xref: /netbsd-src/sys/kern/kern_exec.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: kern_exec.c,v 1.155 2002/08/28 07:16:35 gmcgarry 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 1.155 2002/08/28 07:16:35 gmcgarry 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 };
157 
158 #ifdef LKM
159 /*
160  * Exec lock. Used to control access to execsw[] structures.
161  * This must not be static so that netbsd32 can access it, too.
162  */
163 struct lock exec_lock;
164 
165 static void link_es(struct execsw_entry **, const struct execsw *);
166 #endif /* LKM */
167 
168 /*
169  * check exec:
170  * given an "executable" described in the exec package's namei info,
171  * see what we can do with it.
172  *
173  * ON ENTRY:
174  *	exec package with appropriate namei info
175  *	proc pointer of exec'ing proc
176  *	NO SELF-LOCKED VNODES
177  *
178  * ON EXIT:
179  *	error:	nothing held, etc.  exec header still allocated.
180  *	ok:	filled exec package, executable's vnode (unlocked).
181  *
182  * EXEC SWITCH ENTRY:
183  * 	Locked vnode to check, exec package, proc.
184  *
185  * EXEC SWITCH EXIT:
186  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
187  *	error:	destructive:
188  *			everything deallocated execept exec header.
189  *		non-destructive:
190  *			error code, executable's vnode (unlocked),
191  *			exec header unmodified.
192  */
193 int
194 check_exec(struct proc *p, struct exec_package *epp)
195 {
196 	int		error, i;
197 	struct vnode	*vp;
198 	struct nameidata *ndp;
199 	size_t		resid;
200 
201 	ndp = epp->ep_ndp;
202 	ndp->ni_cnd.cn_nameiop = LOOKUP;
203 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
204 	/* first get the vnode */
205 	if ((error = namei(ndp)) != 0)
206 		return error;
207 	epp->ep_vp = vp = ndp->ni_vp;
208 
209 	/* check access and type */
210 	if (vp->v_type != VREG) {
211 		error = EACCES;
212 		goto bad1;
213 	}
214 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
215 		goto bad1;
216 
217 	/* get attributes */
218 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
219 		goto bad1;
220 
221 	/* Check mount point */
222 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
223 		error = EACCES;
224 		goto bad1;
225 	}
226 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
227 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
228 
229 	/* try to open it */
230 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
231 		goto bad1;
232 
233 	/* unlock vp, since we need it unlocked from here on out. */
234 	VOP_UNLOCK(vp, 0);
235 
236 	/* now we have the file, get the exec header */
237 	uvn_attach(vp, VM_PROT_READ);
238 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
239 			UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
240 	if (error)
241 		goto bad2;
242 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
243 
244 	/*
245 	 * Set up default address space limits.  Can be overridden
246 	 * by individual exec packages.
247 	 *
248 	 * XXX probably shoul be all done in the exec pakages.
249 	 */
250 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
251 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
252 	/*
253 	 * set up the vmcmds for creation of the process
254 	 * address space
255 	 */
256 	error = ENOEXEC;
257 	for (i = 0; i < nexecs && error != 0; i++) {
258 		int newerror;
259 
260 		epp->ep_esch = execsw[i];
261 		newerror = (*execsw[i]->es_check)(p, epp);
262 		/* make sure the first "interesting" error code is saved. */
263 		if (!newerror || error == ENOEXEC)
264 			error = newerror;
265 
266 		/* if es_check call was successful, update epp->ep_es */
267 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
268 			epp->ep_es = execsw[i];
269 
270 		if (epp->ep_flags & EXEC_DESTR && error != 0)
271 			return error;
272 	}
273 	if (!error) {
274 		/* check that entry point is sane */
275 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
276 			error = ENOEXEC;
277 
278 		/* check limits */
279 		if ((epp->ep_tsize > MAXTSIZ) ||
280 		    (epp->ep_dsize >
281 		     (u_quad_t)p->p_rlimit[RLIMIT_DATA].rlim_cur))
282 			error = ENOMEM;
283 
284 		if (!error)
285 			return (0);
286 	}
287 
288 	/*
289 	 * free any vmspace-creation commands,
290 	 * and release their references
291 	 */
292 	kill_vmcmds(&epp->ep_vmcmds);
293 
294 bad2:
295 	/*
296 	 * close and release the vnode, restore the old one, free the
297 	 * pathname buf, and punt.
