xref: /openbsd-src/sys/kern/kern_exec.c (revision db3296cf5c1dd9058ceecc3a29fe4aaa0bd26000)
1 /*	$OpenBSD: kern_exec.c,v 1.80 2003/06/21 00:42:58 tedu Exp $	*/
2 /*	$NetBSD: kern_exec.c,v 1.75 1996/02/09 18:59:28 christos Exp $	*/
3 
4 /*-
5  * Copyright (C) 1993, 1994 Christopher G. Demetriou
6  * Copyright (C) 1992 Wolfgang Solfrank.
7  * Copyright (C) 1992 TooLs GmbH.
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by TooLs GmbH.
21  * 4. The name of TooLs GmbH may not be used to endorse or promote products
22  *    derived from this software without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
25  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
26  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
27  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
30  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/filedesc.h>
39 #include <sys/kernel.h>
40 #include <sys/proc.h>
41 #include <sys/mount.h>
42 #include <sys/malloc.h>
43 #include <sys/namei.h>
44 #include <sys/vnode.h>
45 #include <sys/file.h>
46 #include <sys/acct.h>
47 #include <sys/exec.h>
48 #include <sys/ktrace.h>
49 #include <sys/resourcevar.h>
50 #include <sys/wait.h>
51 #include <sys/mman.h>
52 #include <sys/signalvar.h>
53 #include <sys/stat.h>
54 #include <sys/conf.h>
55 #ifdef SYSVSHM
56 #include <sys/shm.h>
57 #endif
58 
59 #include <sys/syscallargs.h>
60 
61 #include <uvm/uvm_extern.h>
62 
63 #include <machine/cpu.h>
64 #include <machine/reg.h>
65 
66 #include <dev/rndvar.h>
67 
68 /*
69  * Map the shared signal code.
70  */
71 int exec_sigcode_map(struct proc *, struct emul *);
72 
73 /*
74  * stackgap_random specifies if the stackgap should have a random size added
75  * to it. Must be a n^2. If non-zero, the stack gap will be calculated as:
76  * (arc4random() * ALIGNBYTES) & (stackgap_random - 1) + STACKGAPLEN.
77  */
78 int stackgap_random = 64*1024;
79 
80 /*
81  * check exec:
82  * given an "executable" described in the exec package's namei info,
83  * see what we can do with it.
84  *
85  * ON ENTRY:
86  *	exec package with appropriate namei info
87  *	proc pointer of exec'ing proc
88  *	NO SELF-LOCKED VNODES
89  *
90  * ON EXIT:
91  *	error:	nothing held, etc.  exec header still allocated.
92  *	ok:	filled exec package, one locked vnode.
93  *
94  * EXEC SWITCH ENTRY:
95  * 	Locked vnode to check, exec package, proc.
96  *
97  * EXEC SWITCH EXIT:
98  *	ok:	return 0, filled exec package, one locked vnode.
99  *	error:	destructive:
100  *			everything deallocated execept exec header.
101  *		non-destructive:
102  *			error code, locked vnode, exec header unmodified
103  */
104 int
105 check_exec(p, epp)
106 	struct proc *p;
107 	struct exec_package *epp;
108 {
109 	int error, i;
110 	struct vnode *vp;
111 	struct nameidata *ndp;
112 	size_t resid;
113 
114 	ndp = epp->ep_ndp;
115 	ndp->ni_cnd.cn_nameiop = LOOKUP;
116 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
117 	/* first get the vnode */
118 	if ((error = namei(ndp)) != 0)
119 		return (error);
120 	epp->ep_vp = vp = ndp->ni_vp;
121 
122 	/* check for regular file */
123 	if (vp->v_type == VDIR) {
124 		error = EISDIR;
125 		goto bad1;
126 	}
127 	if (vp->v_type != VREG) {
128 		error = EACCES;
129 		goto bad1;
130 	}
131 
132 	/* get attributes */
133 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
134 		goto bad1;
135 
136 	/* Check mount point */
137 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
138 		error = EACCES;
139 		goto bad1;
140 	}
141 
142 	if ((vp->v_mount->mnt_flag & MNT_NOSUID))
143 		epp->ep_vap->va_mode &= ~(VSUID | VSGID);
144 
145 	/* check access.  for root we have to see if any exec bit on */
