xref: /netbsd-src/sys/kern/kern_exec.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: kern_exec.c,v 1.268 2008/02/02 20:42:18 elad 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 1.268 2008/02/02 20:42:18 elad Exp $");
37 
38 #include "opt_ktrace.h"
39 #include "opt_syscall_debug.h"
40 #include "opt_compat_netbsd.h"
41 #include "veriexec.h"
42 #include "opt_pax.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/filedesc.h>
47 #include <sys/kernel.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/malloc.h>
51 #include <sys/kmem.h>
52 #include <sys/namei.h>
53 #include <sys/vnode.h>
54 #include <sys/file.h>
55 #include <sys/acct.h>
56 #include <sys/exec.h>
57 #include <sys/ktrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/wait.h>
60 #include <sys/mman.h>
61 #include <sys/ras.h>
62 #include <sys/signalvar.h>
63 #include <sys/stat.h>
64 #include <sys/syscall.h>
65 #include <sys/kauth.h>
66 #include <sys/lwpctl.h>
67 #include <sys/pax.h>
68 #include <sys/cpu.h>
69 
70 #include <sys/syscallargs.h>
71 #if NVERIEXEC > 0
72 #include <sys/verified_exec.h>
73 #endif /* NVERIEXEC > 0 */
74 
75 #include <uvm/uvm_extern.h>
76 
77 #include <machine/reg.h>
78 
79 #include <compat/common/compat_util.h>
80 
81 static int exec_sigcode_map(struct proc *, const struct emul *);
82 
83 #ifdef DEBUG_EXEC
84 #define DPRINTF(a) uprintf a
85 #else
86 #define DPRINTF(a)
87 #endif /* DEBUG_EXEC */
88 
89 /*
90  * Exec function switch:
91  *
92  * Note that each makecmds function is responsible for loading the
93  * exec package with the necessary functions for any exec-type-specific
94  * handling.
95  *
96  * Functions for specific exec types should be defined in their own
97  * header file.
98  */
99 extern const struct execsw	execsw_builtin[];
100 extern int			nexecs_builtin;
101 static const struct execsw	**execsw = NULL;
102 static int			nexecs;
103 
104 u_int	exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
105 
106 #ifdef LKM
107 /* list of supported emulations */
108 static
109 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
110 struct emul_entry {
111 	LIST_ENTRY(emul_entry)	el_list;
112 	const struct emul	*el_emul;
113 	int			ro_entry;
114 };
115 
116 /* list of dynamically loaded execsw entries */
117 static
118 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
119 struct exec_entry {
120 	LIST_ENTRY(exec_entry)	ex_list;
121 	const struct execsw	*es;
122 };
123 
124 /* structure used for building execw[] */
125 struct execsw_entry {
126 	struct execsw_entry	*next;
127 	const struct execsw	*es;
128 };
129 #endif /* LKM */
130 
131 #ifdef SYSCALL_DEBUG
132 extern const char * const syscallnames[];
133 #endif
134 
135 #ifdef COMPAT_16
136 extern char	sigcode[], esigcode[];
137 struct uvm_object *emul_netbsd_object;
138 #endif
139 
140 #ifndef __HAVE_SYSCALL_INTERN
141 void	syscall(void);
142 #endif
143 
144 /* NetBSD emul struct */
145 const struct emul emul_netbsd = {
146 	"netbsd",
147 	NULL,		/* emulation path */
148 #ifndef __HAVE_MINIMAL_EMUL
149 	EMUL_HAS_SYS___syscall,
150 	NULL,
151 	SYS_syscall,
152 	SYS_NSYSENT,
153 #endif
154 	sysent,
155 #ifdef SYSCALL_DEBUG
156 	syscallnames,
157 #else
158 	NULL,
159 #endif
160 	sendsig,
161 	trapsignal,
162 	NULL,
163 #ifdef COMPAT_16
164 	sigcode,
165 	esigcode,
166 	&emul_netbsd_object,
167 #else
168 	NULL,
169 	NULL,
170 	NULL,
171 #endif
172 	setregs,
173 	NULL,
174 	NULL,
175 	NULL,
176 	NULL,
177 	NULL,
178 #ifdef __HAVE_SYSCALL_INTERN
179 	syscall_intern,
180 #else
181 	syscall,
182 #endif
183 	NULL,
184 	NULL,
185 
186 	uvm_default_mapaddr,
187 	NULL,
188 	sizeof(ucontext_t),
189 	startlwp,
190 };
191 
192 #ifdef LKM
193 /*
194  * Exec lock. Used to control access to execsw[] structures.
195  * This must not be static so that netbsd32 can access it, too.
196  */
197 krwlock_t exec_lock;
198 
199 static void link_es(struct execsw_entry **, const struct execsw *);
200 #endif /* LKM */
201 
202 static kmutex_t sigobject_lock;
203 
204 /*
205  * check exec:
206  * given an "executable" described in the exec package's namei info,
207  * see what we can do with it.
208  *
209  * ON ENTRY:
210  *	exec package with appropriate namei info
211  *	lwp pointer of exec'ing lwp
212  *	NO SELF-LOCKED VNODES
213  *
214  * ON EXIT:
215  *	error:	nothing held, etc.  exec header still allocated.
216  *	ok:	filled exec package, executable's vnode (unlocked).
217  *
218  * EXEC SWITCH ENTRY:
219  * 	Locked vnode to check, exec package, proc.
220  *
221  * EXEC SWITCH EXIT:
222  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
223  *	error:	destructive:
224  *			everything deallocated execept exec header.
225  *		non-destructive:
226  *			error code, executable's vnode (unlocked),
227  *			exec header unmodified.
