xref: /openbsd-src/sys/kern/kern_exec.c (revision daf88648c0e349d5c02e1504293082072c981640)
1 /*	$OpenBSD: kern_exec.c,v 1.100 2006/11/15 17:25:40 jmc 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 = STACKGAP_RANDOM;
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(struct proc *p, struct exec_package *epp)
113 {
114 	int error, i;
115 	struct vnode *vp;
116 	struct nameidata *ndp;
117 	size_t resid;
118 
119 	ndp = epp->ep_ndp;
120 	ndp->ni_cnd.cn_nameiop = LOOKUP;
121 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
122 	/* first get the vnode */
123 	if ((error = namei(ndp)) != 0)
124 		return (error);
125 	epp->ep_vp = vp = ndp->ni_vp;
126 
127 	/* check for regular file */
128 	if (vp->v_type == VDIR) {
129 		error = EISDIR;
130 		goto bad1;
131 	}
132 	if (vp->v_type != VREG) {
133 		error = EACCES;
134 		goto bad1;
135 	}
136 
137 	/* get attributes */
138 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
139 		goto bad1;
140 
141 	/* Check mount point */
142 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
143 		error = EACCES;
144 		goto bad1;
145 	}
146 
147 	if ((vp->v_mount->mnt_flag & MNT_NOSUID))
148 		epp->ep_vap->va_mode &= ~(VSUID | VSGID);
149 
150 	/* check access.  for root we have to see if any exec bit on */
151 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
152 		goto bad1;
153 	if ((epp->ep_vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
154 		error = EACCES;
155 		goto bad1;
156 	}
157 
158 	/* try to open it */
159 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
160 		goto bad1;
161 
162 	/* unlock vp, we need it unlocked from here */
163 	VOP_UNLOCK(vp, 0, p);
164 
165 	/* now we have the file, get the exec header */
166 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
167 	    UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
168 	if (error)
169 		goto bad2;
170 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
171 
172 	/*
173 	 * set up the vmcmds for creation of the process
174 	 * address space
175 	 */
176 	error = ENOEXEC;
177 	for (i = 0; i < nexecs && error != 0; i++) {
178 		int newerror;
179 
180 		if (execsw[i].es_check == NULL)
181 			continue;
182 		newerror = (*execsw[i].es_check)(p, epp);
183 		if (!newerror && !(epp->ep_emul->e_flags & EMUL_ENABLED))
184 			newerror = EPERM;
185 		/* make sure the first "interesting" error code is saved. */
186 		if (!newerror || error == ENOEXEC)
187 			error = newerror;
188 		if (epp->ep_flags & EXEC_DESTR && error != 0)
189 			return (error);
190 	}
191 	if (!error) {
192 		/* check that entry point is sane */
193 		if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
194 			error = ENOEXEC;
195 		}
196 
197 		/* check limits */
198 		if ((epp->ep_tsize > MAXTSIZ) ||
199 		    (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur))
200 			error = ENOMEM;
201 
202 		if (!error)
203 			return (0);
204 	}
205 
206 	/*
207 	 * free any vmspace-creation commands,
208 	 * and release their references
209 	 */
210 	kill_vmcmds(&epp->ep_vmcmds);
211 
212 bad2:
213 	/*
214 	 * close the vnode, free the pathname buf, and punt.
215 	 */
216 	vn_close(vp, FREAD, p->p_ucred, p);
217 	pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf);
218 	return (error);
219 
220 bad1:
221 	/*
222 	 * free the namei pathname buffer, and put the vnode
223 	 * (which we don't yet have open).
224 	 */
225 	pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf);
226 	vput(vp);
227 	return (error);
228 }
229 
230 /*
231  * exec system call
232  */
233 /* ARGSUSED */
234 int
235 sys_execve(struct proc *p, void *v, register_t *retval)
236 {
237 	struct sys_execve_args /* {
238 		syscallarg(const char *) path;
239 		syscallarg(char *const *) argp;
240 		syscallarg(char *const *) envp;
241 	} */ *uap = v;
242 	int error;
243 	struct exec_package pack;
244 	struct nameidata nid;
245 	struct vattr attr;
246 	struct ucred *cred = p->p_ucred;
247 	char *argp;
248 	char * const *cpp, *dp, *sp;
249 	long argc, envc;
250 	size_t len, sgap;
251 #ifdef MACHINE_STACK_GROWS_UP
252 	size_t slen;
253 #endif
254 	char *stack;
255 	struct ps_strings arginfo;
256 	struct vmspace *vm = p->p_vmspace;
257 	char **tmpfap;
258 	extern struct emul emul_native;
259 #if NSYSTRACE > 0
260 	int wassugid =
261 	    ISSET(p->p_flag, P_SUGID) || ISSET(p->p_flag, P_SUGIDEXEC);
262 	size_t pathbuflen;
263 #endif
264 	char *pathbuf = NULL;
265 
266 	/*
267 	 * Cheap solution to complicated problems.
