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/6.4/sys/kern/kern_exec.c 197715 2009-10-02 18:09:56Z simon $");
29
30 #include "opt_hwpmc_hooks.h"
31 #include "opt_ktrace.h"
32 #include "opt_mac.h"
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/eventhandler.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/sysproto.h>
40 #include <sys/signalvar.h>
41 #include <sys/kernel.h>
42 #include <sys/mac.h>
43 #include <sys/mount.h>
44 #include <sys/filedesc.h>
45 #include <sys/fcntl.h>
46 #include <sys/acct.h>
47 #include <sys/exec.h>
48 #include <sys/imgact.h>
49 #include <sys/imgact_elf.h>
50 #include <sys/wait.h>
51 #include <sys/malloc.h>
52 #include <sys/proc.h>
53 #include <sys/pioctl.h>
54 #include <sys/namei.h>
55 #include <sys/resourcevar.h>
56 #include <sys/sf_buf.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
59 #include <sys/shm.h>
60 #include <sys/sysctl.h>
61 #include <sys/vnode.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_pager.h>
75
76 #ifdef HWPMC_HOOKS
77 #include <sys/pmckern.h>
78 #endif
79
80 #include <machine/reg.h>
81
82 #include <security/audit/audit.h>
83
84 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
85
86 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
87 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
88 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
89 static int do_execve(struct thread *td, struct image_args *args,
90 struct mac *mac_p);
91
92 /* XXX This should be vm_size_t. */
93 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
94 NULL, 0, sysctl_kern_ps_strings, "LU", "");
95
96 /* XXX This should be vm_size_t. */
97 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
98 NULL, 0, sysctl_kern_usrstack, "LU", "");
99
100 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
101 NULL, 0, sysctl_kern_stackprot, "I", "");
102
103 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
104 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
105 &ps_arg_cache_limit, 0, "");
106
107 SYSCTL_DECL(_security_bsd);
108
109 static int map_at_zero = 1;
110 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
111 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
112 "Permit processes to map an object at virtual address 0.");
113
114 static int
115 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
116 {
117 struct proc *p;
118 int error;
119
120 p = curproc;
121 #ifdef SCTL_MASK32
122 if (req->flags & SCTL_MASK32) {
123 unsigned int val;
124 val = (unsigned int)p->p_sysent->sv_psstrings;
125 error = SYSCTL_OUT(req, &val, sizeof(val));
126 } else
127 #endif
128 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
129 sizeof(p->p_sysent->sv_psstrings));
130 return error;
131 }
132
133 static int
134 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
135 {
136 struct proc *p;
137 int error;
138
139 p = curproc;
140 #ifdef SCTL_MASK32
141 if (req->flags & SCTL_MASK32) {
142 unsigned int val;
143 val = (unsigned int)p->p_sysent->sv_usrstack;
144 error = SYSCTL_OUT(req, &val, sizeof(val));
145 } else
146 #endif
147 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
148 sizeof(p->p_sysent->sv_usrstack));
149 return error;
150 }
151
152 static int
153 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
154 {
155 struct proc *p;
156
157 p = curproc;
158 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
159 sizeof(p->p_sysent->sv_stackprot)));
160 }
161
162 /*
163 * Each of the items is a pointer to a `const struct execsw', hence the
164 * double pointer here.
165 */
166 static const struct execsw **execsw;
167
168 #ifndef _SYS_SYSPROTO_H_
169 struct execve_args {
170 char *fname;
171 char **argv;
172 char **envv;
173 };
174 #endif
175
176 /*
177 * MPSAFE
178 */
179 int
180 execve(td, uap)
181 struct thread *td;
182 struct execve_args /* {
183 char *fname;
184 char **argv;
185 char **envv;
186 } */ *uap;
187 {
188 int error;
189 struct image_args args;
190
191 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
192 uap->argv, uap->envv);
193
194 if (error == 0)
195 error = kern_execve(td, &args, NULL);
196
197 exec_free_args(&args);
198
199 return (error);
200 }
201
202 #ifndef _SYS_SYSPROTO_H_
203 struct __mac_execve_args {
204 char *fname;
205 char **argv;
206 char **envv;
207 struct mac *mac_p;
208 };
209 #endif
210
211 /*
212 * MPSAFE
213 */
214 int
215 __mac_execve(td, uap)
216 struct thread *td;
217 struct __mac_execve_args /* {
218 char *fname;
219 char **argv;
220 char **envv;
221 struct mac *mac_p;
222 } */ *uap;
223 {
224 #ifdef MAC
225 int error;
226 struct image_args args;
227
228 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
229 uap->argv, uap->envv);
230
231 if (error == 0)
232 error = kern_execve(td, &args, uap->mac_p);
233
234 exec_free_args(&args);
235
236 return (error);
237 #else
238 return (ENOSYS);
239 #endif
240 }
241
242 /*
243 * XXX: kern_execve has the astonishing property of not always
244 * returning to the caller. If sufficiently bad things happen during
245 * the call to do_execve(), it can end up calling exit1(); as a result,
246 * callers must avoid doing anything which they might need to undo
247 * (e.g., allocating memory).