298 	 */
299 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
300 	VOP_CLOSE(vp, FREAD, p->p_ucred, p);
301 	vput(vp);
302 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
303 	return error;
304 
305 bad1:
306 	/*
307 	 * free the namei pathname buffer, and put the vnode
308 	 * (which we don't yet have open).
309 	 */
310 	vput(vp);				/* was still locked */
311 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
312 	return error;
313 }
314 
315 /*
316  * exec system call
317  */
318 /* ARGSUSED */
319 int
320 sys_execve(struct proc *p, void *v, register_t *retval)
321 {
322 	struct sys_execve_args /* {
323 		syscallarg(const char *)	path;
324 		syscallarg(char * const *)	argp;
325 		syscallarg(char * const *)	envp;
326 	} */ *uap = v;
327 	int			error;
328 	u_int			i;
329 	struct exec_package	pack;
330 	struct nameidata	nid;
331 	struct vattr		attr;
332 	struct ucred		*cred;
333 	char			*argp;
334 	char * const		*cpp;
335 	char			*dp, *sp;
336 	long			argc, envc;
337 	size_t			len;
338 	char			*stack;
339 	struct ps_strings	arginfo;
340 	struct vmspace		*vm;
341 	char			**tmpfap;
342 	int			szsigcode;
343 	struct exec_vmcmd	*base_vcp;
344 
345 	/*
346 	 * Lock the process and set the P_INEXEC flag to indicate that
347 	 * it should be left alone until we're done here.  This is
348 	 * necessary to avoid race conditions - e.g. in ptrace() -
349 	 * that might allow a local user to illicitly obtain elevated
350 	 * privileges.
351 	 */
352 	p->p_flag |= P_INEXEC;
353 
354 	cred = p->p_ucred;
355 	base_vcp = NULL;
356 	/*
357 	 * Init the namei data to point the file user's program name.
358 	 * This is done here rather than in check_exec(), so that it's
359 	 * possible to override this settings if any of makecmd/probe
360 	 * functions call check_exec() recursively - for example,
361 	 * see exec_script_makecmds().
362 	 */
363 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
364 
365 	/*
366 	 * initialize the fields of the exec package.
367 	 */
368 	pack.ep_name = SCARG(uap, path);
369 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
370 	pack.ep_hdrlen = exec_maxhdrsz;
371 	pack.ep_hdrvalid = 0;
372 	pack.ep_ndp = &nid;
373 	pack.ep_emul_arg = NULL;
374 	pack.ep_vmcmds.evs_cnt = 0;
375 	pack.ep_vmcmds.evs_used = 0;
376 	pack.ep_vap = &attr;
377 	pack.ep_flags = 0;
378 
379 #ifdef LKM
380 	lockmgr(&exec_lock, LK_SHARED, NULL);
381 #endif
382 
383 	/* see if we can run it. */
384 	if ((error = check_exec(p, &pack)) != 0)
385 		goto freehdr;
386 
387 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
388 
389 	/* allocate an argument buffer */
390 	argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS);
391 #ifdef DIAGNOSTIC
392 	if (argp == (vaddr_t) 0)
393 		panic("execve: argp == NULL");
394 #endif
395 	dp = argp;
396 	argc = 0;
397 
398 	/* copy the fake args list, if there's one, freeing it as we go */
399 	if (pack.ep_flags & EXEC_HASARGL) {
400 		tmpfap = pack.ep_fa;
401 		while (*tmpfap != NULL) {
402 			char *cp;
403 
404 			cp = *tmpfap;
405 			while (*cp)
406 				*dp++ = *cp++;
407 			dp++;
408 
409 			FREE(*tmpfap, M_EXEC);
410 			tmpfap++; argc++;
411 		}
412 		FREE(pack.ep_fa, M_EXEC);
413 		pack.ep_flags &= ~EXEC_HASARGL;
414 	}
415 
416 	/* Now get argv & environment */
417 	if (!(cpp = SCARG(uap, argp))) {
418 		error = EINVAL;
