xref: /openbsd-src/sys/kern/kern_exec.c (revision ff0e7be1ebbcc809ea8ad2b6dafe215824da9e46)
1 /*	$OpenBSD: kern_exec.c,v 1.248 2023/05/30 08:30:01 jsg Exp $	*/
2 /*	$NetBSD: kern_exec.c,v 1.75 1996/02/09 18:59:28 christos Exp $	*/
3 
4 /*-
5  * Copyright (C) 1993, 1994 Christopher G. Demetriou
6  * Copyright (C) 1992 Wolfgang Solfrank.
7  * Copyright (C) 1992 TooLs GmbH.
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by TooLs GmbH.
21  * 4. The name of TooLs GmbH may not be used to endorse or promote products
22  *    derived from this software without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
25  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
26  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
27  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
30  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/filedesc.h>
39 #include <sys/proc.h>
40 #include <sys/user.h>
41 #include <sys/mount.h>
42 #include <sys/malloc.h>
43 #include <sys/pool.h>
44 #include <sys/namei.h>
45 #include <sys/vnode.h>
46 #include <sys/fcntl.h>
47 #include <sys/file.h>
48 #include <sys/acct.h>
49 #include <sys/exec.h>
50 #include <sys/exec_elf.h>
51 #include <sys/ktrace.h>
52 #include <sys/resourcevar.h>
53 #include <sys/mman.h>
54 #include <sys/signalvar.h>
55 #include <sys/stat.h>
56 #include <sys/conf.h>
57 #include <sys/pledge.h>
58 #ifdef SYSVSHM
59 #include <sys/shm.h>
60 #endif
61 
62 #include <sys/syscallargs.h>
63 
64 #include <uvm/uvm_extern.h>
65 #include <machine/tcb.h>
66 
67 #include <sys/timetc.h>
68 
69 struct uvm_object *sigobject;		/* shared sigcode object */
70 vaddr_t sigcode_va;
71 vsize_t sigcode_sz;
72 struct uvm_object *timekeep_object;
73 struct timekeep *timekeep;
74 
75 void	unveil_destroy(struct process *ps);
76 
77 const struct kmem_va_mode kv_exec = {
78 	.kv_wait = 1,
79 	.kv_map = &exec_map
80 };
81 
82 /*
83  * Map the shared signal code.
84  */
85 int exec_sigcode_map(struct process *);
86 
87 /*
88  * Map the shared timekeep page.
89  */
90 int exec_timekeep_map(struct process *);
91 
92 /*
93  * If non-zero, stackgap_random specifies the upper limit of the random gap size
94  * added to the fixed stack position. Must be n^2.
95  */
96 int stackgap_random = STACKGAP_RANDOM;
97 
98 /*
99  * check exec:
100  * given an "executable" described in the exec package's namei info,
101  * see what we can do with it.
102  *
103  * ON ENTRY:
104  *	exec package with appropriate namei info
105  *	proc pointer of exec'ing proc
106  *	NO SELF-LOCKED VNODES
107  *
108  * ON EXIT:
109  *	error:	nothing held, etc.  exec header still allocated.
110  *	ok:	filled exec package, one locked vnode.
111  *
112  * EXEC SWITCH ENTRY:
113  * 	Locked vnode to check, exec package, proc.
114  *
115  * EXEC SWITCH EXIT:
116  *	ok:	return 0, filled exec package, one locked vnode.
117  *	error:	destructive:
118  *			everything deallocated except exec header.
