kern_exec.c revision 1.90 1 /* $NetBSD: kern_exec.c,v 1.90 1998/03/01 02:22:28 fvdl Exp $ */
2
3 /*-
4 * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou
5 * Copyright (C) 1992 Wolfgang Solfrank.
6 * Copyright (C) 1992 TooLs GmbH.
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by TooLs GmbH.
20 * 4. The name of TooLs GmbH may not be used to endorse or promote products
21 * derived from this software without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
29 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
30 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
31 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
32 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 */
34
35 #include "opt_uvm.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/filedesc.h>
40 #include <sys/kernel.h>
41 #include <sys/proc.h>
42 #include <sys/mount.h>
43 #include <sys/malloc.h>
44 #include <sys/namei.h>
45 #include <sys/vnode.h>
46 #include <sys/file.h>
47 #include <sys/acct.h>
48 #include <sys/exec.h>
49 #include <sys/ktrace.h>
50 #include <sys/resourcevar.h>
51 #include <sys/wait.h>
52 #include <sys/mman.h>
53 #include <sys/signalvar.h>
54 #include <sys/stat.h>
55
56 #include <sys/syscallargs.h>
57
58 #include <vm/vm.h>
59 #include <vm/vm_kern.h>
60
61 #if defined(UVM)
62 #include <uvm/uvm_extern.h>
63 #endif
64
65 #include <machine/cpu.h>
66 #include <machine/reg.h>
67
68 /*
69 * check exec:
70 * given an "executable" described in the exec package's namei info,
71 * see what we can do with it.
72 *
73 * ON ENTRY:
74 * exec package with appropriate namei info
75 * proc pointer of exec'ing proc
76 * NO SELF-LOCKED VNODES
77 *
78 * ON EXIT:
79 * error: nothing held, etc. exec header still allocated.
80 * ok: filled exec package, executable's vnode (unlocked).
81 *
82 * EXEC SWITCH ENTRY:
83 * Locked vnode to check, exec package, proc.
84 *
85 * EXEC SWITCH EXIT:
86 * ok: return 0, filled exec package, executable's vnode (unlocked).
87 * error: destructive:
88 * everything deallocated execept exec header.
89 * non-destructive:
90 * error code, executable's vnode (unlocked),
91 * exec header unmodified.
92 */
93 int
94 check_exec(p, epp)
95 struct proc *p;
96 struct exec_package *epp;
97 {
98 int error, i;
99 struct vnode *vp;
100 struct nameidata *ndp;
101 int resid;
102
103 ndp = epp->ep_ndp;
104 ndp->ni_cnd.cn_nameiop = LOOKUP;
105 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
106 /* first get the vnode */
107 if ((error = namei(ndp)) != 0)
108 return error;
109 epp->ep_vp = vp = ndp->ni_vp;
110
111 /* check access and type */
112 if (vp->v_type != VREG) {
113 error = EACCES;
114 goto bad1;
115 }
116 if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
117 goto bad1;
118
119 /* get attributes */
120 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
121 goto bad1;
122
123 /* Check mount point */
124 if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
125 error = EACCES;
126 goto bad1;
127 }
128 if ((vp->v_mount->mnt_flag & MNT_NOSUID) || (p->p_flag & P_TRACED))
129 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
130
131 /* try to open it */
132 if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
133 goto bad1;
134
135 /* unlock vp, since we don't need it locked from here on out. */
136 VOP_UNLOCK(vp, 0);
137
138 /* now we have the file, get the exec header */
139 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
140 UIO_SYSSPACE, 0, p->p_ucred, &resid, p);
141 if (error)
142 goto bad2;
143 epp->ep_hdrvalid = epp->ep_hdrlen - resid;
144
145 /*
146 * set up the vmcmds for creation of the process
147 * address space
148 */
149 error = ENOEXEC;
150 for (i = 0; i < nexecs && error != 0; i++) {
151 int newerror;
152
153 if (execsw[i].es_check == NULL)
154 continue;
155
156 newerror = (*execsw[i].es_check)(p, epp);
157 /* make sure the first "interesting" error code is saved. */
158 if (!newerror || error == ENOEXEC)
159 error = newerror;
160 if (epp->ep_flags & EXEC_DESTR && error != 0)
161 return error;
162 }
163 if (!error) {
164 /* check that entry point is sane */
165 if (epp->ep_entry > VM_MAXUSER_ADDRESS)
166 error = ENOEXEC;
167
168 /* check limits */
169 if ((epp->ep_tsize > MAXTSIZ) ||
170 (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur))
171 error = ENOMEM;
172
173 if (!error)
174 return (0);
175 }
176
177 /*
178 * free any vmspace-creation commands,
179 * and release their references
180 */
181 kill_vmcmds(&epp->ep_vmcmds);
182
183 bad2:
184 /*
185 * unlock and close the vnode, restore the old one, free the
186 * pathname buf, and punt.