248 */
249 int
250 kern_execve(td, args, mac_p)
251 struct thread *td;
252 struct image_args *args;
253 struct mac *mac_p;
254 {
255 struct proc *p = td->td_proc;
256 int error;
257
258 AUDIT_ARG(argv, args->begin_argv, args->argc,
259 args->begin_envv - args->begin_argv);
260 AUDIT_ARG(envv, args->begin_envv, args->envc,
261 args->endp - args->begin_envv);
262 if (p->p_flag & P_HADTHREADS) {
263 PROC_LOCK(p);
264 if (thread_single(SINGLE_BOUNDARY)) {
265 PROC_UNLOCK(p);
266 return (ERESTART); /* Try again later. */
267 }
268 PROC_UNLOCK(p);
269 }
270
271 error = do_execve(td, args, mac_p);
272
273 if (p->p_flag & P_HADTHREADS) {
274 PROC_LOCK(p);
275 /*
276 * If success, we upgrade to SINGLE_EXIT state to
277 * force other threads to suicide.
278 */
279 if (error == 0)
280 thread_single(SINGLE_EXIT);
281 else
282 thread_single_end();
283 PROC_UNLOCK(p);
284 }
285
286 return (error);
287 }
288
289 /*
290 * In-kernel implementation of execve(). All arguments are assumed to be
291 * userspace pointers from the passed thread.
292 *
293 * MPSAFE
294 */
295 static int
296 do_execve(td, args, mac_p)
297 struct thread *td;
298 struct image_args *args;
299 struct mac *mac_p;
300 {
301 struct proc *p = td->td_proc;
302 struct nameidata nd, *ndp;
303 struct ucred *newcred = NULL, *oldcred;
304 struct uidinfo *euip;
305 register_t *stack_base;
306 int error, len, i;
307 struct image_params image_params, *imgp;
308 struct vattr attr;
309 int (*img_first)(struct image_params *);
310 struct pargs *oldargs = NULL, *newargs = NULL;
311 struct sigacts *oldsigacts, *newsigacts;
312 #ifdef KTRACE
313 struct vnode *tracevp = NULL;
314 struct ucred *tracecred = NULL;
315 #endif
316 struct vnode *textvp = NULL;
317 int credential_changing;
318 int vfslocked;
319 int textset;
320 #ifdef MAC
321 struct label *interplabel = NULL;
322 int will_transition;
323 #endif
324 #ifdef HWPMC_HOOKS
325 struct pmckern_procexec pe;
326 #endif
327
328 vfslocked = 0;
329 imgp = &image_params;
330
331 /*
332 * Lock the process and set the P_INEXEC flag to indicate that
333 * it should be left alone until we're done here. This is
334 * necessary to avoid race conditions - e.g. in ptrace() -
335 * that might allow a local user to illicitly obtain elevated
336 * privileges.
337 */
338 PROC_LOCK(p);
339 KASSERT((p->p_flag & P_INEXEC) == 0,
340 ("%s(): process already has P_INEXEC flag", __func__));
341 p->p_flag |= P_INEXEC;
342 PROC_UNLOCK(p);
343
344 /*
345 * Initialize part of the common data
346 */
347 imgp->proc = p;
348 imgp->execlabel = NULL;
349 imgp->attr = &attr;
350 imgp->entry_addr = 0;
351 imgp->vmspace_destroyed = 0;
352 imgp->interpreted = 0;
353 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
354 imgp->auxargs = NULL;
355 imgp->vp = NULL;
356 imgp->object = NULL;
357 imgp->firstpage = NULL;
358 imgp->ps_strings = 0;
359 imgp->auxarg_size = 0;
360 imgp->args = args;
361
362 #ifdef MAC
363 error = mac_execve_enter(imgp, mac_p);
364 if (error)
365 goto exec_fail;
366 #endif
367
368 imgp->image_header = NULL;
369
370 /*
371 * Translate the file name. namei() returns a vnode pointer
372 * in ni_vp amoung other things.
373 *
374 * XXXAUDIT: It would be desirable to also audit the name of the
375 * interpreter if this is an interpreted binary.