119  *		non-destructive:
120  *			error code, locked vnode, exec header unmodified
121  */
122 int
123 check_exec(struct proc *p, struct exec_package *epp)
124 {
125 	int error, i;
126 	struct vnode *vp;
127 	struct nameidata *ndp;
128 	size_t resid;
129 
130 	ndp = epp->ep_ndp;
131 	ndp->ni_cnd.cn_nameiop = LOOKUP;
132 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
133 	if (epp->ep_flags & EXEC_INDIR)
134 		ndp->ni_cnd.cn_flags |= BYPASSUNVEIL;
135 	/* first get the vnode */
136 	if ((error = namei(ndp)) != 0)
137 		return (error);
138 	epp->ep_vp = vp = ndp->ni_vp;
139 
140 	/* check for regular file */
141 	if (vp->v_type != VREG) {
142 		error = EACCES;
143 		goto bad1;
144 	}
145 
146 	/* get attributes */
147 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
148 		goto bad1;
149 
150 	/* Check mount point */
151 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
152 		error = EACCES;
153 		goto bad1;
154 	}
155 
156 	/* SUID programs may not be started with execpromises */
157 	if ((epp->ep_vap->va_mode & (VSUID | VSGID)) &&
158 	    (p->p_p->ps_flags & PS_EXECPLEDGE)) {
159 		error = EACCES;
160 		goto bad1;
161 	}
162 
163 	if ((vp->v_mount->mnt_flag & MNT_NOSUID))
164 		epp->ep_vap->va_mode &= ~(VSUID | VSGID);
165 
166 	/* check access.  for root we have to see if any exec bit on */
167 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
168 		goto bad1;
169 	if ((epp->ep_vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
170 		error = EACCES;
171 		goto bad1;
172 	}
173 
174 	/* try to open it */
175 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
176 		goto bad1;
177 
178 	/* unlock vp, we need it unlocked from here */
179 	VOP_UNLOCK(vp);
180 
181 	/* now we have the file, get the exec header */
182 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
183 	    UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
184 	if (error)
185 		goto bad2;
186 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
187 
188 	/*
189 	 * set up the vmcmds for creation of the process
190 	 * address space
191 	 */
192 	error = ENOEXEC;
193 	for (i = 0; i < nexecs && error != 0; i++) {
194 		int newerror;
195 
196 		if (execsw[i].es_check == NULL)
197 			continue;
198 		newerror = (*execsw[i].es_check)(p, epp);
199 		/* make sure the first "interesting" error code is saved. */
200 		if (!newerror || error == ENOEXEC)
201 			error = newerror;
202 		if (epp->ep_flags & EXEC_DESTR && error != 0)
203 			return (error);
204 	}
205 	if (!error) {
206 		/* check that entry point is sane */
207 		if (epp->ep_entry > VM_MAXUSER_ADDRESS) {
208 			error = ENOEXEC;
209 		}
210 
211 		/* check limits */
212 		if ((epp->ep_tsize > MAXTSIZ) ||
213 		    (epp->ep_dsize > lim_cur(RLIMIT_DATA)))
214 			error = ENOMEM;
215 
216 		if (!error)
217 			return (0);
218 	}
219 
220 	/*
221 	 * free any vmspace-creation commands,
222 	 * and release their references
223 	 */
224 	kill_vmcmds(&epp->ep_vmcmds);
225 
226 bad2:
227 	/*
228 	 * close the vnode, free the pathname buf, and punt.
229 	 */
230 	vn_close(vp, FREAD, p->p_ucred, p);
231 	pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf);
232 	return (error);
233 
234 bad1:
235 	/*
236 	 * free the namei pathname buffer, and put the vnode
237 	 * (which we don't yet have open).
238 	 */
239 	pool_put(&namei_pool, ndp->ni_cnd.cn_pnbuf);
240 	vput(vp);
241 	return (error);
242 }
243 
244 /*
245  * exec system call
246  */
247 int
248 sys_execve(struct proc *p, void *v, register_t *retval)
249 {
250 	struct sys_execve_args /* {
251 		syscallarg(const char *) path;
252 		syscallarg(char *const *) argp;
253 		syscallarg(char *const *) envp;
254 	} */ *uap = v;
255 	int error;
256 	struct exec_package pack;
257 	struct nameidata nid;
258 	struct vattr attr;
259 	struct ucred *cred = p->p_ucred;
260 	char *argp;
261 	char * const *cpp, *dp, *sp;
262 #ifdef KTRACE
263 	char *env_start;
264 #endif
265 	struct process *pr = p->p_p;
266 	long argc, envc;
267 	size_t len, sgap, dstsize;
268 #ifdef MACHINE_STACK_GROWS_UP
269 	size_t slen;
270 #endif
271 	vaddr_t pc = PROC_PC(p);
272 	char *stack;
273 	struct ps_strings arginfo;
274 	struct vmspace *vm = p->p_vmspace;
275 	struct vnode *otvp;
276 
277 	if (vm->vm_execve &&
278 	    (pc >= vm->vm_execve_end || pc < vm->vm_execve)) {
279 		printf("%s(%d): execve %lx outside %lx-%lx\n", pr->ps_comm,
280 		    pr->ps_pid, pc, vm->vm_execve, vm->vm_execve_end);
281 		p->p_p->ps_acflag |= AEXECVE;
282 		sigabort(p);
283 		return (0);
284 	}
285 
286 	/* get other threads to stop */
287 	if ((error = single_thread_set(p, SINGLE_UNWIND, 1)))
288 		return (error);
289 
290 	/*
291 	 * Cheap solution to complicated problems.
292 	 * Mark this process as "leave me alone, I'm execing".
