xref: /openbsd-src/sys/kern/kern_exec.c (revision 5738bc6274bbfbba95d07e5dc8d380a26b687449)
1 /*	$OpenBSD: kern_exec.c,v 1.90 2004/08/24 23:01:26 mickey 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/pool.h>
44 #include <sys/namei.h>
45 #include <sys/vnode.h>
46 #include <sys/file.h>
47 #include <sys/acct.h>
48 #include <sys/exec.h>
49 #include <sys/ktrace.h>
50 #include <sys/resourcevar.h>
51 #include <sys/wait.h>
52 #include <sys/mman.h>
53 #include <sys/signalvar.h>
54 #include <sys/stat.h>
55 #include <sys/conf.h>
56 #ifdef SYSVSHM
57 #include <sys/shm.h>
58 #endif
59 
60 #include <sys/syscallargs.h>
61 
62 #include <uvm/uvm_extern.h>
63 
64 #include <machine/cpu.h>
65 #include <machine/reg.h>
66 
67 #include <dev/rndvar.h>
68 
69 #include "systrace.h"
70 
71 #if NSYSTRACE > 0
72 #include <dev/systrace.h>
73 #endif
74 
75 /*
76  * Map the shared signal code.
77  */
78 int exec_sigcode_map(struct proc *, struct emul *);
79 
80 /*
81  * stackgap_random specifies if the stackgap should have a random size added
82  * to it. Must be a n^2. If non-zero, the stack gap will be calculated as:
83  * (arc4random() * ALIGNBYTES) & (stackgap_random - 1) + STACKGAPLEN.
84  */
85 int stackgap_random = 64*1024;
86 
87 /*
88  * check exec:
89  * given an "executable" described in the exec package's namei info,
90  * see what we can do with it.
91  *
92  * ON ENTRY:
93  *	exec package with appropriate namei info
94  *	proc pointer of exec'ing proc
95  *	NO SELF-LOCKED VNODES
96  *
97  * ON EXIT:
98  *	error:	nothing held, etc.  exec header still allocated.
99  *	ok:	filled exec package, one locked vnode.
100  *
101  * EXEC SWITCH ENTRY:
102  * 	Locked vnode to check, exec package, proc.
103  *
104  * EXEC SWITCH EXIT:
105  *	ok:	return 0, filled exec package, one locked vnode.
106  *	error:	destructive:
107  *			everything deallocated execept exec header.
108  *		non-destructive:
109  *			error code, locked vnode, exec header unmodified
110  */
111 int
112 check_exec(p, epp)
113 	struct proc *p;
114 	struct exec_package *epp;
115 {
116 	int error, i;
117 	struct vnode *vp;
118 	struct nameidata *ndp;
119 	size_t resid;
120 
121 	ndp = epp->ep_ndp;
122 	ndp->ni_cnd.cn_nameiop = LOOKUP;
123 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
124 	/* first get the vnode */
125 	if ((error = namei(ndp)) != 0)
126 		return (error);
127 	epp->ep_vp = vp = ndp->ni_vp;
128 
129 	/* check for regular file */
130 	if (vp->v_type == VDIR) {
131 		error = EISDIR;
132 		goto bad1;
133 	}
134 	if (vp->v_type != VREG) {
135 		error = EACCES;
136 		goto bad1;
137 	}
138 
139 	/* get attributes */
140 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
141 		goto bad1;
142 
143 	/* Check mount point */
144 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
145 		error = EACCES;
146 		goto bad1;
147 	}
148 
149 	if ((vp->v_mount->mnt_flag & MNT_NOSUID))
150 		epp->ep_vap->va_mode &= ~(VSUID | VSGID);
151 
152 	/* check access.  for root we have to see if any exec bit on */
153 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
154 		goto bad1;
155 	if ((epp->ep_vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
156 		error = EACCES;
157 		goto bad1;
158 	}
159 
160 	/* try to open it */
161 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
162 		goto bad1;
163 
164 	/* unlock vp, we need it unlocked from here */
165 	VOP_UNLOCK(vp, 0, p);
166 
167 	/* now we have the file, get the exec header */
168 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
169 	    UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
170 	if (error)
171 		goto bad2;
172 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
173 
174 	/*
175 	 * set up the vmcmds for creation of the process
176 	 * address space
177 	 */
178 	error = ENOEXEC;
179 	for (i = 0; i < nexecs && error != 0; i++) {
180 		int newerror;
181 
182 		if (execsw[i].es_check == NULL)
183 			continue;
184 		newerror = (*execsw[i].es_check)(p, epp);
185 		if (!newerror && !(epp->ep_emul->e_flags & EMUL_ENABLED))
186 			newerror = ENOEXEC;
187 		/* make sure the first "interesting" error code is saved. */
188 		if (!newerror || error == ENOEXEC)
189 			error = newerror;
190 		if (epp->ep_flags & EXEC_DESTR && error != 0)
191 			return (error);
192 	}
193 	if (!error) {
194 		/* check that entry point is sane */
195 		if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
196 			error = ENOEXEC;
197 		}
198 
199 		/* check limits */
200 		if ((epp->ep_tsize > MAXTSIZ) ||
201 		    (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur))
202 			error = ENOMEM;
203 
204 		if (!error)
205 			return (0);
206 	}
207 
208 	/*
209 	 * free any vmspace-creation commands,
210 	 * and release their references
211 	 */
212 	kill_vmcmds(&epp->ep_vmcmds);
213 
214 bad2:
215 	/*
216 	 * close the vnode, free the pathname buf, and punt.
