xref: /netbsd-src/sys/kern/kern_exec.c (revision 28c37e673e4d9b6cbdc7483062b915cc61d1ccf5)
1 /*	$NetBSD: kern_exec.c,v 1.157 2002/09/27 15:37:43 provos Exp $	*/
2 
3 /*-
4  * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
5  * Copyright (C) 1992 Wolfgang Solfrank.
6  * Copyright (C) 1992 TooLs GmbH.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by TooLs GmbH.
20  * 4. The name of TooLs GmbH may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
29  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
30  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
31  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
32  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.157 2002/09/27 15:37:43 provos Exp $");
37 
38 #include "opt_ktrace.h"
39 #include "opt_syscall_debug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/filedesc.h>
44 #include <sys/kernel.h>
45 #include <sys/proc.h>
46 #include <sys/mount.h>
47 #include <sys/malloc.h>
48 #include <sys/namei.h>
49 #include <sys/vnode.h>
50 #include <sys/file.h>
51 #include <sys/acct.h>
52 #include <sys/exec.h>
53 #include <sys/ktrace.h>
54 #include <sys/resourcevar.h>
55 #include <sys/wait.h>
56 #include <sys/mman.h>
57 #include <sys/ras.h>
58 #include <sys/signalvar.h>
59 #include <sys/stat.h>
60 #include <sys/syscall.h>
61 
62 #include <sys/syscallargs.h>
63 
64 #include <uvm/uvm_extern.h>
65 
66 #include <machine/cpu.h>
67 #include <machine/reg.h>
68 
69 #ifdef DEBUG_EXEC
70 #define DPRINTF(a) uprintf a
71 #else
72 #define DPRINTF(a)
73 #endif /* DEBUG_EXEC */
74 
75 /*
76  * Exec function switch:
77  *
78  * Note that each makecmds function is responsible for loading the
79  * exec package with the necessary functions for any exec-type-specific
80  * handling.
81  *
82  * Functions for specific exec types should be defined in their own
83  * header file.
84  */
85 extern const struct execsw	execsw_builtin[];
86 extern int			nexecs_builtin;
87 static const struct execsw	**execsw = NULL;
88 static int			nexecs;
89 
90 u_int	exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
91 
92 #ifdef LKM
93 /* list of supported emulations */
94 static
95 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
96 struct emul_entry {
97 	LIST_ENTRY(emul_entry)	el_list;
98 	const struct emul	*el_emul;
99 	int			ro_entry;
100 };
101 
102 /* list of dynamically loaded execsw entries */
103 static
104 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
105 struct exec_entry {
106 	LIST_ENTRY(exec_entry)	ex_list;
107 	const struct execsw	*es;
108 };
109 
110 /* structure used for building execw[] */
111 struct execsw_entry {
112 	struct execsw_entry	*next;
113 	const struct execsw	*es;
114 };
115 #endif /* LKM */
116 
117 /* NetBSD emul struct */
118 extern char	sigcode[], esigcode[];
119 #ifdef SYSCALL_DEBUG
120 extern const char * const syscallnames[];
121 #endif
122 #ifdef __HAVE_SYSCALL_INTERN
123 void syscall_intern(struct proc *);
124 #else
125 void syscall(void);
126 #endif
127 
128 const struct emul emul_netbsd = {
129 	"netbsd",
130 	NULL,		/* emulation path */
131 #ifndef __HAVE_MINIMAL_EMUL
132 	EMUL_HAS_SYS___syscall,
133 	NULL,
134 	SYS_syscall,
135 	SYS_MAXSYSCALL,
136 #endif
137 	sysent,
138 #ifdef SYSCALL_DEBUG
139 	syscallnames,
140 #else
141 	NULL,
142 #endif
143 	sendsig,
144 	trapsignal,
145 	sigcode,
146 	esigcode,
147 	setregs,
148 	NULL,
149 	NULL,
150 	NULL,
151 #ifdef __HAVE_SYSCALL_INTERN
152 	syscall_intern,
153 #else
154 	syscall,
155 #endif
156 	NULL,
157 	NULL,
158 };
159 
160 #ifdef LKM
161 /*
162  * Exec lock. Used to control access to execsw[] structures.
163  * This must not be static so that netbsd32 can access it, too.
164  */
165 struct lock exec_lock;
166 
167 static void link_es(struct execsw_entry **, const struct execsw *);
168 #endif /* LKM */
169 
170 /*
171  * check exec:
172  * given an "executable" described in the exec package's namei info,
173  * see what we can do with it.
