FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_exec.c
1 /*-
2 * Copyright (c) 1993, David Greenman
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD: releng/5.4/sys/kern/kern_exec.c 145335 2005-04-20 19:11:07Z cvs2svn $");
29
30 #include "opt_ktrace.h"
31 #include "opt_mac.h"
32
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/eventhandler.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/sysproto.h>
39 #include <sys/signalvar.h>
40 #include <sys/kernel.h>
41 #include <sys/mac.h>
42 #include <sys/mount.h>
43 #include <sys/filedesc.h>
44 #include <sys/fcntl.h>
45 #include <sys/acct.h>
46 #include <sys/exec.h>
47 #include <sys/imgact.h>
48 #include <sys/imgact_elf.h>
49 #include <sys/wait.h>
50 #include <sys/malloc.h>
51 #include <sys/proc.h>
52 #include <sys/pioctl.h>
53 #include <sys/namei.h>
54 #include <sys/resourcevar.h>
55 #include <sys/sf_buf.h>
56 #include <sys/sysent.h>
57 #include <sys/shm.h>
58 #include <sys/sysctl.h>
59 #include <sys/vnode.h>
60 #ifdef KTRACE
61 #include <sys/ktrace.h>
62 #endif
63
64 #include <vm/vm.h>
65 #include <vm/vm_param.h>
66 #include <vm/pmap.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_map.h>
69 #include <vm/vm_kern.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_pager.h>
73
74 #include <machine/reg.h>
75
76 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
77
78 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
79 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
80 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
81 static int kern_execve(struct thread *td, char *fname, char **argv,
82 char **envv, struct mac *mac_p);
83 static int do_execve(struct thread *td, char *fname, char **argv,
84 char **envv, struct mac *mac_p);
85
86 /* XXX This should be vm_size_t. */
87 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
88 NULL, 0, sysctl_kern_ps_strings, "LU", "");
89
90 /* XXX This should be vm_size_t. */
91 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
92 NULL, 0, sysctl_kern_usrstack, "LU", "");
93
94 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
95 NULL, 0, sysctl_kern_stackprot, "I", "");
96
97 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
98 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
99 &ps_arg_cache_limit, 0, "");
100
101 static int
102 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
103 {
104 struct proc *p;
105 int error;
106
107 p = curproc;
108 #if defined(__amd64__) || defined(__ia64__)
109 if (req->oldlen == sizeof(unsigned int)) {
110 unsigned int val;
111 val = (unsigned int)p->p_sysent->sv_psstrings;
112 error = SYSCTL_OUT(req, &val, sizeof(val));
113 } else
114 #endif
115 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
116 sizeof(p->p_sysent->sv_psstrings));
117 return error;
118 }
119
120 static int
121 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
122 {
123 struct proc *p;
124 int error;
125
126 p = curproc;
127 #if defined(__amd64__) || defined(__ia64__)
128 if (req->oldlen == sizeof(unsigned int)) {
129 unsigned int val;
130 val = (unsigned int)p->p_sysent->sv_usrstack;
131 error = SYSCTL_OUT(req, &val, sizeof(val));
132 } else
133 #endif
134 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
135 sizeof(p->p_sysent->sv_usrstack));
136 return error;
137 }
138
139 static int
140 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
141 {
142 struct proc *p;
143
144 p = curproc;
145 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
146 sizeof(p->p_sysent->sv_stackprot)));
147 }
148
149 /*
150 * Each of the items is a pointer to a `const struct execsw', hence the
151 * double pointer here.