228  */
229 int
230 /*ARGSUSED*/
231 check_exec(struct lwp *l, struct exec_package *epp)
232 {
233 	int		error, i;
234 	struct vnode	*vp;
235 	struct nameidata *ndp;
236 	size_t		resid;
237 
238 	ndp = epp->ep_ndp;
239 	ndp->ni_cnd.cn_nameiop = LOOKUP;
240 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME | TRYEMULROOT;
241 	/* first get the vnode */
242 	if ((error = namei(ndp)) != 0)
243 		return error;
244 	epp->ep_vp = vp = ndp->ni_vp;
245 
246 	/* check access and type */
247 	if (vp->v_type != VREG) {
248 		error = EACCES;
249 		goto bad1;
250 	}
251 	if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0)
252 		goto bad1;
253 
254 	/* get attributes */
255 	if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0)
256 		goto bad1;
257 
258 	/* Check mount point */
259 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
260 		error = EACCES;
261 		goto bad1;
262 	}
263 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
264 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
265 
266 	/* try to open it */
267 	if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0)
268 		goto bad1;
269 
270 	/* unlock vp, since we need it unlocked from here on out. */
271 	VOP_UNLOCK(vp, 0);
272 
273 #if NVERIEXEC > 0
274 	error = veriexec_verify(l, vp, ndp->ni_cnd.cn_pnbuf,
275 	    epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT,
276 	    NULL);
277 	if (error)
278 		goto bad2;
279 #endif /* NVERIEXEC > 0 */
280 
281 #ifdef PAX_SEGVGUARD
282 	error = pax_segvguard(l, vp, ndp->ni_cnd.cn_pnbuf, false);
283 	if (error)
284 		goto bad2;
285 #endif /* PAX_SEGVGUARD */
286 
287 	/* now we have the file, get the exec header */
288 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
289 			UIO_SYSSPACE, 0, l->l_cred, &resid, NULL);
290 	if (error)
291 		goto bad2;
292 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
293 
294 	/*
295 	 * Set up default address space limits.  Can be overridden
296 	 * by individual exec packages.
297 	 *
298 	 * XXX probably should be all done in the exec packages.
299 	 */
300 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
301 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
302 	/*
303 	 * set up the vmcmds for creation of the process
304 	 * address space
305 	 */
306 	error = ENOEXEC;
307 	for (i = 0; i < nexecs; i++) {
308 		int newerror;
309 
310 		epp->ep_esch = execsw[i];
311 		newerror = (*execsw[i]->es_makecmds)(l, epp);
312 
313 		if (!newerror) {
314 			/* Seems ok: check that entry point is sane */
315 			if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
316 				error = ENOEXEC;
317 				break;
318 			}
319 
320 			/* check limits */
321 			if ((epp->ep_tsize > MAXTSIZ) ||
322 			    (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit
323 						    [RLIMIT_DATA].rlim_cur)) {
324 				error = ENOMEM;
325 				break;
326 			}
327 			return 0;
328 		}
329 
330 		if (epp->ep_emul_root != NULL) {
331 			vrele(epp->ep_emul_root);
332 			epp->ep_emul_root = NULL;
333 		}
334 		if (epp->ep_interp != NULL) {
335 			vrele(epp->ep_interp);
336 			epp->ep_interp = NULL;
337 		}
338 
339 		/* make sure the first "interesting" error code is saved. */
340 		if (error == ENOEXEC)
341 			error = newerror;
342 
343 		if (epp->ep_flags & EXEC_DESTR)
344 			/* Error from "#!" code, tidied up by recursive call */
345 			return error;
346 	}
347 
348 	/* not found, error */
349 
350 	/*
351 	 * free any vmspace-creation commands,
352 	 * and release their references
353 	 */
354 	kill_vmcmds(&epp->ep_vmcmds);
355 
356 bad2:
357 	/*
358 	 * close and release the vnode, restore the old one, free the
359 	 * pathname buf, and punt.
360 	 */
361 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
362 	VOP_CLOSE(vp, FREAD, l->l_cred);
363 	vput(vp);
364 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
365 	return error;
366 
367 bad1:
368 	/*
369 	 * free the namei pathname buffer, and put the vnode
370 	 * (which we don't yet have open).
371 	 */
372 	vput(vp);				/* was still locked */
373 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
374 	return error;
375 }
376 
377 #ifdef __MACHINE_STACK_GROWS_UP
378 #define STACK_PTHREADSPACE NBPG
379 #else
380 #define STACK_PTHREADSPACE 0
381 #endif
382 
383 static int
384 execve_fetch_element(char * const *array, size_t index, char **value)
385 {
386 	return copyin(array + index, value, sizeof(*value));
387 }
388 
389 /*
390  * exec system call
391  */
392 /* ARGSUSED */
393 int
394 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval)
395 {
396 	/* {
397 		syscallarg(const char *)	path;
398 		syscallarg(char * const *)	argp;
399 		syscallarg(char * const *)	envp;
400 	} */
401 
402 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
403 	    SCARG(uap, envp), execve_fetch_element);
404 }
405 
406 int
407 execve1(struct lwp *l, const char *path, char * const *args,
408     char * const *envs, execve_fetch_element_t fetch_element)
409 {
410 	int			error;
411 	struct exec_package	pack;
412 	struct nameidata	nid;
413 	struct vattr		attr;
414 	struct proc		*p;
415 	char			*argp;
416 	char			*dp, *sp;
417 	long			argc, envc;
418 	size_t			i, len;
419 	char			*stack;
420 	struct ps_strings	arginfo;
421 	struct ps_strings	*aip = &arginfo;
422 	struct vmspace		*vm;
423 	struct exec_fakearg	*tmpfap;
424 	int			szsigcode;
425 	struct exec_vmcmd	*base_vcp;
426 	ksiginfo_t		ksi;
427 	ksiginfoq_t		kq;
428 	char			*pathbuf;
429 	size_t			pathbuflen;
430 
431 	p = l->l_proc;
432 
433 	/*
434 	 * Drain existing references and forbid new ones.  The process
435 	 * should be left alone until we're done here.  This is necessary
436 	 * to avoid race conditions - e.g. in ptrace() - that might allow
437 	 * a local user to illicitly obtain elevated privileges.
438 	 */
439 	rw_enter(&p->p_reflock, RW_WRITER);
440 
441 	base_vcp = NULL;
442 	/*
443 	 * Init the namei data to point the file user's program name.
444 	 * This is done here rather than in check_exec(), so that it's
445 	 * possible to override this settings if any of makecmd/probe
446 	 * functions call check_exec() recursively - for example,
447 	 * see exec_script_makecmds().
448 	 */
449 	pathbuf = PNBUF_GET();
450 	error = copyinstr(path, pathbuf, MAXPATHLEN, &pathbuflen);
451 	if (error) {
452 		DPRINTF(("execve: copyinstr path %d", error));
453 		goto clrflg;
454 	}
455 
456 	NDINIT(&nid, LOOKUP, NOFOLLOW | TRYEMULROOT, UIO_SYSSPACE, pathbuf);
457 
458 	/*
459 	 * initialize the fields of the exec package.
