xref: /netbsd-src/sys/kern/kern_exec.c (revision 07bae7edddbb1ce4c926b2e8db425804589074c9)
1 /*	$NetBSD: kern_exec.c,v 1.68 1995/05/01 22:36:45 cgd Exp $	*/
2 
3 /*-
4  * Copyright (C) 1993, 1994 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/param.h>
36 #include <sys/systm.h>
37 #include <sys/filedesc.h>
38 #include <sys/kernel.h>
39 #include <sys/proc.h>
40 #include <sys/mount.h>
41 #include <sys/malloc.h>
42 #include <sys/namei.h>
43 #include <sys/vnode.h>
44 #include <sys/file.h>
45 #include <sys/acct.h>
46 #include <sys/exec.h>
47 #include <sys/ktrace.h>
48 #include <sys/resourcevar.h>
49 #include <sys/wait.h>
50 #include <sys/mman.h>
51 #include <sys/signalvar.h>
52 #include <sys/stat.h>
53 
54 #include <sys/syscallargs.h>
55 
56 #include <vm/vm.h>
57 #include <vm/vm_kern.h>
58 
59 #include <machine/cpu.h>
60 #include <machine/reg.h>
61 
62 /*
63  * check exec:
64  * given an "executable" described in the exec package's namei info,
65  * see what we can do with it.
66  *
67  * ON ENTRY:
68  *	exec package with appropriate namei info
69  *	proc pointer of exec'ing proc
70  *	NO SELF-LOCKED VNODES
71  *
72  * ON EXIT:
73  *	error:	nothing held, etc.  exec header still allocated.
74  *	ok:	filled exec package, one locked vnode.
75  *
76  * EXEC SWITCH ENTRY:
77  * 	Locked vnode to check, exec package, proc.
78  *
79  * EXEC SWITCH EXIT:
80  *	ok:	return 0, filled exec package, one locked vnode.
81  *	error:	destructive:
82  *			everything deallocated execept exec header.
83  *		non-descructive:
84  *			error code, locked vnode, exec header unmodified
85  */
86 int
87 check_exec(p, epp)
88 	struct proc *p;
89 	struct exec_package *epp;
90 {
91 	int error, i;
92 	struct vnode *vp;
93 	char *cp, *ep, *name;
94 	struct nameidata *ndp;
95 	int resid;
96 
97 	ndp = epp->ep_ndp;
98 	ndp->ni_cnd.cn_nameiop = LOOKUP;
99 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
100 	/* first get the vnode */
101 	if (error = namei(ndp))
102 		return error;
103 	epp->ep_vp = vp = ndp->ni_vp;
104 
105 	/* check for regular file */
106 	if (vp->v_type != VREG) {
107 		error = EACCES;
108 		goto bad1;
109 	}
110 
111 	/* get attributes */
112 	if (error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p))
113 		goto bad1;
114 
115 	/* Check mount point */
116 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
117 		error = EACCES;
118 		goto bad1;
119 	}
120 	if ((vp->v_mount->mnt_flag & MNT_NOSUID) || (p->p_flag & P_TRACED))
121 		epp->ep_vap->va_mode &= ~(VSUID | VSGID);
122 
123 	/* check access.  for root we have to see if any exec bit on */
124 	if (error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p))
125 		goto bad1;
126 	if ((epp->ep_vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
127 		error = EACCES;
128 		goto bad1;
129 	}
130 
131 	/* try to open it */
132 	if (error = VOP_OPEN(vp, FREAD, p->p_ucred, p))
133 		goto bad1;
134 
135 	/* now we have the file, get the exec header */
136 	if (error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
137 	    UIO_SYSSPACE, IO_NODELOCKED, p->p_ucred, &resid, p))
138 		goto bad2;
139 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
140 
141 	/*
142 	 * set up the vmcmds for creation of the process
143 	 * address space
144 	 */
145 	error = ENOEXEC;
146 	for (i = 0; i < nexecs && error != 0; i++) {
147 		int newerror;
148 
149 		if (execsw[i].es_check == NULL)
150 			continue;
151 
152 		newerror = (*execsw[i].es_check)(p, epp);
153 		/* make sure the first "interesting" error code is saved. */
154 		if (!newerror || error == ENOEXEC)
155 			error = newerror;
156 		if (epp->ep_flags & EXEC_DESTR && error != 0)
157 			return error;
158 	}
159 	if (!error) {
160 		/* check that entry point is sane */
161 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
162 			error = ENOEXEC;
163 
164 		/* check limits */
165 		if ((epp->ep_tsize > MAXTSIZ) ||
166 		    (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur))
167 			error = ENOMEM;
168 
169 		if (!error)
170 			return (0);
171 	}
172 
173 	/*
174 	 * free any vmspace-creation commands,
175 	 * and release their references
176 	 */
177 	kill_vmcmds(&epp->ep_vmcmds);
178 
179 bad2:
180 	/*
181 	 * unlock and close the vnode, restore the old one, free the
182 	 * pathname buf, and punt.