152 */
153 static const struct execsw **execsw;
154
155 #ifndef _SYS_SYSPROTO_H_
156 struct execve_args {
157 char *fname;
158 char **argv;
159 char **envv;
160 };
161 #endif
162
163 /*
164 * MPSAFE
165 */
166 int
167 execve(td, uap)
168 struct thread *td;
169 struct execve_args /* {
170 char *fname;
171 char **argv;
172 char **envv;
173 } */ *uap;
174 {
175
176 return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL));
177 }
178
179 #ifndef _SYS_SYSPROTO_H_
180 struct __mac_execve_args {
181 char *fname;
182 char **argv;
183 char **envv;
184 struct mac *mac_p;
185 };
186 #endif
187
188 /*
189 * MPSAFE
190 */
191 int
192 __mac_execve(td, uap)
193 struct thread *td;
194 struct __mac_execve_args /* {
195 char *fname;
196 char **argv;
197 char **envv;
198 struct mac *mac_p;
199 } */ *uap;
200 {
201
202 #ifdef MAC
203 return (kern_execve(td, uap->fname, uap->argv, uap->envv,
204 uap->mac_p));
205 #else
206 return (ENOSYS);
207 #endif
208 }
209
210 static int
211 kern_execve(td, fname, argv, envv, mac_p)
212 struct thread *td;
213 char *fname;
214 char **argv;
215 char **envv;
216 struct mac *mac_p;
217 {
218 struct proc *p = td->td_proc;
219 int error;
220
221 if (p->p_flag & P_HADTHREADS) {
222 PROC_LOCK(p);
223 if (thread_single(SINGLE_BOUNDARY)) {
224 PROC_UNLOCK(p);
225 return (ERESTART); /* Try again later. */
226 }
227 PROC_UNLOCK(p);
228 }
229
230 error = do_execve(td, fname, argv, envv, mac_p);
231
232 if (p->p_flag & P_HADTHREADS) {
233 PROC_LOCK(p);
234 /*
235 * If success, we upgrade to SINGLE_EXIT state to
236 * force other threads to suicide.
237 */
238 if (error == 0)
239 thread_single(SINGLE_EXIT);
240 else
241 thread_single_end();
242 PROC_UNLOCK(p);
243 }
244
245 return (error);
246 }
247
248 /*
249 * In-kernel implementation of execve(). All arguments are assumed to be
250 * userspace pointers from the passed thread.
251 *
252 * MPSAFE
253 */
254 static int
255 do_execve(td, fname, argv, envv, mac_p)
256 struct thread *td;
257 char *fname;
258 char **argv;
259 char **envv;
260 struct mac *mac_p;
261 {
262 struct proc *p = td->td_proc;
263 struct nameidata nd, *ndp;
264 struct ucred *newcred = NULL, *oldcred;
265 struct uidinfo *euip;
266 register_t *stack_base;
267 int error, len, i;
268 struct image_params image_params, *imgp;
269 struct vattr attr;
270 int (*img_first)(struct image_params *);
271 struct pargs *oldargs = NULL, *newargs = NULL;
272 struct sigacts *oldsigacts, *newsigacts;
273 #ifdef KTRACE
274 struct vnode *tracevp = NULL;
275 struct ucred *tracecred = NULL;
276 #endif
277 struct vnode *textvp = NULL;
278 int credential_changing;
279 int textset;
280 #ifdef MAC
281 struct label *interplabel = NULL;
282 int will_transition;
283 #endif
284
285 imgp = &image_params;
286
287 /*
288 * Lock the process and set the P_INEXEC flag to indicate that
289 * it should be left alone until we're done here. This is
290 * necessary to avoid race conditions - e.g. in ptrace() -
291 * that might allow a local user to illicitly obtain elevated
292 * privileges.
293 */
294 PROC_LOCK(p);
295 KASSERT((p->p_flag & P_INEXEC) == 0,
296 ("%s(): process already has P_INEXEC flag", __func__));
297 p->p_flag |= P_INEXEC;
298 PROC_UNLOCK(p);
299
300 /*
301 * Initialize part of the common data
302 */
303 imgp->proc = p;
304 imgp->userspace_argv = argv;
305 imgp->userspace_envv = envv;
306 imgp->execlabel = NULL;
307 imgp->attr = &attr;
308 imgp->argc = imgp->envc = 0;
309 imgp->argv0 = NULL;
310 imgp->entry_addr = 0;
311 imgp->vmspace_destroyed = 0;
312 imgp->interpreted = 0;
313 imgp->interpreter_name[0] = '\0';
314 imgp->auxargs = NULL;
315 imgp->vp = NULL;
316 imgp->object = NULL;
317 imgp->firstpage = NULL;
318 imgp->ps_strings = 0;
319 imgp->auxarg_size = 0;
320
321 #ifdef MAC
322 error = mac_execve_enter(imgp, mac_p);
323 if (error) {
324 mtx_lock(&Giant);
325 goto exec_fail;
326 }
327 #endif
328
329 /*
330 * Allocate temporary demand zeroed space for argument and
331 * environment strings
332 */
333 imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX);
334 if (imgp->stringbase == NULL) {
335 error = ENOMEM;
336 mtx_lock(&Giant);
337 goto exec_fail;
338 }
339 imgp->stringp = imgp->stringbase;
340 imgp->stringspace = ARG_MAX;
341 imgp->image_header = NULL;
342
343 /*
344 * Translate the file name. namei() returns a vnode pointer
345 * in ni_vp amoung other things.
