xref: /titanic_44/usr/src/uts/common/os/exec.c (revision 8eea8e29cc4374d1ee24c25a07f45af132db3499)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*	Copyright (c) 1988 AT&T	*/
30 /*	  All Rights Reserved  	*/
31 
32 
33 #include <sys/types.h>
34 #include <sys/param.h>
35 #include <sys/sysmacros.h>
36 #include <sys/systm.h>
37 #include <sys/signal.h>
38 #include <sys/cred_impl.h>
39 #include <sys/policy.h>
40 #include <sys/user.h>
41 #include <sys/errno.h>
42 #include <sys/file.h>
43 #include <sys/vfs.h>
44 #include <sys/vnode.h>
45 #include <sys/mman.h>
46 #include <sys/acct.h>
47 #include <sys/cpuvar.h>
48 #include <sys/proc.h>
49 #include <sys/cmn_err.h>
50 #include <sys/debug.h>
51 #include <sys/pathname.h>
52 #include <sys/vm.h>
53 #include <sys/vtrace.h>
54 #include <sys/exec.h>
55 #include <sys/exechdr.h>
56 #include <sys/kmem.h>
57 #include <sys/prsystm.h>
58 #include <sys/modctl.h>
59 #include <sys/vmparam.h>
60 #include <sys/schedctl.h>
61 #include <sys/utrap.h>
62 #include <sys/systeminfo.h>
63 #include <sys/stack.h>
64 #include <sys/rctl.h>
65 #include <sys/dtrace.h>
66 #include <sys/lwpchan_impl.h>
67 #include <sys/pool.h>
68 #include <sys/sdt.h>
69 
70 #include <c2/audit.h>
71 
72 #include <vm/hat.h>
73 #include <vm/anon.h>
74 #include <vm/as.h>
75 #include <vm/seg.h>
76 #include <vm/seg_vn.h>
77 
78 #define	PRIV_RESET		0x01	/* needs to reset privs */
79 #define	PRIV_SETID		0x02	/* needs to change uids */
80 #define	PRIV_SETUGID		0x04	/* is setuid/setgid/forced privs */
81 #define	PRIV_INCREASE		0x08	/* child runs with more privs */
82 
83 static int execsetid(struct vnode *, struct vattr *, uid_t *, uid_t *);
84 static int hold_execsw(struct execsw *);
85 
86 uint_t auxv_hwcap = 0;	/* auxv AT_SUN_HWCAP value; determined on the fly */
87 #if defined(_SYSCALL32_IMPL)
88 uint_t auxv_hwcap32 = 0;	/* 32-bit version of auxv_hwcap */
89 #endif
90 
91 #if defined(__i386) || defined(__amd64)
92 extern void ldt_free(proc_t *p);
93 extern void ldt_load(void);
94 #endif
95 
96 int exec_lpg_disable = 0;
97 
98 #define	PSUIDFLAGS		(SNOCD|SUGID)
99 
100 /*
101  * exec() - wrapper around exece providing NULL environment pointer
102  */
103 int
104 exec(const char *fname, const char **argp)
105 {
106 	return (exece(fname, argp, NULL));
107 }
108 
109 /*
110  * exece() - system call wrapper around exec_common()
111  */
112 int
113 exece(const char *fname, const char **argp, const char **envp)
114 {
115 	int error;
116 
117 	error = exec_common(fname, argp, envp);
118 	return (error ? (set_errno(error)) : 0);
119 }
120 
121 int
122 exec_common(const char *fname, const char **argp, const char **envp)
123 {
124 	vnode_t *vp = NULL, *dir = NULL, *tmpvp = NULL;
125 	proc_t *p = ttoproc(curthread);
126 	klwp_t *lwp = ttolwp(curthread);
127 	struct user *up = PTOU(p);
128 	long execsz;		/* temporary count of exec size */
129 	int i;
130 	int error;
131 	char exec_file[MAXCOMLEN+1];
132 	struct pathname pn;
133 	struct pathname resolvepn;
134 	struct uarg args;
135 	struct execa ua;
136 	k_sigset_t savedmask;
137 	lwpdir_t *lwpdir = NULL;
138 	lwpdir_t **tidhash;
139 	lwpdir_t *old_lwpdir = NULL;
140 	uint_t old_lwpdir_sz;
141 	lwpdir_t **old_tidhash;
142 	uint_t old_tidhash_sz;
143 	lwpent_t *lep;
144 
145 	/*
146 	 * exec() is not supported for the /proc agent lwp.
147 	 */
148 	if (curthread == p->p_agenttp)
149 		return (ENOTSUP);
150 
151 	if ((error = secpolicy_basic_exec(CRED())) != 0)
152 		return (error);
153 
154 	/*
155 	 * Inform /proc that an exec() has started.
156 	 * Hold signals that are ignored by default so that we will
157 	 * not be interrupted by a signal that will be ignored after
158 	 * successful completion of gexec().
159 	 */
160 	mutex_enter(&p->p_lock);
161 	prexecstart();
162 	schedctl_finish_sigblock(curthread);
163 	savedmask = curthread->t_hold;
164 	sigorset(&curthread->t_hold, &ignoredefault);
165 	mutex_exit(&p->p_lock);
166 
167 	/*
168 	 * Look up path name and remember last component for later.
169 	 * To help coreadm expand its %d token, we attempt to save
170 	 * the directory containing the executable in p_execdir. The
171 	 * first call to lookuppn() may fail and return EINVAL because
172 	 * dirvpp is non-NULL. In that case, we make a second call to
173 	 * lookuppn() with dirvpp set to NULL; p_execdir will be NULL,
174 	 * but coreadm is allowed to expand %d to the empty string and
175 	 * there are other cases in which that failure may occur.
176 	 */
177 	if ((error = pn_get((char *)fname, UIO_USERSPACE, &pn)) != 0)
178 		goto out;
179 	pn_alloc(&resolvepn);
180 	if ((error = lookuppn(&pn, &resolvepn, FOLLOW, &dir, &vp)) != 0) {
181 		pn_free(&resolvepn);
182 		pn_free(&pn);
183 		if (error != EINVAL)
184 			goto out;
185 
186 		dir = NULL;
187 		if ((error = pn_get((char *)fname, UIO_USERSPACE, &pn)) != 0)
188 			goto out;
189 		pn_alloc(&resolvepn);
190 		if ((error = lookuppn(&pn, &resolvepn, FOLLOW, NULLVPP,
191 		    &vp)) != 0) {
192 			pn_free(&resolvepn);
193 			pn_free(&pn);
194 			goto out;
195 		}
196 	}
197 	if (vp == NULL) {
198 		if (dir != NULL)
199 			VN_RELE(dir);
200 		error = ENOENT;
201 		pn_free(&resolvepn);
202 		pn_free(&pn);
203 		goto out;
204 	}
205 	bzero(exec_file, MAXCOMLEN+1);
206 	(void) strncpy(exec_file, pn.pn_path, MAXCOMLEN);
207 	bzero(&args, sizeof (args));
208 	args.pathname = resolvepn.pn_path;
209 	/* don't free resolvepn until we are done with args */
210 	pn_free(&pn);
211 
212 	/*
213 	 * Specific exec handlers, or policies determined via
214 	 * /etc/system may override the historical default.
215 	 */
216 	args.stk_prot = PROT_ZFOD;
217 	args.dat_prot = PROT_ZFOD;
218 
219 	CPU_STATS_ADD_K(sys, sysexec, 1);
220 	DTRACE_PROC1(exec, char *, args.pathname);
221 
222 	ua.fname = fname;
223 	ua.argp = argp;
224 	ua.envp = envp;
225 
226 	if ((error = gexec(&vp, &ua, &args, NULL, 0, &execsz,
227 	    exec_file, p->p_cred)) != 0) {
228 		VN_RELE(vp);
229 		if (dir != NULL)
230 			VN_RELE(dir);
231 		pn_free(&resolvepn);
232 		goto fail;
233 	}
234 
235 	/*
236 	 * Free floating point registers (sun4u only)
237 	 */
238 	ASSERT(lwp != NULL);
239 	lwp_freeregs(lwp, 1);
240 
241 	/*
242 	 * Free device context
243 	 */
244 	if (curthread->t_ctx)
245 		freectx(curthread, 1);
246 
247 	/*
248 	 * Remember file name for accounting; clear any cached DTrace predicate.