268 	 * Mark this process as "leave me alone, I'm execing".
269 	 */
270 	p->p_flag |= P_INEXEC;
271 
272 #if NSYSTRACE > 0
273 	if (ISSET(p->p_flag, P_SYSTRACE)) {
274 		systrace_execve0(p);
275 		pathbuf = pool_get(&namei_pool, PR_WAITOK);
276 		error = copyinstr(SCARG(uap, path), pathbuf, MAXPATHLEN,
277 		    &pathbuflen);
278 		if (error != 0)
279 			goto clrflag;
280 	}
281 #endif
282 	if (pathbuf != NULL) {
283 		NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_SYSSPACE, pathbuf, p);
284 	} else {
285 		NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE,
286 		    SCARG(uap, path), p);
287 	}
288 
289 	/*
290 	 * initialize the fields of the exec package.
291 	 */
292 	if (pathbuf != NULL)
293 		pack.ep_name = pathbuf;
294 	else
295 		pack.ep_name = (char *)SCARG(uap, path);
296 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
297 	pack.ep_hdrlen = exec_maxhdrsz;
298 	pack.ep_hdrvalid = 0;
299 	pack.ep_ndp = &nid;
300 	pack.ep_interp = NULL;
301 	pack.ep_emul_arg = NULL;
302 	VMCMDSET_INIT(&pack.ep_vmcmds);
303 	pack.ep_vap = &attr;
304 	pack.ep_emul = &emul_native;
305 	pack.ep_flags = 0;
306 
307 	/* see if we can run it. */
308 	if ((error = check_exec(p, &pack)) != 0) {
309 		goto freehdr;
310 	}
311 
312 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
313 
314 	/* allocate an argument buffer */
315 	argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS);
316 #ifdef DIAGNOSTIC
317 	if (argp == NULL)
318 		panic("execve: argp == NULL");
319 #endif
320 	dp = argp;
321 	argc = 0;
322 
323 	/* copy the fake args list, if there's one, freeing it as we go */
324 	if (pack.ep_flags & EXEC_HASARGL) {
325 		tmpfap = pack.ep_fa;
326 		while (*tmpfap != NULL) {
327 			char *cp;
328 
329 			cp = *tmpfap;
330 			while (*cp)
331 				*dp++ = *cp++;
332 			dp++;
333 
334 			free(*tmpfap, M_EXEC);
335 			tmpfap++; argc++;
336 		}
337 		FREE(pack.ep_fa, M_EXEC);
338 		pack.ep_flags &= ~EXEC_HASARGL;
339 	}
340 
341 	/* Now get argv & environment */
342 	if (!(cpp = SCARG(uap, argp))) {
343 		error = EFAULT;
344 		goto bad;
345 	}
346 
347 	if (pack.ep_flags & EXEC_SKIPARG)
348 		cpp++;
349 
350 	while (1) {
351 		len = argp + ARG_MAX - dp;
352 		if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
353 			goto bad;
354 		if (!sp)
355 			break;
356 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
357 			if (error == ENAMETOOLONG)
358 				error = E2BIG;
359 			goto bad;
360 		}
361 		dp += len;
362 		cpp++;
363 		argc++;
364 	}
365 
366 	envc = 0;
367 	/* environment does not need to be there */
368 	if ((cpp = SCARG(uap, envp)) != NULL ) {
369 		while (1) {
370 			len = argp + ARG_MAX - dp;
371 			if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
372 				goto bad;
373 			if (!sp)
374 				break;
375 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
376 				if (error == ENAMETOOLONG)
377 					error = E2BIG;
378 				goto bad;
379 			}
380 			dp += len;
381 			cpp++;
382 			envc++;
383 		}
384 	}
385 
386 	dp = (char *)ALIGN(dp);
387 
388 	sgap = STACKGAPLEN;
389 	if (stackgap_random != 0)
390 		sgap += (arc4random() * ALIGNBYTES) & (stackgap_random - 1);
391 	/* Now check if args & environ fit into new stack */
392 	len = ((argc + envc + 2 + pack.ep_emul->e_arglen) * sizeof(char *) +
393 	    sizeof(long) + dp + sgap + sizeof(struct ps_strings)) - argp;
394 
395 	len = ALIGN(len);	/* make the stack "safely" aligned */
396 
397 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
398 		error = ENOMEM;
399 		goto bad;
400 	}
401 
402 	/* adjust "active stack depth" for process VSZ */
403 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
404 
405 	/*
406 	 * Prepare vmspace for remapping. Note that uvmspace_exec can replace
407 	 * p_vmspace!