460 	 */
461 	pack.ep_name = path;
462 	pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP);
463 	pack.ep_hdrlen = exec_maxhdrsz;
464 	pack.ep_hdrvalid = 0;
465 	pack.ep_ndp = &nid;
466 	pack.ep_emul_arg = NULL;
467 	pack.ep_vmcmds.evs_cnt = 0;
468 	pack.ep_vmcmds.evs_used = 0;
469 	pack.ep_vap = &attr;
470 	pack.ep_flags = 0;
471 	pack.ep_emul_root = NULL;
472 	pack.ep_interp = NULL;
473 	pack.ep_esch = NULL;
474 
475 #ifdef LKM
476 	rw_enter(&exec_lock, RW_READER);
477 #endif
478 
479 	/* see if we can run it. */
480 	if ((error = check_exec(l, &pack)) != 0) {
481 		if (error != ENOENT) {
482 			DPRINTF(("execve: check exec failed %d\n", error));
483 		}
484 		goto freehdr;
485 	}
486 
487 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
488 
489 	/* allocate an argument buffer */
490 	argp = (char *) uvm_km_alloc(exec_map, NCARGS, 0,
491 	    UVM_KMF_PAGEABLE|UVM_KMF_WAITVA);
492 #ifdef DIAGNOSTIC
493 	if (argp == NULL)
494 		panic("execve: argp == NULL");
495 #endif
496 	dp = argp;
497 	argc = 0;
498 
499 	/* copy the fake args list, if there's one, freeing it as we go */
500 	if (pack.ep_flags & EXEC_HASARGL) {
501 		tmpfap = pack.ep_fa;
502 		while (tmpfap->fa_arg != NULL) {
503 			const char *cp;
504 
505 			cp = tmpfap->fa_arg;
506 			while (*cp)
507 				*dp++ = *cp++;
508 			dp++;
509 
510 			kmem_free(tmpfap->fa_arg, tmpfap->fa_len);
511 			tmpfap++; argc++;
512 		}
513 		kmem_free(pack.ep_fa, pack.ep_fa_len);
514 		pack.ep_flags &= ~EXEC_HASARGL;
515 	}
516 
517 	/* Now get argv & environment */
518 	if (args == NULL) {
519 		DPRINTF(("execve: null args\n"));
520 		error = EINVAL;
521 		goto bad;
522 	}
523 	/* 'i' will index the argp/envp element to be retrieved */
524 	i = 0;
525 	if (pack.ep_flags & EXEC_SKIPARG)
526 		i++;
527 
528 	while (1) {
529 		len = argp + ARG_MAX - dp;
530 		if ((error = (*fetch_element)(args, i, &sp)) != 0) {
531 			DPRINTF(("execve: fetch_element args %d\n", error));
532 			goto bad;
533 		}
534 		if (!sp)
535 			break;
536 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
537 			DPRINTF(("execve: copyinstr args %d\n", error));
538 			if (error == ENAMETOOLONG)
539 				error = E2BIG;
540 			goto bad;
541 		}
542 		ktrexecarg(dp, len - 1);
543 		dp += len;
544 		i++;
545 		argc++;
546 	}
547 
548 	envc = 0;
549 	/* environment need not be there */
550 	if (envs != NULL) {
551 		i = 0;
552 		while (1) {
553 			len = argp + ARG_MAX - dp;
554 			if ((error = (*fetch_element)(envs, i, &sp)) != 0) {
555 				DPRINTF(("execve: fetch_element env %d\n", error));
556 				goto bad;
557 			}
558 			if (!sp)
559 				break;
560 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
561 				DPRINTF(("execve: copyinstr env %d\n", error));
562 				if (error == ENAMETOOLONG)
563 					error = E2BIG;
564 				goto bad;
565 			}
566 			ktrexecenv(dp, len - 1);
567 			dp += len;
568 			i++;
569 			envc++;
570 		}
571 	}
572 
573 	dp = (char *) ALIGN(dp);
574 
575 	szsigcode = pack.ep_esch->es_emul->e_esigcode -
576 	    pack.ep_esch->es_emul->e_sigcode;
577 
578 #ifdef __MACHINE_STACK_GROWS_UP
579 /* See big comment lower down */
580 #define	RTLD_GAP	32
581 #else
582 #define	RTLD_GAP	0
583 #endif
584 
585 	/* Now check if args & environ fit into new stack */
586 	if (pack.ep_flags & EXEC_32)
587 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
588 		    sizeof(int) + sizeof(int) + dp + RTLD_GAP +
589 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
590 		    - argp;
591 	else
592 		len = ((argc + envc + 2 + pack.ep_esch->es_arglen) *
593 		    sizeof(char *) + sizeof(int) + dp + RTLD_GAP +
594 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
595 		    - argp;
596 
597 #ifdef PAX_ASLR
598 	if (pax_aslr_active(l))
599 		len += (arc4random() % PAGE_SIZE);
600 #endif /* PAX_ASLR */
601 
602 #ifdef STACKLALIGN	/* arm, etc. */
603 	len = STACKALIGN(len);	/* make the stack "safely" aligned */
604 #else
605 	len = ALIGN(len);	/* make the stack "safely" aligned */
606 #endif
607 
608 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
609 		DPRINTF(("execve: stack limit exceeded %zu\n", len));
610 		error = ENOMEM;
611 		goto bad;
612 	}
613 
614 	/* Get rid of other LWPs. */
615 	if (p->p_nlwps > 1) {
616 		mutex_enter(&p->p_smutex);
617 		exit_lwps(l);
618 		mutex_exit(&p->p_smutex);
619 	}
620 	KDASSERT(p->p_nlwps == 1);
621 
622 	/* Destroy any lwpctl info. */
623 	if (p->p_lwpctl != NULL)
624 		lwp_ctl_exit();
625 
626 	/* This is now LWP 1 */
627 	l->l_lid = 1;
628 	p->p_nlwpid = 1;
629 
630 	/* Remove POSIX timers */
631 	timers_free(p, TIMERS_POSIX);
632 
633 	/* adjust "active stack depth" for process VSZ */
634 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
635 
636 	/*
637 	 * Do whatever is necessary to prepare the address space
638 	 * for remapping.  Note that this might replace the current
639 	 * vmspace with another!