249 	 */
250 	up->u_acflag &= ~AFORK;
251 	bcopy(exec_file, up->u_comm, MAXCOMLEN+1);
252 	curthread->t_predcache = NULL;
253 
254 	/*
255 	 * Clear contract template state
256 	 */
257 	lwp_ctmpl_clear(lwp);
258 
259 	/*
260 	 * Save the directory in which we found the executable for expanding
261 	 * the %d token used in core file patterns.
262 	 */
263 	mutex_enter(&p->p_lock);
264 	tmpvp = p->p_execdir;
265 	p->p_execdir = dir;
266 	if (p->p_execdir != NULL)
267 		VN_HOLD(p->p_execdir);
268 	mutex_exit(&p->p_lock);
269 
270 	if (tmpvp != NULL)
271 		VN_RELE(tmpvp);
272 
273 	/*
274 	 * Reset stack state to the user stack, clear set of signals
275 	 * caught on the signal stack, and reset list of signals that
276 	 * restart system calls; the new program's environment should
277 	 * not be affected by detritus from the old program.  Any
278 	 * pending held signals remain held, so don't clear t_hold.
279 	 */
280 	mutex_enter(&p->p_lock);
281 	lwp->lwp_oldcontext = 0;
282 	lwp->lwp_ustack = 0;
283 	lwp->lwp_old_stk_ctl = 0;
284 	sigemptyset(&up->u_signodefer);
285 	sigemptyset(&up->u_sigonstack);
286 	sigemptyset(&up->u_sigresethand);
287 	lwp->lwp_sigaltstack.ss_sp = 0;
288 	lwp->lwp_sigaltstack.ss_size = 0;
289 	lwp->lwp_sigaltstack.ss_flags = SS_DISABLE;
290 
291 	/*
292 	 * Make saved resource limit == current resource limit.
293 	 */
294 	for (i = 0; i < RLIM_NLIMITS; i++) {
295 		/*CONSTCOND*/
296 		if (RLIM_SAVED(i)) {
297 			(void) rctl_rlimit_get(rctlproc_legacy[i], p,
298 			    &up->u_saved_rlimit[i]);
299 		}
300 	}
301 
302 	/*
303 	 * If the action was to catch the signal, then the action
304 	 * must be reset to SIG_DFL.
305 	 */
306 	sigdefault(p);
307 	p->p_flag &= ~(SNOWAIT|SJCTL);
308 	p->p_flag |= (SEXECED|SMSACCT|SMSFORK);
309 	up->u_signal[SIGCLD - 1] = SIG_DFL;
310 
311 	/*
312 	 * Delete the dot4 sigqueues/signotifies.
313 	 */
314 	sigqfree(p);
315 
316 	mutex_exit(&p->p_lock);
317 
318 	mutex_enter(&p->p_pflock);
319 	p->p_prof.pr_base = NULL;
320 	p->p_prof.pr_size = 0;
321 	p->p_prof.pr_off = 0;
322 	p->p_prof.pr_scale = 0;
323 	p->p_prof.pr_samples = 0;
324 	mutex_exit(&p->p_pflock);
325 
326 	ASSERT(curthread->t_schedctl == NULL);
327 
328 #if defined(__i386) || defined(__amd64)
329 	/* If the process uses a private LDT then change it to default */
330 	if (p->p_ldt)
331 		ldt_free(p);
332 #endif	/* __i386 || __amd64 */
333 
334 #if defined(__amd64)
335 	/*
336 	 * Make sure the process has the correct LDT descriptor for its data
337 	 * model.
338 	 */
339 	if (p->p_model == DATAMODEL_LP64)
340 		p->p_ldt_desc = ldt0_default64_desc;
341 	else
342 		p->p_ldt_desc = ldt0_default_desc;
343 
344 	/*
345 	 * Ensure the change of LDT is propagated into the LDTR.
346 	 */
347 	kpreempt_disable();
348 	ldt_load();
349 	kpreempt_enable();
350 #endif /* __amd64 */
351 
352 #if defined(__sparc)
353 	if (p->p_utraps != NULL)
354 		utrap_free(p);
355 #endif	/* __sparc */
356 
357 	/*
358 	 * Close all close-on-exec files.
359 	 */
360 	close_exec(P_FINFO(p));
361 	TRACE_2(TR_FAC_PROC, TR_PROC_EXEC, "proc_exec:p %p up %p", p, up);
362 	setregs(&args);
363 
364 	/* Mark this as an executable vnode */
365 	mutex_enter(&vp->v_lock);
366 	vp->v_flag |= VVMEXEC;
367 	mutex_exit(&vp->v_lock);
368 
369 	VN_RELE(vp);
370 	if (dir != NULL)
371 		VN_RELE(dir);
372 	pn_free(&resolvepn);
373 
374 	/*
375 	 * Allocate a new lwp directory and lwpid hash table if necessary.
376 	 */
377 	if (curthread->t_tid != 1 || p->p_lwpdir_sz != 2) {
378 		lwpdir = kmem_zalloc(2 * sizeof (lwpdir_t), KM_SLEEP);
379 		lwpdir->ld_next = lwpdir + 1;
380 		tidhash = kmem_zalloc(2 * sizeof (lwpdir_t *), KM_SLEEP);
381 		if (p->p_lwpdir != NULL)
382 			lep = p->p_lwpdir[curthread->t_dslot].ld_entry;
383 		else
384 			lep = kmem_zalloc(sizeof (*lep), KM_SLEEP);
385 	}
386 
387 	mutex_enter(&p->p_lock);
388 	prbarrier(p);
389 
390 	/*
391 	 * Reset lwp id to the default value of 1.
392 	 * This is a single-threaded process now
393 	 * and lwp #1 is lwp_wait()able by default.
394 	 * The t_unpark flag should not be inherited.
395 	 */
396 	ASSERT(p->p_lwpcnt == 1 && p->p_zombcnt == 0);
397 	curthread->t_tid = 1;
398 	curthread->t_unpark = 0;
399 	curthread->t_proc_flag |= TP_TWAIT;
400 	curthread->t_proc_flag &= ~TP_DAEMON;	/* daemons shouldn't exec */
401 	p->p_lwpdaemon = 0;			/* but oh well ... */
402 	p->p_lwpid = 1;
403 
404 	/*
405 	 * Install the newly-allocated lwp directory and lwpid hash table
406 	 * and insert the current thread into the new hash table.
407 	 */
408 	if (lwpdir != NULL) {
409 		old_lwpdir = p->p_lwpdir;
410 		old_lwpdir_sz = p->p_lwpdir_sz;
411 		old_tidhash = p->p_tidhash;
412 		old_tidhash_sz = p->p_tidhash_sz;
413 		p->p_lwpdir = p->p_lwpfree = lwpdir;
414 		p->p_lwpdir_sz = 2;
415 		p->p_tidhash = tidhash;
416 		p->p_tidhash_sz = 2;
417 		lep->le_thread = curthread;
418 		lep->le_lwpid = curthread->t_tid;
419 		lep->le_start = curthread->t_start;
420 		lwp_hash_in(p, lep);
421 	}
422 	/*
423 	 * Restore the saved signal mask and
424 	 * inform /proc that the exec() has finished.
425 	 */
426 	curthread->t_hold = savedmask;
427 	prexecend();
428 	mutex_exit(&p->p_lock);
429 	if (old_lwpdir) {
430 		kmem_free(old_lwpdir, old_lwpdir_sz * sizeof (lwpdir_t));
431 		kmem_free(old_tidhash, old_tidhash_sz * sizeof (lwpdir_t *));
432 	}
433 	ASSERT(error == 0);
434 	DTRACE_PROC(exec__success);
435 	return (0);
436 
437 fail:
438 	DTRACE_PROC1(exec__failure, int, error);
439 out:		/* error return */
440 	mutex_enter(&p->p_lock);
441 	curthread->t_hold = savedmask;
442 	prexecend();
443 	mutex_exit(&p->p_lock);
444 	ASSERT(error != 0);
445 	return (error);
446 }
447 
448 
449 /*
450  * Perform generic exec duties and switchout to object-file specific
451  * handler.
