xref: /netbsd-src/sys/kern/kern_exec.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: kern_exec.c,v 1.144 2001/07/29 21:22:42 christos 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 "opt_ktrace.h"
36 #include "opt_syscall_debug.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/filedesc.h>
41 #include <sys/kernel.h>
42 #include <sys/proc.h>
43 #include <sys/mount.h>
44 #include <sys/malloc.h>
45 #include <sys/namei.h>
46 #include <sys/vnode.h>
47 #include <sys/file.h>
48 #include <sys/acct.h>
49 #include <sys/exec.h>
50 #include <sys/ktrace.h>
51 #include <sys/resourcevar.h>
52 #include <sys/wait.h>
53 #include <sys/mman.h>
54 #include <sys/signalvar.h>
55 #include <sys/stat.h>
56 #include <sys/syscall.h>
57 
58 #include <sys/syscallargs.h>
59 
60 #include <uvm/uvm_extern.h>
61 
62 #include <machine/cpu.h>
63 #include <machine/reg.h>
64 
65 #ifdef DEBUG_EXEC
66 #define DPRINTF(a) uprintf a
67 #else
68 #define DPRINTF(a)
69 #endif /* DEBUG_EXEC */
70 
71 /*
72  * Exec function switch:
73  *
74  * Note that each makecmds function is responsible for loading the
75  * exec package with the necessary functions for any exec-type-specific
76  * handling.
77  *
78  * Functions for specific exec types should be defined in their own
79  * header file.
80  */
81 extern const struct execsw	execsw_builtin[];
82 extern int			nexecs_builtin;
83 static const struct execsw	**execsw = NULL;
84 static int			nexecs;
85 
86 int	exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
87 
88 #ifdef LKM
89 /* list of supported emulations */
90 static
91 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
92 struct emul_entry {
93 	LIST_ENTRY(emul_entry)	el_list;
94 	const struct emul	*el_emul;
95 	int			ro_entry;
96 };
97 
98 /* list of dynamically loaded execsw entries */
99 static
100 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
101 struct exec_entry {
102 	LIST_ENTRY(exec_entry)	ex_list;
103 	const struct execsw	*es;
104 };
105 
106 /* structure used for building execw[] */
107 struct execsw_entry {
108 	struct execsw_entry	*next;
109 	const struct execsw	*es;
110 };
111 #endif /* LKM */
112 
113 /* NetBSD emul struct */
114 extern char	sigcode[], esigcode[];
115 #ifdef SYSCALL_DEBUG
116 extern const char * const syscallnames[];
117 #endif
118 #ifdef __HAVE_SYSCALL_INTERN
119 void syscall_intern(struct proc *);
120 #else
121 void syscall(void);
122 #endif
123 
124 const struct emul emul_netbsd = {
125 	"netbsd",
126 	NULL,		/* emulation path */
127 #ifndef __HAVE_MINIMAL_EMUL
128 	EMUL_HAS_SYS___syscall,
129 	NULL,
130 	SYS_syscall,
131 	SYS_MAXSYSCALL,
132 #endif
133 	sysent,
134 #ifdef SYSCALL_DEBUG
135 	syscallnames,
136 #else
137 	NULL,
138 #endif
139 	sendsig,
140 	trapsignal,
141 	sigcode,
142 	esigcode,
143 	NULL,
144 	NULL,
145 	NULL,
146 #ifdef __HAVE_SYSCALL_INTERN
147 	syscall_intern,
148 #else
149 	syscall,
150 #endif
151 };
152 
153 /*
154  * Exec lock. Used to control access to execsw[] structures.
155  * This must not be static so that netbsd32 can access it, too.
156  */
157 struct lock exec_lock;
158 
159 #ifdef LKM
160 static const struct emul * emul_search(const char *);
161 static void link_es(struct execsw_entry **, const struct execsw *);
162 #endif /* LKM */
163 
164 /*
165  * check exec:
166  * given an "executable" described in the exec package's namei info,
167  * see what we can do with it.
168  *
169  * ON ENTRY:
170  *	exec package with appropriate namei info
171  *	proc pointer of exec'ing proc
172  *	NO SELF-LOCKED VNODES
173  *
174  * ON EXIT:
175  *	error:	nothing held, etc.  exec header still allocated.