640 	 */
641 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
642 
643 	/* record proc's vnode, for use by procfs and others */
644         if (p->p_textvp)
645                 vrele(p->p_textvp);
646 	VREF(pack.ep_vp);
647 	p->p_textvp = pack.ep_vp;
648 
649 	/* Now map address space */
650 	vm = p->p_vmspace;
651 	vm->vm_taddr = (void *)pack.ep_taddr;
652 	vm->vm_tsize = btoc(pack.ep_tsize);
653 	vm->vm_daddr = (void*)pack.ep_daddr;
654 	vm->vm_dsize = btoc(pack.ep_dsize);
655 	vm->vm_ssize = btoc(pack.ep_ssize);
656 	vm->vm_maxsaddr = (void *)pack.ep_maxsaddr;
657 	vm->vm_minsaddr = (void *)pack.ep_minsaddr;
658 
659 #ifdef PAX_ASLR
660 	pax_aslr_init(l, vm);
661 #endif /* PAX_ASLR */
662 
663 	/* create the new process's VM space by running the vmcmds */
664 #ifdef DIAGNOSTIC
665 	if (pack.ep_vmcmds.evs_used == 0)
666 		panic("execve: no vmcmds");
667 #endif
668 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
669 		struct exec_vmcmd *vcp;
670 
671 		vcp = &pack.ep_vmcmds.evs_cmds[i];
672 		if (vcp->ev_flags & VMCMD_RELATIVE) {
673 #ifdef DIAGNOSTIC
674 			if (base_vcp == NULL)
675 				panic("execve: relative vmcmd with no base");
676 			if (vcp->ev_flags & VMCMD_BASE)
677 				panic("execve: illegal base & relative vmcmd");
678 #endif
679 			vcp->ev_addr += base_vcp->ev_addr;
680 		}
681 		error = (*vcp->ev_proc)(l, vcp);
682 #ifdef DEBUG_EXEC
683 		if (error) {
684 			size_t j;
685 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
686 			for (j = 0; j <= i; j++)
687 				uprintf(
688 			"vmcmd[%zu] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
689 				    j, vp[j].ev_addr, vp[j].ev_len,
690 				    vp[j].ev_offset, vp[j].ev_prot,
691 				    vp[j].ev_flags);
692 		}
693 #endif /* DEBUG_EXEC */
694 		if (vcp->ev_flags & VMCMD_BASE)
695 			base_vcp = vcp;
696 	}
697 
698 	/* free the vmspace-creation commands, and release their references */
699 	kill_vmcmds(&pack.ep_vmcmds);
700 
701 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
702 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
703 	vput(pack.ep_vp);
704 
705 	/* if an error happened, deallocate and punt */
706 	if (error) {
707 		DPRINTF(("execve: vmcmd %zu failed: %d\n", i - 1, error));
708 		goto exec_abort;
709 	}
710 
711 	/* remember information about the process */
712 	arginfo.ps_nargvstr = argc;
713 	arginfo.ps_nenvstr = envc;
714 
715 	/* set command name & other accounting info */
716 	i = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
717 	(void)memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, i);
718 	p->p_comm[i] = '\0';
719 
720 	dp = PNBUF_GET();
721 	/*
722 	 * If the path starts with /, we don't need to do any work.
723 	 * This handles the majority of the cases.
724 	 * In the future perhaps we could canonicalize it?
725 	 */
726 	if (pathbuf[0] == '/')
727 		(void)strlcpy(pack.ep_path = dp, pathbuf, MAXPATHLEN);
728 #ifdef notyet
729 	/*
730 	 * Although this works most of the time [since the entry was just
731 	 * entered in the cache] we don't use it because it theoretically
732 	 * can fail and it is not the cleanest interface, because there
733 	 * could be races. When the namei cache is re-written, this can
734 	 * be changed to use the appropriate function.
735 	 */
736 	else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p)))
737 		pack.ep_path = dp;
738 #endif
739 	else {
740 #ifdef notyet
741 		printf("Cannot get path for pid %d [%s] (error %d)",
742 		    (int)p->p_pid, p->p_comm, error);
743 #endif
744 		pack.ep_path = NULL;
745 		PNBUF_PUT(dp);
746 	}
747 
748 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
749 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
750 		len - (sizeof(struct ps_strings) + szsigcode));
751 
752 #ifdef __MACHINE_STACK_GROWS_UP
753 	/*
754 	 * The copyargs call always copies into lower addresses
755 	 * first, moving towards higher addresses, starting with
756 	 * the stack pointer that we give.  When the stack grows
757 	 * down, this puts argc/argv/envp very shallow on the
758 	 * stack, right at the first user stack pointer.
759 	 * When the stack grows up, the situation is reversed.
760 	 *
761 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
762 	 * function expects to be called with a single pointer to
763 	 * a region that has a few words it can stash values into,
764 	 * followed by argc/argv/envp.  When the stack grows down,
765 	 * it's easy to decrement the stack pointer a little bit to
766 	 * allocate the space for these few words and pass the new
767 	 * stack pointer to _rtld.  When the stack grows up, however,
768 	 * a few words before argc is part of the signal trampoline, XXX
769 	 * so we have a problem.
770 	 *
771 	 * Instead of changing how _rtld works, we take the easy way
772 	 * out and steal 32 bytes before we call copyargs.
773 	 * This extra space was allowed for when 'len' was calculated.
774 	 */
775 	stack += RTLD_GAP;
776 #endif /* __MACHINE_STACK_GROWS_UP */
777 
778 	/* Now copy argc, args & environ to new stack */
779 	error = (*pack.ep_esch->es_copyargs)(l, &pack, &arginfo, &stack, argp);
780 	if (pack.ep_path) {
781 		PNBUF_PUT(pack.ep_path);
782 		pack.ep_path = NULL;
783 	}
784 	if (error) {
785 		DPRINTF(("execve: copyargs failed %d\n", error));
786 		goto exec_abort;
787 	}
788 	/* Move the stack back to original point */
789 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
790 
791 	/* fill process ps_strings info */
792 	p->p_psstr = (struct ps_strings *)
793 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
794 	    sizeof(struct ps_strings));
795 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
796 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
797 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
798 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
799 
800 	/* copy out the process's ps_strings structure */
801 	if ((error = copyout(aip, (char *)p->p_psstr,
802 	    sizeof(arginfo))) != 0) {
803 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
804 		       aip, (char *)p->p_psstr, (long)sizeof(arginfo)));
805 		goto exec_abort;
806 	}
807 
808 	fdcloseexec(l);		/* handle close on exec */
809 	execsigs(p);		/* reset catched signals */
810 
811 	l->l_ctxlink = NULL;	/* reset ucontext link */
812 
813 
814 	p->p_acflag &= ~AFORK;
815 	mutex_enter(&p->p_mutex);
816 	p->p_flag |= PK_EXEC;
817 	mutex_exit(&p->p_mutex);
818 
819 	/*
820 	 * Stop profiling.