346 */
347 ndp = &nd;
348 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
349 UIO_USERSPACE, fname, td);
350
351 mtx_lock(&Giant);
352 interpret:
353
354 error = namei(ndp);
355 if (error) {
356 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
357 ARG_MAX);
358 goto exec_fail;
359 }
360
361 imgp->vp = ndp->ni_vp;
362 imgp->fname = fname;
363
364 /*
365 * Check file permissions (also 'opens' file)
366 */
367 error = exec_check_permissions(imgp);
368 if (error)
369 goto exec_fail_dealloc;
370
371 if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0)
372 vm_object_reference(imgp->object);
373
374 /*
375 * Set VV_TEXT now so no one can write to the executable while we're
376 * activating it.
377 *
378 * Remember if this was set before and unset it in case this is not
379 * actually an executable image.
380 */
381 textset = imgp->vp->v_vflag & VV_TEXT;
382 imgp->vp->v_vflag |= VV_TEXT;
383
384 error = exec_map_first_page(imgp);
385 if (error)
386 goto exec_fail_dealloc;
387
388 /*
389 * If the current process has a special image activator it
390 * wants to try first, call it. For example, emulating shell
391 * scripts differently.
392 */
393 error = -1;
394 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
395 error = img_first(imgp);
396
397 /*
398 * Loop through the list of image activators, calling each one.
399 * An activator returns -1 if there is no match, 0 on success,
400 * and an error otherwise.
401 */
402 for (i = 0; error == -1 && execsw[i]; ++i) {
403 if (execsw[i]->ex_imgact == NULL ||
404 execsw[i]->ex_imgact == img_first) {
405 continue;
406 }
407 error = (*execsw[i]->ex_imgact)(imgp);
408 }
409
410 if (error) {
411 if (error == -1) {
412 if (textset == 0)
413 imgp->vp->v_vflag &= ~VV_TEXT;
414 error = ENOEXEC;
415 }
416 goto exec_fail_dealloc;
417 }
418
419 /*
420 * Special interpreter operation, cleanup and loop up to try to
421 * activate the interpreter.
422 */
423 if (imgp->interpreted) {
424 exec_unmap_first_page(imgp);
425 /*
426 * VV_TEXT needs to be unset for scripts. There is a short
427 * period before we determine that something is a script where
428 * VV_TEXT will be set. The vnode lock is held over this
429 * entire period so nothing should illegitimately be blocked.
430 */
431 imgp->vp->v_vflag &= ~VV_TEXT;
432 /* free name buffer and old vnode */
433 NDFREE(ndp, NDF_ONLY_PNBUF);
434 #ifdef MAC
435 interplabel = mac_vnode_label_alloc();
436 mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
437 #endif
438 vput(ndp->ni_vp);
439 vm_object_deallocate(imgp->object);
440 imgp->object = NULL;
441 /* set new name to that of the interpreter */
442 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
443 UIO_SYSSPACE, imgp->interpreter_name, td);
444 goto interpret;
445 }
446
447 /*
448 * Copy out strings (args and env) and initialize stack base
449 */
450 if (p->p_sysent->sv_copyout_strings)
451 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
452 else
453 stack_base = exec_copyout_strings(imgp);
454
455 /*
456 * If custom stack fixup routine present for this process
457 * let it do the stack setup.
458 * Else stuff argument count as first item on stack
459 */
460 if (p->p_sysent->sv_fixup != NULL)
461 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
462 else
463 suword(--stack_base, imgp->argc);
464
465 /*
466 * For security and other reasons, the file descriptor table cannot
467 * be shared after an exec.
468 */
469 fdunshare(p, td);
470
471 /*
472 * Malloc things before we need locks.
473 */
474 newcred = crget();
475 euip = uifind(attr.va_uid);
476 i = imgp->endargs - imgp->stringbase;
477 if (ps_arg_cache_limit >= i + sizeof(struct pargs))
478 newargs = pargs_alloc(i);
479
480 /* close files on exec */
481 fdcloseexec(td);
482
483 /* Get a reference to the vnode prior to locking the proc */
484 VREF(ndp->ni_vp);
485
486 /*
487 * For security and other reasons, signal handlers cannot
488 * be shared after an exec. The new process gets a copy of the old
489 * handlers. In execsigs(), the new process will have its signals
490 * reset.