376 */
377 ndp = &nd;
378 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE |
379 AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
380
381 interpret:
382 error = namei(ndp);
383 if (error)
384 goto exec_fail;
385
386 vfslocked = NDHASGIANT(ndp);
387 imgp->vp = ndp->ni_vp;
388
389 /*
390 * Check file permissions (also 'opens' file)
391 */
392 error = exec_check_permissions(imgp);
393 if (error)
394 goto exec_fail_dealloc;
395
396 imgp->object = imgp->vp->v_object;
397 if (imgp->object != NULL)
398 vm_object_reference(imgp->object);
399
400 /*
401 * Set VV_TEXT now so no one can write to the executable while we're
402 * activating it.
403 *
404 * Remember if this was set before and unset it in case this is not
405 * actually an executable image.
406 */
407 textset = imgp->vp->v_vflag & VV_TEXT;
408 imgp->vp->v_vflag |= VV_TEXT;
409
410 error = exec_map_first_page(imgp);
411 if (error)
412 goto exec_fail_dealloc;
413
414 /*
415 * If the current process has a special image activator it
416 * wants to try first, call it. For example, emulating shell
417 * scripts differently.
418 */
419 error = -1;
420 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
421 error = img_first(imgp);
422
423 /*
424 * Loop through the list of image activators, calling each one.
425 * An activator returns -1 if there is no match, 0 on success,
426 * and an error otherwise.
427 */
428 for (i = 0; error == -1 && execsw[i]; ++i) {
429 if (execsw[i]->ex_imgact == NULL ||
430 execsw[i]->ex_imgact == img_first) {
431 continue;
432 }
433 error = (*execsw[i]->ex_imgact)(imgp);
434 }
435
436 if (error) {
437 if (error == -1) {
438 if (textset == 0)
439 imgp->vp->v_vflag &= ~VV_TEXT;
440 error = ENOEXEC;
441 }
442 goto exec_fail_dealloc;
443 }
444
445 /*
446 * Special interpreter operation, cleanup and loop up to try to
447 * activate the interpreter.
448 */
449 if (imgp->interpreted) {
450 exec_unmap_first_page(imgp);
451 /*
452 * VV_TEXT needs to be unset for scripts. There is a short
453 * period before we determine that something is a script where
454 * VV_TEXT will be set. The vnode lock is held over this
455 * entire period so nothing should illegitimately be blocked.
456 */
457 imgp->vp->v_vflag &= ~VV_TEXT;
458 /* free name buffer and old vnode */
459 NDFREE(ndp, NDF_ONLY_PNBUF);
460 #ifdef MAC
461 interplabel = mac_vnode_label_alloc();
462 mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
463 #endif
464 vput(ndp->ni_vp);
465 vm_object_deallocate(imgp->object);
466 imgp->object = NULL;
467 VFS_UNLOCK_GIANT(vfslocked);
468 vfslocked = 0;
469 /* set new name to that of the interpreter */
470 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
471 UIO_SYSSPACE, imgp->interpreter_name, td);
472 goto interpret;
473 }
474
475 /*
476 * Copy out strings (args and env) and initialize stack base
477 */
478 if (p->p_sysent->sv_copyout_strings)
479 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
480 else
481 stack_base = exec_copyout_strings(imgp);
482
483 /*
484 * If custom stack fixup routine present for this process
485 * let it do the stack setup.
486 * Else stuff argument count as first item on stack
487 */
488 if (p->p_sysent->sv_fixup != NULL)
489 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
490 else
491 suword(--stack_base, imgp->args->argc);
492
493 /*
494 * For security and other reasons, the file descriptor table cannot
495 * be shared after an exec.
496 */
497 fdunshare(p, td);
498
499 /*
500 * Malloc things before we need locks.
501 */
502 newcred = crget();
503 euip = uifind(attr.va_uid);
504 i = imgp->args->begin_envv - imgp->args->begin_argv;
505 /* Cache arguments if they fit inside our allowance */
506 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
507 newargs = pargs_alloc(i);
508 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
509 }
510
511 /* close files on exec */
512 VOP_UNLOCK(imgp->vp, 0, td);
513 fdcloseexec(td);
514 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
515
516 /* Get a reference to the vnode prior to locking the proc */
517 VREF(ndp->ni_vp);
518
519 /*
520 * For security and other reasons, signal handlers cannot
521 * be shared after an exec. The new process gets a copy of the old
522 * handlers. In execsigs(), the new process will have its signals
523 * reset.
