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/5.3/sys/kern/kern_exec.c 136588 2004-10-16 08:43:07Z 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/sf_buf.h>
55 #include <sys/sysent.h>
56 #include <sys/shm.h>
57 #include <sys/sysctl.h>
58 #include <sys/user.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 FILEDESC_LOCK(p->p_fd);
470 if (p->p_fd->fd_refcnt > 1) {
471 struct filedesc *tmp;
472
473 tmp = fdcopy(td->td_proc->p_fd);
474 FILEDESC_UNLOCK(p->p_fd);
475 fdfree(td);
476 p->p_fd = tmp;
477 } else
478 FILEDESC_UNLOCK(p->p_fd);
479
480 /*
481 * Malloc things before we need locks.
482 */
483 newcred = crget();
484 euip = uifind(attr.va_uid);
485 i = imgp->endargs - imgp->stringbase;
486 if (ps_arg_cache_limit >= i + sizeof(struct pargs))
487 newargs = pargs_alloc(i);
488
489 /* close files on exec */
490 fdcloseexec(td);
491
492 /* Get a reference to the vnode prior to locking the proc */
493 VREF(ndp->ni_vp);
494
495 /*
496 * For security and other reasons, signal handlers cannot
497 * be shared after an exec. The new process gets a copy of the old
498 * handlers. In execsigs(), the new process will have its signals
499 * reset.
500 */
501 PROC_LOCK(p);
502 if (sigacts_shared(p->p_sigacts)) {
503 oldsigacts = p->p_sigacts;
504 PROC_UNLOCK(p);
505 newsigacts = sigacts_alloc();
506 sigacts_copy(newsigacts, oldsigacts);
507 PROC_LOCK(p);
508 p->p_sigacts = newsigacts;
509 } else
510 oldsigacts = NULL;
511
512 /* Stop profiling */
513 stopprofclock(p);
514
515 /* reset caught signals */
516 execsigs(p);
517
518 /* name this process - nameiexec(p, ndp) */
519 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
520 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
521 p->p_comm[len] = 0;
522
523 /*
524 * mark as execed, wakeup the process that vforked (if any) and tell
525 * it that it now has its own resources back
526 */
527 p->p_flag |= P_EXEC;
528 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
529 p->p_flag &= ~P_PPWAIT;
530 wakeup(p->p_pptr);
531 }
532
533 /*
534 * Implement image setuid/setgid.
535 *
536 * Don't honor setuid/setgid if the filesystem prohibits it or if
537 * the process is being traced.
538 *
539 * XXXMAC: For the time being, use NOSUID to also prohibit
540 * transitions on the file system.
541 */
542 oldcred = p->p_ucred;
543 credential_changing = 0;
544 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
545 attr.va_uid;
546 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
547 attr.va_gid;
548 #ifdef MAC
549 will_transition = mac_execve_will_transition(oldcred, imgp->vp,
550 interplabel, imgp);
551 credential_changing |= will_transition;
552 #endif
553
554 if (credential_changing &&
555 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
556 (p->p_flag & P_TRACED) == 0) {
557 /*
558 * Turn off syscall tracing for set-id programs, except for
559 * root. Record any set-id flags first to make sure that
560 * we do not regain any tracing during a possible block.
561 */
562 setsugid(p);
563 #ifdef KTRACE
564 if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) {
565 mtx_lock(&ktrace_mtx);
566 p->p_traceflag = 0;
567 tracevp = p->p_tracevp;
568 p->p_tracevp = NULL;
569 tracecred = p->p_tracecred;
570 p->p_tracecred = NULL;
571 mtx_unlock(&ktrace_mtx);
572 }
573 #endif
574 /*
575 * Close any file descriptors 0..2 that reference procfs,
576 * then make sure file descriptors 0..2 are in use.
577 *
578 * setugidsafety() may call closef() and then pfind()
579 * which may grab the process lock.
580 * fdcheckstd() may call falloc() which may block to
581 * allocate memory, so temporarily drop the process lock.