217 	 */
218 	vn_close(vp, FREAD, p->p_ucred, p);
219 	pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf);
220 	return (error);
221 
222 bad1:
223 	/*
224 	 * free the namei pathname buffer, and put the vnode
225 	 * (which we don't yet have open).
226 	 */
227 	pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf);
228 	vput(vp);
229 	return (error);
230 }
231 
232 /*
233  * exec system call
234  */
235 /* ARGSUSED */
236 int
237 sys_execve(p, v, retval)
238 	register struct proc *p;
239 	void *v;
240 	register_t *retval;
241 {
242 	struct sys_execve_args /* {
243 		syscallarg(const char *) path;
244 		syscallarg(char *const *) argp;
245 		syscallarg(char *const *) envp;
246 	} */ *uap = v;
247 	int error;
248 	struct exec_package pack;
249 	struct nameidata nid;
250 	struct vattr attr;
251 	struct ucred *cred = p->p_ucred;
252 	char *argp;
253 	char * const *cpp, *dp, *sp;
254 	long argc, envc;
255 	size_t len, sgap;
256 #ifdef MACHINE_STACK_GROWS_UP
257 	size_t slen;
258 #endif
259 	char *stack;
260 	struct ps_strings arginfo;
261 	struct vmspace *vm = p->p_vmspace;
262 	char **tmpfap;
263 	extern struct emul emul_native;
264 #if NSYSTRACE > 0
265 	int wassugid =
266 	    ISSET(p->p_flag, P_SUGID) || ISSET(p->p_flag, P_SUGIDEXEC);
267 	char pathbuf[MAXPATHLEN];
268 	size_t pathbuflen;
269 #endif
270 
271 	/*
272 	 * Cheap solution to complicated problems.
273 	 * Mark this process as "leave me alone, I'm execing".
274 	 */
275 	p->p_flag |= P_INEXEC;
276 
277 #if NSYSTRACE > 0
278 	if (ISSET(p->p_flag, P_SYSTRACE))
279 		systrace_execve0(p);
280 
281 	error = copyinstr(SCARG(uap, path), pathbuf, MAXPATHLEN, &pathbuflen);
282 	if (error != 0)
283 		goto clrflag;
284 
285 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_SYSSPACE, pathbuf, p);
286 #else
287 	/* init the namei data to point the file user's program name */
288 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
289 #endif
290 
291 	/*
292 	 * initialize the fields of the exec package.