174  *
175  * ON ENTRY:
176  *	exec package with appropriate namei info
177  *	proc pointer of exec'ing proc
178  *	NO SELF-LOCKED VNODES
179  *
180  * ON EXIT:
181  *	error:	nothing held, etc.  exec header still allocated.
182  *	ok:	filled exec package, executable's vnode (unlocked).
183  *
184  * EXEC SWITCH ENTRY:
185  * 	Locked vnode to check, exec package, proc.
186  *
187  * EXEC SWITCH EXIT:
188  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
189  *	error:	destructive:
190  *			everything deallocated execept exec header.
191  *		non-destructive:
192  *			error code, executable's vnode (unlocked),
193  *			exec header unmodified.
194  */
195 int
196 check_exec(struct proc *p, struct exec_package *epp)
197 {
198 	int		error, i;
199 	struct vnode	*vp;
200 	struct nameidata *ndp;
201 	size_t		resid;
202 
203 	ndp = epp->ep_ndp;
204 	ndp->ni_cnd.cn_nameiop = LOOKUP;
205 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
206 	/* first get the vnode */
207 	if ((error = namei(ndp)) != 0)
208 		return error;
209 	epp->ep_vp = vp = ndp->ni_vp;
210 
211 	/* check access and type */
212 	if (vp->v_type != VREG) {
213 		error = EACCES;
214 		goto bad1;
215 	}
216 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
217 		goto bad1;
218 
219 	/* get attributes */
220 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
221 		goto bad1;
222 
223 	/* Check mount point */
224 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
225 		error = EACCES;
226 		goto bad1;
227 	}
228 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
229 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
230 
231 	/* try to open it */
232 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
233 		goto bad1;
234 
235 	/* unlock vp, since we need it unlocked from here on out. */
236 	VOP_UNLOCK(vp, 0);
237 
238 	/* now we have the file, get the exec header */
239 	uvn_attach(vp, VM_PROT_READ);
240 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
241 			UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
242 	if (error)
243 		goto bad2;
244 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
245 
246 	/*
247 	 * Set up default address space limits.  Can be overridden
248 	 * by individual exec packages.
249 	 *
250 	 * XXX probably shoul be all done in the exec pakages.
251 	 */
252 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
253 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
254 	/*
255 	 * set up the vmcmds for creation of the process
256 	 * address space
257 	 */
258 	error = ENOEXEC;
259 	for (i = 0; i < nexecs && error != 0; i++) {
260 		int newerror;
261 
262 		epp->ep_esch = execsw[i];
263 		newerror = (*execsw[i]->es_check)(p, epp);
264 		/* make sure the first "interesting" error code is saved. */
265 		if (!newerror || error == ENOEXEC)
266 			error = newerror;
267 
268 		/* if es_check call was successful, update epp->ep_es */
269 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
270 			epp->ep_es = execsw[i];
271 
272 		if (epp->ep_flags & EXEC_DESTR && error != 0)
273 			return error;
274 	}
275 	if (!error) {
276 		/* check that entry point is sane */
277 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
278 			error = ENOEXEC;
279 
280 		/* check limits */
281 		if ((epp->ep_tsize > MAXTSIZ) ||
282 		    (epp->ep_dsize >
283 		     (u_quad_t)p->p_rlimit[RLIMIT_DATA].rlim_cur))
284 			error = ENOMEM;
285 
286 		if (!error)
287 			return (0);
288 	}
289 
290 	/*
291 	 * free any vmspace-creation commands,
292 	 * and release their references
293 	 */
294 	kill_vmcmds(&epp->ep_vmcmds);
295 
296 bad2:
297 	/*
298 	 * close and release the vnode, restore the old one, free the
299 	 * pathname buf, and punt.
300 	 */
301 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
302 	VOP_CLOSE(vp, FREAD, p->p_ucred, p);
303 	vput(vp);
304 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
305 	return error;
306 
307 bad1:
308 	/*
309 	 * free the namei pathname buffer, and put the vnode
310 	 * (which we don't yet have open).