419 		goto bad;
420 	}
421 
422 	if (pack.ep_flags & EXEC_SKIPARG)
423 		cpp++;
424 
425 	while (1) {
426 		len = argp + ARG_MAX - dp;
427 		if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
428 			goto bad;
429 		if (!sp)
430 			break;
431 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
432 			if (error == ENAMETOOLONG)
433 				error = E2BIG;
434 			goto bad;
435 		}
436 		dp += len;
437 		cpp++;
438 		argc++;
439 	}
440 
441 	envc = 0;
442 	/* environment need not be there */
443 	if ((cpp = SCARG(uap, envp)) != NULL ) {
444 		while (1) {
445 			len = argp + ARG_MAX - dp;
446 			if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
447 				goto bad;
448 			if (!sp)
449 				break;
450 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
451 				if (error == ENAMETOOLONG)
452 					error = E2BIG;
453 				goto bad;
454 			}
455 			dp += len;
456 			cpp++;
457 			envc++;
458 		}
459 	}
460 
461 	dp = (char *) ALIGN(dp);
462 
463 	szsigcode = pack.ep_es->es_emul->e_esigcode -
464 	    pack.ep_es->es_emul->e_sigcode;
465 
466 	/* Now check if args & environ fit into new stack */
467 	if (pack.ep_flags & EXEC_32)
468 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
469 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
470 		    szsigcode + sizeof(struct ps_strings)) - argp;
471 	else
472 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
473 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
474 		    szsigcode + sizeof(struct ps_strings)) - argp;
475 
476 	len = ALIGN(len);	/* make the stack "safely" aligned */
477 
478 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
479 		error = ENOMEM;
480 		goto bad;
481 	}
482 
483 	/* adjust "active stack depth" for process VSZ */
484 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
485 
486 	/*
487 	 * Do whatever is necessary to prepare the address space
488 	 * for remapping.  Note that this might replace the current
489 	 * vmspace with another!
490 	 */
491 	uvmspace_exec(p, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
492 
493 	/* Now map address space */
494 	vm = p->p_vmspace;
495 	vm->vm_taddr = (char *) pack.ep_taddr;
496 	vm->vm_tsize = btoc(pack.ep_tsize);
497 	vm->vm_daddr = (char *) pack.ep_daddr;
498 	vm->vm_dsize = btoc(pack.ep_dsize);
499 	vm->vm_ssize = btoc(pack.ep_ssize);
500 	vm->vm_maxsaddr = (char *) pack.ep_maxsaddr;
501 	vm->vm_minsaddr = (char *) pack.ep_minsaddr;
502 
503 	/* create the new process's VM space by running the vmcmds */
504 #ifdef DIAGNOSTIC
505 	if (pack.ep_vmcmds.evs_used == 0)
506 		panic("execve: no vmcmds");
507 #endif
508 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
509 		struct exec_vmcmd *vcp;
510 
511 		vcp = &pack.ep_vmcmds.evs_cmds[i];
512 		if (vcp->ev_flags & VMCMD_RELATIVE) {
513 #ifdef DIAGNOSTIC
514 			if (base_vcp == NULL)
515 				panic("execve: relative vmcmd with no base");
516 			if (vcp->ev_flags & VMCMD_BASE)
517 				panic("execve: illegal base & relative vmcmd");
518 #endif
519 			vcp->ev_addr += base_vcp->ev_addr;
520 		}
521 		error = (*vcp->ev_proc)(p, vcp);
522 #ifdef DEBUG_EXEC
523 		if (error) {
524 			int j;
525 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
526 			for (j = 0; j <= i; j++)
527 				uprintf(
528 			    "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
529 				    j, vp[j].ev_addr, vp[j].ev_len,