293 	 */
294 	atomic_setbits_int(&pr->ps_flags, PS_INEXEC);
295 
296 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
297 	nid.ni_pledge = PLEDGE_EXEC;
298 	nid.ni_unveil = UNVEIL_EXEC;
299 
300 	/*
301 	 * initialize the fields of the exec package.
302 	 */
303 	pack.ep_name = (char *)SCARG(uap, path);
304 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
305 	pack.ep_hdrlen = exec_maxhdrsz;
306 	pack.ep_hdrvalid = 0;
307 	pack.ep_ndp = &nid;
308 	pack.ep_interp = NULL;
309 	pack.ep_args = NULL;
310 	pack.ep_auxinfo = NULL;
311 	VMCMDSET_INIT(&pack.ep_vmcmds);
312 	pack.ep_vap = &attr;
313 	pack.ep_flags = 0;
314 
315 	/* see if we can run it. */
316 	if ((error = check_exec(p, &pack)) != 0) {
317 		goto freehdr;
318 	}
319 
320 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
321 
322 	/* allocate an argument buffer */
323 	argp = km_alloc(NCARGS, &kv_exec, &kp_pageable, &kd_waitok);
324 #ifdef DIAGNOSTIC
325 	if (argp == NULL)
326 		panic("execve: argp == NULL");
327 #endif
328 	dp = argp;
329 	argc = 0;
330 
331 	/*
332 	 * Copy the fake args list, if there's one, freeing it as we go.
333 	 * exec_script_makecmds() allocates either 2 or 3 fake args bounded
334 	 * by MAXINTERP + MAXPATHLEN < NCARGS so no overflow can happen.
335 	 */
336 	if (pack.ep_flags & EXEC_HASARGL) {
337 		dstsize = NCARGS;
338 		for(; pack.ep_fa[argc] != NULL; argc++) {
339 			len = strlcpy(dp, pack.ep_fa[argc], dstsize);
340 			len++;
341 			dp += len; dstsize -= len;
342 			if (pack.ep_fa[argc+1] != NULL)
343 				free(pack.ep_fa[argc], M_EXEC, len);
344 			else
345 				free(pack.ep_fa[argc], M_EXEC, MAXPATHLEN);
346 		}
347 		free(pack.ep_fa, M_EXEC, 4 * sizeof(char *));
348 		pack.ep_flags &= ~EXEC_HASARGL;
349 	}
350 
351 	/* Now get argv & environment */
352 	if (!(cpp = SCARG(uap, argp))) {
353 		error = EFAULT;
354 		goto bad;
355 	}
356 
357 	if (pack.ep_flags & EXEC_SKIPARG)
358 		cpp++;
359 
360 	while (1) {
361 		len = argp + ARG_MAX - dp;
362 		if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
363 			goto bad;
364 		if (!sp)
365 			break;
366 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
367 			if (error == ENAMETOOLONG)
368 				error = E2BIG;
369 			goto bad;
370 		}
371 		dp += len;
372 		cpp++;
373 		argc++;
374 	}
375 
376 	/* must have at least one argument */
377 	if (argc == 0) {
378 		error = EINVAL;
379 		goto bad;
380 	}
381 
382 #ifdef KTRACE
383 	if (KTRPOINT(p, KTR_EXECARGS))
384 		ktrexec(p, KTR_EXECARGS, argp, dp - argp);
385 #endif
386 
387 	envc = 0;
388 	/* environment does not need to be there */
389 	if ((cpp = SCARG(uap, envp)) != NULL ) {
390 #ifdef KTRACE
391 		env_start = dp;
392 #endif
393 		while (1) {
394 			len = argp + ARG_MAX - dp;
395 			if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
396 				goto bad;
397 			if (!sp)
398 				break;
399 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
400 				if (error == ENAMETOOLONG)
401 					error = E2BIG;
402 				goto bad;
403 			}
404 			dp += len;
405 			cpp++;
406 			envc++;
407 		}
408 
409 #ifdef KTRACE
410 		if (KTRPOINT(p, KTR_EXECENV))
411 			ktrexec(p, KTR_EXECENV, env_start, dp - env_start);
412 #endif
413 	}
414 
415 	dp = (char *)(((long)dp + _STACKALIGNBYTES) & ~_STACKALIGNBYTES);
416 
417 	/*
418 	 * If we have enabled random stackgap, the stack itself has already
419 	 * been moved from a random location, but is still aligned to a page
420 	 * boundary.  Provide the lower bits of random placement now.