293 	 */
294 #if NSYSTRACE > 0
295 	pack.ep_name = pathbuf;
296 #else
297 	pack.ep_name = (char *)SCARG(uap, path);
298 #endif
299 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
300 	pack.ep_hdrlen = exec_maxhdrsz;
301 	pack.ep_hdrvalid = 0;
302 	pack.ep_ndp = &nid;
303 	pack.ep_interp = NULL;
304 	pack.ep_emul_arg = NULL;
305 	VMCMDSET_INIT(&pack.ep_vmcmds);
306 	pack.ep_vap = &attr;
307 	pack.ep_emul = &emul_native;
308 	pack.ep_flags = 0;
309 
310 	/* see if we can run it. */
311 	if ((error = check_exec(p, &pack)) != 0) {
312 		goto freehdr;
313 	}
314 
315 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
316 
317 	/* allocate an argument buffer */
318 	argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS);
319 #ifdef DIAGNOSTIC
320 	if (argp == NULL)
321 		panic("execve: argp == NULL");
322 #endif
323 	dp = argp;
324 	argc = 0;
325 
326 	/* copy the fake args list, if there's one, freeing it as we go */
327 	if (pack.ep_flags & EXEC_HASARGL) {
328 		tmpfap = pack.ep_fa;
329 		while (*tmpfap != NULL) {
330 			char *cp;
331 
332 			cp = *tmpfap;
333 			while (*cp)
334 				*dp++ = *cp++;
335 			dp++;
336 
337 			free(*tmpfap, M_EXEC);
338 			tmpfap++; argc++;
339 		}
340 		FREE(pack.ep_fa, M_EXEC);
341 		pack.ep_flags &= ~EXEC_HASARGL;
342 	}
343 
344 	/* Now get argv & environment */
345 	if (!(cpp = SCARG(uap, argp))) {
346 		error = EFAULT;
347 		goto bad;
348 	}
349 
350 	if (pack.ep_flags & EXEC_SKIPARG)
351 		cpp++;
352 
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 		argc++;
367 	}
368 
369 	envc = 0;
370 	/* environment does not need to be there */
371 	if ((cpp = SCARG(uap, envp)) != NULL ) {
372 		while (1) {
373 			len = argp + ARG_MAX - dp;
374 			if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
375 				goto bad;
376 			if (!sp)
377 				break;
378 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
379 				if (error == ENAMETOOLONG)
380 					error = E2BIG;
381 				goto bad;
382 			}
383 			dp += len;
384 			cpp++;
385 			envc++;
386 		}
387 	}
388 
389 	dp = (char *)ALIGN(dp);
390 
391 	sgap = STACKGAPLEN;
392 	if (stackgap_random != 0)
393 		sgap += (arc4random() * ALIGNBYTES) & (stackgap_random - 1);
394 	/* Now check if args & environ fit into new stack */
395 	len = ((argc + envc + 2 + pack.ep_emul->e_arglen) * sizeof(char *) +
396 	    sizeof(long) + dp + sgap + sizeof(struct ps_strings)) - argp;
397 
398 	len = ALIGN(len);	/* make the stack "safely" aligned */
399 
400 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
401 		error = ENOMEM;
402 		goto bad;
403 	}
404 
405 	/* adjust "active stack depth" for process VSZ */
406 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
407 
408 	/*
409 	 * Prepare vmspace for remapping. Note that uvmspace_exec can replace
410 	 * p_vmspace!
411 	 */
412 	uvmspace_exec(p, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
413 
414 	vm = p->p_vmspace;
415 	/* Now map address space */
416 	vm->vm_taddr = (char *)pack.ep_taddr;
417 	vm->vm_tsize = btoc(pack.ep_tsize);
418 	vm->vm_daddr = (char *)pack.ep_daddr;
419 	vm->vm_dsize = btoc(pack.ep_dsize);
420 	vm->vm_ssize = btoc(pack.ep_ssize);
421 	vm->vm_maxsaddr = (char *)pack.ep_maxsaddr;
422 	vm->vm_minsaddr = (char *)pack.ep_minsaddr;
423 
424 	/* create the new process's VM space by running the vmcmds */
425 #ifdef DIAGNOSTIC
426 	if (pack.ep_vmcmds.evs_used == 0)
427 		panic("execve: no vmcmds");
428 #endif
429 	error = exec_process_vmcmds(p, &pack);
430 
431 	/* if an error happened, deallocate and punt */
432 	if (error)
433 		goto exec_abort;
434 
435 	/* remember information about the process */
436 	arginfo.ps_nargvstr = argc;
437 	arginfo.ps_nenvstr = envc;
438 
439 #ifdef MACHINE_STACK_GROWS_UP
440 	stack = (char *)USRSTACK + sizeof(arginfo);
441 	slen = len - sizeof(arginfo);
442 #else
443 	stack = (char *)(USRSTACK - len);
444 #endif
445 	/* Now copy argc, args & environ to new stack */
446 	if (!(*pack.ep_emul->e_copyargs)(&pack, &arginfo, stack, argp))
447 		goto exec_abort;
448 
449 	/* copy out the process's ps_strings structure */
450 	if (copyout(&arginfo, (char *)PS_STRINGS, sizeof(arginfo)))
451 		goto exec_abort;
452 
453 	stopprofclock(p);	/* stop profiling */
454 	fdcloseexec(p);		/* handle close on exec */
455 	execsigs(p);		/* reset catched signals */
456 
457 	/* set command name & other accounting info */
458 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
459 	bcopy(nid.ni_cnd.cn_nameptr, p->p_comm, len);
460 	p->p_comm[len] = 0;
461 	p->p_acflag &= ~AFORK;
462 
463 	/* record proc's vnode, for use by procfs and others */
464 	if (p->p_textvp)
465 		vrele(p->p_textvp);
466 	VREF(pack.ep_vp);
467 	p->p_textvp = pack.ep_vp;
468 
469 	p->p_flag |= P_EXEC;
470 	if (p->p_flag & P_PPWAIT) {
471 		p->p_flag &= ~P_PPWAIT;
472 		wakeup((caddr_t)p->p_pptr);
473 	}
474 
475 	/*
476 	 * If process does execve() while it has a mismatched real,
477 	 * effective, or saved uid/gid, we set P_SUGIDEXEC.