311 	 */
312 	vput(vp);				/* was still locked */
313 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
314 	return error;
315 }
316 
317 /*
318  * exec system call
319  */
320 /* ARGSUSED */
321 int
322 sys_execve(struct proc *p, void *v, register_t *retval)
323 {
324 	struct sys_execve_args /* {
325 		syscallarg(const char *)	path;
326 		syscallarg(char * const *)	argp;
327 		syscallarg(char * const *)	envp;
328 	} */ *uap = v;
329 	int			error;
330 	u_int			i;
331 	struct exec_package	pack;
332 	struct nameidata	nid;
333 	struct vattr		attr;
334 	struct ucred		*cred;
335 	char			*argp;
336 	char * const		*cpp;
337 	char			*dp, *sp;
338 	long			argc, envc;
339 	size_t			len;
340 	char			*stack;
341 	struct ps_strings	arginfo;
342 	struct vmspace		*vm;
343 	char			**tmpfap;
344 	int			szsigcode;
345 	struct exec_vmcmd	*base_vcp;
346 
347 	/*
348 	 * Lock the process and set the P_INEXEC flag to indicate that
349 	 * it should be left alone until we're done here.  This is
350 	 * necessary to avoid race conditions - e.g. in ptrace() -
351 	 * that might allow a local user to illicitly obtain elevated
352 	 * privileges.
353 	 */
354 	p->p_flag |= P_INEXEC;
355 
356 	cred = p->p_ucred;
357 	base_vcp = NULL;
358 	/*
359 	 * Init the namei data to point the file user's program name.
360 	 * This is done here rather than in check_exec(), so that it's
361 	 * possible to override this settings if any of makecmd/probe
362 	 * functions call check_exec() recursively - for example,
363 	 * see exec_script_makecmds().
364 	 */
365 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
366 
367 	/*
368 	 * initialize the fields of the exec package.
369 	 */
370 	pack.ep_name = SCARG(uap, path);
371 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
372 	pack.ep_hdrlen = exec_maxhdrsz;
373 	pack.ep_hdrvalid = 0;
374 	pack.ep_ndp = &nid;
375 	pack.ep_emul_arg = NULL;
376 	pack.ep_vmcmds.evs_cnt = 0;
377 	pack.ep_vmcmds.evs_used = 0;
378 	pack.ep_vap = &attr;
379 	pack.ep_flags = 0;
380 
381 #ifdef LKM
382 	lockmgr(&exec_lock, LK_SHARED, NULL);
383 #endif
384 
385 	/* see if we can run it. */
386 	if ((error = check_exec(p, &pack)) != 0)
387 		goto freehdr;
388 
389 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
390 
391 	/* allocate an argument buffer */
392 	argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS);
393 #ifdef DIAGNOSTIC
394 	if (argp == (vaddr_t) 0)
395 		panic("execve: argp == NULL");
396 #endif
397 	dp = argp;
398 	argc = 0;
399 
400 	/* copy the fake args list, if there's one, freeing it as we go */
401 	if (pack.ep_flags & EXEC_HASARGL) {
402 		tmpfap = pack.ep_fa;
403 		while (*tmpfap != NULL) {
404 			char *cp;
405 
406 			cp = *tmpfap;
407 			while (*cp)
408 				*dp++ = *cp++;
409 			dp++;
410 
411 			FREE(*tmpfap, M_EXEC);
412 			tmpfap++; argc++;
413 		}
414 		FREE(pack.ep_fa, M_EXEC);
415 		pack.ep_flags &= ~EXEC_HASARGL;
416 	}
417 
418 	/* Now get argv & environment */
419 	if (!(cpp = SCARG(uap, argp))) {
420 		error = EINVAL;
421 		goto bad;
422 	}
423 
424 	if (pack.ep_flags & EXEC_SKIPARG)
425 		cpp++;
426 
427 	while (1) {
428 		len = argp + ARG_MAX - dp;
429 		if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
430 			goto bad;
431 		if (!sp)
432 			break;
433 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
434 			if (error == ENAMETOOLONG)
435 				error = E2BIG;
436 			goto bad;
437 		}
438 		dp += len;
439 		cpp++;
440 		argc++;
441 	}
442 
443 	envc = 0;
444 	/* environment need not be there */
445 	if ((cpp = SCARG(uap, envp)) != NULL ) {
446 		while (1) {
447 			len = argp + ARG_MAX - dp;
448 			if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
449 				goto bad;
450 			if (!sp)
451 				break;
452 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
453 				if (error == ENAMETOOLONG)
454 					error = E2BIG;
455 				goto bad;
456 			}
457 			dp += len;
458 			cpp++;
459 			envc++;
460 		}
461 	}
462 
463 	dp = (char *) ALIGN(dp);
464 
465 	szsigcode = pack.ep_es->es_emul->e_esigcode -
466 	    pack.ep_es->es_emul->e_sigcode;
467 
468 	/* Now check if args & environ fit into new stack */
469 	if (pack.ep_flags & EXEC_32)
470 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
471 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
472 		    szsigcode + sizeof(struct ps_strings)) - argp;
473 	else
474 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
475 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
476 		    szsigcode + sizeof(struct ps_strings)) - argp;
477 
478 	len = ALIGN(len);	/* make the stack "safely" aligned */
479 
480 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
481 		error = ENOMEM;
482 		goto bad;
483 	}
484 
485 	/* adjust "active stack depth" for process VSZ */
486 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
487 
488 	/*
489 	 * Do whatever is necessary to prepare the address space
490 	 * for remapping.  Note that this might replace the current
491 	 * vmspace with another!