452  */
453 int
454 gexec(
455 	struct vnode **vpp,
456 	struct execa *uap,
457 	struct uarg *args,
458 	struct intpdata *idatap,
459 	int level,
460 	long *execsz,
461 	caddr_t exec_file,
462 	struct cred *cred)
463 {
464 	struct vnode *vp;
465 	proc_t *pp = ttoproc(curthread);
466 	struct execsw *eswp;
467 	int error = 0;
468 	int suidflags = 0;
469 	ssize_t resid;
470 	uid_t uid, gid;
471 	struct vattr vattr;
472 	char magbuf[MAGIC_BYTES];
473 	int setid;
474 	cred_t *oldcred, *newcred = NULL;
475 	int privflags = 0;
476 
477 	/*
478 	 * If the SNOCD or SUGID flag is set, turn it off and remember the
479 	 * previous setting so we can restore it if we encounter an error.
480 	 */
481 	if (level == 0 && (pp->p_flag & PSUIDFLAGS)) {
482 		mutex_enter(&pp->p_lock);
483 		suidflags = pp->p_flag & PSUIDFLAGS;
484 		pp->p_flag &= ~PSUIDFLAGS;
485 		mutex_exit(&pp->p_lock);
486 	}
487 
488 	if ((error = execpermissions(*vpp, &vattr, args)) != 0)
489 		goto bad;
490 
491 	/* need to open vnode for stateful file systems like rfs */
492 	if ((error = VOP_OPEN(vpp, FREAD, CRED())) != 0)
493 		goto bad;
494 	vp = *vpp;
495 
496 	/*
497 	 * Note: to support binary compatibility with SunOS a.out
498 	 * executables, we read in the first four bytes, as the
499 	 * magic number is in bytes 2-3.
500 	 */
501 	if (error = vn_rdwr(UIO_READ, vp, magbuf, sizeof (magbuf),
502 	    (offset_t)0, UIO_SYSSPACE, 0, (rlim64_t)0, CRED(), &resid))
503 		goto bad;
504 	if (resid != 0)
505 		goto bad;
506 
507 	if ((eswp = findexec_by_hdr(magbuf)) == NULL)
508 		goto bad;
509 
510 	if (level == 0 &&
511 	    (privflags = execsetid(vp, &vattr, &uid, &gid)) != 0) {
512 
513 		newcred = cred = crdup(cred);
514 
515 		/* If we can, drop the PA bit */
516 		if ((privflags & PRIV_RESET) != 0)
517 			priv_adjust_PA(cred);
518 
519 		if (privflags & PRIV_SETID) {
520 			cred->cr_uid = uid;
521 			cred->cr_gid = gid;
522 			cred->cr_suid = uid;
523 			cred->cr_sgid = gid;
524 		}
525 
526 		/*
527 		 * Implement the privilege updates:
528 		 *
529 		 * Restrict with L:
530 		 *
531 		 *	I' = I & L
532 		 *
533 		 *	E' = P' = (I' + F) & A
534 		 *
535 		 * But if running under ptrace, we cap I with P.
536 		 */
537 		if ((privflags & PRIV_RESET) != 0) {
538 			if ((privflags & PRIV_INCREASE) != 0 &&
539 			    (pp->p_proc_flag & P_PR_PTRACE) != 0)
540 				priv_intersect(&CR_OPPRIV(cred),
541 						    &CR_IPRIV(cred));
542 			priv_intersect(&CR_LPRIV(cred), &CR_IPRIV(cred));
543 			CR_EPRIV(cred) = CR_PPRIV(cred) = CR_IPRIV(cred);
544 			priv_adjust_PA(cred);
545 		}
546 	}
547 
548 	/* SunOS 4.x buy-back */
549 	if ((vp->v_vfsp->vfs_flag & VFS_NOSETUID) &&
550 	    (vattr.va_mode & (VSUID|VSGID))) {
551 		cmn_err(CE_NOTE,
552 		    "!%s, uid %d: setuid execution not allowed, dev=%lx",
553 		    exec_file, cred->cr_uid, vp->v_vfsp->vfs_dev);
554 	}
555 
556 	/*
557 	 * execsetid() told us whether or not we had to change the
558 	 * credentials of the process.  In privflags, it told us
559 	 * whether we gained any privileges or executed a set-uid executable.
560 	 */
561 	setid = (privflags & (PRIV_SETUGID|PRIV_INCREASE));
562 
563 	/*
564 	 * Use /etc/system variable to determine if the stack
565 	 * should be marked as executable by default.
566 	 */
567 	if (noexec_user_stack)
568 		args->stk_prot &= ~PROT_EXEC;
569 
570 	args->execswp = eswp; /* Save execsw pointer in uarg for exec_func */
571 
572 	/*
573 	 * Traditionally, the setid flags told the sub processes whether
574 	 * the file just executed was set-uid or set-gid; this caused
575 	 * some confusion as the 'setid' flag did not match the SUGID
576 	 * process flag which is only set when the uids/gids do not match.
577 	 * A script set-gid/set-uid to the real uid/gid would start with
578 	 * /dev/fd/X but an executable would happily trust LD_LIBRARY_PATH.
579 	 * Now we flag those cases where the calling process cannot
580 	 * be trusted to influence the newly exec'ed process, either
581 	 * because it runs with more privileges or when the uids/gids
582 	 * do in fact not match.
583 	 * This also makes the runtime linker agree with the on exec
584 	 * values of SNOCD and SUGID.
585 	 */
586 	error = (*eswp->exec_func)(vp, uap, args, idatap, level, execsz,
587 		(setid & PRIV_INCREASE) != 0 ||
588 		cred->cr_uid != cred->cr_ruid ||
589 		(cred->cr_rgid != cred->cr_gid &&
590 		!supgroupmember(cred->cr_gid, cred)), exec_file, cred);
591 	rw_exit(eswp->exec_lock);
592 	if (error != 0) {
593 		if (newcred != NULL)
594 			crfree(newcred);
595 		goto bad;
596 	}
597 
598 	if (level == 0) {
599 		mutex_enter(&pp->p_crlock);
600 		if (newcred != NULL) {
601 			/*
602 			 * Free the old credentials, and set the new ones.
603 			 * Do this for both the process and the (single) thread.
604 			 */
605 			crfree(pp->p_cred);
606 			pp->p_cred = cred;	/* cred already held for proc */
607 			crhold(cred);		/* hold new cred for thread */
608 			/*
609 			 * DTrace accesses t_cred in probe context.  t_cred
610 			 * must always be either NULL, or point to a valid,
611 			 * allocated cred structure.
612 			 */
613 			oldcred = curthread->t_cred;
614 			curthread->t_cred = cred;
615 			crfree(oldcred);
616 		}
617 		/*
618 		 * On emerging from a successful exec(), the saved
619 		 * uid and gid equal the effective uid and gid.
620 		 */
621 		cred->cr_suid = cred->cr_uid;
622 		cred->cr_sgid = cred->cr_gid;
623 
624 		/*
625 		 * If the real and effective ids do not match, this
626 		 * is a setuid process that should not dump core.
627 		 * The group comparison is tricky; we prevent the code
628 		 * from flagging SNOCD when executing with an effective gid
629 		 * which is a supplementary group.
630 		 */
631 		if (cred->cr_ruid != cred->cr_uid ||
632 		    (cred->cr_rgid != cred->cr_gid &&
633 		    !supgroupmember(cred->cr_gid, cred)) ||
634 		    (privflags & PRIV_INCREASE) != 0)
635 			suidflags = PSUIDFLAGS;
636 		else
637 			suidflags = 0;
638 
639 		mutex_exit(&pp->p_crlock);
640 		if (suidflags) {
641 			mutex_enter(&pp->p_lock);
642 			pp->p_flag |= suidflags;
643 			mutex_exit(&pp->p_lock);
644 		}
645 		if (setid && (pp->p_proc_flag & P_PR_PTRACE) == 0) {
646 			/*
647 			 * If process is traced via /proc, arrange to
648 			 * invalidate the associated /proc vnode.