176  *	ok:	filled exec package, executable's vnode (unlocked).
177  *
178  * EXEC SWITCH ENTRY:
179  * 	Locked vnode to check, exec package, proc.
180  *
181  * EXEC SWITCH EXIT:
182  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
183  *	error:	destructive:
184  *			everything deallocated execept exec header.
185  *		non-destructive:
186  *			error code, executable's vnode (unlocked),
187  *			exec header unmodified.
188  */
189 int
190 check_exec(struct proc *p, struct exec_package *epp)
191 {
192 	int		error, i;
193 	struct vnode	*vp;
194 	struct nameidata *ndp;
195 	size_t		resid;
196 
197 	ndp = epp->ep_ndp;
198 	ndp->ni_cnd.cn_nameiop = LOOKUP;
199 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
200 	/* first get the vnode */
201 	if ((error = namei(ndp)) != 0)
202 		return error;
203 	epp->ep_vp = vp = ndp->ni_vp;
204 
205 	/* check access and type */
206 	if (vp->v_type != VREG) {
207 		error = EACCES;
208 		goto bad1;
209 	}
210 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
211 		goto bad1;
212 
213 	/* get attributes */
214 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
215 		goto bad1;
216 
217 	/* Check mount point */
218 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
219 		error = EACCES;
220 		goto bad1;
221 	}
222 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
223 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
224 
225 	/* try to open it */
226 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
227 		goto bad1;
228 
229 	/* unlock vp, since we need it unlocked from here on out. */
230 	VOP_UNLOCK(vp, 0);
231 
232 	/* now we have the file, get the exec header */
233 	uvn_attach(vp, VM_PROT_READ);
234 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
235 			UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
236 	if (error)
237 		goto bad2;
238 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
239 
240 	/*
241 	 * Set up default address space limits.  Can be overridden
242 	 * by individual exec packages.
243 	 *
244 	 * XXX probably shoul be all done in the exec pakages.
245 	 */
246 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
247 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
248 	/*
249 	 * set up the vmcmds for creation of the process
250 	 * address space
251 	 */
252 	error = ENOEXEC;
253 	for (i = 0; i < nexecs && error != 0; i++) {
254 		int newerror;
255 
256 		epp->ep_esch = execsw[i];
257 		newerror = (*execsw[i]->es_check)(p, epp);
258 		/* make sure the first "interesting" error code is saved. */
259 		if (!newerror || error == ENOEXEC)
260 			error = newerror;
261 
262 		/* if es_check call was successful, update epp->ep_es */
263 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
264 			epp->ep_es = execsw[i];
265 
266 		if (epp->ep_flags & EXEC_DESTR && error != 0)
267 			return error;
268 	}
269 	if (!error) {
270 		/* check that entry point is sane */
271 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
272 			error = ENOEXEC;
273 
274 		/* check limits */
275 		if ((epp->ep_tsize > MAXTSIZ) ||
276 		    (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur))
277 			error = ENOMEM;
278 
279 		if (!error)
280 			return (0);
281 	}
282 
283 	/*
284 	 * free any vmspace-creation commands,
285 	 * and release their references
286 	 */
287 	kill_vmcmds(&epp->ep_vmcmds);
288 
289 bad2:
290 	/*
291 	 * close and release the vnode, restore the old one, free the
292 	 * pathname buf, and punt.
293 	 */
294 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
295 	VOP_CLOSE(vp, FREAD, p->p_ucred, p);
296 	vput(vp);
297 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
298 	return error;
299 
300 bad1:
301 	/*
302 	 * free the namei pathname buffer, and put the vnode
303 	 * (which we don't yet have open).