491 */
492 PROC_LOCK(p);
493 if (sigacts_shared(p->p_sigacts)) {
494 oldsigacts = p->p_sigacts;
495 PROC_UNLOCK(p);
496 newsigacts = sigacts_alloc();
497 sigacts_copy(newsigacts, oldsigacts);
498 PROC_LOCK(p);
499 p->p_sigacts = newsigacts;
500 } else
501 oldsigacts = NULL;
502
503 /* Stop profiling */
504 stopprofclock(p);
505
506 /* reset caught signals */
507 execsigs(p);
508
509 /* name this process - nameiexec(p, ndp) */
510 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
511 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
512 p->p_comm[len] = 0;
513
514 /*
515 * mark as execed, wakeup the process that vforked (if any) and tell
516 * it that it now has its own resources back
517 */
518 p->p_flag |= P_EXEC;
519 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
520 p->p_flag &= ~P_PPWAIT;
521 wakeup(p->p_pptr);
522 }
523
524 /*
525 * Implement image setuid/setgid.
526 *
527 * Don't honor setuid/setgid if the filesystem prohibits it or if
528 * the process is being traced.
529 *
530 * XXXMAC: For the time being, use NOSUID to also prohibit
531 * transitions on the file system.
532 */
533 oldcred = p->p_ucred;
534 credential_changing = 0;
535 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
536 attr.va_uid;
537 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
538 attr.va_gid;
539 #ifdef MAC
540 will_transition = mac_execve_will_transition(oldcred, imgp->vp,
541 interplabel, imgp);
542 credential_changing |= will_transition;
543 #endif
544
545 if (credential_changing &&
546 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
547 (p->p_flag & P_TRACED) == 0) {
548 /*
549 * Turn off syscall tracing for set-id programs, except for
550 * root. Record any set-id flags first to make sure that
551 * we do not regain any tracing during a possible block.
552 */
553 setsugid(p);
554 #ifdef KTRACE
555 if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) {
556 mtx_lock(&ktrace_mtx);
557 p->p_traceflag = 0;
558 tracevp = p->p_tracevp;
559 p->p_tracevp = NULL;
560 tracecred = p->p_tracecred;
561 p->p_tracecred = NULL;
562 mtx_unlock(&ktrace_mtx);
563 }
564 #endif
565 /*
566 * Close any file descriptors 0..2 that reference procfs,
567 * then make sure file descriptors 0..2 are in use.
568 *
569 * setugidsafety() may call closef() and then pfind()
570 * which may grab the process lock.
571 * fdcheckstd() may call falloc() which may block to
572 * allocate memory, so temporarily drop the process lock.
573 */
574 PROC_UNLOCK(p);
575 setugidsafety(td);
576 error = fdcheckstd(td);
577 if (error != 0)
578 goto done1;
579 PROC_LOCK(p);
580 /*
581 * Set the new credentials.
582 */
583 crcopy(newcred, oldcred);
584 if (attr.va_mode & VSUID)
585 change_euid(newcred, euip);
586 if (attr.va_mode & VSGID)
587 change_egid(newcred, attr.va_gid);
588 #ifdef MAC
589 if (will_transition) {
590 mac_execve_transition(oldcred, newcred, imgp->vp,
591 interplabel, imgp);
592 }
593 #endif
594 /*
595 * Implement correct POSIX saved-id behavior.
596 *
597 * XXXMAC: Note that the current logic will save the
598 * uid and gid if a MAC domain transition occurs, even
599 * though maybe it shouldn't.
600 */
601 change_svuid(newcred, newcred->cr_uid);
602 change_svgid(newcred, newcred->cr_gid);
603 p->p_ucred = newcred;
604 newcred = NULL;
605 } else {
606 if (oldcred->cr_uid == oldcred->cr_ruid &&
607 oldcred->cr_gid == oldcred->cr_rgid)
608 p->p_flag &= ~P_SUGID;
609 /*
610 * Implement correct POSIX saved-id behavior.
611 *
612 * XXX: It's not clear that the existing behavior is
613 * POSIX-compliant. A number of sources indicate that the
614 * saved uid/gid should only be updated if the new ruid is
615 * not equal to the old ruid, or the new euid is not equal
616 * to the old euid and the new euid is not equal to the old
617 * ruid. The FreeBSD code always updates the saved uid/gid.