530 				    vp[j].ev_offset, vp[j].ev_prot,
531 				    vp[j].ev_flags);
532 		}
533 #endif /* DEBUG_EXEC */
534 		if (vcp->ev_flags & VMCMD_BASE)
535 			base_vcp = vcp;
536 	}
537 
538 	/* free the vmspace-creation commands, and release their references */
539 	kill_vmcmds(&pack.ep_vmcmds);
540 
541 	/* if an error happened, deallocate and punt */
542 	if (error) {
543 		DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error));
544 		goto exec_abort;
545 	}
546 
547 	/* remember information about the process */
548 	arginfo.ps_nargvstr = argc;
549 	arginfo.ps_nenvstr = envc;
550 
551 	stack = (char *) (vm->vm_minsaddr - len);
552 	/* Now copy argc, args & environ to new stack */
553 	error = (*pack.ep_es->es_copyargs)(p, &pack, &arginfo, &stack, argp);
554 	if (error) {
555 		DPRINTF(("execve: copyargs failed %d\n", error));
556 		goto exec_abort;
557 	}
558 	/* Move the stack back to original point */
559 	stack = (char *) (vm->vm_minsaddr - len);
560 
561 	/* fill process ps_strings info */
562 	p->p_psstr = (struct ps_strings *)(vm->vm_minsaddr
563 		- sizeof(struct ps_strings));
564 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
565 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
566 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
567 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
568 
569 	/* copy out the process's ps_strings structure */
570 	if ((error = copyout(&arginfo, (char *)p->p_psstr,
571 	    sizeof(arginfo))) != 0) {
572 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
573 		       &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo)));
574 		goto exec_abort;
575 	}
576 
577 	/* copy out the process's signal trapoline code */
578 	if (szsigcode) {
579 		if ((error = copyout((char *)pack.ep_es->es_emul->e_sigcode,
580 		    p->p_sigctx.ps_sigcode = (char *)p->p_psstr - szsigcode,
581 		    szsigcode)) != 0) {
582 			DPRINTF(("execve: sig trampoline copyout failed\n"));
583 			goto exec_abort;
584 		}
585 #ifdef PMAP_NEED_PROCWR
586 		/* This is code. Let the pmap do what is needed. */
587 		pmap_procwr(p, (vaddr_t)p->p_sigctx.ps_sigcode, szsigcode);
588 #endif
589 	}
590 
591 	stopprofclock(p);	/* stop profiling */
592 	fdcloseexec(p);		/* handle close on exec */
593 	execsigs(p);		/* reset catched signals */
594 	p->p_ctxlink = NULL;	/* reset ucontext link */
595 
596 	/* set command name & other accounting info */
597 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
598 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
599 	p->p_comm[len] = 0;
600 	p->p_acflag &= ~AFORK;
601 
602 	/* record proc's vnode, for use by procfs and others */
603         if (p->p_textvp)
604                 vrele(p->p_textvp);
605 	VREF(pack.ep_vp);
606 	p->p_textvp = pack.ep_vp;
607 
608 	p->p_flag |= P_EXEC;
609 	if (p->p_flag & P_PPWAIT) {
610 		p->p_flag &= ~P_PPWAIT;
611 		wakeup((caddr_t) p->p_pptr);
612 	}
613 
614 	/*
615 	 * deal with set[ug]id.
616 	 * MNT_NOSUID has already been used to disable s[ug]id.
617 	 */
618 	if ((p->p_flag & P_TRACED) == 0 &&
619 
620 	    (((attr.va_mode & S_ISUID) != 0 &&
621 	      p->p_ucred->cr_uid != attr.va_uid) ||
622 
623 	     ((attr.va_mode & S_ISGID) != 0 &&
624 	      p->p_ucred->cr_gid != attr.va_gid))) {
625 		/*
626 		 * Mark the process as SUGID before we do
627 		 * anything that might block.