408 	 */
409 	uvmspace_exec(p, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
410 
411 	vm = p->p_vmspace;
412 	/* Now map address space */
413 	vm->vm_taddr = (char *)pack.ep_taddr;
414 	vm->vm_tsize = btoc(pack.ep_tsize);
415 	vm->vm_daddr = (char *)pack.ep_daddr;
416 	vm->vm_dsize = btoc(pack.ep_dsize);
417 	vm->vm_dused = 0;
418 	vm->vm_ssize = btoc(pack.ep_ssize);
419 	vm->vm_maxsaddr = (char *)pack.ep_maxsaddr;
420 	vm->vm_minsaddr = (char *)pack.ep_minsaddr;
421 
422 	/* create the new process's VM space by running the vmcmds */
423 #ifdef DIAGNOSTIC
424 	if (pack.ep_vmcmds.evs_used == 0)
425 		panic("execve: no vmcmds");
426 #endif
427 	error = exec_process_vmcmds(p, &pack);
428 
429 	/* if an error happened, deallocate and punt */
430 	if (error)
431 		goto exec_abort;
432 
433 	/* remember information about the process */
434 	arginfo.ps_nargvstr = argc;
435 	arginfo.ps_nenvstr = envc;
436 
437 #ifdef MACHINE_STACK_GROWS_UP
438 	stack = (char *)USRSTACK + sizeof(arginfo);
439 	slen = len - sizeof(arginfo);
440 #else
441 	stack = (char *)(USRSTACK - len);
442 #endif
443 	/* Now copy argc, args & environ to new stack */
444 	if (!(*pack.ep_emul->e_copyargs)(&pack, &arginfo, stack, argp))
445 		goto exec_abort;
446 
447 	/* copy out the process's ps_strings structure */
448 	if (copyout(&arginfo, (char *)PS_STRINGS, sizeof(arginfo)))
449 		goto exec_abort;
450 
451 	stopprofclock(p);	/* stop profiling */
452 	fdcloseexec(p);		/* handle close on exec */
453 	execsigs(p);		/* reset caught signals */
454 
455 	/* set command name & other accounting info */
456 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
457 	bcopy(nid.ni_cnd.cn_nameptr, p->p_comm, len);
458 	p->p_comm[len] = 0;
459 	p->p_acflag &= ~AFORK;
460 
461 	/* record proc's vnode, for use by procfs and others */
462 	if (p->p_textvp)
463 		vrele(p->p_textvp);
464 	VREF(pack.ep_vp);
465 	p->p_textvp = pack.ep_vp;
466 
467 	p->p_flag |= P_EXEC;
468 	if (p->p_flag & P_PPWAIT) {
469 		p->p_flag &= ~P_PPWAIT;
470 		wakeup((caddr_t)p->p_pptr);
471 	}
472 
473 	/*
474 	 * If process does execve() while it has a mismatched real,
475 	 * effective, or saved uid/gid, we set P_SUGIDEXEC.
476 	 */
477 	if (p->p_ucred->cr_uid != p->p_cred->p_ruid ||
478 	    p->p_ucred->cr_uid != p->p_cred->p_svuid ||
479 	    p->p_ucred->cr_gid != p->p_cred->p_rgid ||
480 	    p->p_ucred->cr_gid != p->p_cred->p_svgid)
481 		p->p_flag |= P_SUGIDEXEC;
482 	else
483 		p->p_flag &= ~P_SUGIDEXEC;
484 
485 	/*
486 	 * deal with set[ug]id.