492 	 */
493 	uvmspace_exec(p, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
494 
495 	/* Now map address space */
496 	vm = p->p_vmspace;
497 	vm->vm_taddr = (char *) pack.ep_taddr;
498 	vm->vm_tsize = btoc(pack.ep_tsize);
499 	vm->vm_daddr = (char *) pack.ep_daddr;
500 	vm->vm_dsize = btoc(pack.ep_dsize);
501 	vm->vm_ssize = btoc(pack.ep_ssize);
502 	vm->vm_maxsaddr = (char *) pack.ep_maxsaddr;
503 	vm->vm_minsaddr = (char *) pack.ep_minsaddr;
504 
505 	/* create the new process's VM space by running the vmcmds */
506 #ifdef DIAGNOSTIC
507 	if (pack.ep_vmcmds.evs_used == 0)
508 		panic("execve: no vmcmds");
509 #endif
510 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
511 		struct exec_vmcmd *vcp;
512 
513 		vcp = &pack.ep_vmcmds.evs_cmds[i];
514 		if (vcp->ev_flags & VMCMD_RELATIVE) {
515 #ifdef DIAGNOSTIC
516 			if (base_vcp == NULL)
517 				panic("execve: relative vmcmd with no base");
518 			if (vcp->ev_flags & VMCMD_BASE)
519 				panic("execve: illegal base & relative vmcmd");
520 #endif
521 			vcp->ev_addr += base_vcp->ev_addr;
522 		}
523 		error = (*vcp->ev_proc)(p, vcp);
524 #ifdef DEBUG_EXEC
525 		if (error) {
526 			int j;
527 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
528 			for (j = 0; j <= i; j++)
529 				uprintf(
530 			    "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
531 				    j, vp[j].ev_addr, vp[j].ev_len,
532 				    vp[j].ev_offset, vp[j].ev_prot,
533 				    vp[j].ev_flags);
534 		}
535 #endif /* DEBUG_EXEC */
536 		if (vcp->ev_flags & VMCMD_BASE)
537 			base_vcp = vcp;
538 	}
539 
540 	/* free the vmspace-creation commands, and release their references */
541 	kill_vmcmds(&pack.ep_vmcmds);
542 
543 	/* if an error happened, deallocate and punt */
544 	if (error) {
545 		DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error));
546 		goto exec_abort;
547 	}
548 
549 	/* remember information about the process */
550 	arginfo.ps_nargvstr = argc;
551 	arginfo.ps_nenvstr = envc;
552 
553 	stack = (char *) (vm->vm_minsaddr - len);
554 	/* Now copy argc, args & environ to new stack */
555 	error = (*pack.ep_es->es_copyargs)(p, &pack, &arginfo, &stack, argp);
556 	if (error) {
557 		DPRINTF(("execve: copyargs failed %d\n", error));
558 		goto exec_abort;
559 	}
560 	/* Move the stack back to original point */
561 	stack = (char *) (vm->vm_minsaddr - len);
562 
563 	/* fill process ps_strings info */
564 	p->p_psstr = (struct ps_strings *)(vm->vm_minsaddr
565 		- sizeof(struct ps_strings));
566 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
567 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
568 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
569 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
570 
571 	/* copy out the process's ps_strings structure */
572 	if ((error = copyout(&arginfo, (char *)p->p_psstr,
573 	    sizeof(arginfo))) != 0) {
574 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
575 		       &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo)));
576 		goto exec_abort;
577 	}
578 
579 	/* copy out the process's signal trapoline code */
580 	if (szsigcode) {
581 		if ((error = copyout((char *)pack.ep_es->es_emul->e_sigcode,
582 		    p->p_sigctx.ps_sigcode = (char *)p->p_psstr - szsigcode,
583 		    szsigcode)) != 0) {
584 			DPRINTF(("execve: sig trampoline copyout failed\n"));
585 			goto exec_abort;
586 		}
587 #ifdef PMAP_NEED_PROCWR
588 		/* This is code. Let the pmap do what is needed. */