183 	 */
184 	VOP_UNLOCK(vp);
185 	vn_close(vp, FREAD, p->p_ucred, p);
186 	FREE(ndp->ni_cnd.cn_pnbuf, M_NAMEI);
187 	return error;
188 
189 bad1:
190 	/*
191 	 * free the namei pathname buffer, and put the vnode
192 	 * (which we don't yet have open).
193 	 */
194 	FREE(ndp->ni_cnd.cn_pnbuf, M_NAMEI);
195 	vput(vp);
196 	return error;
197 }
198 
199 /*
200  * exec system call
201  */
202 /* ARGSUSED */
203 execve(p, uap, retval)
204 	register struct proc *p;
205 	register struct execve_args /* {
206 		syscallarg(char *) path;
207 		syscallarg(char * *) argp;
208 		syscallarg(char * *) envp;
209 	} */ *uap;
210 	register_t *retval;
211 {
212 	int error, i;
213 	struct exec_package pack;
214 	struct nameidata nid;
215 	struct vattr attr;
216 	struct ucred *cred = p->p_ucred;
217 	char *argp;
218 	char **cpp, *dp, *sp;
219 	long argc, envc;
220 	size_t len;
221 	char *stack;
222 	struct ps_strings arginfo;
223 	struct vmspace *vm = p->p_vmspace;
224 	char **tmpfap;
225 	int szsigcode;
226 	extern struct emul emul_netbsd;
227 
228 	/*
229 	 * figure out the maximum size of an exec header, if necessary.
230 	 * XXX should be able to keep LKM code from modifying exec switch
231 	 * when we're still using it, but...
232 	 */
233 	if (exec_maxhdrsz == 0) {
234 		for (i = 0; i < nexecs; i++)
235 			if (execsw[i].es_check != NULL
236 			    && execsw[i].es_hdrsz > exec_maxhdrsz)
237 				exec_maxhdrsz = execsw[i].es_hdrsz;
238 	}
239 
240 	/* init the namei data to point the file user's program name */
241 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
242 
243 	/*
244 	 * initialize the fields of the exec package.
245 	 */
246 	pack.ep_name = SCARG(uap, path);
247 	MALLOC(pack.ep_hdr, void *, exec_maxhdrsz, M_EXEC, M_WAITOK);
248 	pack.ep_hdrlen = exec_maxhdrsz;
249 	pack.ep_hdrvalid = 0;
250 	pack.ep_ndp = &nid;
251 	pack.ep_emul_arg = NULL;
252 	pack.ep_vmcmds.evs_cnt = 0;
253 	pack.ep_vmcmds.evs_used = 0;
254 	pack.ep_vap = &attr;
255 	pack.ep_emul = &emul_netbsd;
256 	pack.ep_flags = 0;
257 
258 	/* see if we can run it. */
259 	if (error = check_exec(p, &pack))
260 		goto freehdr;
261 
262 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
263 
264 	/* allocate an argument buffer */
265 	argp = (char *) kmem_alloc_wait(exec_map, NCARGS);
266 #ifdef DIAGNOSTIC
267 	if (argp == (vm_offset_t) 0)
268 		panic("execve: argp == NULL");
269 #endif
270 	dp = argp;
271 	argc = 0;
272 
273 	/* copy the fake args list, if there's one, freeing it as we go */
274 	if (pack.ep_flags & EXEC_HASARGL) {
275 		tmpfap = pack.ep_fa;
276 		while (*tmpfap != NULL) {
277 			char *cp;
278 
279 			cp = *tmpfap;
280 			while (*cp)
281 				*dp++ = *cp++;
282 			*dp++;
283 
284 			FREE(*tmpfap, M_EXEC);
285 			tmpfap++; argc++;
286 		}
287 		FREE(pack.ep_fa, M_EXEC);
288 		pack.ep_flags &= ~EXEC_HASARGL;
289 	}
290 
291 	/* Now get argv & environment */
292 	if (!(cpp = SCARG(uap, argp))) {
293 		error = EINVAL;
294 		goto bad;
295 	}
296 
297 	if (pack.ep_flags & EXEC_SKIPARG)
298 		cpp++;
299 
300 	while (1) {
301 		len = argp + ARG_MAX - dp;
302 		if (error = copyin(cpp, &sp, sizeof(sp)))
303 			goto bad;
304 		if (!sp)
305 			break;
306 		if (error = copyinstr(sp, dp, len, &len)) {
307 			if (error == ENAMETOOLONG)
308 				error = E2BIG;
309 			goto bad;
310 		}
311 		dp += len;
312 		cpp++;
313 		argc++;
314 	}
315 
316 	envc = 0;
317 	if (cpp = SCARG(uap, envp)) {	/* environment need not be there */
318 		while (1) {
319 			len = argp + ARG_MAX - dp;