146 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
147 		goto bad1;
148 	if ((epp->ep_vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
149 		error = EACCES;
150 		goto bad1;
151 	}
152 
153 	/* try to open it */
154 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
155 		goto bad1;
156 
157 	/* unlock vp, we need it unlocked from here */
158 	VOP_UNLOCK(vp, 0, p);
159 
160 	/* now we have the file, get the exec header */
161 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
162 	    UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
163 	if (error)
164 		goto bad2;
165 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
166 
167 	/*
168 	 * set up the vmcmds for creation of the process
169 	 * address space
170 	 */
171 	error = ENOEXEC;
172 	for (i = 0; i < nexecs && error != 0; i++) {
173 		int newerror;
174 
175 		if (execsw[i].es_check == NULL)
176 			continue;
177 		newerror = (*execsw[i].es_check)(p, epp);
178 		/* make sure the first "interesting" error code is saved. */
179 		if (!newerror || error == ENOEXEC)
180 			error = newerror;
181 		if (epp->ep_flags & EXEC_DESTR && error != 0)
182 			return (error);
183 	}
184 	if (!error) {
185 		/* check that entry point is sane */
186 		if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
187 			error = ENOEXEC;
188 		}
189 
190 		/* check limits */
191 		if ((epp->ep_tsize > MAXTSIZ) ||
192 		    (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur))
193 			error = ENOMEM;
194 
195 		if (!error)
196 			return (0);
197 	}
198 
199 	/*
200 	 * free any vmspace-creation commands,
201 	 * and release their references
202 	 */
203 	kill_vmcmds(&epp->ep_vmcmds);
204 
205 bad2:
206 	/*
207 	 * close the vnode, free the pathname buf, and punt.
208 	 */
209 	vn_close(vp, FREAD, p->p_ucred, p);
210 	FREE(ndp->ni_cnd.cn_pnbuf, M_NAMEI);
211 	return (error);
212 
213 bad1:
214 	/*
215 	 * free the namei pathname buffer, and put the vnode
216 	 * (which we don't yet have open).
217 	 */
218 	FREE(ndp->ni_cnd.cn_pnbuf, M_NAMEI);
219 	vput(vp);
220 	return (error);
221 }
222 
223 /*
224  * exec system call
225  */
226 /* ARGSUSED */
227 int
228 sys_execve(p, v, retval)
229 	register struct proc *p;
230 	void *v;
231 	register_t *retval;
232 {
233 	struct sys_execve_args /* {
234 		syscallarg(char *) path;
235 		syscallarg(char * *) argp;
236 		syscallarg(char * *) envp;
237 	} */ *uap = v;
238 	int error, i;
239 	struct exec_package pack;
240 	struct nameidata nid;
241 	struct vattr attr;
242 	struct ucred *cred = p->p_ucred;
243 	char *argp;
244 	char * const *cpp, *dp, *sp;
245 	long argc, envc;
246 	size_t len, sgap;
247 #ifdef MACHINE_STACK_GROWS_UP
248 	size_t slen;
249 #endif
250 	char *stack;
251 	struct ps_strings arginfo;
252 	struct vmspace *vm = p->p_vmspace;
253 	char **tmpfap;
254 	extern struct emul emul_native;
255 
256 	/*
257 	 * Cheap solution to complicated problems.
258 	 * Mark this process as "leave me alone, I'm execing".
259 	 */
260 	p->p_flag |= P_INEXEC;
261 
262 	/*
263 	 * figure out the maximum size of an exec header, if necessary.
264 	 * XXX should be able to keep LKM code from modifying exec switch
265 	 * when we're still using it, but...
266 	 */
267 	if (exec_maxhdrsz == 0) {
268 		for (i = 0; i < nexecs; i++)
269 			if (execsw[i].es_check != NULL
270 			    && execsw[i].es_hdrsz > exec_maxhdrsz)
271 				exec_maxhdrsz = execsw[i].es_hdrsz;
272 	}
273 
274 	/* init the namei data to point the file user's program name */
275 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
276 
277 	/*
278 	 * initialize the fields of the exec package.