582 */
583 PROC_UNLOCK(p);
584 setugidsafety(td);
585 error = fdcheckstd(td);
586 if (error != 0)
587 goto done1;
588 PROC_LOCK(p);
589 /*
590 * Set the new credentials.
591 */
592 crcopy(newcred, oldcred);
593 if (attr.va_mode & VSUID)
594 change_euid(newcred, euip);
595 if (attr.va_mode & VSGID)
596 change_egid(newcred, attr.va_gid);
597 #ifdef MAC
598 if (will_transition) {
599 mac_execve_transition(oldcred, newcred, imgp->vp,
600 interplabel, imgp);
601 }
602 #endif
603 /*
604 * Implement correct POSIX saved-id behavior.
605 *
606 * XXXMAC: Note that the current logic will save the
607 * uid and gid if a MAC domain transition occurs, even
608 * though maybe it shouldn't.
609 */
610 change_svuid(newcred, newcred->cr_uid);
611 change_svgid(newcred, newcred->cr_gid);
612 p->p_ucred = newcred;
613 newcred = NULL;
614 } else {
615 if (oldcred->cr_uid == oldcred->cr_ruid &&
616 oldcred->cr_gid == oldcred->cr_rgid)
617 p->p_flag &= ~P_SUGID;
618 /*
619 * Implement correct POSIX saved-id behavior.
620 *
621 * XXX: It's not clear that the existing behavior is
622 * POSIX-compliant. A number of sources indicate that the
623 * saved uid/gid should only be updated if the new ruid is
624 * not equal to the old ruid, or the new euid is not equal
625 * to the old euid and the new euid is not equal to the old
626 * ruid. The FreeBSD code always updates the saved uid/gid.
627 * Also, this code uses the new (replaced) euid and egid as
628 * the source, which may or may not be the right ones to use.
629 */
630 if (oldcred->cr_svuid != oldcred->cr_uid ||
631 oldcred->cr_svgid != oldcred->cr_gid) {
632 crcopy(newcred, oldcred);
633 change_svuid(newcred, newcred->cr_uid);
634 change_svgid(newcred, newcred->cr_gid);
635 p->p_ucred = newcred;
636 newcred = NULL;
637 }
638 }
639
640 /*
641 * Store the vp for use in procfs. This vnode was referenced prior
642 * to locking the proc lock.
643 */
644 textvp = p->p_textvp;
645 p->p_textvp = ndp->ni_vp;
646
647 /*
648 * Notify others that we exec'd, and clear the P_INEXEC flag
649 * as we're now a bona fide freshly-execed process.
650 */
651 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
652 p->p_flag &= ~P_INEXEC;
653
654 /*
655 * If tracing the process, trap to debugger so breakpoints
656 * can be set before the program executes.
657 * Use tdsignal to deliver signal to current thread, use
658 * psignal may cause the signal to be delivered to wrong thread
659 * because that thread will exit, remember we are going to enter
660 * single thread mode.
661 */
662 if (p->p_flag & P_TRACED)
663 tdsignal(td, SIGTRAP, SIGTARGET_TD);
664
665 /* clear "fork but no exec" flag, as we _are_ execing */
666 p->p_acflag &= ~AFORK;
667
668 /* Free any previous argument cache */
669 oldargs = p->p_args;
670 p->p_args = NULL;
671
672 /* Cache arguments if they fit inside our allowance */
673 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
674 bcopy(imgp->stringbase, newargs->ar_args, i);
675 p->p_args = newargs;
676 newargs = NULL;
677 }
678 PROC_UNLOCK(p);
679
680 /* Set values passed into the program in registers. */
681 if (p->p_sysent->sv_setregs)
682 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
683 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
684 else
685 exec_setregs(td, imgp->entry_addr,
686 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
687
688 done1:
689 /*
690 * Free any resources malloc'd earlier that we didn't use.