478 	 */
479 	if (p->p_ucred->cr_uid != p->p_cred->p_ruid ||
480 	    p->p_ucred->cr_uid != p->p_cred->p_svuid ||
481 	    p->p_ucred->cr_gid != p->p_cred->p_rgid ||
482 	    p->p_ucred->cr_gid != p->p_cred->p_svgid)
483 		p->p_flag |= P_SUGIDEXEC;
484 	else
485 		p->p_flag &= ~P_SUGIDEXEC;
486 
487 	/*
488 	 * deal with set[ug]id.
489 	 * MNT_NOEXEC has already been used to disable s[ug]id.
490 	 */
491 	if ((attr.va_mode & (VSUID | VSGID)) && proc_cansugid(p)) {
492 		int i;
493 
494 		p->p_flag |= P_SUGID;
495 		p->p_flag |= P_SUGIDEXEC;
496 
497 #ifdef KTRACE
498 		/*
499 		 * If process is being ktraced, turn off - unless
500 		 * root set it.
501 		 */
502 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) {
503 			p->p_traceflag = 0;
504 			ktrsettracevnode(p, NULL);
505 		}
506 #endif
507 		p->p_ucred = crcopy(cred);
508 		if (attr.va_mode & VSUID)
509 			p->p_ucred->cr_uid = attr.va_uid;
510 		if (attr.va_mode & VSGID)
511 			p->p_ucred->cr_gid = attr.va_gid;
512 
513 		/*
514 		 * For set[ug]id processes, a few caveats apply to
515 		 * stdin, stdout, and stderr.
516 		 */
517 		for (i = 0; i < 3; i++) {
518 			struct file *fp = NULL;
519 
520 			/*
521 			 * NOTE - This will never return NULL because of
522 			 * unmature fds. The file descriptor table is not
523 			 * shared because we're suid.
524 			 */
525 			fp = fd_getfile(p->p_fd, i);
526 #ifdef PROCFS
527 			/*
528 			 * Close descriptors that are writing to procfs.
529 			 */
530 			if (fp && fp->f_type == DTYPE_VNODE &&
531 			    ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS &&
532 			    (fp->f_flag & FWRITE)) {
533 				fdrelease(p, i);
534 				fp = NULL;
535 			}
536 #endif
537 
538 			/*
539 			 * Ensure that stdin, stdout, and stderr are already
540 			 * allocated.  We do not want userland to accidentally
541 			 * allocate descriptors in this range which has implied
542 			 * meaning to libc.