821 	 */
822 	if ((p->p_stflag & PST_PROFIL) != 0) {
823 		mutex_spin_enter(&p->p_stmutex);
824 		stopprofclock(p);
825 		mutex_spin_exit(&p->p_stmutex);
826 	}
827 
828 	/*
829 	 * It's OK to test PS_PPWAIT unlocked here, as other LWPs have
830 	 * exited and exec()/exit() are the only places it will be cleared.
831 	 */
832 	if ((p->p_sflag & PS_PPWAIT) != 0) {
833 		mutex_enter(&proclist_lock);
834 		mutex_enter(&p->p_smutex);
835 		p->p_sflag &= ~PS_PPWAIT;
836 		cv_broadcast(&p->p_pptr->p_waitcv);
837 		mutex_exit(&p->p_smutex);
838 		mutex_exit(&proclist_lock);
839 	}
840 
841 	/*
842 	 * Deal with set[ug]id.  MNT_NOSUID has already been used to disable
843 	 * s[ug]id.  It's OK to check for PSL_TRACED here as we have blocked
844 	 * out additional references on the process for the moment.
845 	 */
846 	if ((p->p_slflag & PSL_TRACED) == 0 &&
847 
848 	    (((attr.va_mode & S_ISUID) != 0 &&
849 	      kauth_cred_geteuid(l->l_cred) != attr.va_uid) ||
850 
851 	     ((attr.va_mode & S_ISGID) != 0 &&
852 	      kauth_cred_getegid(l->l_cred) != attr.va_gid))) {
853 		/*
854 		 * Mark the process as SUGID before we do
855 		 * anything that might block.
856 		 */
857 		proc_crmod_enter();
858 		proc_crmod_leave(NULL, NULL, true);
859 
860 		/* Make sure file descriptors 0..2 are in use. */
861 		if ((error = fdcheckstd(l)) != 0) {
862 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
863 			goto exec_abort;
864 		}
865 
866 		/*
867 		 * Copy the credential so other references don't see our
868 		 * changes.
869 		 */
870 		l->l_cred = kauth_cred_copy(l->l_cred);
871 #ifdef KTRACE
872 		/*
873 		 * If the persistent trace flag isn't set, turn off.
874 		 */
875 		if (p->p_tracep) {
876 			mutex_enter(&ktrace_lock);
877 			if (!(p->p_traceflag & KTRFAC_PERSISTENT))
878 				ktrderef(p);
879 			mutex_exit(&ktrace_lock);
880 		}
881 #endif
882 		if (attr.va_mode & S_ISUID)
883 			kauth_cred_seteuid(l->l_cred, attr.va_uid);
884 		if (attr.va_mode & S_ISGID)
885 			kauth_cred_setegid(l->l_cred, attr.va_gid);
886 	} else {
887 		if (kauth_cred_geteuid(l->l_cred) ==
888 		    kauth_cred_getuid(l->l_cred) &&
889 		    kauth_cred_getegid(l->l_cred) ==
890 		    kauth_cred_getgid(l->l_cred))
891 			p->p_flag &= ~PK_SUGID;
892 	}
893 
894 	/*
895 	 * Copy the credential so other references don't see our changes.
896 	 * Test to see if this is necessary first, since in the common case
897 	 * we won't need a private reference.
898 	 */
899 	if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) ||
900 	    kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) {
901 		l->l_cred = kauth_cred_copy(l->l_cred);
902 		kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred));
903 		kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred));
904 	}
905 
906 	/* Update the master credentials. */
907 	if (l->l_cred != p->p_cred) {
908 		kauth_cred_t ocred;
909 
910 		kauth_cred_hold(l->l_cred);
911 		mutex_enter(&p->p_mutex);
912 		ocred = p->p_cred;
913 		p->p_cred = l->l_cred;
914 		mutex_exit(&p->p_mutex);
915 		kauth_cred_free(ocred);
916 	}
917 
918 #if defined(__HAVE_RAS)
919 	/*
920 	 * Remove all RASs from the address space.
921 	 */
922 	ras_purgeall();
923 #endif
924 
925 	doexechooks(p);
926 
927 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
928 
929 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
930 
931 	/* notify others that we exec'd */
932 	KNOTE(&p->p_klist, NOTE_EXEC);
933 
934 	/* setup new registers and do misc. setup. */
935 	(*pack.ep_esch->es_emul->e_setregs)(l, &pack, (u_long) stack);
936 	if (pack.ep_esch->es_setregs)
937 		(*pack.ep_esch->es_setregs)(l, &pack, (u_long) stack);
938 
939 	/* map the process's signal trampoline code */
940 	if (exec_sigcode_map(p, pack.ep_esch->es_emul)) {
941 		DPRINTF(("execve: map sigcode failed %d\n", error));
942 		goto exec_abort;
943 	}
944 
945 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
946 
947 	/* The emulation root will usually have been found when we looked
948 	 * for the elf interpreter (or similar), if not look now. */
949 	if (pack.ep_esch->es_emul->e_path != NULL && pack.ep_emul_root == NULL)
950 		emul_find_root(l, &pack);
951 
952 	/* Any old emulation root got removed by fdcloseexec */
953 	rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER);
954 	p->p_cwdi->cwdi_edir = pack.ep_emul_root;
955 	rw_exit(&p->p_cwdi->cwdi_lock);
956 	pack.ep_emul_root = NULL;
957 	if (pack.ep_interp != NULL)
958 		vrele(pack.ep_interp);
959 
960 	/*
961 	 * Call emulation specific exec hook. This can setup per-process
962 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
963 	 *
964 	 * If we are executing process of different emulation than the
965 	 * original forked process, call e_proc_exit() of the old emulation
966 	 * first, then e_proc_exec() of new emulation. If the emulation is
967 	 * same, the exec hook code should deallocate any old emulation
968 	 * resources held previously by this process.
969 	 */
970 	if (p->p_emul && p->p_emul->e_proc_exit
971 	    && p->p_emul != pack.ep_esch->es_emul)
972 		(*p->p_emul->e_proc_exit)(p);
973 
974 	/*
975 	 * Call exec hook. Emulation code may NOT store reference to anything
976 	 * from &pack.