589 		pmap_procwr(p, (vaddr_t)p->p_sigctx.ps_sigcode, szsigcode);
590 #endif
591 	}
592 
593 	stopprofclock(p);	/* stop profiling */
594 	fdcloseexec(p);		/* handle close on exec */
595 	execsigs(p);		/* reset catched signals */
596 	p->p_ctxlink = NULL;	/* reset ucontext link */
597 
598 	/* set command name & other accounting info */
599 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
600 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
601 	p->p_comm[len] = 0;
602 	p->p_acflag &= ~AFORK;
603 
604 	/* record proc's vnode, for use by procfs and others */
605         if (p->p_textvp)
606                 vrele(p->p_textvp);
607 	VREF(pack.ep_vp);
608 	p->p_textvp = pack.ep_vp;
609 
610 	p->p_flag |= P_EXEC;
611 	if (p->p_flag & P_PPWAIT) {
612 		p->p_flag &= ~P_PPWAIT;
613 		wakeup((caddr_t) p->p_pptr);
614 	}
615 
616 	/*
617 	 * deal with set[ug]id.
618 	 * MNT_NOSUID has already been used to disable s[ug]id.
619 	 */
620 	if ((p->p_flag & P_TRACED) == 0 &&
621 
622 	    (((attr.va_mode & S_ISUID) != 0 &&
623 	      p->p_ucred->cr_uid != attr.va_uid) ||
624 
625 	     ((attr.va_mode & S_ISGID) != 0 &&
626 	      p->p_ucred->cr_gid != attr.va_gid))) {
627 		/*
628 		 * Mark the process as SUGID before we do
629 		 * anything that might block.
630 		 */
631 		p_sugid(p);
632 
633 		/* Make sure file descriptors 0..2 are in use. */
634 		if ((error = fdcheckstd(p)) != 0)
635 			goto exec_abort;
636 
637 		p->p_ucred = crcopy(cred);
638 #ifdef KTRACE
639 		/*
640 		 * If process is being ktraced, turn off - unless
641 		 * root set it.
642 		 */
643 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
644 			ktrderef(p);
645 #endif
646 		if (attr.va_mode & S_ISUID)
647 			p->p_ucred->cr_uid = attr.va_uid;
648 		if (attr.va_mode & S_ISGID)
649 			p->p_ucred->cr_gid = attr.va_gid;
650 	} else
651 		p->p_flag &= ~P_SUGID;
652 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
653 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
654 
655 #if defined(__HAVE_RAS)
656 	/*
657 	 * Remove all RASs from the address space.
658 	 */
659 	ras_purgeall(p);
660 #endif
661 
662 	doexechooks(p);
663 
664 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
665 
666 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
667 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
668 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
669 	vput(pack.ep_vp);
670 
671 	/* setup new registers and do misc. setup. */
672 	(*pack.ep_es->es_emul->e_setregs)(p, &pack, (u_long) stack);
673 	if (pack.ep_es->es_setregs)
674 		(*pack.ep_es->es_setregs)(p, &pack, (u_long) stack);
675 
676 	if (p->p_flag & P_TRACED)
677 		psignal(p, SIGTRAP);
678 
679 	free(pack.ep_hdr, M_EXEC);
680 
681 	/*
682 	 * Call emulation specific exec hook. This can setup setup per-process
683 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
684 	 *
685 	 * If we are executing process of different emulation than the
686 	 * original forked process, call e_proc_exit() of the old emulation
687 	 * first, then e_proc_exec() of new emulation. If the emulation is
688 	 * same, the exec hook code should deallocate any old emulation
689 	 * resources held previously by this process.
690 	 */
691 	if (p->p_emul && p->p_emul->e_proc_exit
692 	    && p->p_emul != pack.ep_es->es_emul)
693 		(*p->p_emul->e_proc_exit)(p);
694 
695 	/*
696 	 * Call exec hook. Emulation code may NOT store reference to anything
697 	 * from &pack.