279 	 */
280 	pack.ep_name = (char *)SCARG(uap, path);
281 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
282 	pack.ep_hdrlen = exec_maxhdrsz;
283 	pack.ep_hdrvalid = 0;
284 	pack.ep_ndp = &nid;
285 	pack.ep_emul_arg = NULL;
286 	VMCMDSET_INIT(&pack.ep_vmcmds);
287 	pack.ep_vap = &attr;
288 	pack.ep_emul = &emul_native;
289 	pack.ep_flags = 0;
290 
291 	/* see if we can run it. */
292 	if ((error = check_exec(p, &pack)) != 0) {
293 		goto freehdr;
294 	}
295 
296 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
297 
298 	/* allocate an argument buffer */
299 	argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS);
300 #ifdef DIAGNOSTIC
301 	if (argp == NULL)
302 		panic("execve: argp == NULL");
303 #endif
304 	dp = argp;
305 	argc = 0;
306 
307 	/* copy the fake args list, if there's one, freeing it as we go */
308 	if (pack.ep_flags & EXEC_HASARGL) {
309 		tmpfap = pack.ep_fa;
310 		while (*tmpfap != NULL) {
311 			char *cp;
312 
313 			cp = *tmpfap;
314 			while (*cp)
315 				*dp++ = *cp++;
316 			dp++;
317 
318 			free(*tmpfap, M_EXEC);
319 			tmpfap++; argc++;
320 		}
321 		FREE(pack.ep_fa, M_EXEC);
322 		pack.ep_flags &= ~EXEC_HASARGL;
323 	}
324 
325 	/* Now get argv & environment */
326 	if (!(cpp = SCARG(uap, argp))) {
327 		error = EFAULT;
328 		goto bad;
329 	}
330 
331 	if (pack.ep_flags & EXEC_SKIPARG)
332 		cpp++;
333 
334 	while (1) {
335 		len = argp + ARG_MAX - dp;
336 		if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
337 			goto bad;
338 		if (!sp)
339 			break;
340 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
341 			if (error == ENAMETOOLONG)
342 				error = E2BIG;
343 			goto bad;
344 		}
345 		dp += len;
346 		cpp++;
347 		argc++;
348 	}
349 
350 	envc = 0;
351 	/* environment does not need to be there */
352 	if ((cpp = SCARG(uap, envp)) != NULL ) {
353 		while (1) {
354 			len = argp + ARG_MAX - dp;
355 			if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
356 				goto bad;
357 			if (!sp)
358 				break;
359 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
360 				if (error == ENAMETOOLONG)
361 					error = E2BIG;
362 				goto bad;
363 			}
364 			dp += len;
365 			cpp++;
366 			envc++;
367 		}
368 	}
369 
370 	dp = (char *)ALIGN(dp);
371 
372 	sgap = STACKGAPLEN;
373 	if (stackgap_random != 0)
374 		sgap += (arc4random() * ALIGNBYTES) & (stackgap_random - 1);
375 	/* Now check if args & environ fit into new stack */
376 	len = ((argc + envc + 2 + pack.ep_emul->e_arglen) * sizeof(char *) +
377 	    sizeof(long) + dp + sgap + sizeof(struct ps_strings)) - argp;
378 
379 	len = ALIGN(len);	/* make the stack "safely" aligned */
380 
381 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
382 		error = ENOMEM;
383 		goto bad;
384 	}
385 
386 	/* adjust "active stack depth" for process VSZ */
387 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
388 
389 	/*
390 	 * Prepare vmspace for remapping. Note that uvmspace_exec can replace
391 	 * p_vmspace!