524 */
525 PROC_LOCK(p);
526 if (sigacts_shared(p->p_sigacts)) {
527 oldsigacts = p->p_sigacts;
528 PROC_UNLOCK(p);
529 newsigacts = sigacts_alloc();
530 sigacts_copy(newsigacts, oldsigacts);
531 PROC_LOCK(p);
532 p->p_sigacts = newsigacts;
533 } else
534 oldsigacts = NULL;
535
536 /* Stop profiling */
537 stopprofclock(p);
538
539 /* reset caught signals */
540 execsigs(p);
541
542 /* name this process - nameiexec(p, ndp) */
543 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
544 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
545 p->p_comm[len] = 0;
546
547 /*
548 * mark as execed, wakeup the process that vforked (if any) and tell
549 * it that it now has its own resources back
550 */
551 p->p_flag |= P_EXEC;
552 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
553 p->p_flag &= ~P_PPWAIT;
554 wakeup(p->p_pptr);
555 }
556
557 /*
558 * Implement image setuid/setgid.
559 *
560 * Don't honor setuid/setgid if the filesystem prohibits it or if
561 * the process is being traced.
562 *
563 * XXXMAC: For the time being, use NOSUID to also prohibit
564 * transitions on the file system.
565 */
566 oldcred = p->p_ucred;
567 credential_changing = 0;
568 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
569 attr.va_uid;
570 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
571 attr.va_gid;
572 #ifdef MAC
573 will_transition = mac_execve_will_transition(oldcred, imgp->vp,
574 interplabel, imgp);
575 credential_changing |= will_transition;
576 #endif
577
578 if (credential_changing &&
579 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
580 (p->p_flag & P_TRACED) == 0) {
581 /*
582 * Turn off syscall tracing for set-id programs, except for
583 * root. Record any set-id flags first to make sure that
584 * we do not regain any tracing during a possible block.
585 */
586 setsugid(p);
587 #ifdef KTRACE
588 if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) {
589 mtx_lock(&ktrace_mtx);
590 p->p_traceflag = 0;
591 tracevp = p->p_tracevp;
592 p->p_tracevp = NULL;
593 tracecred = p->p_tracecred;
594 p->p_tracecred = NULL;
595 mtx_unlock(&ktrace_mtx);
596 }
597 #endif
598 /*
599 * Close any file descriptors 0..2 that reference procfs,
600 * then make sure file descriptors 0..2 are in use.
601 *
602 * setugidsafety() may call closef() and then pfind()
603 * which may grab the process lock.
604 * fdcheckstd() may call falloc() which may block to
605 * allocate memory, so temporarily drop the process lock.
606 */
607 PROC_UNLOCK(p);
608 setugidsafety(td);
609 VOP_UNLOCK(imgp->vp, 0, td);
610 error = fdcheckstd(td);
611 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
612 if (error != 0)
613 goto done1;
614 PROC_LOCK(p);
615 /*
616 * Set the new credentials.
617 */
618 crcopy(newcred, oldcred);
619 if (attr.va_mode & VSUID)
620 change_euid(newcred, euip);
621 if (attr.va_mode & VSGID)
622 change_egid(newcred, attr.va_gid);
623 #ifdef MAC
624 if (will_transition) {
625 mac_execve_transition(oldcred, newcred, imgp->vp,
626 interplabel, imgp);
627 }
628 #endif
629 /*
630 * Implement correct POSIX saved-id behavior.
631 *
632 * XXXMAC: Note that the current logic will save the
633 * uid and gid if a MAC domain transition occurs, even
634 * though maybe it shouldn't.
635 */
636 change_svuid(newcred, newcred->cr_uid);
637 change_svgid(newcred, newcred->cr_gid);
638 p->p_ucred = newcred;
639 newcred = NULL;
640 } else {
641 if (oldcred->cr_uid == oldcred->cr_ruid &&
642 oldcred->cr_gid == oldcred->cr_rgid)
643 p->p_flag &= ~P_SUGID;
644 /*
645 * Implement correct POSIX saved-id behavior.
646 *
647 * XXX: It's not clear that the existing behavior is
648 * POSIX-compliant. A number of sources indicate that the
649 * saved uid/gid should only be updated if the new ruid is
650 * not equal to the old ruid, or the new euid is not equal
651 * to the old euid and the new euid is not equal to the old
652 * ruid. The FreeBSD code always updates the saved uid/gid.
653 * Also, this code uses the new (replaced) euid and egid as
654 * the source, which may or may not be the right ones to use.
655 */
656 if (oldcred->cr_svuid != oldcred->cr_uid ||
657 oldcred->cr_svgid != oldcred->cr_gid) {
658 crcopy(newcred, oldcred);
659 change_svuid(newcred, newcred->cr_uid);
660 change_svgid(newcred, newcred->cr_gid);
661 p->p_ucred = newcred;
662 newcred = NULL;
663 }
664 }
665
666 /*
667 * Store the vp for use in procfs. This vnode was referenced prior
668 * to locking the proc lock.