698 	 */
699         if (pack.ep_es->es_emul->e_proc_exec)
700                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
701 
702 	/* update p_emul, the old value is no longer needed */
703 	p->p_emul = pack.ep_es->es_emul;
704 
705 	/* ...and the same for p_execsw */
706 	p->p_execsw = pack.ep_es;
707 
708 #ifdef __HAVE_SYSCALL_INTERN
709 	(*p->p_emul->e_syscall_intern)(p);
710 #endif
711 #ifdef KTRACE
712 	if (KTRPOINT(p, KTR_EMUL))
713 		ktremul(p);
714 #endif
715 
716 #ifdef LKM
717 	lockmgr(&exec_lock, LK_RELEASE, NULL);
718 #endif
719 	p->p_flag &= ~P_INEXEC;
720 	return (EJUSTRETURN);
721 
722  bad:
723 	p->p_flag &= ~P_INEXEC;
724 	/* free the vmspace-creation commands, and release their references */
725 	kill_vmcmds(&pack.ep_vmcmds);
726 	/* kill any opened file descriptor, if necessary */
727 	if (pack.ep_flags & EXEC_HASFD) {
728 		pack.ep_flags &= ~EXEC_HASFD;
729 		(void) fdrelease(p, pack.ep_fd);
730 	}
731 	/* close and put the exec'd file */
732 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
733 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
734 	vput(pack.ep_vp);
735 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
736 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
737 
738  freehdr:
739 	p->p_flag &= ~P_INEXEC;
740 #ifdef LKM
741 	lockmgr(&exec_lock, LK_RELEASE, NULL);
742 #endif
743 
744 	free(pack.ep_hdr, M_EXEC);
745 	return error;
746 
747  exec_abort:
748 	p->p_flag &= ~P_INEXEC;
749 #ifdef LKM
750 	lockmgr(&exec_lock, LK_RELEASE, NULL);
751 #endif
752 
753 	/*
754 	 * the old process doesn't exist anymore.  exit gracefully.
755 	 * get rid of the (new) address space we have created, if any, get rid
756 	 * of our namei data and vnode, and exit noting failure
757 	 */
758 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
759 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
760 	if (pack.ep_emul_arg)
761 		FREE(pack.ep_emul_arg, M_TEMP);
762 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
763 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
764 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
765 	vput(pack.ep_vp);
766 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
767 	free(pack.ep_hdr, M_EXEC);
768 	exit1(p, W_EXITCODE(error, SIGABRT));
769 
770 	/* NOTREACHED */
771 	return 0;
772 }
773 
774 
775 int
776 copyargs(struct proc *p, struct exec_package *pack, struct ps_strings *arginfo,
777     char **stackp, void *argp)
778 {
779 	char	**cpp, *dp, *sp;
780 	size_t	len;
781 	void	*nullp;
782 	long	argc, envc;
783 	int	error;
784 
785 	cpp = (char **)*stackp;
786 	nullp = NULL;
787 	argc = arginfo->ps_nargvstr;
788 	envc = arginfo->ps_nenvstr;
789 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
790 		return error;
791 
792 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
793 	sp = argp;
794 
795 	/* XXX don't copy them out, remap them! */
796 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
797 
798 	for (; --argc >= 0; sp += len, dp += len)
799 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
800 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
801 			return error;
802 
803 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
804 		return error;
805 
806 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
807 
808 	for (; --envc >= 0; sp += len, dp += len)
809 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
810 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
811 			return error;
812 
813 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
814 		return error;
815 
816 	*stackp = (char *)cpp;
817 	return 0;
818 }
819 
820 #ifdef LKM
821 /*
822  * Find an emulation of given name in list of emulations.
823  * Needs to be called with the exec_lock held.
824  */
825 const struct emul *
826 emul_search(const char *name)
827 {
828 	struct emul_entry *it;
829 
830 	LIST_FOREACH(it, &el_head, el_list) {
831 		if (strcmp(name, it->el_emul->e_name) == 0)
832 			return it->el_emul;
833 	}
834 
835 	return NULL;
836 }
837 
838 /*
839  * Add an emulation to list, if it's not there already.
840  */
841 int
842 emul_register(const struct emul *emul, int ro_entry)
843 {
844 	struct emul_entry	*ee;
845 	int			error;
846 
847 	error = 0;
848 	lockmgr(&exec_lock, LK_SHARED, NULL);
849 
850 	if (emul_search(emul->e_name)) {
851 		error = EEXIST;
852 		goto out;
853 	}
854 
855 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
856 		M_EXEC, M_WAITOK);
857 	ee->el_emul = emul;
858 	ee->ro_entry = ro_entry;
859 	LIST_INSERT_HEAD(&el_head, ee, el_list);
860 
861  out:
862 	lockmgr(&exec_lock, LK_RELEASE, NULL);
863 	return error;
864 }
865 
866 /*
867  * Remove emulation with name 'name' from list of supported emulations.