487 	 * MNT_NOEXEC has already been used to disable s[ug]id.
488 	 */
489 	if ((attr.va_mode & (VSUID | VSGID)) && proc_cansugid(p)) {
490 		int i;
491 
492 		p->p_flag |= P_SUGID;
493 		p->p_flag |= P_SUGIDEXEC;
494 
495 #ifdef KTRACE
496 		/*
497 		 * If process is being ktraced, turn off - unless
498 		 * root set it.
499 		 */
500 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) {
501 			p->p_traceflag = 0;
502 			ktrsettracevnode(p, NULL);
503 		}
504 #endif
505 		p->p_ucred = crcopy(cred);
506 		if (attr.va_mode & VSUID)
507 			p->p_ucred->cr_uid = attr.va_uid;
508 		if (attr.va_mode & VSGID)
509 			p->p_ucred->cr_gid = attr.va_gid;
510 
511 		/*
512 		 * For set[ug]id processes, a few caveats apply to
513 		 * stdin, stdout, and stderr.
514 		 */
515 		for (i = 0; i < 3; i++) {
516 			struct file *fp = NULL;
517 
518 			/*
519 			 * NOTE - This will never return NULL because of
520 			 * immature fds. The file descriptor table is not
521 			 * shared because we're suid.
522 			 */
523 			fp = fd_getfile(p->p_fd, i);
524 #ifdef PROCFS
525 			/*
526 			 * Close descriptors that are writing to procfs.
527 			 */
528 			if (fp && fp->f_type == DTYPE_VNODE &&
529 			    ((struct vnode *)(fp->f_data))->v_tag == VT_PROCFS &&
530 			    (fp->f_flag & FWRITE)) {
531 				fdrelease(p, i);
532 				fp = NULL;
533 			}
534 #endif
535 
536 			/*
537 			 * Ensure that stdin, stdout, and stderr are already
538 			 * allocated.  We do not want userland to accidentally
539 			 * allocate descriptors in this range which has implied
540 			 * meaning to libc.
541 			 */
542 			if (fp == NULL) {
543 				short flags = FREAD | (i == 0 ? 0 : FWRITE);
544 				struct vnode *vp;
545 				int indx;
546 
547 				if ((error = falloc(p, &fp, &indx)) != 0)
548 					goto exec_abort;
549 #ifdef DIAGNOSTIC
550 				if (indx != i)
551 					panic("sys_execve: falloc indx != i");
552 #endif
553 				if ((error = cdevvp(getnulldev(), &vp)) != 0) {
554 					fdremove(p->p_fd, indx);
555 					closef(fp, p);
556 					goto exec_abort;
557 				}
558 				if ((error = VOP_OPEN(vp, flags, p->p_ucred, p)) != 0) {
559 					fdremove(p->p_fd, indx);
560 					closef(fp, p);
561 					vrele(vp);
562 					goto exec_abort;
563 				}
564 				if (flags & FWRITE)
565 					vp->v_writecount++;
566 				fp->f_flag = flags;
567 				fp->f_type = DTYPE_VNODE;
568 				fp->f_ops = &vnops;
569 				fp->f_data = (caddr_t)vp;
570 				FILE_SET_MATURE(fp);
571 			}
572 		}
573 	} else
574 		p->p_flag &= ~P_SUGID;
575 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
576 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
577 
578 	if (p->p_flag & P_SUGIDEXEC) {
579 		int i, s = splclock();
580 
581 		timeout_del(&p->p_realit_to);
582 		timerclear(&p->p_realtimer.it_interval);
583 		timerclear(&p->p_realtimer.it_value);
584 		for (i = 0; i < sizeof(p->p_stats->p_timer) /
585 		    sizeof(p->p_stats->p_timer[0]); i++) {
586 			timerclear(&p->p_stats->p_timer[i].it_interval);
587 			timerclear(&p->p_stats->p_timer[i].it_value);
588 		}
589 		splx(s);
590 	}
591 
592 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
593 
594 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