977 	 */
978         if (pack.ep_esch->es_emul->e_proc_exec)
979                 (*pack.ep_esch->es_emul->e_proc_exec)(p, &pack);
980 
981 	/* update p_emul, the old value is no longer needed */
982 	p->p_emul = pack.ep_esch->es_emul;
983 
984 	/* ...and the same for p_execsw */
985 	p->p_execsw = pack.ep_esch;
986 
987 #ifdef __HAVE_SYSCALL_INTERN
988 	(*p->p_emul->e_syscall_intern)(p);
989 #endif
990 	ktremul();
991 
992 	/* Allow new references from the debugger/procfs. */
993 	rw_exit(&p->p_reflock);
994 #ifdef LKM
995 	rw_exit(&exec_lock);
996 #endif
997 
998 	mutex_enter(&proclist_mutex);
999 
1000 	if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) {
1001 		KSI_INIT_EMPTY(&ksi);
1002 		ksi.ksi_signo = SIGTRAP;
1003 		ksi.ksi_lid = l->l_lid;
1004 		kpsignal(p, &ksi, NULL);
1005 	}
1006 
1007 	if (p->p_sflag & PS_STOPEXEC) {
1008 		KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
1009 		p->p_pptr->p_nstopchild++;
1010 		p->p_pptr->p_waited = 0;
1011 		mutex_enter(&p->p_smutex);
1012 		ksiginfo_queue_init(&kq);
1013 		sigclearall(p, &contsigmask, &kq);
1014 		lwp_lock(l);
1015 		l->l_stat = LSSTOP;
1016 		p->p_stat = SSTOP;
1017 		p->p_nrlwps--;
1018 		mutex_exit(&p->p_smutex);
1019 		mutex_exit(&proclist_mutex);
1020 		mi_switch(l);
1021 		ksiginfo_queue_drain(&kq);
1022 		KERNEL_LOCK(l->l_biglocks, l);
1023 	} else {
1024 		mutex_exit(&proclist_mutex);
1025 	}
1026 
1027 	PNBUF_PUT(pathbuf);
1028 	return (EJUSTRETURN);
1029 
1030  bad:
1031 	/* free the vmspace-creation commands, and release their references */
1032 	kill_vmcmds(&pack.ep_vmcmds);
1033 	/* kill any opened file descriptor, if necessary */
1034 	if (pack.ep_flags & EXEC_HASFD) {
1035 		pack.ep_flags &= ~EXEC_HASFD;
1036 		(void) fdrelease(l, pack.ep_fd);
1037 	}
1038 	/* close and put the exec'd file */
1039 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
1040 	VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred);
1041 	vput(pack.ep_vp);
1042 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1043 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
1044 
1045  freehdr:
1046 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1047 	if (pack.ep_emul_root != NULL)
1048 		vrele(pack.ep_emul_root);
1049 	if (pack.ep_interp != NULL)
1050 		vrele(pack.ep_interp);
1051 
1052  clrflg:
1053 	PNBUF_PUT(pathbuf);
1054 	rw_exit(&p->p_reflock);
1055 #ifdef LKM
1056 	rw_exit(&exec_lock);
1057 #endif
1058 
1059 	return error;
1060 
1061  exec_abort:
1062 	PNBUF_PUT(pathbuf);
1063 	rw_exit(&p->p_reflock);
1064 #ifdef LKM
1065 	rw_exit(&exec_lock);
1066 #endif
1067 
1068 	/*
1069 	 * the old process doesn't exist anymore.  exit gracefully.
1070 	 * get rid of the (new) address space we have created, if any, get rid
1071 	 * of our namei data and vnode, and exit noting failure
1072 	 */
1073 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
1074 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
1075 	if (pack.ep_emul_arg)
1076 		FREE(pack.ep_emul_arg, M_TEMP);
1077 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
1078 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
1079 	kmem_free(pack.ep_hdr, pack.ep_hdrlen);
1080 	if (pack.ep_emul_root != NULL)
1081 		vrele(pack.ep_emul_root);
1082 	if (pack.ep_interp != NULL)
1083 		vrele(pack.ep_interp);
1084 
1085 	/* Acquire the sched-state mutex (exit1() will release it). */
1086 	KERNEL_LOCK(1, NULL);	/* XXXSMP */
1087 	mutex_enter(&p->p_smutex);
1088 	exit1(l, W_EXITCODE(error, SIGABRT));
1089 
1090 	/* NOTREACHED */
1091 	return 0;
1092 }
1093 
1094 
1095 int
1096 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo,
1097     char **stackp, void *argp)
1098 {
1099 	char	**cpp, *dp, *sp;
1100 	size_t	len;
1101 	void	*nullp;
1102 	long	argc, envc;
1103 	int	error;
1104 
1105 	cpp = (char **)*stackp;
1106 	nullp = NULL;
1107 	argc = arginfo->ps_nargvstr;
1108 	envc = arginfo->ps_nenvstr;
1109 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
1110 		return error;
1111 
1112 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen);
1113 	sp = argp;
1114 
1115 	/* XXX don't copy them out, remap them! */
1116 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
1117 
1118 	for (; --argc >= 0; sp += len, dp += len)
1119 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1120 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1121 			return error;
1122 
1123 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1124 		return error;
1125 
1126 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
1127 
1128 	for (; --envc >= 0; sp += len, dp += len)
1129 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
1130 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
1131 			return error;
1132 
1133 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
1134 		return error;
1135 
1136 	*stackp = (char *)cpp;
1137 	return 0;
1138 }
1139 
1140 #ifdef LKM
1141 /*
1142  * Find an emulation of given name in list of emulations.
1143  * Needs to be called with the exec_lock held.
1144  */
1145 const struct emul *
1146 emul_search(const char *name)
1147 {
1148 	struct emul_entry *it;
1149 
1150 	LIST_FOREACH(it, &el_head, el_list) {
1151 		if (strcmp(name, it->el_emul->e_name) == 0)
1152 			return it->el_emul;
1153 	}
1154 
1155 	return NULL;
1156 }
1157 
1158 /*
1159  * Add an emulation to list, if it's not there already.
1160  */
1161 int
1162 emul_register(const struct emul *emul, int ro_entry)
1163 {
1164 	struct emul_entry	*ee;
1165 	int			error;
1166 
1167 	error = 0;
1168 	rw_enter(&exec_lock, RW_WRITER);
1169 
1170 	if (emul_search(emul->e_name)) {
1171 		error = EEXIST;
1172 		goto out;
1173 	}
1174 
1175 	ee = kmem_alloc(sizeof(*ee), KM_SLEEP);
1176 	ee->el_emul = emul;
1177 	ee->ro_entry = ro_entry;
1178 	LIST_INSERT_HEAD(&el_head, ee, el_list);
1179 
1180  out:
1181 	rw_exit(&exec_lock);
1182 	return error;
1183 }
1184 
1185 /*
1186  * Remove emulation with name 'name' from list of supported emulations.