669 */
670 textvp = p->p_textvp;
671 p->p_textvp = ndp->ni_vp;
672
673 /*
674 * Notify others that we exec'd, and clear the P_INEXEC flag
675 * as we're now a bona fide freshly-execed process.
676 */
677 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
678 p->p_flag &= ~P_INEXEC;
679
680 /*
681 * If tracing the process, trap to debugger so breakpoints
682 * can be set before the program executes.
683 * Use tdsignal to deliver signal to current thread, use
684 * psignal may cause the signal to be delivered to wrong thread
685 * because that thread will exit, remember we are going to enter
686 * single thread mode.
687 */
688 if (p->p_flag & P_TRACED)
689 tdsignal(td, SIGTRAP, SIGTARGET_TD);
690
691 /* clear "fork but no exec" flag, as we _are_ execing */
692 p->p_acflag &= ~AFORK;
693
694 /*
695 * Free any previous argument cache and replace it with
696 * the new argument cache, if any.
697 */
698 oldargs = p->p_args;
699 p->p_args = newargs;
700 newargs = NULL;
701
702 #ifdef HWPMC_HOOKS
703 /*
704 * Check if system-wide sampling is in effect or if the
705 * current process is using PMCs. If so, do exec() time
706 * processing. This processing needs to happen AFTER the
707 * P_INEXEC flag is cleared.
708 *
709 * The proc lock needs to be released before taking the PMC
710 * SX.
711 */
712 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
713 PROC_UNLOCK(p);
714 pe.pm_credentialschanged = credential_changing;
715 pe.pm_entryaddr = imgp->entry_addr;
716
717 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
718 } else
719 PROC_UNLOCK(p);
720 #else /* !HWPMC_HOOKS */
721 PROC_UNLOCK(p);
722 #endif
723
724 /* Set values passed into the program in registers. */
725 if (p->p_sysent->sv_setregs)
726 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
727 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
728 else
729 exec_setregs(td, imgp->entry_addr,
730 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
731
732 vfs_mark_atime(imgp->vp, td);
733
734 done1:
735 /*
736 * Free any resources malloc'd earlier that we didn't use.
737 */
738 uifree(euip);
739 if (newcred == NULL)
740 crfree(oldcred);
741 else
742 crfree(newcred);
743 VOP_UNLOCK(imgp->vp, 0, td);
744 /*
745 * Handle deferred decrement of ref counts.
746 */
747 if (textvp != NULL) {
748 int tvfslocked;
749
750 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
751 vrele(textvp);
752 VFS_UNLOCK_GIANT(tvfslocked);
753 }
754 if (ndp->ni_vp && error != 0)
755 vrele(ndp->ni_vp);
756 #ifdef KTRACE
757 if (tracevp != NULL) {
758 int tvfslocked;
759
760 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
761 vrele(tracevp);
762 VFS_UNLOCK_GIANT(tvfslocked);
763 }
764 if (tracecred != NULL)
765 crfree(tracecred);
766 #endif
767 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
768 if (oldargs != NULL)
769 pargs_drop(oldargs);
770 if (newargs != NULL)
771 pargs_drop(newargs);
772 if (oldsigacts != NULL)
773 sigacts_free(oldsigacts);
774
775 exec_fail_dealloc:
776
777 /*
778 * free various allocated resources
779 */
780 if (imgp->firstpage != NULL)
781 exec_unmap_first_page(imgp);
782
783 if (imgp->vp != NULL) {
784 NDFREE(ndp, NDF_ONLY_PNBUF);
785 vput(imgp->vp);
786 }
787
788 if (imgp->object != NULL)
789 vm_object_deallocate(imgp->object);
790
791 if (error == 0) {
792 /*
793 * Stop the process here if its stop event mask has
794 * the S_EXEC bit set.