595 	vn_close(pack.ep_vp, FREAD, cred, p);
596 
597 	/*
598 	 * notify others that we exec'd
599 	 */
600 	KNOTE(&p->p_klist, NOTE_EXEC);
601 
602 	/* setup new registers and do misc. setup. */
603 	if (pack.ep_emul->e_fixup != NULL) {
604 		if ((*pack.ep_emul->e_fixup)(p, &pack) != 0)
605 			goto free_pack_abort;
606 	}
607 #ifdef MACHINE_STACK_GROWS_UP
608 	(*pack.ep_emul->e_setregs)(p, &pack, (u_long)stack + slen, retval);
609 #else
610 	(*pack.ep_emul->e_setregs)(p, &pack, (u_long)stack, retval);
611 #endif
612 
613 	/* map the process's signal trampoline code */
614 	if (exec_sigcode_map(p, pack.ep_emul))
615 		goto free_pack_abort;
616 
617 	if (p->p_flag & P_TRACED)
618 		psignal(p, SIGTRAP);
619 
620 	free(pack.ep_hdr, M_EXEC);
621 
622 	/*
623 	 * Call emulation specific exec hook. This can setup per-process
624 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
625 	 *
626 	 * If we are executing process of different emulation than the
627 	 * original forked process, call e_proc_exit() of the old emulation
628 	 * first, then e_proc_exec() of new emulation. If the emulation is
629 	 * same, the exec hook code should deallocate any old emulation
630 	 * resources held previously by this process.
631 	 */
632 	if (p->p_emul && p->p_emul->e_proc_exit &&
633 	    p->p_emul != pack.ep_emul)
634 		(*p->p_emul->e_proc_exit)(p);
635 
636 	p->p_descfd = 255;
637 	if ((pack.ep_flags & EXEC_HASFD) && pack.ep_fd < 255)
638 		p->p_descfd = pack.ep_fd;
639 
640 	/*
641 	 * Call exec hook. Emulation code may NOT store reference to anything
642 	 * from &pack.
643 	 */
644 	if (pack.ep_emul->e_proc_exec)
645 		(*pack.ep_emul->e_proc_exec)(p, &pack);
646 
647 	/* update p_emul, the old value is no longer needed */
648 	p->p_emul = pack.ep_emul;
649 
650 #ifdef KTRACE
651 	if (KTRPOINT(p, KTR_EMUL))
652 		ktremul(p, p->p_emul->e_name);
653 #endif
654 
655 	p->p_flag &= ~P_INEXEC;
656 
657 #if NSYSTRACE > 0
658 	if (ISSET(p->p_flag, P_SYSTRACE) &&
659 	    wassugid && !ISSET(p->p_flag, P_SUGID) &&
660 	    !ISSET(p->p_flag, P_SUGIDEXEC))
661 		systrace_execve1(pathbuf, p);
662 #endif
663 
664 	if (pathbuf != NULL)
665 		pool_put(&namei_pool, pathbuf);
666 
667 	return (0);
668 
669 bad:
670 	/* free the vmspace-creation commands, and release their references */
671 	kill_vmcmds(&pack.ep_vmcmds);
672 	/* kill any opened file descriptor, if necessary */
673 	if (pack.ep_flags & EXEC_HASFD) {
674 		pack.ep_flags &= ~EXEC_HASFD;
675 		(void) fdrelease(p, pack.ep_fd);
676 	}
677 	if (pack.ep_interp != NULL)
678 		pool_put(&namei_pool, pack.ep_interp);
679 	if (pack.ep_emul_arg != NULL)
680 		FREE(pack.ep_emul_arg, M_TEMP);
681 	/* close and put the exec'd file */
682 	vn_close(pack.ep_vp, FREAD, cred, p);
683 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
684 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
685 
686  freehdr:
687 	free(pack.ep_hdr, M_EXEC);
688 #if NSYSTRACE > 0
689  clrflag:
690 #endif
691 	p->p_flag &= ~P_INEXEC;
692 
693 	if (pathbuf != NULL)
694 		pool_put(&namei_pool, pathbuf);
695 
696 	return (error);
697 
698 exec_abort:
699 	/*
700 	 * the old process doesn't exist anymore.  exit gracefully.