649 			 */
650 			if (pp->p_plist || (pp->p_proc_flag & P_PR_TRACE))
651 				args->traceinval = 1;
652 		}
653 		if (pp->p_proc_flag & P_PR_PTRACE)
654 			psignal(pp, SIGTRAP);
655 		if (args->traceinval)
656 			prinvalidate(&pp->p_user);
657 	}
658 
659 	return (0);
660 bad:
661 	if (error == 0)
662 		error = ENOEXEC;
663 
664 	if (suidflags) {
665 		mutex_enter(&pp->p_lock);
666 		pp->p_flag |= suidflags;
667 		mutex_exit(&pp->p_lock);
668 	}
669 	return (error);
670 }
671 
672 extern char *execswnames[];
673 
674 struct execsw *
675 allocate_execsw(char *name, char *magic, size_t magic_size)
676 {
677 	int i, j;
678 	char *ename;
679 	char *magicp;
680 
681 	mutex_enter(&execsw_lock);
682 	for (i = 0; i < nexectype; i++) {
683 		if (execswnames[i] == NULL) {
684 			ename = kmem_alloc(strlen(name) + 1, KM_SLEEP);
685 			(void) strcpy(ename, name);
686 			execswnames[i] = ename;
687 			/*
688 			 * Set the magic number last so that we
689 			 * don't need to hold the execsw_lock in
690 			 * findexectype().
691 			 */
692 			magicp = kmem_alloc(magic_size, KM_SLEEP);
693 			for (j = 0; j < magic_size; j++)
694 				magicp[j] = magic[j];
695 			execsw[i].exec_magic = magicp;
696 			mutex_exit(&execsw_lock);
697 			return (&execsw[i]);
698 		}
699 	}
700 	mutex_exit(&execsw_lock);
701 	return (NULL);
702 }
703 
704 /*
705  * Find the exec switch table entry with the corresponding magic string.
706  */
707 struct execsw *
708 findexecsw(char *magic)
709 {
710 	struct execsw *eswp;
711 
712 	for (eswp = execsw; eswp < &execsw[nexectype]; eswp++) {
713 		ASSERT(eswp->exec_maglen <= MAGIC_BYTES);
714 		if (magic && eswp->exec_maglen != 0 &&
715 		    bcmp(magic, eswp->exec_magic, eswp->exec_maglen) == 0)
716 			return (eswp);
717 	}
718 	return (NULL);
719 }
720 
721 /*
722  * Find the execsw[] index for the given exec header string by looking for the
723  * magic string at a specified offset and length for each kind of executable
724  * file format until one matches.  If no execsw[] entry is found, try to
725  * autoload a module for this magic string.
726  */
727 struct execsw *
728 findexec_by_hdr(char *header)
729 {
730 	struct execsw *eswp;
731 
732 	for (eswp = execsw; eswp < &execsw[nexectype]; eswp++) {
733 		ASSERT(eswp->exec_maglen <= MAGIC_BYTES);
734 		if (header && eswp->exec_maglen != 0 &&
735 		    bcmp(&header[eswp->exec_magoff], eswp->exec_magic,
736 			    eswp->exec_maglen) == 0) {
737 			if (hold_execsw(eswp) != 0)
738 				return (NULL);
739 			return (eswp);
740 		}
741 	}
742 	return (NULL);	/* couldn't find the type */
743 }
744 
745 /*
746  * Find the execsw[] index for the given magic string.  If no execsw[] entry
747  * is found, try to autoload a module for this magic string.
748  */
749 struct execsw *
750 findexec_by_magic(char *magic)
751 {
752 	struct execsw *eswp;
753 
754 	for (eswp = execsw; eswp < &execsw[nexectype]; eswp++) {
755 		ASSERT(eswp->exec_maglen <= MAGIC_BYTES);
756 		if (magic && eswp->exec_maglen != 0 &&
757 		    bcmp(magic, eswp->exec_magic, eswp->exec_maglen) == 0) {
758 			if (hold_execsw(eswp) != 0)
759 				return (NULL);
760 			return (eswp);
761 		}
762 	}
763 	return (NULL);	/* couldn't find the type */
764 }
765 
766 static int
767 hold_execsw(struct execsw *eswp)
768 {
769 	char *name;
770 
771 	rw_enter(eswp->exec_lock, RW_READER);
772 	while (!LOADED_EXEC(eswp)) {
773 		rw_exit(eswp->exec_lock);
774 		name = execswnames[eswp-execsw];
775 		ASSERT(name);
776 		if (modload("exec", name) == -1)
777 			return (-1);
778 		rw_enter(eswp->exec_lock, RW_READER);
779 	}
780 	return (0);
781 }
782 
783 static int
784 execsetid(struct vnode *vp, struct vattr *vattrp, uid_t *uidp, uid_t *gidp)
785 {
786 	proc_t *pp = ttoproc(curthread);
787 	uid_t uid, gid;
788 	cred_t *cr = pp->p_cred;
789 	int privflags = 0;
790 
791 	/*
792 	 * Remember credentials.
793 	 */
794 	uid = cr->cr_uid;
795 	gid = cr->cr_gid;
796 
797 	/* Will try to reset the PRIV_AWARE bit later. */
798 	if ((CR_FLAGS(cr) & (PRIV_AWARE|PRIV_AWARE_INHERIT)) == PRIV_AWARE)
799 		privflags |= PRIV_RESET;
800 
801 	if ((vp->v_vfsp->vfs_flag & VFS_NOSETUID) == 0) {
802 		/*
803 		 * Set-uid root execution only allowed if the limit set
804 		 * holds all unsafe privileges.
805 		 */
806 		if ((vattrp->va_mode & VSUID) && (vattrp->va_uid != 0 ||
807 		    priv_issubset(&priv_unsafe, &CR_LPRIV(cr)))) {
808 			uid = vattrp->va_uid;
809 			privflags |= PRIV_SETUGID;
810 		}
811 		if (vattrp->va_mode & VSGID) {
812 			gid = vattrp->va_gid;
813 			privflags |= PRIV_SETUGID;
814 		}
815 	}
816 
817 	/*
818 	 * Do we need to change our credential anyway?
819 	 * This is the case when E != I or P != I, as
820 	 * we need to do the assignments (with F empty and A full)
821 	 * Or when I is not a subset of L; in that case we need to
822 	 * enforce L.
823 	 *
824 	 *		I' = L & I
825 	 *
826 	 *		E' = P' = (I' + F) & A
827 	 * or
828 	 *		E' = P' = I'
829 	 */
830 	if (!priv_isequalset(&CR_EPRIV(cr), &CR_IPRIV(cr)) ||
831 	    !priv_issubset(&CR_IPRIV(cr), &CR_LPRIV(cr)) ||
832 	    !priv_isequalset(&CR_PPRIV(cr), &CR_IPRIV(cr)))
833 		privflags |= PRIV_RESET;
834 
835 	/*
836 	 * When we introduce the "forced" set then we will need
837 	 * to set PRIV_INCREASE here if I not a subset of P.
838 	 * If the "allowed" set is introduced we will need to do
839 	 * a similar thing; however, it seems more reasonable to
840 	 * have the allowed set reduce "L": script language interpreters
841 	 * would typically have an allowed set of "all".
842 	 */
843 
844 	/*
845 	 * Set setuid/setgid protections if no ptrace() compatibility.
846 	 * For privileged processes, honor setuid/setgid even in
847 	 * the presence of ptrace() compatibility.
848 	 */
849 	if (((pp->p_proc_flag & P_PR_PTRACE) == 0 ||
850 	    PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, (uid == 0))) &&
851 	    (cr->cr_uid != uid ||
852 	    cr->cr_gid != gid ||
853 	    cr->cr_suid != uid ||
854 	    cr->cr_sgid != gid)) {
855 		*uidp = uid;
856 		*gidp = gid;
857 		privflags |= PRIV_SETID;
858 	}
859 	return (privflags);
860 }
861 
862 int
863 execpermissions(struct vnode *vp, struct vattr *vattrp, struct uarg *args)
864 {
865 	int error;
866 	proc_t *p = ttoproc(curthread);
867 
868 	vattrp->va_mask = AT_MODE | AT_UID | AT_GID | AT_SIZE;
869 	if (error = VOP_GETATTR(vp, vattrp, ATTR_EXEC, p->p_cred))
870 		return (error);
871 	/*
872 	 * Check the access mode.