543 			 */
544 			if (fp == NULL) {
545 				short flags = FREAD | (i == 0 ? 0 : FWRITE);
546 				struct vnode *vp;
547 				int indx;
548 
549 				if ((error = falloc(p, &fp, &indx)) != 0)
550 					goto exec_abort;
551 #ifdef DIAGNOSTIC
552 				if (indx != i)
553 					panic("sys_execve: falloc indx != i");
554 #endif
555 				if ((error = cdevvp(getnulldev(), &vp)) != 0) {
556 					fdremove(p->p_fd, indx);
557 					closef(fp, p);
558 					goto exec_abort;
559 				}
560 				if ((error = VOP_OPEN(vp, flags, p->p_ucred, p)) != 0) {
561 					fdremove(p->p_fd, indx);
562 					closef(fp, p);
563 					vrele(vp);
564 					goto exec_abort;
565 				}
566 				if (flags & FWRITE)
567 					vp->v_writecount++;
568 				fp->f_flag = flags;
569 				fp->f_type = DTYPE_VNODE;
570 				fp->f_ops = &vnops;
571 				fp->f_data = (caddr_t)vp;
572 				FILE_SET_MATURE(fp);
573 			}
574 		}
575 	} else
576 		p->p_flag &= ~P_SUGID;
577 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
578 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
579 
580 	if (p->p_flag & P_SUGIDEXEC) {
581 		int i, s = splclock();
582 
583 		timeout_del(&p->p_realit_to);
584 		timerclear(&p->p_realtimer.it_interval);
585 		timerclear(&p->p_realtimer.it_value);
586 		for (i = 0; i < sizeof(p->p_stats->p_timer) /
587 		    sizeof(p->p_stats->p_timer[0]); i++) {
588 			timerclear(&p->p_stats->p_timer[i].it_interval);
589 			timerclear(&p->p_stats->p_timer[i].it_value);
590 		}
591 		splx(s);
592 	}
593 
594 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
595 
596 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
597 	vn_close(pack.ep_vp, FREAD, cred, p);
598 
599 	/*
600 	 * notify others that we exec'd
601 	 */
602 	KNOTE(&p->p_klist, NOTE_EXEC);
603 
604 	/* setup new registers and do misc. setup. */
605 	if (pack.ep_emul->e_fixup != NULL) {
606 		if ((*pack.ep_emul->e_fixup)(p, &pack) != 0)
607 			goto free_pack_abort;
608 	}
609 #ifdef MACHINE_STACK_GROWS_UP
610 	(*pack.ep_emul->e_setregs)(p, &pack, (u_long)stack + slen, retval);
611 #else
612 	(*pack.ep_emul->e_setregs)(p, &pack, (u_long)stack, retval);
613 #endif
614 
615 	/* map the process's signal trampoline code */
616 	if (exec_sigcode_map(p, pack.ep_emul))
617 		goto free_pack_abort;
618 
619 	if (p->p_flag & P_TRACED)
620 		psignal(p, SIGTRAP);
621 
622 	free(pack.ep_hdr, M_EXEC);
623 
624 	/*
625 	 * Call emulation specific exec hook. This can setup per-process
626 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
627 	 *
628 	 * If we are executing process of different emulation than the
629 	 * original forked process, call e_proc_exit() of the old emulation
630 	 * first, then e_proc_exec() of new emulation. If the emulation is
631 	 * same, the exec hook code should deallocate any old emulation
632 	 * resources held previously by this process.
633 	 */
634 	if (p->p_emul && p->p_emul->e_proc_exit &&
635 	    p->p_emul != pack.ep_emul)
636 		(*p->p_emul->e_proc_exit)(p);
637 
638 	/*
639 	 * Call exec hook. Emulation code may NOT store reference to anything
640 	 * from &pack.
641 	 */
642 	if (pack.ep_emul->e_proc_exec)
643 		(*pack.ep_emul->e_proc_exec)(p, &pack);
644 
645 	/* update p_emul, the old value is no longer needed */
646 	p->p_emul = pack.ep_emul;
647 
648 #ifdef KTRACE
649 	if (KTRPOINT(p, KTR_EMUL))
650 		ktremul(p, p->p_emul->e_name);
651 #endif
652 
653 	p->p_flag &= ~P_INEXEC;
654 
655 #if NSYSTRACE > 0
656 	if (ISSET(p->p_flag, P_SYSTRACE) &&
657 	    wassugid && !ISSET(p->p_flag, P_SUGID) &&
658 	    !ISSET(p->p_flag, P_SUGIDEXEC))
659 		systrace_execve1(pathbuf, p);
660 #endif
661 
662 	return (0);
663 
664 bad:
665 	/* free the vmspace-creation commands, and release their references */
666 	kill_vmcmds(&pack.ep_vmcmds);
667 	/* kill any opened file descriptor, if necessary */
668 	if (pack.ep_flags & EXEC_HASFD) {
669 		pack.ep_flags &= ~EXEC_HASFD;
670 		(void) fdrelease(p, pack.ep_fd);
671 	}
672 	if (pack.ep_interp != NULL)
673 		FREE(pack.ep_interp, M_TEMP);
674 	if (pack.ep_emul_arg != NULL)
675 		FREE(pack.ep_emul_arg, M_TEMP);
676 	/* close and put the exec'd file */
677 	vn_close(pack.ep_vp, FREAD, cred, p);
678 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
679 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
680 
681  freehdr:
682 	free(pack.ep_hdr, M_EXEC);
683 #if NSYSTRACE > 0
684  clrflag:
685 #endif
686 	p->p_flag &= ~P_INEXEC;
687 	return (error);
688 
689 exec_abort:
690 	/*
691 	 * the old process doesn't exist anymore.  exit gracefully.