618 * Also, this code uses the new (replaced) euid and egid as
619 * the source, which may or may not be the right ones to use.
620 */
621 if (oldcred->cr_svuid != oldcred->cr_uid ||
622 oldcred->cr_svgid != oldcred->cr_gid) {
623 crcopy(newcred, oldcred);
624 change_svuid(newcred, newcred->cr_uid);
625 change_svgid(newcred, newcred->cr_gid);
626 p->p_ucred = newcred;
627 newcred = NULL;
628 }
629 }
630
631 /*
632 * Store the vp for use in procfs. This vnode was referenced prior
633 * to locking the proc lock.
634 */
635 textvp = p->p_textvp;
636 p->p_textvp = ndp->ni_vp;
637
638 /*
639 * Notify others that we exec'd, and clear the P_INEXEC flag
640 * as we're now a bona fide freshly-execed process.
641 */
642 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
643 p->p_flag &= ~P_INEXEC;
644
645 /*
646 * If tracing the process, trap to debugger so breakpoints
647 * can be set before the program executes.
648 * Use tdsignal to deliver signal to current thread, use
649 * psignal may cause the signal to be delivered to wrong thread
650 * because that thread will exit, remember we are going to enter
651 * single thread mode.
652 */
653 if (p->p_flag & P_TRACED)
654 tdsignal(td, SIGTRAP, SIGTARGET_TD);
655
656 /* clear "fork but no exec" flag, as we _are_ execing */
657 p->p_acflag &= ~AFORK;
658
659 /* Free any previous argument cache */
660 oldargs = p->p_args;
661 p->p_args = NULL;
662
663 /* Cache arguments if they fit inside our allowance */
664 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
665 bcopy(imgp->stringbase, newargs->ar_args, i);
666 p->p_args = newargs;
667 newargs = NULL;
668 }
669 PROC_UNLOCK(p);
670
671 /* Set values passed into the program in registers. */
672 if (p->p_sysent->sv_setregs)
673 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
674 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
675 else
676 exec_setregs(td, imgp->entry_addr,
677 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
678
679 done1:
680 /*
681 * Free any resources malloc'd earlier that we didn't use.
682 */
683 uifree(euip);
684 if (newcred == NULL)
685 crfree(oldcred);
686 else
687 crfree(newcred);
688 /*
689 * Handle deferred decrement of ref counts.
690 */
691 if (textvp != NULL)
692 vrele(textvp);
693 if (ndp->ni_vp && error != 0)
694 vrele(ndp->ni_vp);
695 #ifdef KTRACE
696 if (tracevp != NULL)
697 vrele(tracevp);
698 if (tracecred != NULL)
699 crfree(tracecred);
700 #endif
701 if (oldargs != NULL)
702 pargs_drop(oldargs);
703 if (newargs != NULL)
704 pargs_drop(newargs);
705 if (oldsigacts != NULL)
706 sigacts_free(oldsigacts);
707
708 exec_fail_dealloc:
709
710 /*
711 * free various allocated resources
712 */
713 if (imgp->firstpage != NULL)
714 exec_unmap_first_page(imgp);
715
716 if (imgp->vp != NULL) {
717 NDFREE(ndp, NDF_ONLY_PNBUF);
718 vput(imgp->vp);
719 }
720
721 if (imgp->stringbase != NULL)
722 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
723 ARG_MAX);
724
725 if (imgp->object != NULL)
726 vm_object_deallocate(imgp->object);
727
728 if (error == 0) {
729 /*
730 * Stop the process here if its stop event mask has
731 * the S_EXEC bit set.