868  */
869 int
870 emul_unregister(const char *name)
871 {
872 	const struct proclist_desc *pd;
873 	struct emul_entry	*it;
874 	int			i, error;
875 	struct proc		*ptmp;
876 
877 	error = 0;
878 	lockmgr(&exec_lock, LK_SHARED, NULL);
879 
880 	LIST_FOREACH(it, &el_head, el_list) {
881 		if (strcmp(it->el_emul->e_name, name) == 0)
882 			break;
883 	}
884 
885 	if (!it) {
886 		error = ENOENT;
887 		goto out;
888 	}
889 
890 	if (it->ro_entry) {
891 		error = EBUSY;
892 		goto out;
893 	}
894 
895 	/* test if any execw[] entry is still using this */
896 	for(i=0; i < nexecs; i++) {
897 		if (execsw[i]->es_emul == it->el_emul) {
898 			error = EBUSY;
899 			goto out;
900 		}
901 	}
902 
903 	/*
904 	 * Test if any process is running under this emulation - since
905 	 * emul_unregister() is running quite sendomly, it's better
906 	 * to do expensive check here than to use any locking.
907 	 */
908 	proclist_lock_read();
909 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
910 		LIST_FOREACH(ptmp, pd->pd_list, p_list) {
911 			if (ptmp->p_emul == it->el_emul) {
912 				error = EBUSY;
913 				break;
914 			}
915 		}
916 	}
917 	proclist_unlock_read();
918 
919 	if (error)
920 		goto out;
921 
922 
923 	/* entry is not used, remove it */
924 	LIST_REMOVE(it, el_list);
925 	FREE(it, M_EXEC);
926 
927  out:
928 	lockmgr(&exec_lock, LK_RELEASE, NULL);
929 	return error;
930 }
931 
932 /*
933  * Add execsw[] entry.
934  */
935 int
936 exec_add(struct execsw *esp, const char *e_name)
937 {
938 	struct exec_entry	*it;
939 	int			error;
940 
941 	error = 0;
942 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
943 
944 	if (!esp->es_emul) {
945 		esp->es_emul = emul_search(e_name);
946 		if (!esp->es_emul) {
947 			error = ENOENT;
948 			goto out;
949 		}
950 	}
951 
952 	LIST_FOREACH(it, &ex_head, ex_list) {
953 		/* assume tuple (makecmds, probe_func, emulation) is unique */
954 		if (it->es->es_check == esp->es_check
955 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
956 		    && it->es->es_emul == esp->es_emul) {
957 			error = EEXIST;
958 			goto out;
959 		}
960 	}
961 
962 	/* if we got here, the entry doesn't exist yet */
963 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
964 		M_EXEC, M_WAITOK);
965 	it->es = esp;
966 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
967 
968 	/* update execsw[] */
969 	exec_init(0);
970 
971  out:
972 	lockmgr(&exec_lock, LK_RELEASE, NULL);
973 	return error;
974 }
975 
976 /*
977  * Remove execsw[] entry.
978  */
979 int
980 exec_remove(const struct execsw *esp)
981 {
982 	struct exec_entry	*it;
983 	int			error;
984 
985 	error = 0;
986 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
987 
988 	LIST_FOREACH(it, &ex_head, ex_list) {
989 		/* assume tuple (makecmds, probe_func, emulation) is unique */
990 		if (it->es->es_check == esp->es_check
991 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
992 		    && it->es->es_emul == esp->es_emul)
993 			break;
994 	}
995 	if (!it) {
996 		error = ENOENT;
997 		goto out;
998 	}
999 
1000 	/* remove item from list and free resources */
1001 	LIST_REMOVE(it, ex_list);
1002 	FREE(it, M_EXEC);
1003 
1004 	/* update execsw[] */
1005 	exec_init(0);
1006 
1007  out:
1008 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1009 	return error;
1010 }
1011 
1012 static void
1013 link_es(struct execsw_entry **listp, const struct execsw *esp)
1014 {
1015 	struct execsw_entry *et, *e1;
1016 
1017 	MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry),
1018 			M_TEMP, M_WAITOK);
1019 	et->next = NULL;
1020 	et->es = esp;
1021 	if (*listp == NULL) {
1022 		*listp = et;
1023 		return;
1024 	}
1025 
1026 	switch(et->es->es_prio) {
1027 	case EXECSW_PRIO_FIRST:
1028 		/* put new entry as the first */
1029 		et->next = *listp;
1030 		*listp = et;
1031 		break;
1032 	case EXECSW_PRIO_ANY:
1033 		/* put new entry after all *_FIRST and *_ANY entries */
1034 		for(e1 = *listp; e1->next
1035 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
1036 			e1 = e1->next);
1037 		et->next = e1->next;
1038 		e1->next = et;
1039 		break;
1040 	case EXECSW_PRIO_LAST:
1041 		/* put new entry as the last one */
1042 		for(e1 = *listp; e1->next; e1 = e1->next);
1043 		e1->next = et;
1044 		break;
1045 	default:
1046 #ifdef DIAGNOSTIC
1047 		panic("execw[] entry with unknown priority %d found",
1048 			et->es->es_prio);
1049 #endif
1050 		break;
1051 	}
1052 }
1053 
1054 /*
1055  * Initialize exec structures. If init_boot is true, also does necessary
1056  * one-time initialization (it's called from main() that way).