304 	 */
305 	vput(vp);				/* was still locked */
306 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
307 	return error;
308 }
309 
310 /*
311  * exec system call
312  */
313 /* ARGSUSED */
314 int
315 sys_execve(struct proc *p, void *v, register_t *retval)
316 {
317 	struct sys_execve_args /* {
318 		syscallarg(const char *)	path;
319 		syscallarg(char * const *)	argp;
320 		syscallarg(char * const *)	envp;
321 	} */ *uap = v;
322 	int			error, i;
323 	struct exec_package	pack;
324 	struct nameidata	nid;
325 	struct vattr		attr;
326 	struct ucred		*cred;
327 	char			*argp;
328 	char * const		*cpp;
329 	char			*dp, *sp;
330 	long			argc, envc;
331 	size_t			len;
332 	char			*stack;
333 	struct ps_strings	arginfo;
334 	struct vmspace		*vm;
335 	char			**tmpfap;
336 	int			szsigcode;
337 	struct exec_vmcmd	*base_vcp;
338 
339 	cred = p->p_ucred;
340 	base_vcp = NULL;
341 	/*
342 	 * Init the namei data to point the file user's program name.
343 	 * This is done here rather than in check_exec(), so that it's
344 	 * possible to override this settings if any of makecmd/probe
345 	 * functions call check_exec() recursively - for example,
346 	 * see exec_script_makecmds().
347 	 */
348 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
349 
350 	/*
351 	 * initialize the fields of the exec package.
352 	 */
353 	pack.ep_name = SCARG(uap, path);
354 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
355 	pack.ep_hdrlen = exec_maxhdrsz;
356 	pack.ep_hdrvalid = 0;
357 	pack.ep_ndp = &nid;
358 	pack.ep_emul_arg = NULL;
359 	pack.ep_vmcmds.evs_cnt = 0;
360 	pack.ep_vmcmds.evs_used = 0;
361 	pack.ep_vap = &attr;
362 	pack.ep_flags = 0;
363 
364 	lockmgr(&exec_lock, LK_SHARED, NULL);
365 
366 	/* see if we can run it. */
367 	if ((error = check_exec(p, &pack)) != 0)
368 		goto freehdr;
369 
370 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
371 
372 	/* allocate an argument buffer */
373 	argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS);
374 #ifdef DIAGNOSTIC
375 	if (argp == (vaddr_t) 0)
376 		panic("execve: argp == NULL");
377 #endif
378 	dp = argp;
379 	argc = 0;
380 
381 	/* copy the fake args list, if there's one, freeing it as we go */
382 	if (pack.ep_flags & EXEC_HASARGL) {
383 		tmpfap = pack.ep_fa;
384 		while (*tmpfap != NULL) {
385 			char *cp;
386 
387 			cp = *tmpfap;
388 			while (*cp)
389 				*dp++ = *cp++;
390 			dp++;
391 
392 			FREE(*tmpfap, M_EXEC);
393 			tmpfap++; argc++;
394 		}
395 		FREE(pack.ep_fa, M_EXEC);
396 		pack.ep_flags &= ~EXEC_HASARGL;
397 	}
398 
399 	/* Now get argv & environment */
400 	if (!(cpp = SCARG(uap, argp))) {
401 		error = EINVAL;
402 		goto bad;
403 	}
404 
405 	if (pack.ep_flags & EXEC_SKIPARG)
406 		cpp++;
407 
408 	while (1) {
409 		len = argp + ARG_MAX - dp;
410 		if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
411 			goto bad;
412 		if (!sp)
413 			break;
414 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
415 			if (error == ENAMETOOLONG)
416 				error = E2BIG;
417 			goto bad;
418 		}
419 		dp += len;
420 		cpp++;
421 		argc++;
422 	}
423 
424 	envc = 0;
425 	/* environment need not be there */
426 	if ((cpp = SCARG(uap, envp)) != NULL ) {
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 			envc++;
441 		}
442 	}
443 
444 	dp = (char *) ALIGN(dp);
445 
446 	szsigcode = pack.ep_es->es_emul->e_esigcode -
447 	    pack.ep_es->es_emul->e_sigcode;
448 
449 	/* Now check if args & environ fit into new stack */
450 	if (pack.ep_flags & EXEC_32)
451 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
452 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
453 		    szsigcode + sizeof(struct ps_strings)) - argp;
454 	else
455 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
456 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
457 		    szsigcode + sizeof(struct ps_strings)) - argp;
458 
459 	len = ALIGN(len);	/* make the stack "safely" aligned */
460 
461 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
462 		error = ENOMEM;
463 		goto bad;
464 	}
465 
466 	/* adjust "active stack depth" for process VSZ */
467 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
468 
469 	/*
470 	 * Do whatever is necessary to prepare the address space
471 	 * for remapping.  Note that this might replace the current
472 	 * vmspace with another!