628 		 */
629 		p_sugid(p);
630 
631 		/* Make sure file descriptors 0..2 are in use. */
632 		if ((error = fdcheckstd(p)) != 0)
633 			goto exec_abort;
634 
635 		p->p_ucred = crcopy(cred);
636 #ifdef KTRACE
637 		/*
638 		 * If process is being ktraced, turn off - unless
639 		 * root set it.
640 		 */
641 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
642 			ktrderef(p);
643 #endif
644 		if (attr.va_mode & S_ISUID)
645 			p->p_ucred->cr_uid = attr.va_uid;
646 		if (attr.va_mode & S_ISGID)
647 			p->p_ucred->cr_gid = attr.va_gid;
648 	} else
649 		p->p_flag &= ~P_SUGID;
650 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
651 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
652 
653 #if defined(__HAVE_RAS)
654 	/*
655 	 * Remove all RASs from the address space.
656 	 */
657 	ras_purgeall(p);
658 #endif
659 
660 	doexechooks(p);
661 
662 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
663 
664 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
665 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
666 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
667 	vput(pack.ep_vp);
668 
669 	/* setup new registers and do misc. setup. */
670 	(*pack.ep_es->es_emul->e_setregs)(p, &pack, (u_long) stack);
671 	if (pack.ep_es->es_setregs)
672 		(*pack.ep_es->es_setregs)(p, &pack, (u_long) stack);
673 
674 	if (p->p_flag & P_TRACED)
675 		psignal(p, SIGTRAP);
676 
677 	free(pack.ep_hdr, M_EXEC);
678 
679 	/*
680 	 * Call emulation specific exec hook. This can setup setup per-process
681 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
682 	 *
683 	 * If we are executing process of different emulation than the
684 	 * original forked process, call e_proc_exit() of the old emulation
685 	 * first, then e_proc_exec() of new emulation. If the emulation is
686 	 * same, the exec hook code should deallocate any old emulation
687 	 * resources held previously by this process.
688 	 */
689 	if (p->p_emul && p->p_emul->e_proc_exit
690 	    && p->p_emul != pack.ep_es->es_emul)
691 		(*p->p_emul->e_proc_exit)(p);
692 
693 	/*
694 	 * Call exec hook. Emulation code may NOT store reference to anything
695 	 * from &pack.
696 	 */
697         if (pack.ep_es->es_emul->e_proc_exec)
698                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
699 
700 	/* update p_emul, the old value is no longer needed */
701 	p->p_emul = pack.ep_es->es_emul;
702 
703 	/* ...and the same for p_execsw */
704 	p->p_execsw = pack.ep_es;
705 
706 #ifdef __HAVE_SYSCALL_INTERN
707 	(*p->p_emul->e_syscall_intern)(p);
708 #endif
709 #ifdef KTRACE
710 	if (KTRPOINT(p, KTR_EMUL))
711 		ktremul(p);
712 #endif
713 
714 #ifdef LKM
715 	lockmgr(&exec_lock, LK_RELEASE, NULL);
716 #endif
717 	p->p_flag &= ~P_INEXEC;
718 	return (EJUSTRETURN);
719 
720  bad:
721 	p->p_flag &= ~P_INEXEC;
722 	/* free the vmspace-creation commands, and release their references */
723 	kill_vmcmds(&pack.ep_vmcmds);
724 	/* kill any opened file descriptor, if necessary */
725 	if (pack.ep_flags & EXEC_HASFD) {
726 		pack.ep_flags &= ~EXEC_HASFD;
727 		(void) fdrelease(p, pack.ep_fd);
728 	}
729 	/* close and put the exec'd file */
730 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
731 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
732 	vput(pack.ep_vp);
733 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
734 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
735 
736  freehdr:
737 	p->p_flag &= ~P_INEXEC;
738 #ifdef LKM
739 	lockmgr(&exec_lock, LK_RELEASE, NULL);
740 #endif
741 
742 	free(pack.ep_hdr, M_EXEC);
743 	return error;
744 
745  exec_abort:
746 	p->p_flag &= ~P_INEXEC;
747 #ifdef LKM
748 	lockmgr(&exec_lock, LK_RELEASE, NULL);
749 #endif
750 
751 	/*
752 	 * the old process doesn't exist anymore.  exit gracefully.