873 	 * If VPROC, ask /proc if the file is an object file.
874 	 */
875 	if ((error = VOP_ACCESS(vp, VEXEC, 0, p->p_cred)) != 0 ||
876 	    !(vp->v_type == VREG || (vp->v_type == VPROC && pr_isobject(vp))) ||
877 	    (vp->v_vfsp->vfs_flag & VFS_NOEXEC) != 0 ||
878 	    (vattrp->va_mode & (VEXEC|(VEXEC>>3)|(VEXEC>>6))) == 0) {
879 		if (error == 0)
880 			error = EACCES;
881 		return (error);
882 	}
883 
884 	if ((p->p_plist || (p->p_proc_flag & (P_PR_PTRACE|P_PR_TRACE))) &&
885 	    (error = VOP_ACCESS(vp, VREAD, 0, p->p_cred))) {
886 		/*
887 		 * If process is under ptrace(2) compatibility,
888 		 * fail the exec(2).
889 		 */
890 		if (p->p_proc_flag & P_PR_PTRACE)
891 			goto bad;
892 		/*
893 		 * Process is traced via /proc.
894 		 * Arrange to invalidate the /proc vnode.
895 		 */
896 		args->traceinval = 1;
897 	}
898 	return (0);
899 bad:
900 	if (error == 0)
901 		error = ENOEXEC;
902 	return (error);
903 }
904 
905 /*
906  * Map a section of an executable file into the user's
907  * address space.
908  */
909 int
910 execmap(struct vnode *vp, caddr_t addr, size_t len, size_t zfodlen,
911     off_t offset, int prot, int page, uint_t szc)
912 {
913 	int error = 0;
914 	off_t oldoffset;
915 	caddr_t zfodbase, oldaddr;
916 	size_t end, oldlen;
917 	size_t zfoddiff;
918 	label_t ljb;
919 	proc_t *p = ttoproc(curthread);
920 
921 	oldaddr = addr;
922 	addr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK);
923 	if (len) {
924 		oldlen = len;
925 		len += ((size_t)oldaddr - (size_t)addr);
926 		oldoffset = offset;
927 		offset = (off_t)((uintptr_t)offset & PAGEMASK);
928 		if (page) {
929 			spgcnt_t  prefltmem, availm, npages;
930 			int preread;
931 			uint_t mflag = MAP_PRIVATE | MAP_FIXED;
932 
933 			if ((prot & (PROT_WRITE | PROT_EXEC)) == PROT_EXEC) {
934 				mflag |= MAP_TEXT;
935 			} else {
936 				mflag |= MAP_INITDATA;
937 			}
938 
939 			if (valid_usr_range(addr, len, prot, p->p_as,
940 			    p->p_as->a_userlimit) != RANGE_OKAY) {
941 				error = ENOMEM;
942 				goto bad;
943 			}
944 			if (error = VOP_MAP(vp, (offset_t)offset,
945 			    p->p_as, &addr, len, prot, PROT_ALL,
946 			    mflag, CRED()))
947 				goto bad;
948 
949 			/*
950 			 * If the segment can fit, then we prefault
951 			 * the entire segment in.  This is based on the
952 			 * model that says the best working set of a
953 			 * small program is all of its pages.
954 			 */
955 			npages = (spgcnt_t)btopr(len);
956 			prefltmem = freemem - desfree;
957 			preread =
958 			    (npages < prefltmem && len < PGTHRESH) ? 1 : 0;
959 
960 			/*
961 			 * If we aren't prefaulting the segment,
962 			 * increment "deficit", if necessary to ensure
963 			 * that pages will become available when this
964 			 * process starts executing.
965 			 */
966 			availm = freemem - lotsfree;
967 			if (preread == 0 && npages > availm &&
968 			    deficit < lotsfree) {
969 				deficit += MIN((pgcnt_t)(npages - availm),
970 				    lotsfree - deficit);
971 			}
972 
973 			if (preread) {
974 				TRACE_2(TR_FAC_PROC, TR_EXECMAP_PREREAD,
975 				    "execmap preread:freemem %d size %lu",
976 				    freemem, len);
977 				(void) as_fault(p->p_as->a_hat, p->p_as,
978 				    (caddr_t)addr, len, F_INVAL, S_READ);
979 			}
980 		} else {
981 			if (valid_usr_range(addr, len, prot, p->p_as,
982 			    p->p_as->a_userlimit) != RANGE_OKAY) {
983 				error = ENOMEM;
984 				goto bad;
985 			}
986 
987 			if (error = as_map(p->p_as, addr, len,
988 			    segvn_create, zfod_argsp))
989 				goto bad;
990 			/*
991 			 * Read in the segment in one big chunk.
992 			 */
993 			if (error = vn_rdwr(UIO_READ, vp, (caddr_t)oldaddr,
994 			    oldlen, (offset_t)oldoffset, UIO_USERSPACE, 0,
995 			    (rlim64_t)0, CRED(), (ssize_t *)0))
996 				goto bad;
997 			/*
998 			 * Now set protections.
999 			 */
1000 			if (prot != PROT_ZFOD) {
1001 				(void) as_setprot(p->p_as, (caddr_t)addr,
1002 				    len, prot);
1003 			}
1004 		}
1005 	}
1006 
1007 	if (zfodlen) {
1008 		end = (size_t)addr + len;
1009 		zfodbase = (caddr_t)roundup(end, PAGESIZE);
1010 		zfoddiff = (uintptr_t)zfodbase - end;
1011 		if (zfoddiff) {
1012 			if (on_fault(&ljb)) {
1013 				no_fault();
1014 				error = EFAULT;
1015 				goto bad;
1016 			}
1017 			uzero((void *)end, zfoddiff);
1018 			no_fault();
1019 		}
1020 		if (zfodlen > zfoddiff) {
1021 			struct segvn_crargs crargs =
1022 			    SEGVN_ZFOD_ARGS(PROT_ZFOD, PROT_ALL);
1023 
1024 			zfodlen -= zfoddiff;
1025 			if (valid_usr_range(zfodbase, zfodlen, prot, p->p_as,
1026 			    p->p_as->a_userlimit) != RANGE_OKAY) {
1027 				error = ENOMEM;
1028 				goto bad;
1029 			}
1030 			crargs.szc = szc;
1031 			if (error = as_map(p->p_as, (caddr_t)zfodbase,
1032 			    zfodlen, segvn_create, &crargs))
1033 				goto bad;
1034 			if (prot != PROT_ZFOD) {
1035 				(void) as_setprot(p->p_as, (caddr_t)zfodbase,
1036 				    zfodlen, prot);
1037 			}
1038 		}
1039 	}
1040 	return (0);
1041 bad:
1042 	return (error);
1043 }
1044 
1045 void
1046 setexecenv(struct execenv *ep)
1047 {
1048 	proc_t *p = ttoproc(curthread);
1049 	klwp_t *lwp = ttolwp(curthread);
1050 	struct vnode *vp;
1051 
1052 	p->p_bssbase = ep->ex_bssbase;
1053 	p->p_brkbase = ep->ex_brkbase;
1054 	p->p_brksize = ep->ex_brksize;
1055 	if (p->p_exec)
1056 		VN_RELE(p->p_exec);	/* out with the old */
1057 	vp = p->p_exec = ep->ex_vp;
1058 	if (vp != NULL)
1059 		VN_HOLD(vp);		/* in with the new */
1060 
1061 	lwp->lwp_sigaltstack.ss_sp = 0;
1062 	lwp->lwp_sigaltstack.ss_size = 0;
1063 	lwp->lwp_sigaltstack.ss_flags = SS_DISABLE;
1064 }
1065 
1066 int
1067 execopen(struct vnode **vpp, int *fdp)
1068 {
1069 	struct vnode *vp = *vpp;
1070 	file_t *fp;
1071 	int error = 0;
1072 	int filemode = FREAD;
1073 
1074 	VN_HOLD(vp);		/* open reference */
1075 	if (error = falloc(NULL, filemode, &fp, fdp)) {
1076 		VN_RELE(vp);
1077 		*fdp = -1;	/* just in case falloc changed value */
1078 		return (error);
1079 	}
1080 	if (error = VOP_OPEN(&vp, filemode, CRED())) {
1081 		VN_RELE(vp);
1082 		setf(*fdp, NULL);
1083 		unfalloc(fp);
1084 		*fdp = -1;
1085 		return (error);
1086 	}
1087 	*vpp = vp;		/* vnode should not have changed */
1088 	fp->f_vnode = vp;
1089 	mutex_exit(&fp->f_tlock);
1090 	setf(*fdp, fp);
1091 	return (0);
1092 }
1093 
1094 int
1095 execclose(int fd)
1096 {
1097 	return (closeandsetf(fd, NULL));
1098 }
1099 
1100 
1101 /*
1102  * noexec stub function.