473 	 */
474 	uvmspace_exec(p, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
475 
476 	/* Now map address space */
477 	vm = p->p_vmspace;
478 	vm->vm_taddr = (char *) pack.ep_taddr;
479 	vm->vm_tsize = btoc(pack.ep_tsize);
480 	vm->vm_daddr = (char *) pack.ep_daddr;
481 	vm->vm_dsize = btoc(pack.ep_dsize);
482 	vm->vm_ssize = btoc(pack.ep_ssize);
483 	vm->vm_maxsaddr = (char *) pack.ep_maxsaddr;
484 	vm->vm_minsaddr = (char *) pack.ep_minsaddr;
485 
486 	/* create the new process's VM space by running the vmcmds */
487 #ifdef DIAGNOSTIC
488 	if (pack.ep_vmcmds.evs_used == 0)
489 		panic("execve: no vmcmds");
490 #endif
491 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
492 		struct exec_vmcmd *vcp;
493 
494 		vcp = &pack.ep_vmcmds.evs_cmds[i];
495 		if (vcp->ev_flags & VMCMD_RELATIVE) {
496 #ifdef DIAGNOSTIC
497 			if (base_vcp == NULL)
498 				panic("execve: relative vmcmd with no base");
499 			if (vcp->ev_flags & VMCMD_BASE)
500 				panic("execve: illegal base & relative vmcmd");
501 #endif
502 			vcp->ev_addr += base_vcp->ev_addr;
503 		}
504 		error = (*vcp->ev_proc)(p, vcp);
505 #ifdef DEBUG_EXEC
506 		if (error) {
507 			int j;
508 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
509 			for (j = 0; j <= i; j++)
510 				uprintf(
511 			    "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
512 				    j, vp[j].ev_addr, vp[j].ev_len,
513 				    vp[j].ev_offset, vp[j].ev_prot,
514 				    vp[j].ev_flags);
515 		}
516 #endif /* DEBUG_EXEC */
517 		if (vcp->ev_flags & VMCMD_BASE)
518 			base_vcp = vcp;
519 	}
520 
521 	/* free the vmspace-creation commands, and release their references */
522 	kill_vmcmds(&pack.ep_vmcmds);
523 
524 	/* if an error happened, deallocate and punt */
525 	if (error) {
526 		DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error));
527 		goto exec_abort;
528 	}
529 
530 	/* remember information about the process */
531 	arginfo.ps_nargvstr = argc;
532 	arginfo.ps_nenvstr = envc;
533 
534 	stack = (char *) (vm->vm_minsaddr - len);
535 	/* Now copy argc, args & environ to new stack */
536 	error = (*pack.ep_es->es_copyargs)(&pack, &arginfo, &stack, argp);
537 	if (error) {
538 		DPRINTF(("execve: copyargs failed %d\n", error));
539 		goto exec_abort;
540 	}
541 	/* Move the stack back to original point */
542 	stack = (char *) (vm->vm_minsaddr - len);
543 
544 	/* fill process ps_strings info */
545 	p->p_psstr = (struct ps_strings *)(vm->vm_minsaddr
546 		- sizeof(struct ps_strings));
547 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
548 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
549 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
550 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
551 
552 	/* copy out the process's ps_strings structure */
553 	if ((error = copyout(&arginfo, (char *)p->p_psstr,
554 	    sizeof(arginfo))) != 0) {
555 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
556 		       &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo)));
557 		goto exec_abort;
558 	}
559 
560 	/* copy out the process's signal trapoline code */
561 	if (szsigcode) {
562 		if ((error = copyout((char *)pack.ep_es->es_emul->e_sigcode,
563 		    p->p_sigctx.ps_sigcode = (char *)p->p_psstr - szsigcode,
564 		    szsigcode)) != 0) {
565 			DPRINTF(("execve: sig trampoline copyout failed\n"));
566 			goto exec_abort;
567 		}
568 #ifdef PMAP_NEED_PROCWR
569 		/* This is code. Let the pmap do what is needed. */
570 		pmap_procwr(p, (vaddr_t)p->p_sigctx.ps_sigcode, szsigcode);
571 #endif
572 	}
573 
574 	stopprofclock(p);	/* stop profiling */
575 	fdcloseexec(p);		/* handle close on exec */
576 	execsigs(p);		/* reset catched signals */
577 	p->p_ctxlink = NULL;	/* reset ucontext link */
578 
579 	/* set command name & other accounting info */
580 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
581 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
582 	p->p_comm[len] = 0;
583 	p->p_acflag &= ~AFORK;
584 
585 	/* record proc's vnode, for use by procfs and others */
586         if (p->p_textvp)
587                 vrele(p->p_textvp);
588 	VREF(pack.ep_vp);
589 	p->p_textvp = pack.ep_vp;
590 
591 	p->p_flag |= P_EXEC;
592 	if (p->p_flag & P_PPWAIT) {
593 		p->p_flag &= ~P_PPWAIT;
594 		wakeup((caddr_t) p->p_pptr);
595 	}
596 
597 	/*
598 	 * deal with set[ug]id.