1057  * Once system is multiuser, this should be called with exec_lock held,
1058  * i.e. via exec_{add|remove}().
1059  */
1060 int
1061 exec_init(int init_boot)
1062 {
1063 	const struct execsw	**new_es, * const *old_es;
1064 	struct execsw_entry	*list, *e1;
1065 	struct exec_entry	*e2;
1066 	int			i, es_sz;
1067 
1068 	if (init_boot) {
1069 		/* do one-time initializations */
1070 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1071 
1072 		/* register compiled-in emulations */
1073 		for(i=0; i < nexecs_builtin; i++) {
1074 			if (execsw_builtin[i].es_emul)
1075 				emul_register(execsw_builtin[i].es_emul, 1);
1076 		}
1077 #ifdef DIAGNOSTIC
1078 		if (i == 0)
1079 			panic("no emulations found in execsw_builtin[]");
1080 #endif
1081 	}
1082 
1083 	/*
1084 	 * Build execsw[] array from builtin entries and entries added
1085 	 * at runtime.
1086 	 */
1087 	list = NULL;
1088 	for(i=0; i < nexecs_builtin; i++)
1089 		link_es(&list, &execsw_builtin[i]);
1090 
1091 	/* Add dynamically loaded entries */
1092 	es_sz = nexecs_builtin;
1093 	LIST_FOREACH(e2, &ex_head, ex_list) {
1094 		link_es(&list, e2->es);
1095 		es_sz++;
1096 	}
1097 
1098 	/*
1099 	 * Now that we have sorted all execw entries, create new execsw[]
1100 	 * and free no longer needed memory in the process.
1101 	 */
1102 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
1103 	for(i=0; list; i++) {
1104 		new_es[i] = list->es;
1105 		e1 = list->next;
1106 		FREE(list, M_TEMP);
1107 		list = e1;
1108 	}
1109 
1110 	/*
1111 	 * New execsw[] array built, now replace old execsw[] and free
1112 	 * used memory.
1113 	 */
1114 	old_es = execsw;
1115 	execsw = new_es;
1116 	nexecs = es_sz;
1117 	if (old_es)
1118 		free((void *)old_es, M_EXEC);
1119 
1120 	/*
1121 	 * Figure out the maximum size of an exec header.
1122 	 */
1123 	exec_maxhdrsz = 0;
1124 	for (i = 0; i < nexecs; i++) {
1125 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1126 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1127 	}
1128 
1129 	return 0;
1130 }
1131 #endif
1132 
1133 #ifndef LKM
1134 /*
1135  * Simplified exec_init() for kernels without LKMs. Only initialize
1136  * exec_maxhdrsz and execsw[].
1137  */
1138 int
1139 exec_init(int init_boot)
1140 {
1141 	int i;
1142 
1143 #ifdef DIAGNOSTIC
1144 	if (!init_boot)
1145 		panic("exec_init(): called with init_boot == 0");
1146 #endif
1147 
1148 	/* do one-time initializations */
1149 	nexecs = nexecs_builtin;
1150 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
1151 
1152 	/*
1153 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1154 	 */
1155 	exec_maxhdrsz = 0;
1156 	for(i=0; i < nexecs; i++) {
1157 		execsw[i] = &execsw_builtin[i];
1158 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1159 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1160 	}
1161 
1162 	return 0;
1163 
1164 }
1165 #endif /* !LKM */
1166