320 			if (error = copyin(cpp, &sp, sizeof(sp)))
321 				goto bad;
322 			if (!sp)
323 				break;
324 			if (error = copyinstr(sp, dp, len, &len)) {
325 				if (error == ENAMETOOLONG)
326 					error = E2BIG;
327 				goto bad;
328 			}
329 			dp += len;
330 			cpp++;
331 			envc++;
332 		}
333 	}
334 
335 	dp = (char *) ALIGN(dp);
336 
337 	szsigcode = pack.ep_emul->e_esigcode - pack.ep_emul->e_sigcode;
338 
339 	/* Now check if args & environ fit into new stack */
340 	len = ((argc + envc + 2 + pack.ep_emul->e_arglen) * sizeof(char *) +
341 	    sizeof(long) + dp + STACKGAPLEN + szsigcode +
342 	    sizeof(struct ps_strings)) - argp;
343 
344 	len = ALIGN(len);	/* make the stack "safely" aligned */
345 
346 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
347 		error = ENOMEM;
348 		goto bad;
349 	}
350 
351 	/* adjust "active stack depth" for process VSZ */
352 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
353 
354 	/* Unmap old program */
355 #ifdef sparc
356 	kill_user_windows(p);		/* before stack addresses go away */
357 #endif
358 	/* Kill shared memory and unmap old program */
359 #ifdef SYSVSHM
360 	if (vm->vm_shm)
361 		shmexit(p);
362 #endif
363 	vm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
364 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
365 
366 	/* Now map address space */
367 	vm->vm_taddr = (char *) pack.ep_taddr;
368 	vm->vm_tsize = btoc(pack.ep_tsize);
369 	vm->vm_daddr = (char *) pack.ep_daddr;
370 	vm->vm_dsize = btoc(pack.ep_dsize);
371 	vm->vm_ssize = btoc(pack.ep_ssize);
372 	vm->vm_maxsaddr = (char *) pack.ep_maxsaddr;
373 
374 	/* create the new process's VM space by running the vmcmds */
375 #ifdef DIAGNOSTIC
376 	if (pack.ep_vmcmds.evs_used == 0)
377 		panic("execve: no vmcmds");
378 #endif
379 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
380 		struct exec_vmcmd *vcp;
381 
382 		vcp = &pack.ep_vmcmds.evs_cmds[i];
383 		error = (*vcp->ev_proc)(p, vcp);
384 	}
385 
386 	/* free the vmspace-creation commands, and release their references */
387 	kill_vmcmds(&pack.ep_vmcmds);
388 
389 	/* if an error happened, deallocate and punt */
390 	if (error)
391 		goto exec_abort;
392 
393 	/* remember information about the process */
394 	arginfo.ps_nargvstr = argc;
395 	arginfo.ps_nenvstr = envc;
396 
397 	stack = (char *) (USRSTACK - len);
398 	/* Now copy argc, args & environ to new stack */
399 	if (!(*pack.ep_emul->e_copyargs)(&pack, &arginfo, stack, argp))
400 		goto exec_abort;
401 
402 	/* copy out the process's ps_strings structure */
403 	if (copyout(&arginfo, (char *) PS_STRINGS, sizeof(arginfo)))
404 		goto exec_abort;
405 
406 	/* copy out the process's signal trapoline code */
407 	if (szsigcode && copyout((char *) pack.ep_emul->e_sigcode,
408 				 ((char *) PS_STRINGS) - szsigcode,
409 				 szsigcode))
410 		goto exec_abort;
411 
412 	fdcloseexec(p);		/* handle close on exec */
413 	execsigs(p);		/* reset catched signals */
414 
415 	/* set command name & other accounting info */
416 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
417 	bcopy(nid.ni_cnd.cn_nameptr, p->p_comm, len);
418 	p->p_comm[len] = 0;
419 	p->p_acflag &= ~AFORK;
420 
421 	/* record proc's vnode, for use by procfs and others */
422         if (p->p_textvp)
423                 vrele(p->p_textvp);
424 	VREF(pack.ep_vp);
425 	p->p_textvp = pack.ep_vp;
426 
427 	p->p_flag |= P_EXEC;
428 	if (p->p_flag & P_PPWAIT) {
429 		p->p_flag &= ~P_PPWAIT;
430 		wakeup((caddr_t) p->p_pptr);
431 	}
432 
433 	/*
434 	 * deal with set[ug]id.