599 	 * MNT_NOSUID has already been used to disable s[ug]id.
600 	 */
601 	if ((p->p_flag & P_TRACED) == 0 &&
602 
603 	    (((attr.va_mode & S_ISUID) != 0 &&
604 	      p->p_ucred->cr_uid != attr.va_uid) ||
605 
606 	     ((attr.va_mode & S_ISGID) != 0 &&
607 	      p->p_ucred->cr_gid != attr.va_gid))) {
608 		/*
609 		 * Mark the process as SUGID before we do
610 		 * anything that might block.
611 		 */
612 		p_sugid(p);
613 
614 		p->p_ucred = crcopy(cred);
615 #ifdef KTRACE
616 		/*
617 		 * If process is being ktraced, turn off - unless
618 		 * root set it.
619 		 */
620 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
621 			ktrderef(p);
622 #endif
623 		if (attr.va_mode & S_ISUID)
624 			p->p_ucred->cr_uid = attr.va_uid;
625 		if (attr.va_mode & S_ISGID)
626 			p->p_ucred->cr_gid = attr.va_gid;
627 	} else
628 		p->p_flag &= ~P_SUGID;
629 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
630 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
631 
632 	doexechooks(p);
633 
634 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
635 
636 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
637 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
638 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
639 	vput(pack.ep_vp);
640 
641 	/* setup new registers and do misc. setup. */
642 	(*pack.ep_es->es_setregs)(p, &pack, (u_long) stack);
643 
644 	if (p->p_flag & P_TRACED)
645 		psignal(p, SIGTRAP);
646 
647 	free(pack.ep_hdr, M_EXEC);
648 
649 	/*
650 	 * Call emulation specific exec hook. This can setup setup per-process
651 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
652 	 *
653 	 * If we are executing process of different emulation than the
654 	 * original forked process, call e_proc_exit() of the old emulation
655 	 * first, then e_proc_exec() of new emulation. If the emulation is
656 	 * same, the exec hook code should deallocate any old emulation
657 	 * resources held previously by this process.
658 	 */
659 	if (p->p_emul && p->p_emul->e_proc_exit
660 	    && p->p_emul != pack.ep_es->es_emul)
661 		(*p->p_emul->e_proc_exit)(p);
662 
663 	/*
664 	 * Call exec hook. Emulation code may NOT store reference to anything
665 	 * from &pack.