732 */
733 STOPEVENT(p, S_EXEC, 0);
734 goto done2;
735 }
736
737 exec_fail:
738 /* we're done here, clear P_INEXEC */
739 PROC_LOCK(p);
740 p->p_flag &= ~P_INEXEC;
741 PROC_UNLOCK(p);
742
743 if (imgp->vmspace_destroyed) {
744 /* sorry, no more process anymore. exit gracefully */
745 #ifdef MAC
746 mac_execve_exit(imgp);
747 if (interplabel != NULL)
748 mac_vnode_label_free(interplabel);
749 #endif
750 exit1(td, W_EXITCODE(0, SIGABRT));
751 /* NOT REACHED */
752 error = 0;
753 }
754 done2:
755 #ifdef MAC
756 mac_execve_exit(imgp);
757 if (interplabel != NULL)
758 mac_vnode_label_free(interplabel);
759 #endif
760 mtx_unlock(&Giant);
761 return (error);
762 }
763
764 int
765 exec_map_first_page(imgp)
766 struct image_params *imgp;
767 {
768 int rv, i;
769 int initial_pagein;
770 vm_page_t ma[VM_INITIAL_PAGEIN];
771 vm_object_t object;
772
773 GIANT_REQUIRED;
774
775 if (imgp->firstpage != NULL)
776 exec_unmap_first_page(imgp);
777
778 VOP_GETVOBJECT(imgp->vp, &object);
779 VM_OBJECT_LOCK(object);
780 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
781 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
782 initial_pagein = VM_INITIAL_PAGEIN;
783 if (initial_pagein > object->size)
784 initial_pagein = object->size;
785 for (i = 1; i < initial_pagein; i++) {
786 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
787 if (ma[i]->valid)
788 break;
789 vm_page_lock_queues();
790 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
791 vm_page_unlock_queues();
792 break;
793 }
794 vm_page_busy(ma[i]);
795 vm_page_unlock_queues();
796 } else {
797 ma[i] = vm_page_alloc(object, i,
798 VM_ALLOC_NORMAL);
799 if (ma[i] == NULL)
800 break;
801 }
802 }
803 initial_pagein = i;
804 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
805 ma[0] = vm_page_lookup(object, 0);
806 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
807 (ma[0]->valid == 0)) {
808 if (ma[0]) {
809 vm_page_lock_queues();
810 pmap_remove_all(ma[0]);
811 vm_page_free(ma[0]);
812 vm_page_unlock_queues();
813 }
814 VM_OBJECT_UNLOCK(object);
815 return (EIO);
816 }
817 }
818 vm_page_lock_queues();
819 vm_page_hold(ma[0]);
820 vm_page_wakeup(ma[0]);
821 vm_page_unlock_queues();
822 VM_OBJECT_UNLOCK(object);
823
824 imgp->firstpage = sf_buf_alloc(ma[0], 0);
825 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
826
827 return (0);
828 }
829
830 void
831 exec_unmap_first_page(imgp)
832 struct image_params *imgp;
833 {
834 vm_page_t m;
835
836 if (imgp->firstpage != NULL) {
837 m = sf_buf_page(imgp->firstpage);
838 sf_buf_free(imgp->firstpage);
839 imgp->firstpage = NULL;
840 vm_page_lock_queues();
841 vm_page_unhold(m);
842 vm_page_unlock_queues();
843 }
844 }
845
846 /*
847 * Destroy old address space, and allocate a new stack
848 * The new stack is only SGROWSIZ large because it is grown
849 * automatically in trap.c.
850 */
851 int
852 exec_new_vmspace(imgp, sv)
853 struct image_params *imgp;
854 struct sysentvec *sv;
855 {
856 int error;
857 struct proc *p = imgp->proc;
858 struct vmspace *vmspace = p->p_vmspace;
859 vm_offset_t stack_addr;
860 vm_map_t map;
861
862 GIANT_REQUIRED;
863
864 imgp->vmspace_destroyed = 1;
865
866 /* Called with Giant held, do not depend on it! */
867 EVENTHANDLER_INVOKE(process_exec, p);
868
869 /*
870 * Here is as good a place as any to do any resource limit cleanups.
871 * This is needed if a 64 bit binary exec's a 32 bit binary - the
872 * data size limit may need to be changed to a value that makes
873 * sense for the 32 bit binary.
874 */
875 if (sv->sv_fixlimits != NULL)
876 sv->sv_fixlimits(imgp);
877
878 /*
879 * Blow away entire process VM, if address space not shared,
880 * otherwise, create a new VM space so that other threads are
881 * not disrupted
882 */
883 map = &vmspace->vm_map;
884 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
885 vm_map_max(map) == sv->sv_maxuser) {
886 shmexit(vmspace);
887 pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
888 vm_map_max(map));
889 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
890 } else {
891 vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
892 vmspace = p->p_vmspace;
893 map = &vmspace->vm_map;
894 }
895
896 /* Allocate a new stack */
897 stack_addr = sv->sv_usrstack - maxssiz;
898 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
899 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
900 if (error)
901 return (error);
902
903 #ifdef __ia64__
904 /* Allocate a new register stack */
905 stack_addr = IA64_BACKINGSTORE;
906 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
907 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
908 if (error)
909 return (error);
910 #endif
911
912 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
913 * VM_STACK case, but they are still used to monitor the size of the
914 * process stack so we can check the stack rlimit.
