
833 				args->argc++;
834 			}
835 		}
836 	}
837 
838 	args->begin_envv = args->endp;
839 
840 	/*
841 	 * extract environment strings.  envv may be NULL.
842 	 */
843 	if (envv && error == 0) {
844 		while ((envp = (caddr_t) (intptr_t) fuword(envv++))) {
845 			if (envp == (caddr_t) -1) {
846 				error = EFAULT;
847 				break;
848 			}
849 			error = copyinstr(envp, args->endp, args->space,
850 			    &length);
851 			if (error) {
852 				if (error == ENAMETOOLONG)
853 					error = E2BIG;
854 				break;
855 			}
856 			args->space -= length;
857 			args->endp += length;
858 			args->envc++;
859 		}
860 	}
861 	return (error);
862 }
863 
864 void
865 exec_free_args(struct image_args *args)
866 {
867 	if (args->buf) {
868 		objcache_put(exec_objcache, args->buf);
869 		args->buf = NULL;
870 	}
871 }
872 
873 /*
874  * Copy strings out to the new process address space, constructing
875  *	new arg and env vector tables. Return a pointer to the base
876  *	so that it can be used as the initial stack pointer.
877  */
878 register_t *
879 exec_copyout_strings(struct image_params *imgp)
880 {
881 	int argc, envc, sgap;
882 	char **vectp;
883 	char *stringp, *destp;
884 	register_t *stack_base;
885 	struct ps_strings *arginfo;
886 	int szsigcode;
887 
888 	/*
889 	 * Calculate string base and vector table pointers.
890 	 * Also deal with signal trampoline code for this exec type.
891 	 */
892 	arginfo = (struct ps_strings *)PS_STRINGS;
893 	szsigcode = *(imgp->proc->p_sysent->sv_szsigcode);
894 	if (stackgap_random != 0)
895 		sgap = ALIGN(karc4random() & (stackgap_random - 1));
896 	else
897 		sgap = 0;
898 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap -
899 	    roundup((ARG_MAX - imgp->args->space), sizeof(char *));
900 
901 	/*
902 	 * install sigcode
903 	 */
904 	if (szsigcode)
905 		copyout(imgp->proc->p_sysent->sv_sigcode,
906 		    ((caddr_t)arginfo - szsigcode), szsigcode);
907 
908 	/*
909 	 * If we have a valid auxargs ptr, prepare some room
910 	 * on the stack.
911 	 *
912 	 * The '+ 2' is for the null pointers at the end of each of the
913 	 * arg and env vector sets, and 'AT_COUNT*2' is room for the
914 	 * ELF Auxargs data.
915 	 */
916 	if (imgp->auxargs) {
917 		vectp = (char **)(destp - (imgp->args->argc +
918 			imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*));
919 	} else {
920 		vectp = (char **)(destp - (imgp->args->argc +
921 			imgp->args->envc + 2) * sizeof(char*));
922 	}
923 
924 	/*
925 	 * NOTE: don't bother aligning the stack here for GCC 2.x, it will
926 	 * be done in crt1.o.  Note that GCC 3.x aligns the stack in main.
927 	 */
928 
929 	/*
930 	 * vectp also becomes our initial stack base
931 	 */
932 	stack_base = (register_t *)vectp;
933 
934 	stringp = imgp->args->begin_argv;
935 	argc = imgp->args->argc;
936 	envc = imgp->args->envc;
937 
938 	/*
939 	 * Copy out strings - arguments and environment.
940 	 */
941 	copyout(stringp, destp, ARG_MAX - imgp->args->space);
942 
943 	/*
944 	 * Fill in "ps_strings" struct for ps, w, etc.
945 	 */
946 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
947 	suword(&arginfo->ps_nargvstr, argc);
948 
949 	/*
950 	 * Fill in argument portion of vector table.
951 	 */
952 	for (; argc > 0; --argc) {
953 		suword(vectp++, (long)(intptr_t)destp);
954 		while (*stringp++ != 0)
955 			destp++;
956 		destp++;
957 	}
958 
959 	/* a null vector table pointer separates the argp's from the envp's */
960 	suword(vectp++, 0);
961 
962 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
963 	suword(&arginfo->ps_nenvstr, envc);
964 
965 	/*
966 	 * Fill in environment portion of vector table.