691 */
692 uifree(euip);
693 if (newcred == NULL)
694 crfree(oldcred);
695 else
696 crfree(newcred);
697 /*
698 * Handle deferred decrement of ref counts.
699 */
700 if (textvp != NULL)
701 vrele(textvp);
702 if (ndp->ni_vp && error != 0)
703 vrele(ndp->ni_vp);
704 #ifdef KTRACE
705 if (tracevp != NULL)
706 vrele(tracevp);
707 if (tracecred != NULL)
708 crfree(tracecred);
709 #endif
710 if (oldargs != NULL)
711 pargs_drop(oldargs);
712 if (newargs != NULL)
713 pargs_drop(newargs);
714 if (oldsigacts != NULL)
715 sigacts_free(oldsigacts);
716
717 exec_fail_dealloc:
718
719 /*
720 * free various allocated resources
721 */
722 if (imgp->firstpage != NULL)
723 exec_unmap_first_page(imgp);
724
725 if (imgp->vp != NULL) {
726 NDFREE(ndp, NDF_ONLY_PNBUF);
727 vput(imgp->vp);
728 }
729
730 if (imgp->stringbase != NULL)
731 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase,
732 ARG_MAX);
733
734 if (imgp->object != NULL)
735 vm_object_deallocate(imgp->object);
736
737 if (error == 0) {
738 /*
739 * Stop the process here if its stop event mask has
740 * the S_EXEC bit set.
741 */
742 STOPEVENT(p, S_EXEC, 0);
743 goto done2;
744 }
745
746 exec_fail:
747 /* we're done here, clear P_INEXEC */
748 PROC_LOCK(p);
749 p->p_flag &= ~P_INEXEC;
750 PROC_UNLOCK(p);
751
752 if (imgp->vmspace_destroyed) {
753 /* sorry, no more process anymore. exit gracefully */
754 #ifdef MAC
755 mac_execve_exit(imgp);
756 if (interplabel != NULL)
757 mac_vnode_label_free(interplabel);
758 #endif
759 exit1(td, W_EXITCODE(0, SIGABRT));
760 /* NOT REACHED */
761 error = 0;
762 }
763 done2:
764 #ifdef MAC
765 mac_execve_exit(imgp);
766 if (interplabel != NULL)
767 mac_vnode_label_free(interplabel);
768 #endif
769 mtx_unlock(&Giant);
770 return (error);
771 }
772
773 int
774 exec_map_first_page(imgp)
775 struct image_params *imgp;
776 {
777 int rv, i;
778 int initial_pagein;
779 vm_page_t ma[VM_INITIAL_PAGEIN];
780 vm_object_t object;
781
782 GIANT_REQUIRED;
783
784 if (imgp->firstpage != NULL)
785 exec_unmap_first_page(imgp);
786
787 VOP_GETVOBJECT(imgp->vp, &object);
788 VM_OBJECT_LOCK(object);
789 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
790 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
791 initial_pagein = VM_INITIAL_PAGEIN;
792 if (initial_pagein > object->size)
793 initial_pagein = object->size;
794 for (i = 1; i < initial_pagein; i++) {
795 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
796 if (ma[i]->valid)
797 break;
798 vm_page_lock_queues();
799 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
800 vm_page_unlock_queues();
801 break;
802 }
803 vm_page_busy(ma[i]);
804 vm_page_unlock_queues();
805 } else {
806 ma[i] = vm_page_alloc(object, i,
807 VM_ALLOC_NORMAL);
808 if (ma[i] == NULL)
809 break;
810 }
811 }
812 initial_pagein = i;
813 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
814 ma[0] = vm_page_lookup(object, 0);
815 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
816 (ma[0]->valid == 0)) {
817 if (ma[0]) {
818 vm_page_lock_queues();
819 pmap_remove_all(ma[0]);
820 vm_page_free(ma[0]);
821 vm_page_unlock_queues();
822 }
823 VM_OBJECT_UNLOCK(object);
824 return (EIO);
825 }
826 }
827 vm_page_lock_queues();
828 vm_page_hold(ma[0]);
829 vm_page_wakeup(ma[0]);
830 vm_page_unlock_queues();
831 VM_OBJECT_UNLOCK(object);
832
833 imgp->firstpage = sf_buf_alloc(ma[0], 0);
834 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
835
836 return (0);
837 }
838
839 void
840 exec_unmap_first_page(imgp)
841 struct image_params *imgp;
842 {
843 vm_page_t m;
844
845 if (imgp->firstpage != NULL) {
846 m = sf_buf_page(imgp->firstpage);
847 sf_buf_free(imgp->firstpage);
848 imgp->firstpage = NULL;
849 vm_page_lock_queues();
850 vm_page_unhold(m);
851 vm_page_unlock_queues();
852 }
853 }
854
855 /*
856 * Destroy old address space, and allocate a new stack
857 * The new stack is only SGROWSIZ large because it is grown
858 * automatically in trap.c.