795 */
796 STOPEVENT(p, S_EXEC, 0);
797 goto done2;
798 }
799
800 exec_fail:
801 /* we're done here, clear P_INEXEC */
802 PROC_LOCK(p);
803 p->p_flag &= ~P_INEXEC;
804 PROC_UNLOCK(p);
805
806 if (imgp->vmspace_destroyed) {
807 /* sorry, no more process anymore. exit gracefully */
808 #ifdef MAC
809 mac_execve_exit(imgp);
810 if (interplabel != NULL)
811 mac_vnode_label_free(interplabel);
812 #endif
813 VFS_UNLOCK_GIANT(vfslocked);
814 exec_free_args(args);
815 exit1(td, W_EXITCODE(0, SIGABRT));
816 /* NOT REACHED */
817 error = 0;
818 }
819 done2:
820 #ifdef MAC
821 mac_execve_exit(imgp);
822 if (interplabel != NULL)
823 mac_vnode_label_free(interplabel);
824 #endif
825 VFS_UNLOCK_GIANT(vfslocked);
826 return (error);
827 }
828
829 int
830 exec_map_first_page(imgp)
831 struct image_params *imgp;
832 {
833 int rv, i;
834 int initial_pagein;
835 vm_page_t ma[VM_INITIAL_PAGEIN];
836 vm_object_t object;
837
838 if (imgp->firstpage != NULL)
839 exec_unmap_first_page(imgp);
840
841 object = imgp->vp->v_object;
842 if (object == NULL)
843 return (EACCES);
844 VM_OBJECT_LOCK(object);
845 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
846 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
847 initial_pagein = VM_INITIAL_PAGEIN;
848 if (initial_pagein > object->size)
849 initial_pagein = object->size;
850 for (i = 1; i < initial_pagein; i++) {
851 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
852 if (ma[i]->valid)
853 break;
854 vm_page_lock_queues();
855 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
856 vm_page_unlock_queues();
857 break;
858 }
859 vm_page_busy(ma[i]);
860 vm_page_unlock_queues();
861 } else {
862 ma[i] = vm_page_alloc(object, i,
863 VM_ALLOC_NORMAL);
864 if (ma[i] == NULL)
865 break;
866 }
867 }
868 initial_pagein = i;
869 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
870 ma[0] = vm_page_lookup(object, 0);
871 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
872 (ma[0]->valid == 0)) {
873 if (ma[0]) {
874 vm_page_lock_queues();
875 pmap_remove_all(ma[0]);
876 vm_page_free(ma[0]);
877 vm_page_unlock_queues();
878 }
879 VM_OBJECT_UNLOCK(object);
880 return (EIO);
881 }
882 }
883 vm_page_lock_queues();
884 vm_page_hold(ma[0]);
885 vm_page_wakeup(ma[0]);
886 vm_page_unlock_queues();
887 VM_OBJECT_UNLOCK(object);
888
889 imgp->firstpage = sf_buf_alloc(ma[0], 0);
890 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
891
892 return (0);
893 }
894
895 void
896 exec_unmap_first_page(imgp)
897 struct image_params *imgp;
898 {
899 vm_page_t m;
900
901 if (imgp->firstpage != NULL) {
902 m = sf_buf_page(imgp->firstpage);
903 sf_buf_free(imgp->firstpage);
904 imgp->firstpage = NULL;
905 vm_page_lock_queues();
906 vm_page_unhold(m);
907 vm_page_unlock_queues();
908 }
909 }
910
911 /*
912 * Destroy old address space, and allocate a new stack
913 * The new stack is only SGROWSIZ large because it is grown
914 * automatically in trap.c.
915 */
916 int
917 exec_new_vmspace(imgp, sv)
918 struct image_params *imgp;
919 struct sysentvec *sv;
920 {
921 int error;
922 struct proc *p = imgp->proc;
923 struct vmspace *vmspace = p->p_vmspace;
924 vm_offset_t sv_minuser, stack_addr;
925 vm_map_t map;
926
927 imgp->vmspace_destroyed = 1;
928
929 /* Called with Giant held, do not depend on it! */
930 EVENTHANDLER_INVOKE(process_exec, p);
931
932 /*
933 * Blow away entire process VM, if address space not shared,
934 * otherwise, create a new VM space so that other threads are
935 * not disrupted
936 */
937 map = &vmspace->vm_map;
938 if (map_at_zero)
939 sv_minuser = sv->sv_minuser;
940 else
941 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
942 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
943 vm_map_max(map) == sv->sv_maxuser) {
944 shmexit(vmspace);
945 pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
946 vm_map_max(map));
947 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
948 } else {
949 vmspace_exec(p, sv_minuser, sv->sv_maxuser);
950 vmspace = p->p_vmspace;
951 map = &vmspace->vm_map;
952 }
953
954 /* Allocate a new stack */
955 stack_addr = sv->sv_usrstack - maxssiz;
956 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
957 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
958 if (error)
959 return (error);
960
961 #ifdef __ia64__
962 /* Allocate a new register stack */
963 stack_addr = IA64_BACKINGSTORE;
964 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
965 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
966 if (error)
967 return (error);
968 #endif
969
970 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
971 * VM_STACK case, but they are still used to monitor the size of the
972 * process stack so we can check the stack rlimit.
973 */
974 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
975 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
976
977 return (0);
978 }
979
980 /*
981 * Copy out argument and environment strings from the old process
982 * address space into the temporary string buffer.