967 	 */
968 	for (; envc > 0; --envc) {
969 		suword(vectp++, (long)(intptr_t)destp);
970 		while (*stringp++ != 0)
971 			destp++;
972 		destp++;
973 	}
974 
975 	/* end of vector table is a null pointer */
976 	suword(vectp, 0);
977 
978 	return (stack_base);
979 }
980 
981 /*
982  * Check permissions of file to execute.
983  *	Return 0 for success or error code on failure.
984  */
985 int
986 exec_check_permissions(struct image_params *imgp, struct mount *topmnt)
987 {
988 	struct proc *p = imgp->proc;
989 	struct vnode *vp = imgp->vp;
990 	struct vattr *attr = imgp->attr;
991 	int error;
992 
993 	/* Get file attributes */
994 	error = VOP_GETATTR(vp, attr);
995 	if (error)
996 		return (error);
997 
998 	/*
999 	 * 1) Check if file execution is disabled for the filesystem that this
1000 	 *	file resides on.
1001 	 * 2) Insure that at least one execute bit is on - otherwise root
1002 	 *	will always succeed, and we don't want to happen unless the
1003 	 *	file really is executable.
1004 	 * 3) Insure that the file is a regular file.
1005 	 */
1006 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1007 	    ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) ||
1008 	    ((attr->va_mode & 0111) == 0) ||
1009 	    (attr->va_type != VREG)) {
1010 		return (EACCES);
1011 	}
1012 
1013 	/*
1014 	 * Zero length files can't be exec'd
1015 	 */
1016 	if (attr->va_size == 0)
1017 		return (ENOEXEC);
1018 
1019 	/*
1020 	 *  Check for execute permission to file based on current credentials.
1021 	 */
1022 	error = VOP_EACCESS(vp, VEXEC, p->p_ucred);
1023 	if (error)
1024 		return (error);
1025 
1026 	/*
1027 	 * Check number of open-for-writes on the file and deny execution
1028 	 * if there are any.
1029 	 */
1030 	if (vp->v_writecount)
1031 		return (ETXTBSY);
1032 
1033 	/*
1034 	 * Call filesystem specific open routine, which allows us to read,
1035 	 * write, and mmap the file.  Without the VOP_OPEN we can only
1036 	 * stat the file.
1037 	 */
1038 	error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL);
1039 	if (error)
1040 		return (error);
1041 
1042 	return (0);
1043 }
1044 
1045 /*
1046  * Exec handler registration
1047  */
1048 int
1049 exec_register(const struct execsw *execsw_arg)
1050 {
1051 	const struct execsw **es, **xs, **newexecsw;
1052 	int count = 2;	/* New slot and trailing NULL */
1053 
1054 	if (execsw)
1055 		for (es = execsw; *es; es++)
1056 			count++;
1057 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1058 	xs = newexecsw;
1059 	if (execsw)
1060 		for (es = execsw; *es; es++)
1061 			*xs++ = *es;
1062 	*xs++ = execsw_arg;
1063 	*xs = NULL;
1064 	if (execsw)
1065 		kfree(execsw, M_TEMP);
1066 	execsw = newexecsw;
1067 	return 0;
1068 }
1069 
1070 int
1071 exec_unregister(const struct execsw *execsw_arg)
1072 {
1073 	const struct execsw **es, **xs, **newexecsw;
1074 	int count = 1;
1075 
1076 	if (execsw == NULL)
1077 		panic("unregister with no handlers left?");
1078 
1079 	for (es = execsw; *es; es++) {
1080 		if (*es == execsw_arg)
1081 			break;
1082 	}
1083 	if (*es == NULL)
1084 		return ENOENT;
1085 	for (es = execsw; *es; es++)
1086 		if (*es != execsw_arg)
1087 			count++;
1088 	newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1089 	xs = newexecsw;
1090 	for (es = execsw; *es; es++)
1091 		if (*es != execsw_arg)
1092 			*xs++ = *es;
1093 	*xs = NULL;
1094 	if (execsw)
1095 		kfree(execsw, M_TEMP);
1096 	execsw = newexecsw;
1097 	return 0;
1098 }
1099