187 */
188 VOP_CLOSE(vp, FREAD, p->p_ucred, p);
189 vrele(vp);
190 FREE(ndp->ni_cnd.cn_pnbuf, M_NAMEI);
191 return error;
192
193 bad1:
194 /*
195 * free the namei pathname buffer, and put the vnode
196 * (which we don't yet have open).
197 */
198 vput(vp); /* was still locked */
199 FREE(ndp->ni_cnd.cn_pnbuf, M_NAMEI);
200 return error;
201 }
202
203 /*
204 * exec system call
205 */
206 /* ARGSUSED */
207 int
208 sys_execve(p, v, retval)
209 register struct proc *p;
210 void *v;
211 register_t *retval;
212 {
213 register struct sys_execve_args /* {
214 syscallarg(const char *) path;
215 syscallarg(char * const *) argp;
216 syscallarg(char * const *) envp;
217 } */ *uap = v;
218 int error, i;
219 struct exec_package pack;
220 struct nameidata nid;
221 struct vattr attr;
222 struct ucred *cred = p->p_ucred;
223 char *argp;
224 char * const *cpp;
225 char *dp, *sp;
226 long argc, envc;
227 size_t len;
228 char *stack;
229 struct ps_strings arginfo;
230 struct vmspace *vm;
231 char **tmpfap;
232 int szsigcode;
233 extern struct emul emul_netbsd;
234
235 /*
236 * figure out the maximum size of an exec header, if necessary.
237 * XXX should be able to keep LKM code from modifying exec switch
238 * when we're still using it, but...
239 */
240 if (exec_maxhdrsz == 0) {
241 for (i = 0; i < nexecs; i++)
242 if (execsw[i].es_check != NULL
243 && execsw[i].es_hdrsz > exec_maxhdrsz)
244 exec_maxhdrsz = execsw[i].es_hdrsz;
245 }
246
247 /* init the namei data to point the file user's program name */
248 /* XXX cgd 960926: why do this here? most will be clobbered. */
249 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p);
250
251 /*
252 * initialize the fields of the exec package.
253 */
254 pack.ep_name = SCARG(uap, path);
255 MALLOC(pack.ep_hdr, void *, exec_maxhdrsz, M_EXEC, M_WAITOK);
256 pack.ep_hdrlen = exec_maxhdrsz;
257 pack.ep_hdrvalid = 0;
258 pack.ep_ndp = &nid;
259 pack.ep_emul_arg = NULL;
260 pack.ep_vmcmds.evs_cnt = 0;
261 pack.ep_vmcmds.evs_used = 0;
262 pack.ep_vap = &attr;
263 pack.ep_emul = &emul_netbsd;
264 pack.ep_flags = 0;
265
266 /* see if we can run it. */
267 if ((error = check_exec(p, &pack)) != 0)
268 goto freehdr;
269
270 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
271
272 /* allocate an argument buffer */
273 #if defined(UVM)
274 argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS);
275 #else
276 argp = (char *) kmem_alloc_wait(exec_map, NCARGS);
277 #endif
278 #ifdef DIAGNOSTIC
279 if (argp == (vm_offset_t) 0)
280 panic("execve: argp == NULL");
281 #endif
282 dp = argp;
283 argc = 0;
284
285 /* copy the fake args list, if there's one, freeing it as we go */
286 if (pack.ep_flags & EXEC_HASARGL) {
287 tmpfap = pack.ep_fa;
288 while (*tmpfap != NULL) {
289 char *cp;
290
291 cp = *tmpfap;
292 while (*cp)
293 *dp++ = *cp++;
294 dp++;
295
296 FREE(*tmpfap, M_EXEC);
297 tmpfap++; argc++;
298 }
299 FREE(pack.ep_fa, M_EXEC);
300 pack.ep_flags &= ~EXEC_HASARGL;
301 }
302
303 /* Now get argv & environment */
304 if (!(cpp = SCARG(uap, argp))) {
305 error = EINVAL;
306 goto bad;
307 }
308
309 if (pack.ep_flags & EXEC_SKIPARG)
310 cpp++;
311
312 while (1) {
313 len = argp + ARG_MAX - dp;
314 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
315 goto bad;
316 if (!sp)
317 break;
318 if ((error = copyinstr(sp, dp, len, &len)) != 0) {
319 if (error == ENAMETOOLONG)
320 error = E2BIG;
321 goto bad;
322 }
323 dp += len;
324 cpp++;
325 argc++;
326 }
327
328 envc = 0;
329 /* environment need not be there */
330 if ((cpp = SCARG(uap, envp)) != NULL ) {
331 while (1) {
332 len = argp + ARG_MAX - dp;
333 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0)
334 goto bad;
335 if (!sp)
336 break;
337 if ((error = copyinstr(sp, dp, len, &len)) != 0) {
338 if (error == ENAMETOOLONG)
339 error = E2BIG;
340 goto bad;
341 }
342 dp += len;
343 cpp++;
344 envc++;
345 }
346 }
347
348 dp = (char *) ALIGN(dp);
349
350 szsigcode = pack.ep_emul->e_esigcode - pack.ep_emul->e_sigcode;
351
352 /* Now check if args & environ fit into new stack */
353 len = ((argc + envc + 2 + pack.ep_emul->e_arglen) * sizeof(char *) +
354 sizeof(long) + dp + STACKGAPLEN + szsigcode +
355 sizeof(struct ps_strings)) - argp;
356
357 len = ALIGN(len); /* make the stack "safely" aligned */
358
359 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
360 error = ENOMEM;
361 goto bad;
362 }
363
364 /* adjust "active stack depth" for process VSZ */
365 pack.ep_ssize = len; /* maybe should go elsewhere, but... */
366
367 /*
368 * Do whatever is necessary to prepare the address space
369 * for remapping. Note that this might replace the current
370 * vmspace with another!