859 */
860 int
861 exec_new_vmspace(imgp, sv)
862 struct image_params *imgp;
863 struct sysentvec *sv;
864 {
865 int error;
866 struct proc *p = imgp->proc;
867 struct vmspace *vmspace = p->p_vmspace;
868 vm_offset_t stack_addr;
869 vm_map_t map;
870
871 GIANT_REQUIRED;
872
873 imgp->vmspace_destroyed = 1;
874
875 /* Called with Giant held, do not depend on it! */
876 EVENTHANDLER_INVOKE(process_exec, p);
877
878 /*
879 * Here is as good a place as any to do any resource limit cleanups.
880 * This is needed if a 64 bit binary exec's a 32 bit binary - the
881 * data size limit may need to be changed to a value that makes
882 * sense for the 32 bit binary.
883 */
884 if (sv->sv_fixlimits != NULL)
885 sv->sv_fixlimits(imgp);
886
887 /*
888 * Blow away entire process VM, if address space not shared,
889 * otherwise, create a new VM space so that other threads are
890 * not disrupted
891 */
892 map = &vmspace->vm_map;
893 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
894 vm_map_max(map) == sv->sv_maxuser) {
895 shmexit(vmspace);
896 pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
897 vm_map_max(map));
898 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
899 } else {
900 vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
901 vmspace = p->p_vmspace;
902 map = &vmspace->vm_map;
903 }
904
905 /* Allocate a new stack */
906 stack_addr = sv->sv_usrstack - maxssiz;
907 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
908 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
909 if (error)
910 return (error);
911
912 #ifdef __ia64__
913 /* Allocate a new register stack */
914 stack_addr = IA64_BACKINGSTORE;
915 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
916 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
917 if (error)
918 return (error);
919 #endif
920
921 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
922 * VM_STACK case, but they are still used to monitor the size of the
923 * process stack so we can check the stack rlimit.
924 */
925 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
926 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
927
928 return (0);
929 }
930
931 /*
932 * Copy out argument and environment strings from the old process
933 * address space into the temporary string buffer.