666 	 */
667         if (pack.ep_es->es_emul->e_proc_exec)
668                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
669 
670 	/* update p_emul, the old value is no longer needed */
671 	p->p_emul = pack.ep_es->es_emul;
672 #ifdef __HAVE_SYSCALL_INTERN
673 	(*p->p_emul->e_syscall_intern)(p);
674 #endif
675 #ifdef KTRACE
676 	if (KTRPOINT(p, KTR_EMUL))
677 		ktremul(p);
678 #endif
679 
680 	lockmgr(&exec_lock, LK_RELEASE, NULL);
681 
682 	return (EJUSTRETURN);
683 
684  bad:
685 	/* free the vmspace-creation commands, and release their references */
686 	kill_vmcmds(&pack.ep_vmcmds);
687 	/* kill any opened file descriptor, if necessary */
688 	if (pack.ep_flags & EXEC_HASFD) {
689 		pack.ep_flags &= ~EXEC_HASFD;
690 		(void) fdrelease(p, pack.ep_fd);
691 	}
692 	/* close and put the exec'd file */
693 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
694 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
695 	vput(pack.ep_vp);
696 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
697 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
698 
699  freehdr:
700 	lockmgr(&exec_lock, LK_RELEASE, NULL);
701 
702 	free(pack.ep_hdr, M_EXEC);
703 	return error;
704 
705  exec_abort:
706 	lockmgr(&exec_lock, LK_RELEASE, NULL);
707 
708 	/*
709 	 * the old process doesn't exist anymore.  exit gracefully.
710 	 * get rid of the (new) address space we have created, if any, get rid
711 	 * of our namei data and vnode, and exit noting failure
712 	 */
713 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
714 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
715 	if (pack.ep_emul_arg)
716 		FREE(pack.ep_emul_arg, M_TEMP);
717 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
718 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
719 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
720 	vput(pack.ep_vp);
721 	uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS);
722 	free(pack.ep_hdr, M_EXEC);
723 	exit1(p, W_EXITCODE(error, SIGABRT));
724 
725 	/* NOTREACHED */
726 	return 0;
727 }
728 
729 
730 int
731 copyargs(struct exec_package *pack, struct ps_strings *arginfo,
732     char **stackp, void *argp)
733 {
734 	char	**cpp, *dp, *sp;
735 	size_t	len;
736 	void	*nullp;
737 	long	argc, envc;
738 	int	error;
739 
740 	cpp = (char **)*stackp;
741 	nullp = NULL;
742 	argc = arginfo->ps_nargvstr;
743 	envc = arginfo->ps_nenvstr;
744 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
745 		return error;
746 
747 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
748 	sp = argp;
749 
750 	/* XXX don't copy them out, remap them! */
751 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
752 
753 	for (; --argc >= 0; sp += len, dp += len)
754 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
755 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
756 			return error;
757 
758 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
759 		return error;
760 
761 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
762 
763 	for (; --envc >= 0; sp += len, dp += len)
764 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
765 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
766 			return error;
767 
768 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
769 		return error;
770 
771 	*stackp = (char *)cpp;
772 	return 0;
773 }
774 
775 #ifdef LKM
776 /*
777  * Find an emulation of given name in list of emulations.
778  */
779 static const struct emul *
780 emul_search(const char *name)
781 {
782 	struct emul_entry *it;
783 
784 	LIST_FOREACH(it, &el_head, el_list) {
785 		if (strcmp(name, it->el_emul->e_name) == 0)
786 			return it->el_emul;
787 	}
788 
789 	return NULL;
790 }
791 
792 /*
793  * Add an emulation to list, if it's not there already.
794  */
795 int
796 emul_register(const struct emul *emul, int ro_entry)
797 {
798 	struct emul_entry	*ee;
799 	int			error;
800 
801 	error = 0;
802 	lockmgr(&exec_lock, LK_SHARED, NULL);
803 
804 	if (emul_search(emul->e_name)) {
805 		error = EEXIST;
806 		goto out;
807 	}
808 
809 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
810 		M_EXEC, M_WAITOK);
811 	ee->el_emul = emul;
812 	ee->ro_entry = ro_entry;
813 	LIST_INSERT_HEAD(&el_head, ee, el_list);
814 
815  out:
816 	lockmgr(&exec_lock, LK_RELEASE, NULL);
817 	return error;
818 }
819 
820 /*
821  * Remove emulation with name 'name' from list of supported emulations.