983 */
984 int
985 exec_copyin_args(struct image_args *args, char *fname,
986 enum uio_seg segflg, char **argv, char **envv)
987 {
988 char *argp, *envp;
989 int error;
990 size_t length;
991
992 error = 0;
993
994 bzero(args, sizeof(*args));
995 if (argv == NULL)
996 return (EFAULT);
997 /*
998 * Allocate temporary demand zeroed space for argument and
999 * environment strings:
1000 *
1001 * o ARG_MAX for argument and environment;
1002 * o MAXSHELLCMDLEN for the name of interpreters.
1003 */
1004 args->buf = (char *) kmem_alloc_wait(exec_map,
1005 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1006 if (args->buf == NULL)
1007 return (ENOMEM);
1008 args->begin_argv = args->buf;
1009 args->endp = args->begin_argv;
1010 args->stringspace = ARG_MAX;
1011
1012 args->fname = args->buf + ARG_MAX;
1013
1014 /*
1015 * Copy the file name.
1016 */
1017 error = (segflg == UIO_SYSSPACE) ?
1018 copystr(fname, args->fname, PATH_MAX, &length) :
1019 copyinstr(fname, args->fname, PATH_MAX, &length);
1020 if (error != 0)
1021 return (error);
1022
1023 /*
1024 * extract arguments first
1025 */
1026 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1027 if (argp == (caddr_t) -1)
1028 return (EFAULT);
1029 if ((error = copyinstr(argp, args->endp,
1030 args->stringspace, &length))) {
1031 if (error == ENAMETOOLONG)
1032 return (E2BIG);
1033 return (error);
1034 }
1035 args->stringspace -= length;
1036 args->endp += length;
1037 args->argc++;
1038 }
1039
1040 args->begin_envv = args->endp;
1041
1042 /*
1043 * extract environment strings
1044 */
1045 if (envv) {
1046 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1047 if (envp == (caddr_t)-1)
1048 return (EFAULT);
1049 if ((error = copyinstr(envp, args->endp,
1050 args->stringspace, &length))) {
1051 if (error == ENAMETOOLONG)
1052 return (E2BIG);
1053 return (error);
1054 }
1055 args->stringspace -= length;
1056 args->endp += length;
1057 args->envc++;
1058 }
1059 }
1060
1061 return (0);
1062 }
1063
1064 void
1065 exec_free_args(struct image_args *args)
1066 {
1067
1068 if (args->buf) {
1069 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1070 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1071 args->buf = NULL;
1072 }
1073 }
1074
1075 /*
1076 * Copy strings out to the new process address space, constructing
1077 * new arg and env vector tables. Return a pointer to the base
1078 * so that it can be used as the initial stack pointer.
1079 */
1080 register_t *
1081 exec_copyout_strings(imgp)
1082 struct image_params *imgp;
1083 {
1084 int argc, envc;
1085 char **vectp;
1086 char *stringp, *destp;
1087 register_t *stack_base;
1088 struct ps_strings *arginfo;
1089 struct proc *p;
1090 int szsigcode;
1091
1092 /*
1093 * Calculate string base and vector table pointers.
1094 * Also deal with signal trampoline code for this exec type.
1095 */
1096 p = imgp->proc;
1097 szsigcode = 0;
1098 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1099 if (p->p_sysent->sv_szsigcode != NULL)
1100 szsigcode = *(p->p_sysent->sv_szsigcode);
1101 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1102 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1103
1104 /*
1105 * install sigcode
1106 */
1107 if (szsigcode)
1108 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1109 szsigcode), szsigcode);
1110
1111 /*
1112 * If we have a valid auxargs ptr, prepare some room
1113 * on the stack.
1114 */
1115 if (imgp->auxargs) {
1116 /*
1117 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1118 * lower compatibility.
1119 */
1120 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1121 (AT_COUNT * 2);
1122 /*
1123 * The '+ 2' is for the null pointers at the end of each of
1124 * the arg and env vector sets,and imgp->auxarg_size is room
1125 * for argument of Runtime loader.
1126 */
1127 vectp = (char **)(destp - (imgp->args->argc +
1128 imgp->args->envc + 2 + imgp->auxarg_size) *
1129 sizeof(char *));
1130
1131 } else {
1132 /*
1133 * The '+ 2' is for the null pointers at the end of each of
1134 * the arg and env vector sets
1135 */
1136 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1137 sizeof(char *));
1138 }
1139
1140 /*
1141 * vectp also becomes our initial stack base
1142 */
1143 stack_base = (register_t *)vectp;
1144
1145 stringp = imgp->args->begin_argv;
1146 argc = imgp->args->argc;
1147 envc = imgp->args->envc;
1148
1149 /*
1150 * Copy out strings - arguments and environment.
1151 */
1152 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1153
1154 /*
1155 * Fill in "ps_strings" struct for ps, w, etc.