392 	 */
393 	uvmspace_exec(p, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
394 
395 	vm = p->p_vmspace;
396 	/* Now map address space */
397 	vm->vm_taddr = (char *)pack.ep_taddr;
398 	vm->vm_tsize = btoc(pack.ep_tsize);
399 	vm->vm_daddr = (char *)pack.ep_daddr;
400 	vm->vm_dsize = btoc(pack.ep_dsize);
401 	vm->vm_ssize = btoc(pack.ep_ssize);
402 	vm->vm_maxsaddr = (char *)pack.ep_maxsaddr;
403 
404 	/* create the new process's VM space by running the vmcmds */
405 #ifdef DIAGNOSTIC
406 	if (pack.ep_vmcmds.evs_used == 0)
407 		panic("execve: no vmcmds");
408 #endif
409 	error = exec_process_vmcmds(p, &pack);
410 
411 	/* if an error happened, deallocate and punt */
412 	if (error)
413 		goto exec_abort;
414 
415 	/* remember information about the process */
416 	arginfo.ps_nargvstr = argc;
417 	arginfo.ps_nenvstr = envc;
418 
419 #ifdef MACHINE_STACK_GROWS_UP
420 	stack = (char *)USRSTACK + sizeof(arginfo);
421 	slen = len - sizeof(arginfo);
422 #else
423 	stack = (char *)(USRSTACK - len);
424 #endif
425 	/* Now copy argc, args & environ to new stack */
426 	if (!(*pack.ep_emul->e_copyargs)(&pack, &arginfo, stack, argp))
427 		goto exec_abort;
428 
429 	/* copy out the process's ps_strings structure */
430 	if (copyout(&arginfo, (char *)PS_STRINGS, sizeof(arginfo)))
431 		goto exec_abort;
432 
433 	stopprofclock(p);	/* stop profiling */
434 	fdcloseexec(p);		/* handle close on exec */
435 	execsigs(p);		/* reset catched signals */
436 
437 	/* set command name & other accounting info */
438 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
439 	bcopy(nid.ni_cnd.cn_nameptr, p->p_comm, len);
440 	p->p_comm[len] = 0;
441 	p->p_acflag &= ~AFORK;
442 
443 	/* record proc's vnode, for use by procfs and others */
444 	if (p->p_textvp)
445 		vrele(p->p_textvp);
446 	VREF(pack.ep_vp);
447 	p->p_textvp = pack.ep_vp;
448 
449 	p->p_flag |= P_EXEC;
450 	if (p->p_flag & P_PPWAIT) {
451 		p->p_flag &= ~P_PPWAIT;
452 		wakeup((caddr_t)p->p_pptr);
453 	}
454 
455 	/*
456 	 * If process does execve() while it has a mismatched real,
457 	 * effective, or saved uid/gid, we set P_SUGIDEXEC.
458 	 */
459 	if (p->p_ucred->cr_uid != p->p_cred->p_ruid ||
460 	    p->p_ucred->cr_uid != p->p_cred->p_svuid ||
461 	    p->p_ucred->cr_gid != p->p_cred->p_rgid ||
462 	    p->p_ucred->cr_gid != p->p_cred->p_svgid)
463 		p->p_flag |= P_SUGIDEXEC;
464 	else
465 		p->p_flag &= ~P_SUGIDEXEC;
466 
467 	/*
468 	 * deal with set[ug]id.
469 	 * MNT_NOEXEC has already been used to disable s[ug]id.
470 	 */
471 	if ((attr.va_mode & (VSUID | VSGID)) && proc_cansugid(p)) {
472 		int i;
473 
474 		p->p_flag |= P_SUGID;
475 		p->p_flag |= P_SUGIDEXEC;
476 
477 #ifdef KTRACE
478 		/*
479 		 * If process is being ktraced, turn off - unless
480 		 * root set it.
481 		 */
482 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) {
483 			p->p_traceflag = 0;
484 			ktrsettracevnode(p, NULL);
485 		}
486 #endif
487 		p->p_ucred = crcopy(cred);
488 		if (attr.va_mode & VSUID)
489 			p->p_ucred->cr_uid = attr.va_uid;
490 		if (attr.va_mode & VSGID)
491 			p->p_ucred->cr_gid = attr.va_gid;
492 
493 		/*
494 		 * For set[ug]id processes, a few caveats apply to
495 		 * stdin, stdout, and stderr.
496 		 */
497 		for (i = 0; i < 3; i++) {
498 			struct file *fp = NULL;
499 
500 			/*
501 			 * NOTE - This will never return NULL because of
502 			 * unmature fds. The file descriptor table is not
503 			 * shared because we're suid.
504 			 */
505 			fp = fd_getfile(p->p_fd, i);
506 #ifdef PROCFS
507 			/*
508 			 * Close descriptors that are writing to procfs.
509 			 */
510 			if (fp && fp->f_type == DTYPE_VNODE &&
511 			    ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS &&
512 			    (fp->f_flag & FWRITE)) {
513 				fdrelease(p, i);
514 				fp = NULL;
515 			}
516 #endif
517 
518 			/*
519 			 * Ensure that stdin, stdout, and stderr are already
520 			 * allocated.  We do not want userland to accidentally
521 			 * allocate descriptors in this range which has implied
522 			 * meaning to libc.