822  */
823 int
824 emul_unregister(const char *name)
825 {
826 	const struct proclist_desc *pd;
827 	struct emul_entry	*it;
828 	int			i, error;
829 	struct proc		*ptmp;
830 
831 	error = 0;
832 	lockmgr(&exec_lock, LK_SHARED, NULL);
833 
834 	LIST_FOREACH(it, &el_head, el_list) {
835 		if (strcmp(it->el_emul->e_name, name) == 0)
836 			break;
837 	}
838 
839 	if (!it) {
840 		error = ENOENT;
841 		goto out;
842 	}
843 
844 	if (it->ro_entry) {
845 		error = EBUSY;
846 		goto out;
847 	}
848 
849 	/* test if any execw[] entry is still using this */
850 	for(i=0; i < nexecs; i++) {
851 		if (execsw[i]->es_emul == it->el_emul) {
852 			error = EBUSY;
853 			goto out;
854 		}
855 	}
856 
857 	/*
858 	 * Test if any process is running under this emulation - since
859 	 * emul_unregister() is running quite sendomly, it's better
860 	 * to do expensive check here than to use any locking.
861 	 */
862 	proclist_lock_read();
863 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
864 		LIST_FOREACH(ptmp, pd->pd_list, p_list) {
865 			if (ptmp->p_emul == it->el_emul) {
866 				error = EBUSY;
867 				break;
868 			}
869 		}
870 	}
871 	proclist_unlock_read();
872 
873 	if (error)
874 		goto out;
875 
876 
877 	/* entry is not used, remove it */
878 	LIST_REMOVE(it, el_list);
879 	FREE(it, M_EXEC);
880 
881  out:
882 	lockmgr(&exec_lock, LK_RELEASE, NULL);
883 	return error;
884 }
885 
886 /*
887  * Add execsw[] entry.
888  */
889 int
890 exec_add(struct execsw *esp, const char *e_name)
891 {
892 	struct exec_entry	*it;
893 	int			error;
894 
895 	error = 0;
896 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
897 
898 	if (!esp->es_emul) {
899 		esp->es_emul = emul_search(e_name);
900 		if (!esp->es_emul) {
901 			error = ENOENT;
902 			goto out;
903 		}
904 	}
905 
906 	LIST_FOREACH(it, &ex_head, ex_list) {
907 		/* assume tuple (makecmds, probe_func, emulation) is unique */
908 		if (it->es->es_check == esp->es_check
909 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
910 		    && it->es->es_emul == esp->es_emul) {
911 			error = EEXIST;
912 			goto out;
913 		}
914 	}
915 
916 	/* if we got here, the entry doesn't exist yet */
917 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
918 		M_EXEC, M_WAITOK);
919 	it->es = esp;
920 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
921 
922 	/* update execsw[] */
923 	exec_init(0);
924 
925  out:
926 	lockmgr(&exec_lock, LK_RELEASE, NULL);
927 	return error;
928 }
929 
930 /*
931  * Remove execsw[] entry.
932  */
933 int
934 exec_remove(const struct execsw *esp)
935 {
936 	struct exec_entry	*it;
937 	int			error;
938 
939 	error = 0;
940 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
941 
942 	LIST_FOREACH(it, &ex_head, ex_list) {
943 		/* assume tuple (makecmds, probe_func, emulation) is unique */
944 		if (it->es->es_check == esp->es_check
945 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
946 		    && it->es->es_emul == esp->es_emul)
947 			break;
948 	}
949 	if (!it) {
950 		error = ENOENT;
951 		goto out;
952 	}
953 
954 	/* remove item from list and free resources */
955 	LIST_REMOVE(it, ex_list);
956 	FREE(it, M_EXEC);
957 
958 	/* update execsw[] */
959 	exec_init(0);
960 
961  out:
962 	lockmgr(&exec_lock, LK_RELEASE, NULL);
963 	return error;
964 }
965 
966 static void
967 link_es(struct execsw_entry **listp, const struct execsw *esp)
968 {
969 	struct execsw_entry *et, *e1;
970 
971 	MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry),
972 			M_TEMP, M_WAITOK);
973 	et->next = NULL;
974 	et->es = esp;
975 	if (*listp == NULL) {
976 		*listp = et;
977 		return;
978 	}
979 
980 	switch(et->es->es_prio) {
981 	case EXECSW_PRIO_FIRST:
982 		/* put new entry as the first */
983 		et->next = *listp;
984 		*listp = et;
985 		break;
986 	case EXECSW_PRIO_ANY:
987 		/* put new entry after all *_FIRST and *_ANY entries */
988 		for(e1 = *listp; e1->next
989 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
990 			e1 = e1->next);
991 		et->next = e1->next;
992 		e1->next = et;
993 		break;
994 	case EXECSW_PRIO_LAST:
995 		/* put new entry as the last one */
996 		for(e1 = *listp; e1->next; e1 = e1->next);
997 		e1->next = et;
998 		break;
999 	default:
1000 #ifdef DIAGNOSTIC
1001 		panic("execw[] entry with unknown priority %d found\n",
1002 			et->es->es_prio);
1003 #endif
1004 		break;
1005 	}
1006 }
1007 
1008 /*
1009  * Initialize exec structures. If init_boot is true, also does necessary
1010  * one-time initialization (it's called from main() that way).