421 	 */
422 	if (stackgap_random == 0) {
423 		sgap = 0;
424 	} else {
425 		sgap = arc4random() & PAGE_MASK;
426 		sgap = (sgap + _STACKALIGNBYTES) & ~_STACKALIGNBYTES;
427 	}
428 
429 	/* Now check if args & environ fit into new stack */
430 	len = ((argc + envc + 2 + ELF_AUX_WORDS) * sizeof(char *) +
431 	    sizeof(long) + dp + sgap + sizeof(struct ps_strings)) - argp;
432 
433 	len = (len + _STACKALIGNBYTES) &~ _STACKALIGNBYTES;
434 
435 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
436 		error = ENOMEM;
437 		goto bad;
438 	}
439 
440 	/* adjust "active stack depth" for process VSZ */
441 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
442 
443 	/*
444 	 * we're committed: any further errors will kill the process, so
445 	 * kill the other threads now.
446 	 */
447 	single_thread_set(p, SINGLE_EXIT, 1);
448 
449 	/*
450 	 * Prepare vmspace for remapping. Note that uvmspace_exec can replace
451 	 * ps_vmspace!
452 	 */
453 	uvmspace_exec(p, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
454 
455 	vm = pr->ps_vmspace;
456 	/* Now map address space */
457 	vm->vm_taddr = (char *)trunc_page(pack.ep_taddr);
458 	vm->vm_tsize = atop(round_page(pack.ep_taddr + pack.ep_tsize) -
459 	    trunc_page(pack.ep_taddr));
460 	vm->vm_daddr = (char *)trunc_page(pack.ep_daddr);
461 	vm->vm_dsize = atop(round_page(pack.ep_daddr + pack.ep_dsize) -
462 	    trunc_page(pack.ep_daddr));
463 	vm->vm_dused = 0;
464 	vm->vm_ssize = atop(round_page(pack.ep_ssize));
465 	vm->vm_maxsaddr = (char *)pack.ep_maxsaddr;
466 	vm->vm_minsaddr = (char *)pack.ep_minsaddr;
467 
468 	/* create the new process's VM space by running the vmcmds */
469 #ifdef DIAGNOSTIC
470 	if (pack.ep_vmcmds.evs_used == 0)
471 		panic("execve: no vmcmds");
472 #endif
473 	error = exec_process_vmcmds(p, &pack);
474 
475 	/* if an error happened, deallocate and punt */
476 	if (error)
477 		goto exec_abort;
478 
479 #ifdef MACHINE_STACK_GROWS_UP
480 	pr->ps_strings = (vaddr_t)vm->vm_maxsaddr + sgap;
481         if (uvm_map_protect(&vm->vm_map, (vaddr_t)vm->vm_maxsaddr,
482             trunc_page(pr->ps_strings), PROT_NONE, 0, TRUE, FALSE))
483                 goto exec_abort;
484 #else
485 	pr->ps_strings = (vaddr_t)vm->vm_minsaddr - sizeof(arginfo) - sgap;
486         if (uvm_map_protect(&vm->vm_map,
487             round_page(pr->ps_strings + sizeof(arginfo)),
488             (vaddr_t)vm->vm_minsaddr, PROT_NONE, 0, TRUE, FALSE))
489                 goto exec_abort;
490 #endif
491 
492 	memset(&arginfo, 0, sizeof(arginfo));
493 
494 	/* remember information about the process */
495 	arginfo.ps_nargvstr = argc;
496 	arginfo.ps_nenvstr = envc;
497 
498 #ifdef MACHINE_STACK_GROWS_UP
499 	stack = (char *)vm->vm_maxsaddr + sizeof(arginfo) + sgap;
500 	slen = len - sizeof(arginfo) - sgap;
501 #else
502 	stack = (char *)(vm->vm_minsaddr - len);
503 #endif
504 	/* Now copy argc, args & environ to new stack */
505 	if (!copyargs(&pack, &arginfo, stack, argp))
506 		goto exec_abort;
507 
508 	pr->ps_auxinfo = (vaddr_t)pack.ep_auxinfo;
509 
510 	/* copy out the process's ps_strings structure */
511 	if (copyout(&arginfo, (char *)pr->ps_strings, sizeof(arginfo)))
512 		goto exec_abort;
513 
514 	stopprofclock(pr);	/* stop profiling */
515 	fdcloseexec(p);		/* handle close on exec */
516 	execsigs(p);		/* reset caught signals */
517 	TCB_SET(p, NULL);	/* reset the TCB address */
518 	pr->ps_kbind_addr = 0;	/* reset the kbind bits */
519 	pr->ps_kbind_cookie = 0;
520 	arc4random_buf(&pr->ps_sigcookie, sizeof pr->ps_sigcookie);
521 
522 	/* set command name & other accounting info */
523 	memset(pr->ps_comm, 0, sizeof(pr->ps_comm));
524 	strlcpy(pr->ps_comm, nid.ni_cnd.cn_nameptr, sizeof(pr->ps_comm));
525 	pr->ps_acflag &= ~AFORK;
526 
527 	/* record proc's vnode, for use by sysctl */
528 	otvp = pr->ps_textvp;
529 	vref(pack.ep_vp);
530 	pr->ps_textvp = pack.ep_vp;
531 	if (otvp)
532 		vrele(otvp);
533 
534 	/*
535 	 * XXX As a transition mechanism, we don't enforce branch
536 	 * target control flow integrity on partitions mounted with
537 	 * the wxallowed flag.