523 			 */
524 			if (fp == NULL) {
525 				short flags = FREAD | (i == 0 ? 0 : FWRITE);
526 				struct vnode *vp;
527 				int indx;
528 
529 				if ((error = falloc(p, &fp, &indx)) != 0)
530 					goto exec_abort;
531 #ifdef DIAGNOSTIC
532 				if (indx != i)
533 					panic("sys_execve: falloc indx != i");
534 #endif
535 				if ((error = cdevvp(getnulldev(), &vp)) != 0) {
536 					fdremove(p->p_fd, indx);
537 					closef(fp, p);
538 					goto exec_abort;
539 				}
540 				if ((error = VOP_OPEN(vp, flags, p->p_ucred, p)) != 0) {
541 					fdremove(p->p_fd, indx);
542 					closef(fp, p);
543 					vrele(vp);
544 					goto exec_abort;
545 				}
546 				if (flags & FWRITE)
547 					vp->v_writecount++;
548 				fp->f_flag = flags;
549 				fp->f_type = DTYPE_VNODE;
550 				fp->f_ops = &vnops;
551 				fp->f_data = (caddr_t)vp;
552 				FILE_SET_MATURE(fp);
553 			}
554 		}
555 	} else
556 		p->p_flag &= ~P_SUGID;
557 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
558 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
559 
560 	if (p->p_flag & P_SUGIDEXEC) {
561 		int i, s = splclock();
562 
563 		timeout_del(&p->p_realit_to);
564 		timerclear(&p->p_realtimer.it_interval);
565 		timerclear(&p->p_realtimer.it_value);
566 		for (i = 0; i < sizeof(p->p_stats->p_timer) /
567 		    sizeof(p->p_stats->p_timer[0]); i++) {
568 			timerclear(&p->p_stats->p_timer[i].it_interval);
569 			timerclear(&p->p_stats->p_timer[i].it_value);
570 		}
571 		splx(s);
572 	}
573 
574 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
575 
576 	FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
577 	vn_close(pack.ep_vp, FREAD, cred, p);
578 
579 	/*
580 	 * notify others that we exec'd
581 	 */
582 	KNOTE(&p->p_klist, NOTE_EXEC);
583 
584 	/* setup new registers and do misc. setup. */
585 	if (pack.ep_emul->e_fixup != NULL) {
586 		if ((*pack.ep_emul->e_fixup)(p, &pack) != 0)
587 			goto free_pack_abort;
588 	}
589 #ifdef MACHINE_STACK_GROWS_UP
590 	(*pack.ep_emul->e_setregs)(p, &pack, (u_long)stack + slen, retval);
591 #else
592 	(*pack.ep_emul->e_setregs)(p, &pack, (u_long)stack, retval);
593 #endif
594 
595 	/* map the process's signal trampoline code */
596 	if (exec_sigcode_map(p, pack.ep_emul))
597 		goto exec_abort;
598 
599 	if (p->p_flag & P_TRACED)
600 		psignal(p, SIGTRAP);
601 
602 	free(pack.ep_hdr, M_EXEC);
603 
604 	/*
605 	 * Call emulation specific exec hook. This can setup per-process
606 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
607 	 *
608 	 * If we are executing process of different emulation than the
609 	 * original forked process, call e_proc_exit() of the old emulation
610 	 * first, then e_proc_exec() of new emulation. If the emulation is
611 	 * same, the exec hook code should deallocate any old emulation
612 	 * resources held previously by this process.
613 	 */
614 	if (p->p_emul && p->p_emul->e_proc_exit &&
615 	    p->p_emul != pack.ep_emul)
616 		(*p->p_emul->e_proc_exit)(p);
617 
618 	/*
619 	 * Call exec hook. Emulation code may NOT store reference to anything
620 	 * from &pack.
621 	 */
622 	if (pack.ep_emul->e_proc_exec)
623 		(*pack.ep_emul->e_proc_exec)(p, &pack);
624 
625 	/* update p_emul, the old value is no longer needed */
626 	p->p_emul = pack.ep_emul;
627 
628 #ifdef KTRACE
629 	if (KTRPOINT(p, KTR_EMUL))
630 		ktremul(p, p->p_emul->e_name);
631 #endif
632 	p->p_flag &= ~P_INEXEC;
633 	return (0);
634 
635 bad:
636 	/* free the vmspace-creation commands, and release their references */
637 	kill_vmcmds(&pack.ep_vmcmds);
638 	/* kill any opened file descriptor, if necessary */
639 	if (pack.ep_flags & EXEC_HASFD) {
640 		pack.ep_flags &= ~EXEC_HASFD;
641 		(void) fdrelease(p, pack.ep_fd);
642 	}
643 	/* close and put the exec'd file */
644 	vn_close(pack.ep_vp, FREAD, cred, p);
645 	FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
646 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
647 
648 freehdr:
649 	free(pack.ep_hdr, M_EXEC);
650 	p->p_flag &= ~P_INEXEC;
651 	return (error);
652 
653 exec_abort:
654 	/*
655 	 * the old process doesn't exist anymore.  exit gracefully.