934 */
935 int
936 exec_extract_strings(imgp)
937 struct image_params *imgp;
938 {
939 char **argv, **envv;
940 char *argp, *envp;
941 int error;
942 size_t length;
943
944 /*
945 * extract arguments first
946 */
947
948 argv = imgp->userspace_argv;
949
950 if (argv) {
951 argp = (caddr_t)(intptr_t)fuword(argv);
952 if (argp == (caddr_t)-1)
953 return (EFAULT);
954 if (argp)
955 argv++;
956 if (imgp->argv0)
957 argp = imgp->argv0;
958 if (argp) {
959 do {
960 if (argp == (caddr_t)-1)
961 return (EFAULT);
962 if ((error = copyinstr(argp, imgp->stringp,
963 imgp->stringspace, &length))) {
964 if (error == ENAMETOOLONG)
965 return (E2BIG);
966 return (error);
967 }
968 imgp->stringspace -= length;
969 imgp->stringp += length;
970 imgp->argc++;
971 } while ((argp = (caddr_t)(intptr_t)fuword(argv++)));
972 }
973 } else
974 return (EFAULT);
975
976 imgp->endargs = imgp->stringp;
977
978 /*
979 * extract environment strings
980 */
981
982 envv = imgp->userspace_envv;
983
984 if (envv) {
985 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
986 if (envp == (caddr_t)-1)
987 return (EFAULT);
988 if ((error = copyinstr(envp, imgp->stringp,
989 imgp->stringspace, &length))) {
990 if (error == ENAMETOOLONG)
991 return (E2BIG);
992 return (error);
993 }
994 imgp->stringspace -= length;
995 imgp->stringp += length;
996 imgp->envc++;
997 }
998 }
999
1000 return (0);
1001 }
1002
1003 /*
1004 * Copy strings out to the new process address space, constructing
1005 * new arg and env vector tables. Return a pointer to the base
1006 * so that it can be used as the initial stack pointer.
1007 */
1008 register_t *
1009 exec_copyout_strings(imgp)
1010 struct image_params *imgp;
1011 {
1012 int argc, envc;
1013 char **vectp;
1014 char *stringp, *destp;
1015 register_t *stack_base;
1016 struct ps_strings *arginfo;
1017 struct proc *p;
1018 int szsigcode;
1019
1020 /*
1021 * Calculate string base and vector table pointers.
1022 * Also deal with signal trampoline code for this exec type.
1023 */
1024 p = imgp->proc;
1025 szsigcode = 0;
1026 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1027 if (p->p_sysent->sv_szsigcode != NULL)
1028 szsigcode = *(p->p_sysent->sv_szsigcode);
1029 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1030 roundup((ARG_MAX - imgp->stringspace), sizeof(char *));
1031
1032 /*
1033 * install sigcode
1034 */
1035 if (szsigcode)
1036 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1037 szsigcode), szsigcode);
1038
1039 /*
1040 * If we have a valid auxargs ptr, prepare some room
1041 * on the stack.
1042 */
1043 if (imgp->auxargs) {
1044 /*
1045 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1046 * lower compatibility.
1047 */
1048 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1049 (AT_COUNT * 2);
1050 /*
1051 * The '+ 2' is for the null pointers at the end of each of
1052 * the arg and env vector sets,and imgp->auxarg_size is room
1053 * for argument of Runtime loader.
1054 */
1055 vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 +
1056 imgp->auxarg_size) * sizeof(char *));
1057
1058 } else
1059 /*
1060 * The '+ 2' is for the null pointers at the end of each of
1061 * the arg and env vector sets
1062 */
1063 vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) *
1064 sizeof(char *));
1065
1066 /*
1067 * vectp also becomes our initial stack base
1068 */
1069 stack_base = (register_t *)vectp;
1070
1071 stringp = imgp->stringbase;
1072 argc = imgp->argc;
1073 envc = imgp->envc;
1074
1075 /*
1076 * Copy out strings - arguments and environment.
1077 */
1078 copyout(stringp, destp, ARG_MAX - imgp->stringspace);
1079
1080 /*
1081 * Fill in "ps_strings" struct for ps, w, etc.
1082 */
1083 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1084 suword(&arginfo->ps_nargvstr, argc);
1085
1086 /*
1087 * Fill in argument portion of vector table.
1088 */
1089 for (; argc > 0; --argc) {
1090 suword(vectp++, (long)(intptr_t)destp);
1091 while (*stringp++ != 0)
1092 destp++;
1093 destp++;
1094 }
1095
1096 /* a null vector table pointer separates the argp's from the envp's */
1097 suword(vectp++, 0);
1098
1099 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1100 suword(&arginfo->ps_nenvstr, envc);
1101
1102 /*
1103 * Fill in environment portion of vector table.