753 	 * get rid of the (new) address space we have created, if any, get rid
754 	 * of our namei data and vnode, and exit noting failure
755 	 */
756 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
757 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
758 	if (pack.ep_emul_arg)
759 		FREE(pack.ep_emul_arg, M_TEMP);
760 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
761 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
762 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
763 	vput(pack.ep_vp);
764 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
765 	free(pack.ep_hdr, M_EXEC);
766 	exit1(p, W_EXITCODE(error, SIGABRT));
767 
768 	/* NOTREACHED */
769 	return 0;
770 }
771 
772 
773 int
774 copyargs(struct proc *p, struct exec_package *pack, struct ps_strings *arginfo,
775     char **stackp, void *argp)
776 {
777 	char	**cpp, *dp, *sp;
778 	size_t	len;
779 	void	*nullp;
780 	long	argc, envc;
781 	int	error;
782 
783 	cpp = (char **)*stackp;
784 	nullp = NULL;
785 	argc = arginfo->ps_nargvstr;
786 	envc = arginfo->ps_nenvstr;
787 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
788 		return error;
789 
790 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
791 	sp = argp;
792 
793 	/* XXX don't copy them out, remap them! */
794 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
795 
796 	for (; --argc >= 0; sp += len, dp += len)
797 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
798 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
799 			return error;
800 
801 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
802 		return error;
803 
804 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
805 
806 	for (; --envc >= 0; sp += len, dp += len)
807 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
808 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
809 			return error;
810 
811 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
812 		return error;
813 
814 	*stackp = (char *)cpp;
815 	return 0;
816 }
817 
818 #ifdef LKM
819 /*
820  * Find an emulation of given name in list of emulations.
821  * Needs to be called with the exec_lock held.
822  */
823 const struct emul *
824 emul_search(const char *name)
825 {
826 	struct emul_entry *it;
827 
828 	LIST_FOREACH(it, &el_head, el_list) {
829 		if (strcmp(name, it->el_emul->e_name) == 0)
830 			return it->el_emul;
831 	}
832 
833 	return NULL;
834 }
835 
836 /*
837  * Add an emulation to list, if it's not there already.
838  */
839 int
840 emul_register(const struct emul *emul, int ro_entry)
841 {
842 	struct emul_entry	*ee;
843 	int			error;
844 
845 	error = 0;
846 	lockmgr(&exec_lock, LK_SHARED, NULL);
847 
848 	if (emul_search(emul->e_name)) {
849 		error = EEXIST;
850 		goto out;
851 	}
852 
853 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
854 		M_EXEC, M_WAITOK);
855 	ee->el_emul = emul;
856 	ee->ro_entry = ro_entry;
857 	LIST_INSERT_HEAD(&el_head, ee, el_list);
858 
859  out:
860 	lockmgr(&exec_lock, LK_RELEASE, NULL);
861 	return error;
862 }
863 
864 /*
865  * Remove emulation with name 'name' from list of supported emulations.
866  */
867 int
868 emul_unregister(const char *name)
869 {
870 	const struct proclist_desc *pd;
871 	struct emul_entry	*it;
872 	int			i, error;
873 	struct proc		*ptmp;
874 
875 	error = 0;
876 	lockmgr(&exec_lock, LK_SHARED, NULL);
877 
878 	LIST_FOREACH(it, &el_head, el_list) {
879 		if (strcmp(it->el_emul->e_name, name) == 0)
880 			break;
881 	}
882 
883 	if (!it) {
884 		error = ENOENT;
885 		goto out;
886 	}
887 
888 	if (it->ro_entry) {
889 		error = EBUSY;
890 		goto out;
891 	}
892 
893 	/* test if any execw[] entry is still using this */
894 	for(i=0; i < nexecs; i++) {
895 		if (execsw[i]->es_emul == it->el_emul) {
896 			error = EBUSY;
897 			goto out;
898 		}
899 	}
900 
901 	/*
902 	 * Test if any process is running under this emulation - since
903 	 * emul_unregister() is running quite sendomly, it's better
904 	 * to do expensive check here than to use any locking.