656 	 * get rid of the (new) address space we have created, if any, get rid
657 	 * of our namei data and vnode, and exit noting failure
658 	 */
659 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
660 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
661 	if (pack.ep_emul_arg)
662 		FREE(pack.ep_emul_arg, M_TEMP);
663 	FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
664 	vn_close(pack.ep_vp, FREAD, cred, p);
665 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
666 
667 free_pack_abort:
668 	free(pack.ep_hdr, M_EXEC);
669 	exit1(p, W_EXITCODE(0, SIGABRT));
670 
671 	/* NOTREACHED */
672 	p->p_flag &= ~P_INEXEC;
673 	return (0);
674 }
675 
676 
677 void *
678 copyargs(pack, arginfo, stack, argp)
679 	struct exec_package *pack;
680 	struct ps_strings *arginfo;
681 	void *stack;
682 	void *argp;
683 {
684 	char **cpp = stack;
685 	char *dp, *sp;
686 	size_t len;
687 	void *nullp = NULL;
688 	long argc = arginfo->ps_nargvstr;
689 	int envc = arginfo->ps_nenvstr;
690 
691 	if (copyout(&argc, cpp++, sizeof(argc)))
692 		return (NULL);
693 
694 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_emul->e_arglen);
695 	sp = argp;
696 
697 	/* XXX don't copy them out, remap them! */
698 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
699 
700 	for (; --argc >= 0; sp += len, dp += len)
701 		if (copyout(&dp, cpp++, sizeof(dp)) ||
702 		    copyoutstr(sp, dp, ARG_MAX, &len))
703 			return (NULL);
704 
705 	if (copyout(&nullp, cpp++, sizeof(nullp)))
706 		return (NULL);
707 
708 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
709 
710 	for (; --envc >= 0; sp += len, dp += len)
711 		if (copyout(&dp, cpp++, sizeof(dp)) ||
712 		    copyoutstr(sp, dp, ARG_MAX, &len))
713 			return (NULL);
714 
715 	if (copyout(&nullp, cpp++, sizeof(nullp)))
716 		return (NULL);
717 
718 	return (cpp);
719 }
720 
721 int
722 exec_sigcode_map(struct proc *p, struct emul *e)
723 {
724 	vsize_t sz;
725 
726 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
727 
728 	/*
729 	 * If we don't have a sigobject for this emulation, create one.
730 	 *
731 	 * sigobject is an anonymous memory object (just like SYSV shared
732 	 * memory) that we keep a permanent reference to and that we map
733 	 * in all processes that need this sigcode. The creation is simple,
734 	 * we create an object, add a permanent reference to it, map it in
735 	 * kernel space, copy out the sigcode to it and unmap it.
736 	 * Then we map it with PROT_READ|PROT_EXEC into the process just
737 	 * the way sys_mmap would map it.
738 	 */
739 	if (e->e_sigobject == NULL) {
740 		vaddr_t va;
741 		int r;
742 
743 		e->e_sigobject = uao_create(sz, 0);
744 		uao_reference(e->e_sigobject);	/* permanent reference */
745 
746 		va = vm_map_min(kernel_map);	/* hint */
747 		if ((r = uvm_map(kernel_map, &va, round_page(sz), e->e_sigobject,
748 		    0, 0, UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
749 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
750 			printf("kernel mapping failed %d\n", r);
751 			return (ENOMEM);
752 		}
753 		memcpy((void *)va, e->e_sigcode, sz);
754 		uvm_unmap(kernel_map, va, va + round_page(sz));
755 	}
756 
757 	/* Just a hint to uvm_mmap where to put it. */
758 	p->p_sigcode = uvm_map_hint(p, VM_PROT_READ|VM_PROT_EXECUTE);
759 	uao_reference(e->e_sigobject);
760 	if (uvm_map(&p->p_vmspace->vm_map, &p->p_sigcode, round_page(sz),
761 	    e->e_sigobject, 0, 0, UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX,
762 	    UVM_INH_SHARE, UVM_ADV_RANDOM, 0))) {
763 		printf("user mapping failed\n");
764 		return (ENOMEM);
765 	}
766 
767 	return (0);
768 }
769