538 	 */
539 	if (pr->ps_textvp->v_mount &&
540 	    (pr->ps_textvp->v_mount->mnt_flag & MNT_WXALLOWED))
541 		pack.ep_flags |= EXEC_NOBTCFI;
542 	/* XXX XXX But enable it for chrome. */
543 	if (strcmp(p->p_p->ps_comm, "chrome") == 0)
544 		pack.ep_flags &= ~EXEC_NOBTCFI;
545 
546 	atomic_setbits_int(&pr->ps_flags, PS_EXEC);
547 	if (pr->ps_flags & PS_PPWAIT) {
548 		atomic_clearbits_int(&pr->ps_flags, PS_PPWAIT);
549 		atomic_clearbits_int(&pr->ps_pptr->ps_flags, PS_ISPWAIT);
550 		wakeup(pr->ps_pptr);
551 	}
552 
553 	/*
554 	 * If process does execve() while it has a mismatched real,
555 	 * effective, or saved uid/gid, we set PS_SUGIDEXEC.
556 	 */
557 	if (cred->cr_uid != cred->cr_ruid ||
558 	    cred->cr_uid != cred->cr_svuid ||
559 	    cred->cr_gid != cred->cr_rgid ||
560 	    cred->cr_gid != cred->cr_svgid)
561 		atomic_setbits_int(&pr->ps_flags, PS_SUGIDEXEC);
562 	else
563 		atomic_clearbits_int(&pr->ps_flags, PS_SUGIDEXEC);
564 
565 	if (pr->ps_flags & PS_EXECPLEDGE) {
566 		pr->ps_pledge = pr->ps_execpledge;
567 		atomic_setbits_int(&pr->ps_flags, PS_PLEDGE);
568 	} else {
569 		atomic_clearbits_int(&pr->ps_flags, PS_PLEDGE);
570 		pr->ps_pledge = 0;
571 		/* XXX XXX XXX XXX */
572 		/* Clear our unveil paths out so the child
573 		 * starts afresh
574 		 */
575 		unveil_destroy(pr);
576 		pr->ps_uvdone = 0;
577 	}
578 
579 	/*
580 	 * deal with set[ug]id.
581 	 * MNT_NOEXEC has already been used to disable s[ug]id.
582 	 */
583 	if ((attr.va_mode & (VSUID | VSGID)) && proc_cansugid(p)) {
584 		int i;
585 
586 		atomic_setbits_int(&pr->ps_flags, PS_SUGID|PS_SUGIDEXEC);
587 
588 #ifdef KTRACE
589 		/*
590 		 * If process is being ktraced, turn off - unless
591 		 * root set it.
592 		 */
593 		if (pr->ps_tracevp && !(pr->ps_traceflag & KTRFAC_ROOT))
594 			ktrcleartrace(pr);
595 #endif
596 		p->p_ucred = cred = crcopy(cred);
597 		if (attr.va_mode & VSUID)
598 			cred->cr_uid = attr.va_uid;
599 		if (attr.va_mode & VSGID)
600 			cred->cr_gid = attr.va_gid;
601 
602 		/*
603 		 * For set[ug]id processes, a few caveats apply to
604 		 * stdin, stdout, and stderr.
605 		 */
606 		error = 0;
607 		fdplock(p->p_fd);
608 		for (i = 0; i < 3; i++) {
609 			struct file *fp = NULL;
610 
611 			/*
612 			 * NOTE - This will never return NULL because of
613 			 * immature fds. The file descriptor table is not
614 			 * shared because we're suid.
615 			 */
616 			fp = fd_getfile(p->p_fd, i);
617 
618 			/*
619 			 * Ensure that stdin, stdout, and stderr are already
620 			 * allocated.  We do not want userland to accidentally
621 			 * allocate descriptors in this range which has implied
622 			 * meaning to libc.