915 */
916 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
917 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
918
919 return (0);
920 }
921
922 /*
923 * Copy out argument and environment strings from the old process
924 * address space into the temporary string buffer.
925 */
926 int
927 exec_extract_strings(imgp)
928 struct image_params *imgp;
929 {
930 char **argv, **envv;
931 char *argp, *envp;
932 int error;
933 size_t length;
934
935 /*
936 * extract arguments first
937 */
938
939 argv = imgp->userspace_argv;
940
941 if (argv) {
942 argp = (caddr_t)(intptr_t)fuword(argv);
943 if (argp == (caddr_t)-1)
944 return (EFAULT);
945 if (argp)
946 argv++;
947 if (imgp->argv0)
948 argp = imgp->argv0;
949 if (argp) {
950 do {
951 if (argp == (caddr_t)-1)
952 return (EFAULT);
953 if ((error = copyinstr(argp, imgp->stringp,
954 imgp->stringspace, &length))) {
955 if (error == ENAMETOOLONG)
956 return (E2BIG);
957 return (error);
958 }
959 imgp->stringspace -= length;
960 imgp->stringp += length;
961 imgp->argc++;
962 } while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
963 }
964 } else
965 return (EFAULT);
966
967 imgp->endargs = imgp->stringp;
968
969 /*
970 * extract environment strings
971 */
972
973 envv = imgp->userspace_envv;
974
975 if (envv) {
976 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
977 if (envp == (caddr_t)-1)
978 return (EFAULT);
979 if ((error = copyinstr(envp, imgp->stringp,
980 imgp->stringspace, &length))) {
981 if (error == ENAMETOOLONG)
982 return (E2BIG);
983 return (error);
984 }
985 imgp->stringspace -= length;
986 imgp->stringp += length;
987 imgp->envc++;
988 }
989 }
990
991 return (0);
992 }
993
994 /*
995 * Copy strings out to the new process address space, constructing
996 * new arg and env vector tables. Return a pointer to the base
997 * so that it can be used as the initial stack pointer.
998 */
999 register_t *
1000 exec_copyout_strings(imgp)
1001 struct image_params *imgp;
1002 {
1003 int argc, envc;
1004 char **vectp;
1005 char *stringp, *destp;
1006 register_t *stack_base;
1007 struct ps_strings *arginfo;
1008 struct proc *p;
1009 int szsigcode;
1010
1011 /*
1012 * Calculate string base and vector table pointers.
1013 * Also deal with signal trampoline code for this exec type.
1014 */
1015 p = imgp->proc;
1016 szsigcode = 0;
1017 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1018 if (p->p_sysent->sv_szsigcode != NULL)
1019 szsigcode = *(p->p_sysent->sv_szsigcode);
1020 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1021 roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
1022
1023 /*
1024 * install sigcode
1025 */
1026 if (szsigcode)
1027 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1028 szsigcode), szsigcode);
1029
1030 /*
1031 * If we have a valid auxargs ptr, prepare some room
1032 * on the stack.
1033 */
1034 if (imgp->auxargs) {
1035 /*
1036 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1037 * lower compatibility.
1038 */
1039 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1040 (AT_COUNT * 2);
1041 /*
1042 * The '+ 2' is for the null pointers at the end of each of
1043 * the arg and env vector sets,and imgp->auxarg_size is room
1044 * for argument of Runtime loader.
1045 */
1046 vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1047 imgp->auxarg_size) * sizeof(char *));
1048
1049 } else
1050 /*
1051 * The '+ 2' is for the null pointers at the end of each of
1052 * the arg and env vector sets
1053 */
1054 vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1055 sizeof(char *));
1056
1057 /*
1058 * vectp also becomes our initial stack base
1059 */
1060 stack_base = (register_t *)vectp;
1061
1062 stringp = imgp->stringbase;
1063 argc = imgp->argc;
1064 envc = imgp->envc;
1065
1066 /*
1067 * Copy out strings - arguments and environment.
1068 */
1069 copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1070
1071 /*
1072 * Fill in "ps_strings" struct for ps, w, etc.
1073 */
1074 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1075 suword(&arginfo->ps_nargvstr, argc);
1076
1077 /*
1078 * Fill in argument portion of vector table.