1187  */
1188 int
1189 emul_unregister(const char *name)
1190 {
1191 	const struct proclist_desc *pd;
1192 	struct emul_entry	*it;
1193 	int			i, error;
1194 	struct proc		*ptmp;
1195 
1196 	error = 0;
1197 	rw_enter(&exec_lock, RW_WRITER);
1198 
1199 	LIST_FOREACH(it, &el_head, el_list) {
1200 		if (strcmp(it->el_emul->e_name, name) == 0)
1201 			break;
1202 	}
1203 
1204 	if (!it) {
1205 		error = ENOENT;
1206 		goto out;
1207 	}
1208 
1209 	if (it->ro_entry) {
1210 		error = EBUSY;
1211 		goto out;
1212 	}
1213 
1214 	/* test if any execw[] entry is still using this */
1215 	for(i=0; i < nexecs; i++) {
1216 		if (execsw[i]->es_emul == it->el_emul) {
1217 			error = EBUSY;
1218 			goto out;
1219 		}
1220 	}
1221 
1222 	/*
1223 	 * Test if any process is running under this emulation - since
1224 	 * emul_unregister() is running quite sendomly, it's better
1225 	 * to do expensive check here than to use any locking.
1226 	 */
1227 	mutex_enter(&proclist_lock);
1228 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
1229 		PROCLIST_FOREACH(ptmp, pd->pd_list) {
1230 			if (ptmp->p_emul == it->el_emul) {
1231 				error = EBUSY;
1232 				break;
1233 			}
1234 		}
1235 	}
1236 	mutex_exit(&proclist_lock);
1237 
1238 	if (error)
1239 		goto out;
1240 
1241 
1242 	/* entry is not used, remove it */
1243 	LIST_REMOVE(it, el_list);
1244 	kmem_free(it, sizeof(*it));
1245 
1246  out:
1247 	rw_exit(&exec_lock);
1248 	return error;
1249 }
1250 
1251 /*
1252  * Add execsw[] entry.
1253  */
1254 int
1255 exec_add(struct execsw *esp, const char *e_name)
1256 {
1257 	struct exec_entry	*it;
1258 	int			error;
1259 
1260 	error = 0;
1261 	rw_enter(&exec_lock, RW_WRITER);
1262 
1263 	if (!esp->es_emul) {
1264 		esp->es_emul = emul_search(e_name);
1265 		if (!esp->es_emul) {
1266 			error = ENOENT;
1267 			goto out;
1268 		}
1269 	}
1270 
1271 	LIST_FOREACH(it, &ex_head, ex_list) {
1272 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1273 		if (it->es->es_makecmds == esp->es_makecmds
1274 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1275 		    && it->es->es_emul == esp->es_emul) {
1276 			error = EEXIST;
1277 			goto out;
1278 		}
1279 	}
1280 
1281 	/* if we got here, the entry doesn't exist yet */
1282 	it = kmem_alloc(sizeof(*it), KM_SLEEP);
1283 	it->es = esp;
1284 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
1285 
1286 	/* update execsw[] */
1287 	exec_init(0);
1288 
1289  out:
1290 	rw_exit(&exec_lock);
1291 	return error;
1292 }
1293 
1294 /*
1295  * Remove execsw[] entry.
1296  */
1297 int
1298 exec_remove(const struct execsw *esp)
1299 {
1300 	struct exec_entry	*it;
1301 	int			error;
1302 
1303 	error = 0;
1304 	rw_enter(&exec_lock, RW_WRITER);
1305 
1306 	LIST_FOREACH(it, &ex_head, ex_list) {
1307 		/* assume tuple (makecmds, probe_func, emulation) is unique */
1308 		if (it->es->es_makecmds == esp->es_makecmds
1309 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
1310 		    && it->es->es_emul == esp->es_emul)
1311 			break;
1312 	}
1313 	if (!it) {
1314 		error = ENOENT;
1315 		goto out;
1316 	}
1317 
1318 	/* remove item from list and free resources */
1319 	LIST_REMOVE(it, ex_list);
1320 	kmem_free(it, sizeof(*it));
1321 
1322 	/* update execsw[] */
1323 	exec_init(0);
1324 
1325  out:
1326 	rw_exit(&exec_lock);
1327 	return error;
1328 }
1329 
1330 static void
1331 link_es(struct execsw_entry **listp, const struct execsw *esp)
1332 {
1333 	struct execsw_entry *et, *e1;
1334 
1335 	et = (struct execsw_entry *) malloc(sizeof(struct execsw_entry),
1336 			M_TEMP, M_WAITOK);
1337 	et->next = NULL;
1338 	et->es = esp;
1339 	if (*listp == NULL) {
1340 		*listp = et;
1341 		return;
1342 	}
1343 
1344 	switch(et->es->es_prio) {
1345 	case EXECSW_PRIO_FIRST:
1346 		/* put new entry as the first */
1347 		et->next = *listp;
1348 		*listp = et;
1349 		break;
1350 	case EXECSW_PRIO_ANY:
1351 		/* put new entry after all *_FIRST and *_ANY entries */
1352 		for(e1 = *listp; e1->next
1353 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
1354 			e1 = e1->next);
1355 		et->next = e1->next;
1356 		e1->next = et;
1357 		break;
1358 	case EXECSW_PRIO_LAST:
1359 		/* put new entry as the last one */
1360 		for(e1 = *listp; e1->next; e1 = e1->next);
1361 		e1->next = et;
1362 		break;
1363 	default:
1364 #ifdef DIAGNOSTIC
1365 		panic("execw[] entry with unknown priority %d found",
1366 			et->es->es_prio);
1367 #else
1368 		free(et, M_TEMP);
1369 #endif
1370 		break;
1371 	}
1372 }
1373 
1374 /*
1375  * Initialize exec structures. If init_boot is true, also does necessary
1376  * one-time initialization (it's called from main() that way).
1377  * Once system is multiuser, this should be called with exec_lock held,
1378  * i.e. via exec_{add|remove}().