1103  */
1104 /*ARGSUSED*/
1105 int
1106 noexec(
1107     struct vnode *vp,
1108     struct execa *uap,
1109     struct uarg *args,
1110     struct intpdata *idatap,
1111     int level,
1112     long *execsz,
1113     int setid,
1114     caddr_t exec_file,
1115     struct cred *cred)
1116 {
1117 	cmn_err(CE_WARN, "missing exec capability for %s", uap->fname);
1118 	return (ENOEXEC);
1119 }
1120 
1121 /*
1122  * Support routines for building a user stack.
1123  *
1124  * execve(path, argv, envp) must construct a new stack with the specified
1125  * arguments and environment variables (see exec_args() for a description
1126  * of the user stack layout).  To do this, we copy the arguments and
1127  * environment variables from the old user address space into the kernel,
1128  * free the old as, create the new as, and copy our buffered information
1129  * to the new stack.  Our kernel buffer has the following structure:
1130  *
1131  *	+-----------------------+ <--- stk_base + stk_size
1132  *	| string offsets	|
1133  *	+-----------------------+ <--- stk_offp
1134  *	|			|
1135  *	| STK_AVAIL() space	|
1136  *	|			|
1137  *	+-----------------------+ <--- stk_strp
1138  *	| strings		|
1139  *	+-----------------------+ <--- stk_base
1140  *
1141  * When we add a string, we store the string's contents (including the null
1142  * terminator) at stk_strp, and we store the offset of the string relative to
1143  * stk_base at --stk_offp.  At strings are added, stk_strp increases and
1144  * stk_offp decreases.  The amount of space remaining, STK_AVAIL(), is just
1145  * the difference between these pointers.  If we run out of space, we return
1146  * an error and exec_args() starts all over again with a buffer twice as large.
1147  * When we're all done, the kernel buffer looks like this:
1148  *
1149  *	+-----------------------+ <--- stk_base + stk_size
1150  *	| argv[0] offset	|
1151  *	+-----------------------+
1152  *	| ...			|
1153  *	+-----------------------+
1154  *	| argv[argc-1] offset	|
1155  *	+-----------------------+
1156  *	| envp[0] offset	|
1157  *	+-----------------------+
1158  *	| ...			|
1159  *	+-----------------------+
1160  *	| envp[envc-1] offset	|
1161  *	+-----------------------+
1162  *	| AT_SUN_PLATFORM offset|
1163  *	+-----------------------+
1164  *	| AT_SUN_EXECNAME offset|
1165  *	+-----------------------+ <--- stk_offp
1166  *	|			|
1167  *	| STK_AVAIL() space	|
1168  *	|			|
1169  *	+-----------------------+ <--- stk_strp
1170  *	| AT_SUN_EXECNAME offset|
1171  *	+-----------------------+
1172  *	| AT_SUN_PLATFORM offset|
1173  *	+-----------------------+
1174  *	| envp[envc-1] string	|
1175  *	+-----------------------+
1176  *	| ...			|
1177  *	+-----------------------+
1178  *	| envp[0] string	|
1179  *	+-----------------------+
1180  *	| argv[argc-1] string	|
1181  *	+-----------------------+
1182  *	| ...			|
1183  *	+-----------------------+
1184  *	| argv[0] string	|
1185  *	+-----------------------+ <--- stk_base
1186  */
1187 
1188 #define	STK_AVAIL(args)		((char *)(args)->stk_offp - (args)->stk_strp)
1189 
1190 /*
1191  * Add a string to the stack.
1192  */
1193 static int
1194 stk_add(uarg_t *args, const char *sp, enum uio_seg segflg)
1195 {
1196 	int error;
1197 	size_t len;
1198 
1199 	if (STK_AVAIL(args) < sizeof (int))
1200 		return (E2BIG);
1201 	*--args->stk_offp = args->stk_strp - args->stk_base;
1202 
1203 	if (segflg == UIO_USERSPACE) {
1204 		error = copyinstr(sp, args->stk_strp, STK_AVAIL(args), &len);
1205 		if (error != 0)
1206 			return (error);
1207 	} else {
1208 		len = strlen(sp) + 1;
1209 		if (len > STK_AVAIL(args))
1210 			return (E2BIG);
1211 		bcopy(sp, args->stk_strp, len);
1212 	}
1213 
1214 	args->stk_strp += len;
1215 
1216 	return (0);
1217 }
1218 
1219 static int
1220 stk_getptr(uarg_t *args, char *src, char **dst)
1221 {
1222 	int error;
1223 
1224 	if (args->from_model == DATAMODEL_NATIVE) {
1225 		ulong_t ptr;
1226 		error = fulword(src, &ptr);
1227 		*dst = (caddr_t)ptr;
1228 	} else {
1229 		uint32_t ptr;
1230 		error = fuword32(src, &ptr);
1231 		*dst = (caddr_t)(uintptr_t)ptr;
1232 	}
1233 	return (error);
1234 }
1235 
1236 static int
1237 stk_putptr(uarg_t *args, char *addr, char *value)
1238 {
1239 	if (args->to_model == DATAMODEL_NATIVE)
1240 		return (sulword(addr, (ulong_t)value));
1241 	else
1242 		return (suword32(addr, (uint32_t)(uintptr_t)value));
1243 }
1244 
1245 static int
1246 stk_copyin(execa_t *uap, uarg_t *args, intpdata_t *intp, void **auxvpp)
1247 {
1248 	char *sp;
1249 	int argc, error;
1250 	int argv_empty = 0;
1251 	size_t ptrsize = args->from_ptrsize;
1252 	size_t size, pad;
1253 	char *argv = (char *)uap->argp;
1254 	char *envp = (char *)uap->envp;
1255 
1256 	/*
1257 	 * Copy interpreter's name and argument to argv[0] and argv[1].
1258 	 */
1259 	if (intp != NULL && intp->intp_name != NULL) {
1260 		if ((error = stk_add(args, intp->intp_name, UIO_SYSSPACE)) != 0)
1261 			return (error);
1262 		if (intp->intp_arg != NULL &&
1263 		    (error = stk_add(args, intp->intp_arg, UIO_SYSSPACE)) != 0)
1264 			return (error);
1265 		if (args->fname != NULL)
1266 			error = stk_add(args, args->fname, UIO_SYSSPACE);
1267 		else
1268 			error = stk_add(args, uap->fname, UIO_USERSPACE);
1269 		if (error)
1270 			return (error);
1271 
1272 		/*
1273 		 * Check for an empty argv[].
1274 		 */
1275 		if (stk_getptr(args, argv, &sp))
1276 			return (EFAULT);
1277 		if (sp == NULL)
1278 			argv_empty = 1;
1279 
1280 		argv += ptrsize;		/* ignore original argv[0] */
1281 	}
1282 
1283 	if (argv_empty == 0) {
1284 		/*
1285 		 * Add argv[] strings to the stack.
1286 		 */
1287 		for (;;) {
1288 			if (stk_getptr(args, argv, &sp))
1289 				return (EFAULT);
1290 			if (sp == NULL)
1291 				break;
1292 			if ((error = stk_add(args, sp, UIO_USERSPACE)) != 0)
1293 				return (error);
1294 			argv += ptrsize;
1295 		}
1296 	}
1297 	argc = (int *)(args->stk_base + args->stk_size) - args->stk_offp;
1298 	args->arglen = args->stk_strp - args->stk_base;
1299 
1300 	/*
1301 	 * Add environ[] strings to the stack.