623 			 */
624 			if (fp == NULL) {
625 				short flags = FREAD | (i == 0 ? 0 : FWRITE);
626 				struct vnode *vp;
627 				int indx;
628 
629 				if ((error = falloc(p, &fp, &indx)) != 0)
630 					break;
631 #ifdef DIAGNOSTIC
632 				if (indx != i)
633 					panic("sys_execve: falloc indx != i");
634 #endif
635 				if ((error = cdevvp(getnulldev(), &vp)) != 0) {
636 					fdremove(p->p_fd, indx);
637 					closef(fp, p);
638 					break;
639 				}
640 				if ((error = VOP_OPEN(vp, flags, cred, p)) != 0) {
641 					fdremove(p->p_fd, indx);
642 					closef(fp, p);
643 					vrele(vp);
644 					break;
645 				}
646 				if (flags & FWRITE)
647 					vp->v_writecount++;
648 				fp->f_flag = flags;
649 				fp->f_type = DTYPE_VNODE;
650 				fp->f_ops = &vnops;
651 				fp->f_data = (caddr_t)vp;
652 				fdinsert(p->p_fd, indx, 0, fp);
653 			}
654 			FRELE(fp, p);
655 		}
656 		fdpunlock(p->p_fd);
657 		if (error)
658 			goto exec_abort;
659 	} else
660 		atomic_clearbits_int(&pr->ps_flags, PS_SUGID);
661 
662 	/*
663 	 * Reset the saved ugids and update the process's copy of the
664 	 * creds if the creds have been changed
665 	 */
666 	if (cred->cr_uid != cred->cr_svuid ||
667 	    cred->cr_gid != cred->cr_svgid) {
668 		/* make sure we have unshared ucreds */
669 		p->p_ucred = cred = crcopy(cred);
670 		cred->cr_svuid = cred->cr_uid;
671 		cred->cr_svgid = cred->cr_gid;
672 	}
673 
674 	if (pr->ps_ucred != cred) {
675 		struct ucred *ocred;
676 
677 		ocred = pr->ps_ucred;
678 		crhold(cred);
679 		pr->ps_ucred = cred;
680 		crfree(ocred);
681 	}
682 
683 	if (pr->ps_flags & PS_SUGIDEXEC) {
684 		cancel_all_itimers();
685 	}
686 
687 	/* reset CPU time usage for the thread, but not the process */
688 	timespecclear(&p->p_tu.tu_runtime);
689 	p->p_tu.tu_uticks = p->p_tu.tu_sticks = p->p_tu.tu_iticks = 0;
690 
691 	memset(p->p_name, 0, sizeof p->p_name);
692 
693 	km_free(argp, NCARGS, &kv_exec, &kp_pageable);
694 
695 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
696 	vn_close(pack.ep_vp, FREAD, cred, p);
697 
698 	/*
699 	 * notify others that we exec'd
700 	 */
701 	knote_locked(&pr->ps_klist, NOTE_EXEC);
702 
703 	/* map the process's timekeep page, needs to be before exec_elf_fixup */
704 	if (exec_timekeep_map(pr))
705 		goto free_pack_abort;
706 
707 	/* setup new registers and do misc. setup. */
708 	if (exec_elf_fixup(p, &pack) != 0)
709 		goto free_pack_abort;
710 #ifdef MACHINE_STACK_GROWS_UP
711 	setregs(p, &pack, (u_long)stack + slen, &arginfo);
712 #else
713 	setregs(p, &pack, (u_long)stack, &arginfo);
714 #endif
715 
716 	/* map the process's signal trampoline code */
717 	if (exec_sigcode_map(pr))
718 		goto free_pack_abort;
719 
720 #ifdef __HAVE_EXEC_MD_MAP
721 	/* perform md specific mappings that process might need */
722 	if (exec_md_map(p, &pack))
723 		goto free_pack_abort;
724 #endif
725 
726 	if (pr->ps_flags & PS_TRACED)
727 		psignal(p, SIGTRAP);
728 
729 	free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen);
730 
731 	p->p_descfd = 255;
732 	if ((pack.ep_flags & EXEC_HASFD) && pack.ep_fd < 255)
733 		p->p_descfd = pack.ep_fd;
734 
735 	if (pack.ep_flags & EXEC_WXNEEDED)
736 		atomic_setbits_int(&p->p_p->ps_flags, PS_WXNEEDED);
737 	else
738 		atomic_clearbits_int(&p->p_p->ps_flags, PS_WXNEEDED);
739 
740 	atomic_clearbits_int(&pr->ps_flags, PS_INEXEC);
741 	single_thread_clear(p, P_SUSPSIG);
742 
743 	/* setregs() sets up all the registers, so just 'return' */
744 	return EJUSTRETURN;
745 
746 bad:
747 	/* free the vmspace-creation commands, and release their references */