371 */
372 #if defined(UVM)
373 uvmspace_exec(p);
374 #else
375 vmspace_exec(p);
376 #endif
377
378 /* Now map address space */
379 vm = p->p_vmspace;
380 vm->vm_taddr = (char *) pack.ep_taddr;
381 vm->vm_tsize = btoc(pack.ep_tsize);
382 vm->vm_daddr = (char *) pack.ep_daddr;
383 vm->vm_dsize = btoc(pack.ep_dsize);
384 vm->vm_ssize = btoc(pack.ep_ssize);
385 vm->vm_maxsaddr = (char *) pack.ep_maxsaddr;
386
387 /* create the new process's VM space by running the vmcmds */
388 #ifdef DIAGNOSTIC
389 if (pack.ep_vmcmds.evs_used == 0)
390 panic("execve: no vmcmds");
391 #endif
392 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
393 struct exec_vmcmd *vcp;
394
395 vcp = &pack.ep_vmcmds.evs_cmds[i];
396 error = (*vcp->ev_proc)(p, vcp);
397 }
398
399 /* free the vmspace-creation commands, and release their references */
400 kill_vmcmds(&pack.ep_vmcmds);
401
402 /* if an error happened, deallocate and punt */
403 if (error)
404 goto exec_abort;
405
406 /* remember information about the process */
407 arginfo.ps_nargvstr = argc;
408 arginfo.ps_nenvstr = envc;
409
410 stack = (char *) (USRSTACK - len);
411 /* Now copy argc, args & environ to new stack */
412 if (!(*pack.ep_emul->e_copyargs)(&pack, &arginfo, stack, argp))
413 goto exec_abort;
414
415 /* copy out the process's ps_strings structure */
416 if (copyout(&arginfo, (char *) PS_STRINGS, sizeof(arginfo)))
417 goto exec_abort;
418
419 /* copy out the process's signal trapoline code */
420 if (szsigcode && copyout((char *) pack.ep_emul->e_sigcode,
421 ((char *) PS_STRINGS) - szsigcode,
422 szsigcode))
423 goto exec_abort;
424
425 fdcloseexec(p); /* handle close on exec */
426 execsigs(p); /* reset catched signals */
427
428 /* set command name & other accounting info */
429 len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
430 bcopy(nid.ni_cnd.cn_nameptr, p->p_comm, len);
431 p->p_comm[len] = 0;
432 p->p_acflag &= ~AFORK;
433
434 /* record proc's vnode, for use by procfs and others */
435 if (p->p_textvp)
436 vrele(p->p_textvp);
437 VREF(pack.ep_vp);
438 p->p_textvp = pack.ep_vp;
439
440 p->p_flag |= P_EXEC;
441 if (p->p_flag & P_PPWAIT) {
442 p->p_flag &= ~P_PPWAIT;
443 wakeup((caddr_t) p->p_pptr);
444 }
445
446 /*
447 * deal with set[ug]id.
448 * MNT_NOEXEC and P_TRACED have already been used to disable s[ug]id.
449 */
450 if (((attr.va_mode & S_ISUID) != 0 && p->p_ucred->cr_uid != attr.va_uid)
451 || ((attr.va_mode & S_ISGID) != 0 && p->p_ucred->cr_gid != attr.va_gid)){
452 p->p_ucred = crcopy(cred);
453 #ifdef KTRACE
454 /*
455 * If process is being ktraced, turn off - unless
456 * root set it.