1379  */
1380 int
1381 exec_init(int init_boot)
1382 {
1383 	const struct execsw	**new_es, * const *old_es;
1384 	struct execsw_entry	*list, *e1;
1385 	struct exec_entry	*e2;
1386 	int			i, es_sz;
1387 
1388 	if (init_boot) {
1389 		/* do one-time initializations */
1390 		rw_init(&exec_lock);
1391 		mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE);
1392 
1393 		/* register compiled-in emulations */
1394 		for(i=0; i < nexecs_builtin; i++) {
1395 			if (execsw_builtin[i].es_emul)
1396 				emul_register(execsw_builtin[i].es_emul, 1);
1397 		}
1398 #ifdef DIAGNOSTIC
1399 		if (i == 0)
1400 			panic("no emulations found in execsw_builtin[]");
1401 #endif
1402 	}
1403 
1404 	/*
1405 	 * Build execsw[] array from builtin entries and entries added
1406 	 * at runtime.
1407 	 */
1408 	list = NULL;
1409 	for(i=0; i < nexecs_builtin; i++)
1410 		link_es(&list, &execsw_builtin[i]);
1411 
1412 	/* Add dynamically loaded entries */
1413 	es_sz = nexecs_builtin;
1414 	LIST_FOREACH(e2, &ex_head, ex_list) {
1415 		link_es(&list, e2->es);
1416 		es_sz++;
1417 	}
1418 
1419 	/*
1420 	 * Now that we have sorted all execw entries, create new execsw[]
1421 	 * and free no longer needed memory in the process.
1422 	 */
1423 	new_es = kmem_alloc(es_sz * sizeof(struct execsw *), KM_SLEEP);
1424 	for(i=0; list; i++) {
1425 		new_es[i] = list->es;
1426 		e1 = list->next;
1427 		free(list, M_TEMP);
1428 		list = e1;
1429 	}
1430 
1431 	/*
1432 	 * New execsw[] array built, now replace old execsw[] and free
1433 	 * used memory.
1434 	 */
1435 	old_es = execsw;
1436 	if (old_es)
1437 		/*XXXUNCONST*/
1438 		kmem_free(__UNCONST(old_es), nexecs * sizeof(struct execsw *));
1439 	execsw = new_es;
1440 	nexecs = es_sz;
1441 
1442 	/*
1443 	 * Figure out the maximum size of an exec header.
1444 	 */
1445 	exec_maxhdrsz = 0;
1446 	for (i = 0; i < nexecs; i++) {
1447 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
1448 			exec_maxhdrsz = execsw[i]->es_hdrsz;
1449 	}
1450 
1451 	return 0;
1452 }
1453 #endif
1454 
1455 #ifndef LKM
1456 /*
1457  * Simplified exec_init() for kernels without LKMs. Only initialize
1458  * exec_maxhdrsz and execsw[].
1459  */
1460 int
1461 exec_init(int init_boot)
1462 {
1463 	int i;
1464 
1465 #ifdef DIAGNOSTIC
1466 	if (!init_boot)
1467 		panic("exec_init(): called with init_boot == 0");
1468 #endif
1469 
1470 	/* do one-time initializations */
1471 	nexecs = nexecs_builtin;
1472 	execsw = kmem_alloc(nexecs * sizeof(struct execsw *), KM_SLEEP);
1473 
1474 	/*
1475 	 * Fill in execsw[] and figure out the maximum size of an exec header.
1476 	 */
1477 	exec_maxhdrsz = 0;
1478 	for(i=0; i < nexecs; i++) {
1479 		execsw[i] = &execsw_builtin[i];
1480 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
1481 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
1482 	}
1483 
1484 	return 0;
1485 
1486 }
1487 #endif /* !LKM */
1488 
1489 static int
1490 exec_sigcode_map(struct proc *p, const struct emul *e)
1491 {
1492 	vaddr_t va;
1493 	vsize_t sz;
1494 	int error;
1495 	struct uvm_object *uobj;
1496 
1497 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
1498 
1499 	if (e->e_sigobject == NULL || sz == 0) {
1500 		return 0;
1501 	}
1502 
1503 	/*
1504 	 * If we don't have a sigobject for this emulation, create one.
1505 	 *
1506 	 * sigobject is an anonymous memory object (just like SYSV shared
1507 	 * memory) that we keep a permanent reference to and that we map
1508 	 * in all processes that need this sigcode. The creation is simple,
1509 	 * we create an object, add a permanent reference to it, map it in
1510 	 * kernel space, copy out the sigcode to it and unmap it.
1511 	 * We map it with PROT_READ|PROT_EXEC into the process just
1512 	 * the way sys_mmap() would map it.
1513 	 */
1514 
1515 	uobj = *e->e_sigobject;
1516 	if (uobj == NULL) {
1517 		mutex_enter(&sigobject_lock);
1518 		if ((uobj = *e->e_sigobject) == NULL) {
1519 			uobj = uao_create(sz, 0);
1520 			(*uobj->pgops->pgo_reference)(uobj);
1521 			va = vm_map_min(kernel_map);
1522 			if ((error = uvm_map(kernel_map, &va, round_page(sz),
1523 			    uobj, 0, 0,
1524 			    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
1525 			    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
1526 				printf("kernel mapping failed %d\n", error);
1527 				(*uobj->pgops->pgo_detach)(uobj);
1528 				mutex_exit(&sigobject_lock);
1529 				return (error);
1530 			}
1531 			memcpy((void *)va, e->e_sigcode, sz);
1532 #ifdef PMAP_NEED_PROCWR
1533 			pmap_procwr(&proc0, va, sz);
1534 #endif
1535 			uvm_unmap(kernel_map, va, va + round_page(sz));
1536 			*e->e_sigobject = uobj;
1537 		}
1538 		mutex_exit(&sigobject_lock);
1539 	}
1540 
1541 	/* Just a hint to uvm_map where to put it. */
1542 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
1543 	    round_page(sz));
1544 
1545 #ifdef __alpha__
1546 	/*
1547 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
1548 	 * which causes the above calculation to put the sigcode at
1549 	 * an invalid address.  Put it just below the text instead.
1550 	 */
1551 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
1552 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
1553 	}
1554 #endif
1555 
1556 	(*uobj->pgops->pgo_reference)(uobj);
1557 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
1558 			uobj, 0, 0,
1559 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
1560 				    UVM_ADV_RANDOM, 0));
1561 	if (error) {
1562 		(*uobj->pgops->pgo_detach)(uobj);
1563 		return (error);
1564 	}
1565 	p->p_sigctx.ps_sigcode = (void *)va;
1566 	return (0);
1567 }
1568