1156 */
1157 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1158 suword(&arginfo->ps_nargvstr, argc);
1159
1160 /*
1161 * Fill in argument portion of vector table.
1162 */
1163 for (; argc > 0; --argc) {
1164 suword(vectp++, (long)(intptr_t)destp);
1165 while (*stringp++ != 0)
1166 destp++;
1167 destp++;
1168 }
1169
1170 /* a null vector table pointer separates the argp's from the envp's */
1171 suword(vectp++, 0);
1172
1173 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1174 suword(&arginfo->ps_nenvstr, envc);
1175
1176 /*
1177 * Fill in environment portion of vector table.
1178 */
1179 for (; envc > 0; --envc) {
1180 suword(vectp++, (long)(intptr_t)destp);
1181 while (*stringp++ != 0)
1182 destp++;
1183 destp++;
1184 }
1185
1186 /* end of vector table is a null pointer */
1187 suword(vectp, 0);
1188
1189 return (stack_base);
1190 }
1191
1192 /*
1193 * Check permissions of file to execute.
1194 * Called with imgp->vp locked.
1195 * Return 0 for success or error code on failure.
1196 */
1197 int
1198 exec_check_permissions(imgp)
1199 struct image_params *imgp;
1200 {
1201 struct vnode *vp = imgp->vp;
1202 struct vattr *attr = imgp->attr;
1203 struct thread *td;
1204 int error;
1205
1206 td = curthread; /* XXXKSE */
1207
1208 /* Get file attributes */
1209 error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1210 if (error)
1211 return (error);
1212
1213 #ifdef MAC
1214 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1215 if (error)
1216 return (error);
1217 #endif
1218
1219 /*
1220 * 1) Check if file execution is disabled for the filesystem that this
1221 * file resides on.
1222 * 2) Insure that at least one execute bit is on - otherwise root
1223 * will always succeed, and we don't want to happen unless the
1224 * file really is executable.
1225 * 3) Insure that the file is a regular file.
1226 */
1227 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1228 ((attr->va_mode & 0111) == 0) ||
1229 (attr->va_type != VREG))
1230 return (EACCES);
1231
1232 /*
1233 * Zero length files can't be exec'd
1234 */
1235 if (attr->va_size == 0)
1236 return (ENOEXEC);
1237
1238 /*
1239 * Check for execute permission to file based on current credentials.
1240 */
1241 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1242 if (error)
1243 return (error);
1244
1245 /*
1246 * Check number of open-for-writes on the file and deny execution
1247 * if there are any.
1248 */
1249 if (vp->v_writecount)
1250 return (ETXTBSY);
1251
1252 /*
1253 * Call filesystem specific open routine (which does nothing in the
1254 * general case).
1255 */
1256 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1257 return (error);
1258 }
1259
1260 /*
1261 * Exec handler registration
1262 */
1263 int
1264 exec_register(execsw_arg)
1265 const struct execsw *execsw_arg;
1266 {
1267 const struct execsw **es, **xs, **newexecsw;
1268 int count = 2; /* New slot and trailing NULL */
1269
1270 if (execsw)
1271 for (es = execsw; *es; es++)
1272 count++;
1273 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1274 if (newexecsw == NULL)
1275 return (ENOMEM);
1276 xs = newexecsw;
1277 if (execsw)
1278 for (es = execsw; *es; es++)
1279 *xs++ = *es;
1280 *xs++ = execsw_arg;
1281 *xs = NULL;
1282 if (execsw)
1283 free(execsw, M_TEMP);
1284 execsw = newexecsw;
1285 return (0);
1286 }
1287
1288 int
1289 exec_unregister(execsw_arg)
1290 const struct execsw *execsw_arg;
1291 {
1292 const struct execsw **es, **xs, **newexecsw;
1293 int count = 1;
1294
1295 if (execsw == NULL)
1296 panic("unregister with no handlers left?\n");
1297
1298 for (es = execsw; *es; es++) {
1299 if (*es == execsw_arg)
1300 break;
1301 }
1302 if (*es == NULL)
1303 return (ENOENT);
1304 for (es = execsw; *es; es++)
1305 if (*es != execsw_arg)
1306 count++;
1307 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1308 if (newexecsw == NULL)
1309 return (ENOMEM);
1310 xs = newexecsw;
1311 for (es = execsw; *es; es++)
1312 if (*es != execsw_arg)
1313 *xs++ = *es;
1314 *xs = NULL;
1315 if (execsw)
1316 free(execsw, M_TEMP);
1317 execsw = newexecsw;
1318 return (0);
1319 }
Cache object: 42718ed50af4512e3a1bc43fc26f5ab7
|