457 */
458 if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) {
459 vrele(p->p_tracep);
460 p->p_tracep = NULL;
461 p->p_traceflag = 0;
462 }
463 #endif
464 if (attr.va_mode & S_ISUID)
465 p->p_ucred->cr_uid = attr.va_uid;
466 if (attr.va_mode & S_ISGID)
467 p->p_ucred->cr_gid = attr.va_gid;
468 p->p_flag |= P_SUGID;
469 } else
470 p->p_flag &= ~P_SUGID;
471 p->p_cred->p_svuid = p->p_ucred->cr_uid;
472 p->p_cred->p_svgid = p->p_ucred->cr_gid;
473
474 #if defined(UVM)
475 uvm_km_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
476 #else
477 kmem_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
478 #endif
479
480 FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
481 VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
482 vrele(pack.ep_vp);
483
484 /* setup new registers and do misc. setup. */
485 (*pack.ep_emul->e_setregs)(p, &pack, (u_long) stack);
486
487 if (p->p_flag & P_TRACED)
488 psignal(p, SIGTRAP);
489
490 p->p_emul = pack.ep_emul;
491 FREE(pack.ep_hdr, M_EXEC);
492
493 #ifdef KTRACE
494 if (KTRPOINT(p, KTR_EMUL))
495 ktremul(p->p_tracep, p->p_emul->e_name);
496 #endif
497
498 return (EJUSTRETURN);
499
500 bad:
501 /* free the vmspace-creation commands, and release their references */
502 kill_vmcmds(&pack.ep_vmcmds);
503 /* kill any opened file descriptor, if necessary */
504 if (pack.ep_flags & EXEC_HASFD) {
505 pack.ep_flags &= ~EXEC_HASFD;
506 (void) fdrelease(p, pack.ep_fd);
507 }
508 /* close and put the exec'd file */
509 VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
510 vrele(pack.ep_vp);
511 FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
512 #if defined(UVM)
513 uvm_km_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
514 #else
515 kmem_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
516 #endif
517
518 freehdr:
519 FREE(pack.ep_hdr, M_EXEC);
520 return error;
521
522 exec_abort:
523 /*
524 * the old process doesn't exist anymore. exit gracefully.
525 * get rid of the (new) address space we have created, if any, get rid
526 * of our namei data and vnode, and exit noting failure
527 */
528 #if defined(UVM)
529 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
530 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
531 #else
532 vm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
533 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
534 #endif
535 if (pack.ep_emul_arg)
536 FREE(pack.ep_emul_arg, M_TEMP);
537 FREE(nid.ni_cnd.cn_pnbuf, M_NAMEI);
538 VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
539 vrele(pack.ep_vp);
540 #if defined(UVM)
541 uvm_km_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
542 #else
543 kmem_free_wakeup(exec_map, (vm_offset_t) argp, NCARGS);
544 #endif
545 FREE(pack.ep_hdr, M_EXEC);
546 exit1(p, W_EXITCODE(0, SIGABRT));
547 exit1(p, -1);
548
549 /* NOTREACHED */
550 return 0;
551 }
552
553
554 void *
555 copyargs(pack, arginfo, stack, argp)
556 struct exec_package *pack;
557 struct ps_strings *arginfo;
558 void *stack;
559 void *argp;
560 {
561 char **cpp = stack;
562 char *dp, *sp;
563 size_t len;
564 void *nullp = NULL;
565 int argc = arginfo->ps_nargvstr;
566 int envc = arginfo->ps_nenvstr;
567
568 if (copyout(&argc, cpp++, sizeof(argc)))
569 return NULL;
570
571 dp = (char *) (cpp + argc + envc + 2 + pack->ep_emul->e_arglen);
572 sp = argp;
573
574 /* XXX don't copy them out, remap them! */
575 arginfo->ps_argvstr = cpp; /* remember location of argv for later */
576
577 for (; --argc >= 0; sp += len, dp += len)
578 if (copyout(&dp, cpp++, sizeof(dp)) ||
579 copyoutstr(sp, dp, ARG_MAX, &len))
580 return NULL;
581
582 if (copyout(&nullp, cpp++, sizeof(nullp)))
583 return NULL;
584
585 arginfo->ps_envstr = cpp; /* remember location of envp for later */
586
587 for (; --envc >= 0; sp += len, dp += len)
588 if (copyout(&dp, cpp++, sizeof(dp)) ||
589 copyoutstr(sp, dp, ARG_MAX, &len))
590 return NULL;
591
592 if (copyout(&nullp, cpp++, sizeof(nullp)))
593 return NULL;
594
595 return cpp;
596 }
597