1302 	 */
1303 	if (envp != NULL) {
1304 		for (;;) {
1305 			if (stk_getptr(args, envp, &sp))
1306 				return (EFAULT);
1307 			if (sp == NULL)
1308 				break;
1309 			if ((error = stk_add(args, sp, UIO_USERSPACE)) != 0)
1310 				return (error);
1311 			envp += ptrsize;
1312 		}
1313 	}
1314 	args->na = (int *)(args->stk_base + args->stk_size) - args->stk_offp;
1315 	args->ne = args->na - argc;
1316 
1317 	/*
1318 	 * Add AT_SUN_PLATFORM and AT_SUN_EXECNAME strings to the stack.
1319 	 */
1320 	if (auxvpp != NULL && *auxvpp != NULL) {
1321 		if ((error = stk_add(args, platform, UIO_SYSSPACE)) != 0)
1322 			return (error);
1323 		if ((error = stk_add(args, args->pathname, UIO_SYSSPACE)) != 0)
1324 			return (error);
1325 	}
1326 
1327 	/*
1328 	 * Compute the size of the stack.  This includes all the pointers,
1329 	 * the space reserved for the aux vector, and all the strings.
1330 	 * The total number of pointers is args->na (which is argc + envc)
1331 	 * plus 4 more: (1) a pointer's worth of space for argc; (2) the NULL
1332 	 * after the last argument (i.e. argv[argc]); (3) the NULL after the
1333 	 * last environment variable (i.e. envp[envc]); and (4) the NULL after
1334 	 * all the strings, at the very top of the stack.
1335 	 */
1336 	size = (args->na + 4) * args->to_ptrsize + args->auxsize +
1337 	    (args->stk_strp - args->stk_base);
1338 
1339 	/*
1340 	 * Pad the string section with zeroes to align the stack size.
1341 	 */
1342 	pad = P2NPHASE(size, args->stk_align);
1343 
1344 	if (STK_AVAIL(args) < pad)
1345 		return (E2BIG);
1346 
1347 	args->usrstack_size = size + pad;
1348 
1349 	while (pad-- != 0)
1350 		*args->stk_strp++ = 0;
1351 
1352 	args->nc = args->stk_strp - args->stk_base;
1353 
1354 	return (0);
1355 }
1356 
1357 static int
1358 stk_copyout(uarg_t *args, char *usrstack, void **auxvpp, user_t *up)
1359 {
1360 	size_t ptrsize = args->to_ptrsize;
1361 	ssize_t pslen;
1362 	char *kstrp = args->stk_base;
1363 	char *ustrp = usrstack - args->nc - ptrsize;
1364 	char *usp = usrstack - args->usrstack_size;
1365 	int *offp = (int *)(args->stk_base + args->stk_size);
1366 	int envc = args->ne;
1367 	int argc = args->na - envc;
1368 	int i;
1369 
1370 	/*
1371 	 * Record argc for /proc.
1372 	 */
1373 	up->u_argc = argc;
1374 
1375 	/*
1376 	 * Put argc on the stack.  Note that even though it's an int,
1377 	 * it always consumes ptrsize bytes (for alignment).
1378 	 */
1379 	if (stk_putptr(args, usp, (char *)(uintptr_t)argc))
1380 		return (-1);
1381 
1382 	/*
1383 	 * Add argc space (ptrsize) to usp and record argv for /proc.
1384 	 */
1385 	up->u_argv = (uintptr_t)(usp += ptrsize);
1386 
1387 	/*
1388 	 * Put the argv[] pointers on the stack.
1389 	 */
1390 	for (i = 0; i < argc; i++, usp += ptrsize)
1391 		if (stk_putptr(args, usp, &ustrp[*--offp]))
1392 			return (-1);
1393 
1394 	/*
1395 	 * Copy arguments to u_psargs.
1396 	 */
1397 	pslen = MIN(args->arglen, PSARGSZ) - 1;
1398 	for (i = 0; i < pslen; i++)
1399 		up->u_psargs[i] = (kstrp[i] == '\0' ? ' ' : kstrp[i]);
1400 	while (i < PSARGSZ)
1401 		up->u_psargs[i++] = '\0';
1402 
1403 	/*
1404 	 * Add space for argv[]'s NULL terminator (ptrsize) to usp and
1405 	 * record envp for /proc.
1406 	 */
1407 	up->u_envp = (uintptr_t)(usp += ptrsize);
1408 
1409 	/*
1410 	 * Put the envp[] pointers on the stack.
1411 	 */
1412 	for (i = 0; i < envc; i++, usp += ptrsize)
1413 		if (stk_putptr(args, usp, &ustrp[*--offp]))
1414 			return (-1);
1415 
1416 	/*
1417 	 * Add space for envp[]'s NULL terminator (ptrsize) to usp and
1418 	 * remember where the stack ends, which is also where auxv begins.
1419 	 */
1420 	args->stackend = usp += ptrsize;
1421 
1422 	/*
1423 	 * Put all the argv[], envp[], and auxv strings on the stack.
1424 	 */
1425 	if (copyout(args->stk_base, ustrp, args->nc))
1426 		return (-1);
1427 
1428 	/*
1429 	 * Fill in the aux vector now that we know the user stack addresses
1430 	 * for the AT_SUN_PLATFORM and AT_SUN_EXECNAME strings.
1431 	 */
1432 	if (auxvpp != NULL && *auxvpp != NULL) {
1433 		if (args->to_model == DATAMODEL_NATIVE) {
1434 			auxv_t **a = (auxv_t **)auxvpp;
1435 			ADDAUX(*a, AT_SUN_PLATFORM, (long)&ustrp[*--offp])
1436 			ADDAUX(*a, AT_SUN_EXECNAME, (long)&ustrp[*--offp])
1437 		} else {
1438 			auxv32_t **a = (auxv32_t **)auxvpp;
1439 			ADDAUX(*a,
1440 			    AT_SUN_PLATFORM, (int)(uintptr_t)&ustrp[*--offp])
1441 			ADDAUX(*a,
1442 			    AT_SUN_EXECNAME, (int)(uintptr_t)&ustrp[*--offp]);
1443 		}
1444 	}
1445 
1446 	return (0);
1447 }
1448 
1449 #ifdef DEBUG
1450 int mpss_brkpgszsel = 0;
1451 int mpss_stkpgszsel = 0;
1452 #endif
1453 
1454 /*
1455  * Initialize a new user stack with the specified arguments and environment.
1456  * The initial user stack layout is as follows:
1457  *
1458  *	User Stack
1459  *	+---------------+ <--- curproc->p_usrstack
1460  *	| NULL		|
1461  *	+---------------+
1462  *	|		|
1463  *	| auxv strings	|
1464  *	|		|
1465  *	+---------------+
1466  *	|		|
1467  *	| envp strings	|
1468  *	|		|
1469  *	+---------------+
1470  *	|		|
1471  *	| argv strings	|
1472  *	|		|
1473  *	+---------------+ <--- ustrp
1474  *	|		|
1475  *	| aux vector	|
1476  *	|		|
1477  *	+---------------+ <--- auxv
1478  *	| NULL		|
1479  *	+---------------+
1480  *	| envp[envc-1]	|
1481  *	+---------------+
1482  *	| ...		|
1483  *	+---------------+
1484  *	| envp[0]	|
1485  *	+---------------+ <--- envp[]
1486  *	| NULL		|
1487  *	+---------------+
1488  *	| argv[argc-1]	|
1489  *	+---------------+
1490  *	| ...		|
1491  *	+---------------+
1492  *	| argv[0]	|
1493  *	+---------------+ <--- argv[]
1494  *	| argc		|
1495  *	+---------------+ <--- stack base
1496  */
1497 int
1498 exec_args(execa_t *uap, uarg_t *args, intpdata_t *intp, void **auxvpp)
1499 {
1500 	size_t size;
1501 	int error;
1502 	proc_t *p = ttoproc(curthread);
1503 	user_t *up = PTOU(p);
1504 	char *usrstack;
1505 	rctl_entity_p_t e;
1506 
1507 	struct as *as;
1508 
1509 	args->from_model = p->p_model;
1510 	if (p->p_model == DATAMODEL_NATIVE) {
1511 		args->from_ptrsize = sizeof (long);
1512 	} else {
1513 		args->from_ptrsize = sizeof (int32_t);
1514 	}
1515 
1516 	if (args->to_model == DATAMODEL_NATIVE) {
1517 		args->to_ptrsize = sizeof (long);
1518 		args->ncargs = NCARGS;
1519 		args->stk_align = STACK_ALIGN;
1520 		usrstack = (char *)USRSTACK;
1521 	} else {
1522 		args->to_ptrsize = sizeof (int32_t);
1523 		args->ncargs = NCARGS32;
1524 		args->stk_align = STACK_ALIGN32;
1525 		usrstack = (char *)USRSTACK32;
1526 	}
1527 
1528 	ASSERT(P2PHASE((uintptr_t)usrstack, args->stk_align) == 0);
1529 
1530 #if defined(__sparc)
1531 	/*
1532 	 * Make sure user register windows are empty before
1533 	 * attempting to make a new stack.