905 	 */
906 	proclist_lock_read();
907 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
908 		LIST_FOREACH(ptmp, pd->pd_list, p_list) {
909 			if (ptmp->p_emul == it->el_emul) {
910 				error = EBUSY;
911 				break;
912 			}
913 		}
914 	}
915 	proclist_unlock_read();
916 
917 	if (error)
918 		goto out;
919 
920 
921 	/* entry is not used, remove it */
922 	LIST_REMOVE(it, el_list);
923 	FREE(it, M_EXEC);
924 
925  out:
926 	lockmgr(&exec_lock, LK_RELEASE, NULL);
927 	return error;
928 }
929 
930 /*
931  * Add execsw[] entry.
932  */
933 int
934 exec_add(struct execsw *esp, const char *e_name)
935 {
936 	struct exec_entry	*it;
937 	int			error;
938 
939 	error = 0;
940 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
941 
942 	if (!esp->es_emul) {
943 		esp->es_emul = emul_search(e_name);
944 		if (!esp->es_emul) {
945 			error = ENOENT;
946 			goto out;
947 		}
948 	}
949 
950 	LIST_FOREACH(it, &ex_head, ex_list) {
951 		/* assume tuple (makecmds, probe_func, emulation) is unique */
952 		if (it->es->es_check == esp->es_check
953 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
954 		    && it->es->es_emul == esp->es_emul) {
955 			error = EEXIST;
956 			goto out;
957 		}
958 	}
959 
960 	/* if we got here, the entry doesn't exist yet */
961 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
962 		M_EXEC, M_WAITOK);
963 	it->es = esp;
964 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
965 
966 	/* update execsw[] */
967 	exec_init(0);
968 
969  out:
970 	lockmgr(&exec_lock, LK_RELEASE, NULL);
971 	return error;
972 }
973 
974 /*
975  * Remove execsw[] entry.
976  */
977 int
978 exec_remove(const struct execsw *esp)
979 {
980 	struct exec_entry	*it;
981 	int			error;
982 
983 	error = 0;
984 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
985 
986 	LIST_FOREACH(it, &ex_head, ex_list) {
987 		/* assume tuple (makecmds, probe_func, emulation) is unique */
988 		if (it->es->es_check == esp->es_check
989 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
990 		    && it->es->es_emul == esp->es_emul)
991 			break;
992 	}
993 	if (!it) {
994 		error = ENOENT;
995 		goto out;
996 	}
997 
998 	/* remove item from list and free resources */
999 	LIST_REMOVE(it, ex_list);
1000 	FREE(it, M_EXEC);
1001 
1002 	/* update execsw[] */
1003 	exec_init(0);
1004 
1005  out:
1006 	lockmgr(&exec_lock, LK_RELEASE, NULL);
1007 	return error;
1008 }
1009 
1010 static void
1011 link_es(struct execsw_entry **listp, const struct execsw *esp)
1012 {
1013 	struct execsw_entry *et, *e1;
1014 
1015 	MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry),
1016 			M_TEMP, M_WAITOK);
1017 	et->next = NULL;
1018 	et->es = esp;
1019 	if (*listp == NULL) {
1020 		*listp = et;
1021 		return;
1022 	}
1023 
1024 	switch(et->es->es_prio) {
1025 	case EXECSW_PRIO_FIRST:
1026 		/* put new entry as the first */
1027 		et->next = *listp;
1028 		*listp = et;
1029 		break;
1030 	case EXECSW_PRIO_ANY:
1031 		/* put new entry after all *_FIRST and *_ANY entries */
1032 		for(e1 = *listp; e1->next
1033 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
1034 			e1 = e1->next);
1035 		et->next = e1->next;
1036 		e1->next = et;
1037 		break;
1038 	case EXECSW_PRIO_LAST:
1039 		/* put new entry as the last one */
1040 		for(e1 = *listp; e1->next; e1 = e1->next);
1041 		e1->next = et;
1042 		break;
1043 	default:
1044 #ifdef DIAGNOSTIC
1045 		panic("execw[] entry with unknown priority %d found\n",
1046 			et->es->es_prio);
1047 #endif
1048 		break;
1049 	}
1050 }
1051 
1052 /*
1053  * Initialize exec structures. If init_boot is true, also does necessary
1054  * one-time initialization (it's called from main() that way).