692 	 * get rid of the (new) address space we have created, if any, get rid
693 	 * of our namei data and vnode, and exit noting failure
694 	 */
695 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
696 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
697 	if (pack.ep_interp != NULL)
698 		FREE(pack.ep_interp, M_TEMP);
699 	if (pack.ep_emul_arg != NULL)
700 		FREE(pack.ep_emul_arg, M_TEMP);
701 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
702 	vn_close(pack.ep_vp, FREAD, cred, p);
703 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
704 
705 free_pack_abort:
706 	free(pack.ep_hdr, M_EXEC);
707 	exit1(p, W_EXITCODE(0, SIGABRT));
708 
709 	/* NOTREACHED */
710 	p->p_flag &= ~P_INEXEC;
711 	return (0);
712 }
713 
714 
715 void *
716 copyargs(pack, arginfo, stack, argp)
717 	struct exec_package *pack;
718 	struct ps_strings *arginfo;
719 	void *stack;
720 	void *argp;
721 {
722 	char **cpp = stack;
723 	char *dp, *sp;
724 	size_t len;
725 	void *nullp = NULL;
726 	long argc = arginfo->ps_nargvstr;
727 	int envc = arginfo->ps_nenvstr;
728 
729 	if (copyout(&argc, cpp++, sizeof(argc)))
730 		return (NULL);
731 
732 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_emul->e_arglen);
733 	sp = argp;
734 
735 	/* XXX don't copy them out, remap them! */
736 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
737 
738 	for (; --argc >= 0; sp += len, dp += len)
739 		if (copyout(&dp, cpp++, sizeof(dp)) ||
740 		    copyoutstr(sp, dp, ARG_MAX, &len))
741 			return (NULL);
742 
743 	if (copyout(&nullp, cpp++, sizeof(nullp)))
744 		return (NULL);
745 
746 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
747 
748 	for (; --envc >= 0; sp += len, dp += len)
749 		if (copyout(&dp, cpp++, sizeof(dp)) ||
750 		    copyoutstr(sp, dp, ARG_MAX, &len))
751 			return (NULL);
752 
753 	if (copyout(&nullp, cpp++, sizeof(nullp)))
754 		return (NULL);
755 
756 	return (cpp);
757 }
758 
759 int
760 exec_sigcode_map(struct proc *p, struct emul *e)
761 {
762 	vsize_t sz;
763 
764 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
765 
766 	/*
767 	 * If we don't have a sigobject for this emulation, create one.
768 	 *
769 	 * sigobject is an anonymous memory object (just like SYSV shared
770 	 * memory) that we keep a permanent reference to and that we map
771 	 * in all processes that need this sigcode. The creation is simple,
772 	 * we create an object, add a permanent reference to it, map it in
773 	 * kernel space, copy out the sigcode to it and unmap it.
774 	 * Then we map it with PROT_READ|PROT_EXEC into the process just
775 	 * the way sys_mmap would map it.
776 	 */
777 	if (e->e_sigobject == NULL) {
778 		vaddr_t va;
779 		int r;
780 
781 		e->e_sigobject = uao_create(sz, 0);
782 		uao_reference(e->e_sigobject);	/* permanent reference */
783 
784 		va = vm_map_min(kernel_map);	/* hint */
785 		if ((r = uvm_map(kernel_map, &va, round_page(sz), e->e_sigobject,
786 		    0, 0, UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
787 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
788 			printf("kernel mapping failed %d\n", r);
789 			uao_detach(e->e_sigobject);
790 			return (ENOMEM);
791 		}
792 		memcpy((void *)va, e->e_sigcode, sz);
793 		uvm_unmap(kernel_map, va, va + round_page(sz));
794 	}
795 
796 	/* Just a hint to uvm_mmap where to put it. */
797 	p->p_sigcode = uvm_map_hint(p, VM_PROT_READ|VM_PROT_EXECUTE);
798 	uao_reference(e->e_sigobject);
799 	if (uvm_map(&p->p_vmspace->vm_map, &p->p_sigcode, round_page(sz),
800 	    e->e_sigobject, 0, 0, UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX,
801 	    UVM_INH_SHARE, UVM_ADV_RANDOM, 0))) {
802 		printf("user mapping failed\n");
803 		uao_detach(e->e_sigobject);
804 		return (ENOMEM);
805 	}
806 
807 	return (0);
808 }
809