748 	kill_vmcmds(&pack.ep_vmcmds);
749 	/* kill any opened file descriptor, if necessary */
750 	if (pack.ep_flags & EXEC_HASFD) {
751 		pack.ep_flags &= ~EXEC_HASFD;
752 		fdplock(p->p_fd);
753 		/* fdrelease unlocks p->p_fd. */
754 		(void) fdrelease(p, pack.ep_fd);
755 	}
756 	if (pack.ep_interp != NULL)
757 		pool_put(&namei_pool, pack.ep_interp);
758 	free(pack.ep_args, M_TEMP, sizeof *pack.ep_args);
759 	/* close and put the exec'd file */
760 	vn_close(pack.ep_vp, FREAD, cred, p);
761 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
762 	km_free(argp, NCARGS, &kv_exec, &kp_pageable);
763 
764 freehdr:
765 	free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen);
766 	atomic_clearbits_int(&pr->ps_flags, PS_INEXEC);
767 	single_thread_clear(p, P_SUSPSIG);
768 
769 	return (error);
770 
771 exec_abort:
772 	/*
773 	 * the old process doesn't exist anymore.  exit gracefully.
774 	 * get rid of the (new) address space we have created, if any, get rid
775 	 * of our namei data and vnode, and exit noting failure
776 	 */
777 	uvm_unmap(&vm->vm_map, VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
778 	if (pack.ep_interp != NULL)
779 		pool_put(&namei_pool, pack.ep_interp);
780 	free(pack.ep_args, M_TEMP, sizeof *pack.ep_args);
781 	pool_put(&namei_pool, nid.ni_cnd.cn_pnbuf);
782 	vn_close(pack.ep_vp, FREAD, cred, p);
783 	km_free(argp, NCARGS, &kv_exec, &kp_pageable);
784 
785 free_pack_abort:
786 	free(pack.ep_hdr, M_EXEC, pack.ep_hdrlen);
787 	exit1(p, 0, SIGABRT, EXIT_NORMAL);
788 
789 	/* NOTREACHED */
790 	atomic_clearbits_int(&pr->ps_flags, PS_INEXEC);
791 
792 	return (0);
793 }
794 
795 
796 int
797 copyargs(struct exec_package *pack, struct ps_strings *arginfo, void *stack,
798     void *argp)
799 {
800 	char **cpp = stack;
801 	char *dp, *sp;
802 	size_t len;
803 	void *nullp = NULL;
804 	long argc = arginfo->ps_nargvstr;
805 	int envc = arginfo->ps_nenvstr;
806 
807 	if (copyout(&argc, cpp++, sizeof(argc)))
808 		return (0);
809 
810 	dp = (char *) (cpp + argc + envc + 2 + ELF_AUX_WORDS);
811 	sp = argp;
812 
813 	/* XXX don't copy them out, remap them! */
814 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
815 
816 	for (; --argc >= 0; sp += len, dp += len)
817 		if (copyout(&dp, cpp++, sizeof(dp)) ||
818 		    copyoutstr(sp, dp, ARG_MAX, &len))
819 			return (0);
820 
821 	if (copyout(&nullp, cpp++, sizeof(nullp)))
822 		return (0);
823 
824 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
825 
826 	for (; --envc >= 0; sp += len, dp += len)
827 		if (copyout(&dp, cpp++, sizeof(dp)) ||
828 		    copyoutstr(sp, dp, ARG_MAX, &len))
829 			return (0);
830 
831 	if (copyout(&nullp, cpp++, sizeof(nullp)))
832 		return (0);
833 
834 	/* if this process needs auxinfo, note where to place it */
835 	if (pack->ep_args != NULL)
836 		pack->ep_auxinfo = cpp;
837 
838 	return (1);
839 }
840 
841 int
842 exec_sigcode_map(struct process *pr)
843 {
844 	extern char sigcode[], esigcode[], sigcoderet[];
845 	vsize_t sz;
846 
847 	sz = (vaddr_t)esigcode - (vaddr_t)sigcode;
848 
849 	/*
850 	 * If we don't have a sigobject yet, create one.
851 	 *
852 	 * sigobject is an anonymous memory object (just like SYSV shared
853 	 * memory) that we keep a permanent reference to and that we map
854 	 * in all processes that need this sigcode. The creation is simple,
855 	 * we create an object, add a permanent reference to it, map it in
856 	 * kernel space, copy out the sigcode to it and unmap it.  Then we map
857 	 * it with PROT_EXEC into the process just the way sys_mmap would map it.