1104 */
1105 for (; envc > 0; --envc) {
1106 suword(vectp++, (long)(intptr_t)destp);
1107 while (*stringp++ != 0)
1108 destp++;
1109 destp++;
1110 }
1111
1112 /* end of vector table is a null pointer */
1113 suword(vectp, 0);
1114
1115 return (stack_base);
1116 }
1117
1118 /*
1119 * Check permissions of file to execute.
1120 * Called with imgp->vp locked.
1121 * Return 0 for success or error code on failure.
1122 */
1123 int
1124 exec_check_permissions(imgp)
1125 struct image_params *imgp;
1126 {
1127 struct vnode *vp = imgp->vp;
1128 struct vattr *attr = imgp->attr;
1129 struct thread *td;
1130 int error;
1131
1132 td = curthread; /* XXXKSE */
1133
1134 /* Get file attributes */
1135 error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1136 if (error)
1137 return (error);
1138
1139 #ifdef MAC
1140 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1141 if (error)
1142 return (error);
1143 #endif
1144
1145 /*
1146 * 1) Check if file execution is disabled for the filesystem that this
1147 * file resides on.
1148 * 2) Insure that at least one execute bit is on - otherwise root
1149 * will always succeed, and we don't want to happen unless the
1150 * file really is executable.
1151 * 3) Insure that the file is a regular file.
1152 */
1153 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1154 ((attr->va_mode & 0111) == 0) ||
1155 (attr->va_type != VREG))
1156 return (EACCES);
1157
1158 /*
1159 * Zero length files can't be exec'd
1160 */
1161 if (attr->va_size == 0)
1162 return (ENOEXEC);
1163
1164 /*
1165 * Check for execute permission to file based on current credentials.
1166 */
1167 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1168 if (error)
1169 return (error);
1170
1171 /*
1172 * Check number of open-for-writes on the file and deny execution
1173 * if there are any.
1174 */
1175 if (vp->v_writecount)
1176 return (ETXTBSY);
1177
1178 /*
1179 * Call filesystem specific open routine (which does nothing in the
1180 * general case).
1181 */
1182 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1183 return (error);
1184 }
1185
1186 /*
1187 * Exec handler registration
1188 */
1189 int
1190 exec_register(execsw_arg)
1191 const struct execsw *execsw_arg;
1192 {
1193 const struct execsw **es, **xs, **newexecsw;
1194 int count = 2; /* New slot and trailing NULL */
1195
1196 if (execsw)
1197 for (es = execsw; *es; es++)
1198 count++;
1199 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1200 if (newexecsw == NULL)
1201 return (ENOMEM);
1202 xs = newexecsw;
1203 if (execsw)
1204 for (es = execsw; *es; es++)
1205 *xs++ = *es;
1206 *xs++ = execsw_arg;
1207 *xs = NULL;
1208 if (execsw)
1209 free(execsw, M_TEMP);
1210 execsw = newexecsw;
1211 return (0);
1212 }
1213
1214 int
1215 exec_unregister(execsw_arg)
1216 const struct execsw *execsw_arg;
1217 {
1218 const struct execsw **es, **xs, **newexecsw;
1219 int count = 1;
1220
1221 if (execsw == NULL)
1222 panic("unregister with no handlers left?\n");
1223
1224 for (es = execsw; *es; es++) {
1225 if (*es == execsw_arg)
1226 break;
1227 }
1228 if (*es == NULL)
1229 return (ENOENT);
1230 for (es = execsw; *es; es++)
1231 if (*es != execsw_arg)
1232 count++;
1233 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1234 if (newexecsw == NULL)
1235 return (ENOMEM);
1236 xs = newexecsw;
1237 for (es = execsw; *es; es++)
1238 if (*es != execsw_arg)
1239 *xs++ = *es;
1240 *xs = NULL;
1241 if (execsw)
1242 free(execsw, M_TEMP);
1243 execsw = newexecsw;
1244 return (0);
1245 }
Cache object: c3bde3ebf717701dae9ee06a0032c55e
|