1534 	 */
1535 	(void) flush_user_windows_to_stack(NULL);
1536 #endif
1537 
1538 	for (size = PAGESIZE; ; size *= 2) {
1539 		args->stk_size = size;
1540 		args->stk_base = kmem_alloc(size, KM_SLEEP);
1541 		args->stk_strp = args->stk_base;
1542 		args->stk_offp = (int *)(args->stk_base + size);
1543 		error = stk_copyin(uap, args, intp, auxvpp);
1544 		if (error == 0)
1545 			break;
1546 		kmem_free(args->stk_base, size);
1547 		if (error != E2BIG && error != ENAMETOOLONG)
1548 			return (error);
1549 		if (size >= args->ncargs)
1550 			return (E2BIG);
1551 	}
1552 
1553 	size = args->usrstack_size;
1554 
1555 	ASSERT(error == 0);
1556 	ASSERT(P2PHASE(size, args->stk_align) == 0);
1557 	ASSERT((ssize_t)STK_AVAIL(args) >= 0);
1558 
1559 	if (size > args->ncargs) {
1560 		kmem_free(args->stk_base, args->stk_size);
1561 		return (E2BIG);
1562 	}
1563 
1564 	/*
1565 	 * Leave only the current lwp and force the other lwps to exit.
1566 	 * If another lwp beat us to the punch by calling exit(), bail out.
1567 	 */
1568 	if ((error = exitlwps(0)) != 0) {
1569 		kmem_free(args->stk_base, args->stk_size);
1570 		return (error);
1571 	}
1572 
1573 	/*
1574 	 * Revoke any doors created by the process.
1575 	 */
1576 	if (p->p_door_list)
1577 		door_exit();
1578 
1579 	/*
1580 	 * Release schedctl data structures.
1581 	 */
1582 	if (p->p_pagep)
1583 		schedctl_proc_cleanup();
1584 
1585 	/*
1586 	 * Clean up any DTrace helpers for the process.
1587 	 */
1588 	if (p->p_dtrace_helpers != NULL) {
1589 		ASSERT(dtrace_helpers_cleanup != NULL);
1590 		(*dtrace_helpers_cleanup)();
1591 	}
1592 
1593 	mutex_enter(&p->p_lock);
1594 	/*
1595 	 * Cleanup the DTrace provider associated with this process.
1596 	 */
1597 	if (p->p_dtrace_probes) {
1598 		ASSERT(dtrace_fasttrap_exec_ptr != NULL);
1599 		dtrace_fasttrap_exec_ptr(p);
1600 	}
1601 	mutex_exit(&p->p_lock);
1602 
1603 	/*
1604 	 * discard the lwpchan cache.
1605 	 */
1606 	if (p->p_lcp != NULL)
1607 		lwpchan_destroy_cache(1);
1608 
1609 	/*
1610 	 * Delete the POSIX timers.
1611 	 */
1612 	if (p->p_itimer != NULL)
1613 		timer_exit();
1614 
1615 #ifdef C2_AUDIT
1616 	if (audit_active)
1617 		audit_exec(args->stk_base, args->stk_base + args->arglen,
1618 		    args->na - args->ne, args->ne);
1619 #endif
1620 
1621 	/*
1622 	 * Ensure that we don't change resource associations while we
1623 	 * change address spaces.
1624 	 */
1625 	mutex_enter(&p->p_lock);
1626 	pool_barrier_enter();
1627 	mutex_exit(&p->p_lock);
1628 
1629 	/*
1630 	 * Destroy the old address space and create a new one.
1631 	 * From here on, any errors are fatal to the exec()ing process.
1632 	 * On error we return -1, which means the caller must SIGKILL
1633 	 * the process.
1634 	 */
1635 	relvm();
1636 
1637 	mutex_enter(&p->p_lock);
1638 	pool_barrier_exit();
1639 	mutex_exit(&p->p_lock);
1640 
1641 	up->u_execsw = args->execswp;
1642 
1643 	p->p_brkbase = NULL;
1644 	p->p_brksize = 0;
1645 	p->p_stksize = 0;
1646 	p->p_model = args->to_model;
1647 	p->p_usrstack = usrstack;
1648 	p->p_stkprot = args->stk_prot;
1649 	p->p_datprot = args->dat_prot;
1650 
1651 	/*
1652 	 * Reset resource controls such that all controls are again active as
1653 	 * well as appropriate to the potentially new address model for the
1654 	 * process.
1655 	 */
1656 	e.rcep_p.proc = p;
1657 	e.rcep_t = RCENTITY_PROCESS;
1658 	rctl_set_reset(p->p_rctls, p, &e);
1659 
1660 	if (exec_lpg_disable == 0) {
1661 #ifdef DEBUG
1662 		uint_t pgsizes = page_num_pagesizes();
1663 		uint_t szc;
1664 #endif
1665 		p->p_brkpageszc = args->brkpageszc;
1666 		p->p_stkpageszc = args->stkpageszc;
1667 
1668 		if (p->p_brkpageszc == 0) {
1669 			p->p_brkpageszc = page_szc(map_pgsz(MAPPGSZ_HEAP,
1670 			    p, 0, 0, NULL));
1671 		}
1672 		if (p->p_stkpageszc == 0) {
1673 			p->p_stkpageszc = page_szc(map_pgsz(MAPPGSZ_STK,
1674 			    p, 0, 0, NULL));
1675 		}
1676 
1677 #ifdef DEBUG
1678 		if (mpss_brkpgszsel != 0) {
1679 			if (mpss_brkpgszsel == -1) {
1680 				szc = ((uint_t)gethrtime() >> 8) % pgsizes;
1681 			} else {
1682 				szc = mpss_brkpgszsel % pgsizes;
1683 			}
1684 			p->p_brkpageszc = szc;
1685 		}
1686 
1687 		if (mpss_stkpgszsel != 0) {
1688 			if (mpss_stkpgszsel == -1) {
1689 				szc = ((uint_t)gethrtime() >> 7) % pgsizes;
1690 			} else {
1691 				szc = mpss_stkpgszsel % pgsizes;
1692 			}
1693 			p->p_stkpageszc = szc;
1694 		}
1695 
1696 #endif
1697 		mutex_enter(&p->p_lock);
1698 		p->p_flag |= SAUTOLPG;	/* kernel controls page sizes */
1699 		mutex_exit(&p->p_lock);
1700 
1701 	} else {
1702 		p->p_brkpageszc = 0;
1703 		p->p_stkpageszc = 0;
1704 	}
1705 
1706 	exec_set_sp(size);
1707 
1708 	as = as_alloc();
1709 	p->p_as = as;
1710 	if (p->p_model == DATAMODEL_ILP32)
1711 		as->a_userlimit = (caddr_t)USERLIMIT32;
1712 	(void) hat_setup(as->a_hat, HAT_ALLOC);
1713 
1714 	/*
1715 	 * Finally, write out the contents of the new stack.
1716 	 */
1717 	error = stk_copyout(args, usrstack, auxvpp, up);
1718 	kmem_free(args->stk_base, args->stk_size);
1719 	return (error);
1720 }
1721