1011  * Once system is multiuser, this should be called with exec_lock hold,
1012  * i.e. via exec_{add|remove}().
1013  */
1014 int
1015 exec_init(int init_boot)
1016 {
1017 	const struct execsw	**new_es, * const *old_es;
1018 	struct execsw_entry	*list, *e1;
1019 	struct exec_entry	*e2;
1020 	int			i, es_sz;
1021 
1022 	if (init_boot) {
1023 		/* do one-time initializations */
1024 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1025 
1026 		/* register compiled-in emulations */
1027 		for(i=0; i < nexecs_builtin; i++) {
1028 			if (execsw_builtin[i].es_emul)
1029 				emul_register(execsw_builtin[i].es_emul, 1);
1030 		}
1031 #ifdef DIAGNOSTIC
1032 		if (i == 0)
1033 			panic("no emulations found in execsw_builtin[]\n");
1034 #endif
1035 	}
1036 
1037 	/*
1038 	 * Build execsw[] array from builtin entries and entries added
1039 	 * at runtime.
1040 	 */
1041 	list = NULL;
1042 	for(i=0; i < nexecs_builtin; i++)
1043 		link_es(&list, &execsw_builtin[i]);
1044 
1045 	/* Add dynamically loaded entries */
1046 	es_sz = nexecs_builtin;
1047 	LIST_FOREACH(e2, &ex_head, ex_list) {
1048 		link_es(&list, e2->es);
1049 		es_sz++;
1050 	}
1051 
1052 	/*
1053 	 * Now that we have sorted all execw entries, create new execsw[]
1054 	 * and free no longer needed memory in the process.
1055 	 */
1056 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
1057 	for(i=0; list; i++) {
1058 		new_es[i] = list->es;
1059 		e1 = list->next;
1060 		FREE(list, M_TEMP);
1061 		list = e1;
1062 	}
1063 
1064 	/*
1065 	 * New execsw[] array built, now replace old execsw[] and free
1066 	 * used memory.
1067 	 */
1068 	old_es = execsw;
1069 	execsw = new_es;
1070 	nexecs = es_sz;
1071 	if (old_es)
1072 		free((void *)old_es, M_EXEC);
1073 
1074 	/*
1075 	 * Figure out the maximum size of an exec header.
1076 	 */
1077 	exec_maxhdrsz = 0;
1078 	for (i = 0; i < nexecs; i++) {
1079 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1080 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1081 	}
1082 
1083 	return 0;
1084 }
1085 #endif
1086 
1087 #ifndef LKM
1088 /*
1089  * Simplified exec_init() for kernels without LKMs. Only initialize
1090  * exec_maxhdrsz and execsw[].
1091  */
1092 int
1093 exec_init(int init_boot)
1094 {
1095 	int i;
1096 
1097 #ifdef DIAGNOSTIC
1098 	if (!init_boot)
1099 		panic("exec_init(): called with init_boot == 0");
1100 #endif
1101 
1102 	/* do one-time initializations */
1103 	lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
1104 
1105 	nexecs = nexecs_builtin;
1106 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
1107 
1108 	/*
1109 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1110 	 */
1111 	exec_maxhdrsz = 0;
1112 	for(i=0; i < nexecs; i++) {
1113 		execsw[i] = &execsw_builtin[i];
1114 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1115 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1116 	}
1117 
1118 	return 0;
1119 
1120 }
1121 #endif /* !LKM */
1122