1079 */
1080 for (; argc > 0; --argc) {
1081 suword(vectp++, (long)(intptr_t)destp);
1082 while (*stringp++ != 0)
1083 destp++;
1084 destp++;
1085 }
1086
1087 /* a null vector table pointer separates the argp's from the envp's */
1088 suword(vectp++, 0);
1089
1090 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1091 suword(&arginfo->ps_nenvstr, envc);
1092
1093 /*
1094 * Fill in environment portion of vector table.
1095 */
1096 for (; envc > 0; --envc) {
1097 suword(vectp++, (long)(intptr_t)destp);
1098 while (*stringp++ != 0)
1099 destp++;
1100 destp++;
1101 }
1102
1103 /* end of vector table is a null pointer */
1104 suword(vectp, 0);
1105
1106 return (stack_base);
1107 }
1108
1109 /*
1110 * Check permissions of file to execute.
1111 * Called with imgp->vp locked.
1112 * Return 0 for success or error code on failure.
1113 */
1114 int
1115 exec_check_permissions(imgp)
1116 struct image_params *imgp;
1117 {
1118 struct vnode *vp = imgp->vp;
1119 struct vattr *attr = imgp->attr;
1120 struct thread *td;
1121 int error;
1122
1123 td = curthread; /* XXXKSE */
1124
1125 /* Get file attributes */
1126 error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1127 if (error)
1128 return (error);
1129
1130 #ifdef MAC
1131 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1132 if (error)
1133 return (error);
1134 #endif
1135
1136 /*
1137 * 1) Check if file execution is disabled for the filesystem that this
1138 * file resides on.
1139 * 2) Insure that at least one execute bit is on - otherwise root
1140 * will always succeed, and we don't want to happen unless the
1141 * file really is executable.
1142 * 3) Insure that the file is a regular file.
1143 */
1144 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1145 ((attr->va_mode & 0111) == 0) ||
1146 (attr->va_type != VREG))
1147 return (EACCES);
1148
1149 /*
1150 * Zero length files can't be exec'd
1151 */
1152 if (attr->va_size == 0)
1153 return (ENOEXEC);
1154
1155 /*
1156 * Check for execute permission to file based on current credentials.
1157 */
1158 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1159 if (error)
1160 return (error);
1161
1162 /*
1163 * Check number of open-for-writes on the file and deny execution
1164 * if there are any.
1165 */
1166 if (vp->v_writecount)
1167 return (ETXTBSY);
1168
1169 /*
1170 * Call filesystem specific open routine (which does nothing in the
1171 * general case).
1172 */
1173 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1174 return (error);
1175 }
1176
1177 /*
1178 * Exec handler registration
1179 */
1180 int
1181 exec_register(execsw_arg)
1182 const struct execsw *execsw_arg;
1183 {
1184 const struct execsw **es, **xs, **newexecsw;
1185 int count = 2; /* New slot and trailing NULL */
1186
1187 if (execsw)
1188 for (es = execsw; *es; es++)
1189 count++;
1190 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1191 if (newexecsw == NULL)
1192 return (ENOMEM);
1193 xs = newexecsw;
1194 if (execsw)
1195 for (es = execsw; *es; es++)
1196 *xs++ = *es;
1197 *xs++ = execsw_arg;
1198 *xs = NULL;
1199 if (execsw)
1200 free(execsw, M_TEMP);
1201 execsw = newexecsw;
1202 return (0);
1203 }
1204
1205 int
1206 exec_unregister(execsw_arg)
1207 const struct execsw *execsw_arg;
1208 {
1209 const struct execsw **es, **xs, **newexecsw;
1210 int count = 1;
1211
1212 if (execsw == NULL)
1213 panic("unregister with no handlers left?\n");
1214
1215 for (es = execsw; *es; es++) {
1216 if (*es == execsw_arg)
1217 break;
1218 }
1219 if (*es == NULL)
1220 return (ENOENT);
1221 for (es = execsw; *es; es++)
1222 if (*es != execsw_arg)
1223 count++;
1224 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1225 if (newexecsw == NULL)
1226 return (ENOMEM);
1227 xs = newexecsw;
1228 for (es = execsw; *es; es++)
1229 if (*es != execsw_arg)
1230 *xs++ = *es;
1231 *xs = NULL;
1232 if (execsw)
1233 free(execsw, M_TEMP);
1234 execsw = newexecsw;
1235 return (0);
1236 }
Cache object: 699f4049adf90ca7c8ec337049dc3726
|