701 	 * get rid of the (new) address space we have created, if any, get rid
702 	 * of our namei data and vnode, and exit noting failure
703 	 */
704 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
705 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
706 	if (pack.ep_interp != NULL)
707 		pool_put(&namei_pool, pack.ep_interp);
708 	if (pack.ep_emul_arg != NULL)
709 		FREE(pack.ep_emul_arg, M_TEMP);
710 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
711 	vn_close(pack.ep_vp, FREAD, cred, p);
712 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
713 
714 free_pack_abort:
715 	free(pack.ep_hdr, M_EXEC);
716 	exit1(p, W_EXITCODE(0, SIGABRT), EXIT_NORMAL);
717 
718 	/* NOTREACHED */
719 	p->p_flag &= ~P_INEXEC;
720 	if (pathbuf != NULL)
721 		pool_put(&namei_pool, pathbuf);
722 
723 	return (0);
724 }
725 
726 
727 void *
728 copyargs(struct exec_package *pack, struct ps_strings *arginfo, void *stack,
729     void *argp)
730 {
731 	char **cpp = stack;
732 	char *dp, *sp;
733 	size_t len;
734 	void *nullp = NULL;
735 	long argc = arginfo->ps_nargvstr;
736 	int envc = arginfo->ps_nenvstr;
737 
738 	if (copyout(&argc, cpp++, sizeof(argc)))
739 		return (NULL);
740 
741 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_emul->e_arglen);
742 	sp = argp;
743 
744 	/* XXX don't copy them out, remap them! */
745 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
746 
747 	for (; --argc >= 0; sp += len, dp += len)
748 		if (copyout(&dp, cpp++, sizeof(dp)) ||
749 		    copyoutstr(sp, dp, ARG_MAX, &len))
750 			return (NULL);
751 
752 	if (copyout(&nullp, cpp++, sizeof(nullp)))
753 		return (NULL);
754 
755 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
756 
757 	for (; --envc >= 0; sp += len, dp += len)
758 		if (copyout(&dp, cpp++, sizeof(dp)) ||
759 		    copyoutstr(sp, dp, ARG_MAX, &len))
760 			return (NULL);
761 
762 	if (copyout(&nullp, cpp++, sizeof(nullp)))
763 		return (NULL);
764 
765 	return (cpp);
766 }
767 
768 int
769 exec_sigcode_map(struct proc *p, struct emul *e)
770 {
771 	vsize_t sz;
772 
773 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
774 
775 	/*
776 	 * If we don't have a sigobject for this emulation, create one.
777 	 *
778 	 * sigobject is an anonymous memory object (just like SYSV shared
779 	 * memory) that we keep a permanent reference to and that we map
780 	 * in all processes that need this sigcode. The creation is simple,
781 	 * we create an object, add a permanent reference to it, map it in
782 	 * kernel space, copy out the sigcode to it and unmap it.
783 	 * Then we map it with PROT_READ|PROT_EXEC into the process just
784 	 * the way sys_mmap would map it.
785 	 */
786 	if (e->e_sigobject == NULL) {
787 		vaddr_t va;
788 		int r;
789 
790 		e->e_sigobject = uao_create(sz, 0);
791 		uao_reference(e->e_sigobject);	/* permanent reference */
792 
793 		va = vm_map_min(kernel_map);	/* hint */
794 		if ((r = uvm_map(kernel_map, &va, round_page(sz), e->e_sigobject,
795 		    0, 0, UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
796 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
797 			printf("kernel mapping failed %d\n", r);
798 			uao_detach(e->e_sigobject);
799 			return (ENOMEM);
800 		}
801 		memcpy((void *)va, e->e_sigcode, sz);
802 		uvm_unmap(kernel_map, va, va + round_page(sz));
803 	}
804 
805 	/* Just a hint to uvm_mmap where to put it. */
806 	p->p_sigcode = uvm_map_hint(p, VM_PROT_READ|VM_PROT_EXECUTE);
807 	uao_reference(e->e_sigobject);
808 	if (uvm_map(&p->p_vmspace->vm_map, &p->p_sigcode, round_page(sz),
809 	    e->e_sigobject, 0, 0, UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX,
810 	    UVM_INH_SHARE, UVM_ADV_RANDOM, 0))) {
811 		printf("user mapping failed\n");
812 		uao_detach(e->e_sigobject);
813 		return (ENOMEM);
814 	}
815 
816 	return (0);
817 }
818