1055  * Once system is multiuser, this should be called with exec_lock held,
1056  * i.e. via exec_{add|remove}().
1057  */
1058 int
1059 exec_init(int init_boot)
1060 {
1061 	const struct execsw	**new_es, * const *old_es;
1062 	struct execsw_entry	*list, *e1;
1063 	struct exec_entry	*e2;
1064 	int			i, es_sz;
1065 
1066 	if (init_boot) {
1067 		/* do one-time initializations */
1068 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1069 
1070 		/* register compiled-in emulations */
1071 		for(i=0; i < nexecs_builtin; i++) {
1072 			if (execsw_builtin[i].es_emul)
1073 				emul_register(execsw_builtin[i].es_emul, 1);
1074 		}
1075 #ifdef DIAGNOSTIC
1076 		if (i == 0)
1077 			panic("no emulations found in execsw_builtin[]\n");
1078 #endif
1079 	}
1080 
1081 	/*
1082 	 * Build execsw[] array from builtin entries and entries added
1083 	 * at runtime.
1084 	 */
1085 	list = NULL;
1086 	for(i=0; i < nexecs_builtin; i++)
1087 		link_es(&list, &execsw_builtin[i]);
1088 
1089 	/* Add dynamically loaded entries */
1090 	es_sz = nexecs_builtin;
1091 	LIST_FOREACH(e2, &ex_head, ex_list) {
1092 		link_es(&list, e2->es);
1093 		es_sz++;
1094 	}
1095 
1096 	/*
1097 	 * Now that we have sorted all execw entries, create new execsw[]
1098 	 * and free no longer needed memory in the process.
1099 	 */
1100 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
1101 	for(i=0; list; i++) {
1102 		new_es[i] = list->es;
1103 		e1 = list->next;
1104 		FREE(list, M_TEMP);
1105 		list = e1;
1106 	}
1107 
1108 	/*
1109 	 * New execsw[] array built, now replace old execsw[] and free
1110 	 * used memory.
1111 	 */
1112 	old_es = execsw;
1113 	execsw = new_es;
1114 	nexecs = es_sz;
1115 	if (old_es)
1116 		free((void *)old_es, M_EXEC);
1117 
1118 	/*
1119 	 * Figure out the maximum size of an exec header.
1120 	 */
1121 	exec_maxhdrsz = 0;
1122 	for (i = 0; i < nexecs; i++) {
1123 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1124 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1125 	}
1126 
1127 	return 0;
1128 }
1129 #endif
1130 
1131 #ifndef LKM
1132 /*
1133  * Simplified exec_init() for kernels without LKMs. Only initialize
1134  * exec_maxhdrsz and execsw[].
1135  */
1136 int
1137 exec_init(int init_boot)
1138 {
1139 	int i;
1140 
1141 #ifdef DIAGNOSTIC
1142 	if (!init_boot)
1143 		panic("exec_init(): called with init_boot == 0");
1144 #endif
1145 
1146 	/* do one-time initializations */
1147 	nexecs = nexecs_builtin;
1148 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
1149 
1150 	/*
1151 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1152 	 */
1153 	exec_maxhdrsz = 0;
1154 	for(i=0; i < nexecs; i++) {
1155 		execsw[i] = &execsw_builtin[i];
1156 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1157 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1158 	}
1159 
1160 	return 0;
1161 
1162 }
1163 #endif /* !LKM */
1164