858 	 */
859 	if (sigobject == NULL) {
860 		extern int sigfillsiz;
861 		extern u_char sigfill[];
862 		size_t off, left;
863 		vaddr_t va;
864 		int r;
865 
866 		sigobject = uao_create(sz, 0);
867 		uao_reference(sigobject);	/* permanent reference */
868 
869 		if ((r = uvm_map(kernel_map, &va, round_page(sz), sigobject,
870 		    0, 0, UVM_MAPFLAG(PROT_READ | PROT_WRITE, PROT_READ | PROT_WRITE,
871 		    MAP_INHERIT_SHARE, MADV_RANDOM, 0)))) {
872 			uao_detach(sigobject);
873 			return (ENOMEM);
874 		}
875 
876 		for (off = 0, left = round_page(sz); left != 0;
877 		    off += sigfillsiz) {
878 			size_t chunk = ulmin(left, sigfillsiz);
879 			memcpy((caddr_t)va + off, sigfill, chunk);
880 			left -= chunk;
881 		}
882 		memcpy((caddr_t)va, sigcode, sz);
883 
884 		(void) uvm_map_protect(kernel_map, va, round_page(sz),
885 		    PROT_READ, 0, FALSE, FALSE);
886 		sigcode_va = va;
887 		sigcode_sz = round_page(sz);
888 	}
889 
890 	pr->ps_sigcode = 0; /* no hint */
891 	uao_reference(sigobject);
892 	if (uvm_map(&pr->ps_vmspace->vm_map, &pr->ps_sigcode, round_page(sz),
893 	    sigobject, 0, 0, UVM_MAPFLAG(PROT_EXEC,
894 	    PROT_READ | PROT_WRITE | PROT_EXEC, MAP_INHERIT_COPY,
895 	    MADV_RANDOM, UVM_FLAG_COPYONW | UVM_FLAG_SYSCALL))) {
896 		uao_detach(sigobject);
897 		return (ENOMEM);
898 	}
899 	uvm_map_immutable(&pr->ps_vmspace->vm_map, pr->ps_sigcode,
900 	    pr->ps_sigcode + round_page(sz), 1);
901 
902 	/* Calculate PC at point of sigreturn entry */
903 	pr->ps_sigcoderet = pr->ps_sigcode + (sigcoderet - sigcode);
904 
905 	return (0);
906 }
907 
908 int
909 exec_timekeep_map(struct process *pr)
910 {
911 	size_t timekeep_sz = round_page(sizeof(struct timekeep));
912 
913 	/*
914 	 * Similar to the sigcode object
915 	 */
916 	if (timekeep_object == NULL) {
917 		vaddr_t va = 0;
918 
919 		timekeep_object = uao_create(timekeep_sz, 0);
920 		uao_reference(timekeep_object);
921 
922 		if (uvm_map(kernel_map, &va, timekeep_sz, timekeep_object,
923 		    0, 0, UVM_MAPFLAG(PROT_READ | PROT_WRITE, PROT_READ | PROT_WRITE,
924 		    MAP_INHERIT_SHARE, MADV_RANDOM, 0))) {
925 			uao_detach(timekeep_object);
926 			timekeep_object = NULL;
927 			return (ENOMEM);
928 		}
929 		if (uvm_fault_wire(kernel_map, va, va + timekeep_sz,
930 		    PROT_READ | PROT_WRITE)) {
931 			uvm_unmap(kernel_map, va, va + timekeep_sz);
932 			uao_detach(timekeep_object);
933 			timekeep_object = NULL;
934 			return (ENOMEM);
935 		}
936 
937 		timekeep = (struct timekeep *)va;
938 		timekeep->tk_version = TK_VERSION;
939 	}
940 
941 	pr->ps_timekeep = 0; /* no hint */
942 	uao_reference(timekeep_object);
943 	if (uvm_map(&pr->ps_vmspace->vm_map, &pr->ps_timekeep, timekeep_sz,
944 	    timekeep_object, 0, 0, UVM_MAPFLAG(PROT_READ, PROT_READ,
945 	    MAP_INHERIT_COPY, MADV_RANDOM, 0))) {
946 		uao_detach(timekeep_object);
947 		return (ENOMEM);
948 	}
949 	uvm_map_immutable(&pr->ps_vmspace->vm_map, pr->ps_timekeep,
950 	    pr->ps_timekeep + timekeep_sz, 1);
951 
952 	return (0);
953 }
954