435 	 * MNT_NOEXEC and P_TRACED have already been used to disable s[ug]id.
436 	 */
437 	p->p_flag &= ~P_SUGID;
438 	if (((attr.va_mode & VSUID) != 0 &&
439 	    p->p_ucred->cr_uid != attr.va_uid)
440 	    || (attr.va_mode & VSGID) != 0 &&
441 	    p->p_ucred->cr_gid != attr.va_gid) {
442 		p->p_ucred = crcopy(cred);
443 #ifdef KTRACE
444 		/*
445 		 * If process is being ktraced, turn off - unless
446 		 * root set it.
447 		 */
448 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) {
449 			vrele(p->p_tracep);
450 			p->p_tracep = NULL;
451 			p->p_traceflag = 0;
452 		}
453 #endif
454 		if (attr.va_mode & VSUID)
455 			p->p_ucred->cr_uid = attr.va_uid;
456 		if (attr.va_mode & VSGID)
457 			p->p_ucred->cr_gid = attr.va_gid;
458 		p->p_flag |= P_SUGID;
459 	}
460 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
461 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
462 
463 	kmem_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
464 
465 	FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
466 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
467 	vput(pack.ep_vp);
468 
469 	/* setup new registers and do misc. setup. */
470 	(*pack.ep_emul->e_setregs)(p, &pack, (u_long) stack, retval);
471 
472 	if (p->p_flag & P_TRACED)
473 		psignal(p, SIGTRAP);
474 
475 	p->p_emul = pack.ep_emul;
476 	FREE(pack.ep_hdr, M_EXEC);
477 	return 0;
478 
479 bad:
480 	/* free the vmspace-creation commands, and release their references */
481 	kill_vmcmds(&pack.ep_vmcmds);
482 	/* kill any opened file descriptor, if necessary */
483 	if (pack.ep_flags & EXEC_HASFD) {
484 		pack.ep_flags &= ~EXEC_HASFD;
485 		(void) fdrelease(p, pack.ep_fd);
486 	}
487 	/* close and put the exec'd file */
488 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
489 	vput(pack.ep_vp);
490 	FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
491 	kmem_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
492 
493 freehdr:
494 	FREE(pack.ep_hdr, M_EXEC);
495 	return error;
496 
497 exec_abort:
498 	/*
499 	 * the old process doesn't exist anymore.  exit gracefully.
500 	 * get rid of the (new) address space we have created, if any, get rid
501 	 * of our namei data and vnode, and exit noting failure
502 	 */
503 	vm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
504 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
505 	FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
506 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
507 	vput(pack.ep_vp);
508 	kmem_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
509 	FREE(pack.ep_hdr, M_EXEC);
510 	exit1(p, W_EXITCODE(0, SIGABRT));
511 	exit1(p, -1);
512 
513 	/* NOTREACHED */
514 	return 0;
515 }
516 
517 
518 void *
519 copyargs(pack, arginfo, stack, argp)
520 	struct exec_package *pack;
521 	struct ps_strings *arginfo;
522 	void *stack;
523 	void *argp;
524 {
525 	char **cpp = stack;
526 	char *dp, *sp;
527 	size_t len;
528 	void *nullp = NULL;
529 	int argc = arginfo->ps_nargvstr;
530 	int envc = arginfo->ps_nenvstr;
531 
532 	if (copyout(&argc, cpp++, sizeof(argc)))
533 		return NULL;
534 
535 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_emul->e_arglen);
536 	sp = argp;
537 
538 	/* XXX don't copy them out, remap them! */
539 	arginfo->ps_argvstr = dp; /* remember location of argv for later */
540 
541 	for (; --argc >= 0; sp += len, dp += len)
542 		if (copyout(&dp, cpp++, sizeof(dp)) ||
543 		    copyoutstr(sp, dp, ARG_MAX, &len))
544 			return NULL;
545 
546 	if (copyout(&nullp, cpp++, sizeof(nullp)))
547 		return NULL;
548 
549 	arginfo->ps_envstr = dp; /* remember location of envp for later */
550 
551 	for (; --envc >= 0; sp += len, dp += len)
552 		if (copyout(&dp, cpp++, sizeof(dp)) ||
553 		    copyoutstr(sp, dp, ARG_MAX, &len))
554 			return NULL;
555 
556 	if (copyout(&nullp, cpp++, sizeof(nullp)))
557 		return NULL;
558 
559 	return cpp;
560 }
561